{ "metadata": { "title": "Artificial Intelligence Timeline", "description": "Key events in the history and future of artificial intelligence, from ancient automata to projected milestones", "version": "1.2.0", "sources": [ "AI Deep Timeline (Globaïa compilation)", "Epoch AI", "Our World in Data", "Wikipedia Timeline of AI" ], "lastUpdated": "2026-04-24", "dateNote": "Each event has an ISO `date` field at day/month/year precision (Wikidata or Wikipedia). Items without a `date` are year-only (the existing `year` field remains the source of truth). `yearShifted: true` flags events where the verified ISO date lies in a different year than the original fractional `year` — review needed.", "count": 529 }, "events": [ { "id": "tictactoe", "year": -1300, "title": "Tic-Tac-Toe", "description": "One of the earliest known rule-based games, later foundational for search trees, game theory, and early machine learning demonstrations.", "category": "game_playing", "significance": "minor", "tags": [ "game", "strategy", "ancient", "rules" ], "references": [ { "title": "Tic-tac-toe", "url": "https://en.wikipedia.org/wiki/Tic-tac-toe", "type": "wikipedia" }, { "title": "OXO (1952) — A.S. Douglas's EDSAC tic-tac-toe, an early game-playing program", "url": "https://en.wikipedia.org/wiki/OXO", "type": "wikipedia" } ] }, { "id": "talos", "year": -400, "title": "Talos", "description": "A mythic bronze automaton from Greek tradition, one of the earliest imagined artificial beings. Talos guarded the island of Crete, patrolling its shores three times daily.", "category": "mythology_fiction", "significance": "notable", "tags": [ "myth", "Greek", "automaton", "guardian", "bronze", "legend" ], "references": [ { "title": "Talos — Wikipedia", "url": "https://en.wikipedia.org/wiki/Talos", "type": "wikipedia" }, { "title": "Argonautica (Apollonius of Rhodes) — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/830", "type": "book" } ] }, { "id": "aristotle-organon", "year": -350, "title": "Aristotle's Organon", "description": "Foundational treatise on formal logic, introducing syllogisms and deductive reasoning that would later underpin symbolic AI and knowledge representation.", "category": "logic_mathematics", "significance": "notable", "tags": [ "logic", "reasoning", "philosophy", "Greek", "deduction", "syllogism" ], "references": [ { "title": "Prior Analytics (Aristotle)", "url": "https://en.wikipedia.org/wiki/Prior_Analytics", "type": "wikipedia" }, { "title": "Aristotle's Logic (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/aristotle-logic/", "type": "docs" }, { "title": "The Organon (Internet Archive)", "url": "https://archive.org/details/AristotleOrganon", "type": "archive" } ] }, { "id": "ktesibios-clock", "year": -250, "title": "Ktesibios's Water Clock", "description": "An early feedback-controlled, self-regulating mechanism. The clepsydra maintained constant water flow through ingenious float valves, a precursor to cybernetic control systems.", "category": "automata_robotics", "significance": "notable", "tags": [ "feedback", "mechanism", "engineering", "water", "control", "cybernetics" ], "references": [ { "title": "Ctesibius", "url": "https://en.wikipedia.org/wiki/Ctesibius", "type": "wikipedia" }, { "title": "Clepsydra (water clock)", "url": "https://en.wikipedia.org/wiki/Water_clock", "type": "wikipedia" } ] }, { "id": "abacus", "year": -190, "title": "Abacus", "description": "An early cognitive aid for arithmetic, representing the beginning of external computational tools and a distant ancestor of digital calculators.", "category": "computing_hardware", "significance": "notable", "tags": [ "calculation", "tool", "arithmetic", "ancient" ], "references": [ { "title": "Abacus", "url": "https://en.wikipedia.org/wiki/Abacus", "type": "wikipedia" }, { "title": "Computer History Museum: Abacus", "url": "https://www.computerhistory.org/revolution/calculators/1", "type": "archive" } ] }, { "id": "antikythera", "year": -125, "title": "Antikythera Mechanism", "description": "A geared astronomical computing device recovered from a shipwreck, capable of predicting eclipses and planetary positions. One of history's most remarkable proto-computers.", "category": "computing_hardware", "significance": "notable", "tags": [ "astronomy", "computer", "ancient", "Greek", "gears", "eclipse", "shipwreck" ], "references": [ { "title": "Antikythera mechanism", "url": "https://en.wikipedia.org/wiki/Antikythera_mechanism", "type": "wikipedia" }, { "title": "Freeth et al., Decoding the ancient Greek astronomical calculator", "url": "https://www.nature.com/articles/nature05357", "type": "paper" } ] }, { "id": "al-jazari", "year": 1206, "title": "al-Jazari's Automata", "description": "Ismail al-Jazari documented over 50 mechanical devices including programmable automata, a humanoid musician band, and water-powered mechanisms in his Book of Knowledge of Ingenious Mechanical Devices.", "category": "automata_robotics", "significance": "notable", "tags": [ "automaton", "Islamic", "engineering", "medieval", "programmable", "musician" ], "references": [ { "title": "Ismail al-Jazari", "url": "https://en.wikipedia.org/wiki/Ismail_al-Jazari", "type": "wikipedia" }, { "title": "Book of Knowledge of Ingenious Mechanical Devices", "url": "https://en.wikipedia.org/wiki/Book_of_Knowledge_of_Ingenious_Mechanical_Devices", "type": "wikipedia" }, { "title": "The Book of Knowledge of Ingenious Mechanical Devices (Internet Archive)", "url": "https://archive.org/details/TheBookOfKnowledgeOfIngeniousMechanicalDevices", "type": "archive" } ] }, { "id": "copper-knights", "year": 1220, "title": "Lancelot's Copper Knights", "description": "A medieval literary example of imagined robotic beings in the Lancelot-Grail cycle, copper automata that guard a castle and must be defeated by the hero.", "category": "mythology_fiction", "significance": "minor", "tags": [ "literature", "medieval", "fiction", "Arthurian", "knight", "guardian" ], "references": [ { "title": "Lancelot-Grail — Wikipedia", "url": "https://en.wikipedia.org/wiki/Lancelot-Grail", "type": "wikipedia" } ] }, { "id": "hesdin", "year": 1300, "title": "Hesdin Mechanical Park", "description": "A courtly landscape of lifelike automata, trick devices, and engineered marvels built for Count Robert II of Artois at Hesdin Castle.", "category": "automata_robotics", "significance": "minor", "tags": [ "automaton", "medieval", "courtly", "park", "French" ], "references": [ { "title": "Park of Hesdin", "url": "https://en.wikipedia.org/wiki/Hesdin", "type": "wikipedia" }, { "title": "Medieval Automata (history)", "url": "https://en.wikipedia.org/wiki/Automaton#Medieval_automata", "type": "wikipedia" } ] }, { "id": "llull-ars-magna", "year": 1305, "title": "Ramon Llull's Ars Magna", "description": "An early combinatorial logic system designed to generate knowledge through rotating symbolic disks. A remarkable precursor to computational reasoning and formal ontologies.", "category": "logic_mathematics", "significance": "notable", "tags": [ "combinatorics", "reasoning", "medieval", "Spanish", "knowledge generation" ], "references": [ { "title": "Ars Magna (Ramon Llull)", "url": "https://en.wikipedia.org/wiki/Ars_Magna_(Ramon_Llull)", "type": "wikipedia" }, { "title": "Ramon Llull (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/llull/", "type": "docs" } ] }, { "id": "religious-automata", "year": 1352, "title": "Religious Automata", "description": "Church-linked mechanical figures including the Strasbourg astronomical clock's procession of apostles, blending theology with engineering spectacle.", "category": "automata_robotics", "significance": "minor", "tags": [ "church", "clock", "medieval", "Strasbourg", "religion" ], "references": [ { "title": "Automaton", "url": "https://en.wikipedia.org/wiki/Automaton", "type": "wikipedia" }, { "title": "Smithsonian: Medieval robots of religion", "url": "https://www.smithsonianmag.com/history/medieval-robots-180959224/", "type": "archive" } ] }, { "id": "davinci-robot", "year": 1495, "title": "da Vinci's Robot Knight", "description": "Leonardo da Vinci's humanoid mechanical knight, designed with articulated limbs driven by pulleys and cables. It could sit, stand, raise its visor, and move its arms.", "category": "automata_robotics", "significance": "notable", "tags": [ "Leonardo", "Renaissance", "humanoid", "mechanism", "Italian" ], "references": [ { "title": "Leonardo's robot", "url": "https://en.wikipedia.org/wiki/Leonardo%27s_robot", "type": "wikipedia" }, { "title": "Codex Atlanticus (Internet Archive)", "url": "https://archive.org/details/codex-atlanticus", "type": "archive" } ] }, { "id": "golem", "year": 1580, "title": "Golem of Prague", "description": "A legendary artificial being animated through sacred language by Rabbi Loew. The Golem raises enduring questions about creation, control, and the unintended consequences of artificial life.", "category": "mythology_fiction", "significance": "notable", "tags": [ "Jewish", "Prague", "legend", "creation", "clay", "Rabbi Loew" ], "references": [ { "title": "Golem — Wikipedia", "url": "https://en.wikipedia.org/wiki/Golem", "type": "wikipedia" }, { "title": "Golem of Prague (Judah Loew ben Bezalel) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Judah_Loew_ben_Bezalel", "type": "wikipedia" } ] }, { "id": "hobbes-leviathan", "year": 1651, "title": "Hobbes's Leviathan", "description": "Thomas Hobbes proposed a mechanistic account of reasoning as computation: 'by ratiocination, I mean computation.' A philosophical foundation for viewing thought as information processing.", "category": "philosophy_mind", "significance": "notable", "tags": [ "computation", "philosophy", "mechanism", "thought", "British" ], "references": [ { "title": "Leviathan", "url": "https://www.gutenberg.org/ebooks/3207", "type": "book" }, { "title": "Leviathan (Hobbes book)", "url": "https://en.wikipedia.org/wiki/Leviathan_(Hobbes_book)", "type": "wikipedia" } ], "date": "1651-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q193034" }, { "id": "leibniz-combinatoria", "year": 1666, "title": "Leibniz's Calculus Ratiocinator", "description": "In Dissertatio de Arte Combinatoria, Gottfried Wilhelm Leibniz envisioned a universal symbolic language and a calculus of reasoning that would let disputes be settled by computation. The dream of a 'calculus of thought' foreshadows symbolic AI by three centuries.", "category": "logic_mathematics", "significance": "notable", "tags": [ "leibniz", "symbolic", "reasoning", "universal language", "combinatorics" ], "references": [ { "title": "Dissertatio de Arte Combinatoria (Internet Archive)", "url": "https://archive.org/details/gottfriedwilhelm00leibuoft", "type": "archive" }, { "title": "Gottfried Wilhelm Leibniz (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/leibniz/", "type": "docs" }, { "title": "Characteristica universalis (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Characteristica_universalis", "type": "wikipedia" } ] }, { "id": "consciousness-mill", "year": 1714, "title": "The Consciousness Mill", "description": "Leibniz's thought experiment: if you could walk inside a thinking machine like a mill, you would see only mechanical parts pushing each other, never perception or consciousness itself.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Leibniz", "consciousness", "thought experiment", "perception", "German" ], "references": [ { "title": "Monadology", "url": "https://en.wikipedia.org/wiki/Monadology", "type": "wikipedia" } ] }, { "id": "lagado-engine", "year": 1726, "title": "Lagado Book-Writing Engine", "description": "Jonathan Swift's satirical machine in Gulliver's Travels that mechanically generates books by randomly combining words, a prescient mockery of algorithmic creativity.", "category": "mythology_fiction", "significance": "minor", "tags": [ "satire", "Swift", "literature", "generation", "book", "fiction" ], "references": [ { "title": "Gulliver's Travels — Wikipedia", "url": "https://en.wikipedia.org/wiki/Gulliver%27s_Travels", "type": "wikipedia" }, { "title": "Gulliver's Travels — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/829", "type": "book" } ], "date": "1726-10-28", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q181488" }, { "id": "vaucanson-duck", "year": 1738, "title": "de Vaucanson's Duck", "description": "Jacques de Vaucanson's celebrated automaton that appeared to eat, digest, and excrete grain. It intensified Enlightenment debates about life, mechanism, and the boundary between animate and artificial.", "category": "automata_robotics", "significance": "notable", "tags": [ "automaton", "Enlightenment", "digestion", "French", "spectacle" ], "references": [ { "title": "Digesting Duck", "url": "https://en.wikipedia.org/wiki/Digesting_Duck", "type": "wikipedia" }, { "title": "Jacques de Vaucanson", "url": "https://en.wikipedia.org/wiki/Jacques_de_Vaucanson", "type": "wikipedia" } ] }, { "id": "mechanical-turk", "year": 1770, "title": "The Mechanical Turk", "description": "Wolfgang von Kempelen's chess-playing pseudo-automaton toured Europe for decades, defeating Napoleon and Benjamin Franklin. It concealed a human operator, but catalyzed dreams of thinking machines.", "category": "game_playing", "significance": "notable", "tags": [ "chess", "hoax", "spectacle", "Napoleon", "Kempelen" ], "references": [ { "title": "The Turk (chess-playing automaton)", "url": "https://en.wikipedia.org/wiki/Mechanical_Turk", "type": "wikipedia" }, { "title": "Shannon, 'Programming a Computer for Playing Chess' (Philosophical Magazine, 1950)", "url": "https://www.pi.infn.it/~carosi/chess/shannon.txt", "type": "paper" } ], "date": "1769-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q275924", "yearShifted": true, "originalYear": 1770 }, { "id": "jaquet-droz", "year": 1774, "title": "Jaquet-Droz Automata", "description": "Pierre Jaquet-Droz and his son built three programmable automata: The Writer (40 characters, programmable text), The Musician, and The Draughtsman. Among the finest clockwork machines ever built.", "category": "automata_robotics", "significance": "notable", "tags": [ "automaton", "clockwork", "programmable", "Swiss", "writer", "musician" ], "references": [ { "title": "Jaquet-Droz automata", "url": "https://en.wikipedia.org/wiki/Jaquet-Droz_automata", "type": "wikipedia" }, { "title": "Pierre Jaquet-Droz", "url": "https://en.wikipedia.org/wiki/Pierre_Jaquet-Droz", "type": "wikipedia" } ] }, { "id": "frankenstein", "year": 1818, "title": "Frankenstein", "description": "Mary Shelley's novel is a foundational meditation on artificial life, the hubris of creation, moral responsibility, and technological overreach. Its themes remain central to AI ethics.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "book", "literature", "Shelley", "creation", "ethics", "monster", "Gothic" ], "references": [ { "title": "Frankenstein — Wikipedia", "url": "https://en.wikipedia.org/wiki/Frankenstein", "type": "wikipedia" }, { "title": "Frankenstein; or, The Modern Prometheus — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/84", "type": "book" } ], "date": "1818-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q150827" }, { "id": "computational-creativity", "year": 1821, "title": "Computational Creativity", "description": "Dietrich Winkel's componium, an early machine for generative musical variation, marks the beginning of machine-aided creative production.", "category": "art_creativity", "significance": "minor", "tags": [ "music", "generation", "composition", "mechanical" ], "references": [ { "title": "Componium", "url": "https://en.wikipedia.org/wiki/Componium", "type": "wikipedia" }, { "title": "Dietrich Nikolaus Winkel and the Componium", "url": "https://publicdomainreview.org/collection/winkel-componium/", "type": "archive" } ] }, { "id": "babbage", "year": 1822, "title": "Babbage's Mechanical Computer", "description": "Charles Babbage's Difference Engine, and later his Analytical Engine (1837), laid the conceptual foundation for programmable computation. Ada Lovelace wrote the first algorithm for it in 1843.", "category": "computing_hardware", "significance": "notable", "tags": [ "computer", "engine", "Ada Lovelace", "programming", "Victorian", "British" ], "references": [ { "title": "Analytical Engine", "url": "https://en.wikipedia.org/wiki/Analytical_Engine", "type": "wikipedia" }, { "title": "Difference engine", "url": "https://en.wikipedia.org/wiki/Difference_engine", "type": "wikipedia" }, { "title": "Computer History Museum: Babbage", "url": "https://www.computerhistory.org/babbage/", "type": "archive" } ], "date": "1822-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q628007" }, { "id": "artist-beautiful", "year": 1844, "title": "\"The Artist of the Beautiful\"", "description": "Nathaniel Hawthorne's story about a watchmaker who creates a mechanical butterfly, exploring themes of mechanism versus beauty, invention, and the maker's inner vision.", "category": "mythology_fiction", "significance": "minor", "tags": [ "literature", "Hawthorne", "mechanism", "beauty", "American", "fiction" ], "references": [ { "title": "The Artist of the Beautiful — Wikipedia", "url": "https://en.wikipedia.org/wiki/The_Artist_of_the_Beautiful", "type": "wikipedia" }, { "title": "Mosses from an Old Manse (includes The Artist of the Beautiful) — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/9231", "type": "book" } ] }, { "id": "boolean-algebra", "year": 1854, "title": "Boolean Algebra", "description": "George Boole's The Laws of Thought formalized logic as algebra, creating the mathematical foundation that would later enable digital circuits and all of modern computing.", "category": "logic_mathematics", "significance": "notable", "tags": [ "logic", "Boole", "mathematics", "digital", "circuits", "binary" ], "references": [ { "title": "An Investigation of the Laws of Thought (1854)", "url": "https://archive.org/details/investigationofl00boolrich", "type": "archive" }, { "title": "Boolean algebra (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Boolean_algebra", "type": "wikipedia" }, { "title": "George Boole (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/boole/", "type": "docs" } ] }, { "id": "darwin-machines", "year": 1863, "title": "\"Darwin among the Machines\"", "description": "Samuel Butler's essay warned that machines might evolve beyond human control, anticipating arguments about technological autonomy and existential risk from AI.", "category": "ethics_safety", "significance": "notable", "tags": [ "evolution", "Butler", "autonomy", "essay", "existential risk", "Victorian" ], "references": [ { "title": "Darwin among the Machines (Samuel Butler, 1863)", "url": "https://en.wikipedia.org/wiki/Darwin_among_the_Machines", "type": "wikipedia" }, { "title": "Butler's original letter (The Press, Christchurch, NZ)", "url": "https://nzetc.victoria.ac.nz/tm/scholarly/tei-ButFir-t1-g1-t1-g1-t4-body.html", "type": "archive" } ], "date": "1863-06-13", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Darwin_among_the_Machines" }, { "id": "steam-man", "year": 1868, "title": "The Steam Man of the Prairies", "description": "Edward S. Ellis's dime novel featured a steam-powered mechanical humanoid, one of the earliest robot characters in American popular fiction.", "category": "mythology_fiction", "significance": "minor", "tags": [ "fiction", "dime novel", "robot", "American", "steam" ], "references": [ { "title": "The Steam Man of the Prairies — Wikipedia", "url": "https://en.wikipedia.org/wiki/The_Steam_Man_of_the_Prairies", "type": "wikipedia" }, { "title": "The Huge Hunter; or, The Steam Man of the Prairies — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/7506", "type": "book" } ] }, { "id": "frege-begriffsschrift", "year": 1879, "title": "Frege's Begriffsschrift", "description": "Gottlob Frege introduced the first fully formal system of predicate logic with quantifiers, variables, and rigorous proof rules. Begriffsschrift ('concept script') is the ancestor of every modern logic programming language and automated theorem prover.", "category": "logic_mathematics", "significance": "notable", "tags": [ "frege", "predicate logic", "quantifiers", "formal logic", "foundations" ], "references": [ { "title": "Begriffsschrift (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Begriffsschrift", "type": "wikipedia" }, { "title": "Gottlob Frege (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/frege/", "type": "docs" }, { "title": "Begriffsschrift (Internet Archive)", "url": "https://archive.org/details/begriffsschrifte00freg", "type": "archive" } ], "date": "1879-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q814390" }, { "id": "tower-hanoi", "year": 1883, "title": "Tower of Hanoi", "description": "Edouard Lucas's mathematical puzzle became a canonical example of recursion, planning, and computational problem-solving in computer science education.", "category": "logic_mathematics", "significance": "minor", "tags": [ "puzzle", "recursion", "mathematics", "education" ], "references": [ { "title": "Tower of Hanoi (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Tower_of_Hanoi", "type": "wikipedia" } ], "date": "1883-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q213593" }, { "id": "peano-axioms", "year": 1889, "title": "Peano Axioms", "description": "Giuseppe Peano published a small set of axioms defining the natural numbers in purely formal terms. His notation and axiomatic method became a template for the formalization of mathematics underlying computer science.", "category": "logic_mathematics", "significance": "notable", "tags": [ "peano", "axioms", "arithmetic", "foundations", "formalism" ], "references": [ { "title": "Peano axioms (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Peano_axioms", "type": "wikipedia" }, { "title": "Dedekind's Contributions to the Foundations of Mathematics (SEP)", "url": "https://plato.stanford.edu/entries/dedekind-foundations/", "type": "docs" } ] }, { "id": "hollerith-tabulator", "year": 1890, "title": "Hollerith Tabulating Machine", "description": "Herman Hollerith's electromechanical tabulator processed the 1890 U.S. Census using punched cards, reducing years of manual work to months. His company eventually became IBM, and punched cards dominated data processing for the next 70 years.", "category": "computing_hardware", "significance": "notable", "tags": [ "hollerith", "punch-card", "census", "ibm", "tabulator" ], "references": [ { "title": "Herman Hollerith", "url": "https://en.wikipedia.org/wiki/Herman_Hollerith", "type": "wikipedia" }, { "title": "Computer History Museum: Hollerith", "url": "https://www.computerhistory.org/revolution/punched-cards/2/", "type": "archive" } ] }, { "id": "electric-bob", "year": 1893, "title": "Electric Bob's Big Black Ostrich", "description": "A late 19th-century fictional mechanized animal reflecting popular fascination with engineered mobility and electrical power.", "category": "mythology_fiction", "significance": "minor", "tags": [ "fiction", "electrical", "Victorian", "animal" ], "references": [ { "title": "Edisonade — Wikipedia", "url": "https://en.wikipedia.org/wiki/Edisonade", "type": "wikipedia" } ] }, { "id": "tesla-boat", "year": 1898, "title": "Tesla's \"Borrowed Mind\"", "description": "Nikola Tesla demonstrated a radio-controlled boat at Madison Square Garden, describing it as having a 'borrowed mind.' An early concept of remote artificial agency.", "category": "automata_robotics", "significance": "notable", "tags": [ "radio", "remote control", "Tesla", "demonstration", "New York" ], "references": [ { "title": "Teleautomaton (Tesla)", "url": "https://en.wikipedia.org/wiki/Teleautomaton", "type": "wikipedia" }, { "title": "US Patent 613,809 — Method of and Apparatus for Controlling Mechanism of Moving Vessels or Vehicles", "url": "https://patents.google.com/patent/US613809A", "type": "docs" } ] }, { "id": "soul-search", "year": 1907, "title": "Searches for the Soul", "description": "Early 20th-century attempts to localize or test the existence of mind and consciousness in relation to mechanism, presaging the hard problem of consciousness in AI.", "category": "philosophy_mind", "significance": "minor", "tags": [ "consciousness", "mechanism", "mind", "soul" ], "references": [ { "title": "21 grams experiment", "url": "https://en.wikipedia.org/wiki/21_grams_experiment", "type": "wikipedia" } ], "date": "1907-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q3393257" }, { "id": "tiktok", "year": 1907, "title": "Tik-Tok of Oz", "description": "L. Frank Baum's mechanical man from the Oz books, described as a 'Patent Double-Action, Extra-Responsive, Thought-Creating, Perfect-Talking Mechanical Man.' An early fictional robot character.", "category": "mythology_fiction", "significance": "minor", "tags": [ "Oz", "fiction", "robot", "Baum", "book", "children" ], "references": [ { "title": "Tik-Tok (Oz) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Tik-Tok_(Oz)", "type": "wikipedia" }, { "title": "Ozma of Oz — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/33361", "type": "book" } ] }, { "id": "principia-mathematica", "year": 1910, "title": "Principia Mathematica", "description": "Bertrand Russell and Alfred North Whitehead published the first volume of Principia Mathematica, an attempt to derive all of mathematics from logic. Its formal machinery inspired Goedel, Turing, and the founders of computer science.", "category": "logic_mathematics", "significance": "notable", "tags": [ "russell", "whitehead", "logicism", "foundations", "proof" ], "references": [ { "title": "Principia Mathematica (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Principia_Mathematica", "type": "wikipedia" }, { "title": "Principia Mathematica, vol. 1 (Internet Archive)", "url": "https://archive.org/details/PrincipiaMathematicaVolumeI", "type": "archive" }, { "title": "Principia Mathematica (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/principia-mathematica/", "type": "docs" } ] }, { "id": "ocr", "year": 1913, "title": "Optical Character Recognition", "description": "Early efforts to make machines identify written symbols automatically, beginning with Emanuel Goldberg's statistical machine. A founding challenge of computer vision.", "category": "computer_vision", "significance": "notable", "tags": [ "vision", "pattern recognition", "reading", "Goldberg", "text" ], "references": [ { "title": "Optical character recognition", "url": "https://en.wikipedia.org/wiki/Optical_character_recognition", "type": "wikipedia" }, { "title": "Emanuel Goldberg and his Knowledge Machine (Buckland, 2006)", "url": "https://people.ischool.berkeley.edu/~buckland/goldbib.html", "type": "archive" } ] }, { "id": "rur", "year": 1920, "title": "Rossum's Universal Robots", "description": "Karel Čapek's play R.U.R. introduced the word 'robot' (from Czech robota, meaning forced labor) to the world. The robots ultimately rebel and destroy humanity.", "category": "mythology_fiction", "significance": "notable", "tags": [ "theater", "play", "Czech", "Čapek", "labor", "rebellion", "robot origin", "culture" ], "references": [ { "title": "R.U.R. — Wikipedia", "url": "https://en.wikipedia.org/wiki/R.U.R.", "type": "wikipedia" }, { "title": "R.U.R. (Rossum's Universal Robots) — Project Gutenberg", "url": "https://www.gutenberg.org/ebooks/59112", "type": "book" } ], "date": "1921-01-25", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1164094", "yearShifted": true, "originalYear": 1920 }, { "id": "metropolis", "year": 1927, "title": "Metropolis", "description": "Fritz Lang's film featured the iconic Maschinenmensch (Machine-Human), linking automation to class struggle, deception, and artificial humanity. A defining image of the robot in cinema.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "movie", "cinema", "Lang", "dystopia", "robot", "German", "culture", "silent film" ], "references": [ { "title": "Metropolis (1927 film) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Metropolis_(1927_film)", "type": "wikipedia" }, { "title": "Metropolis (1927) — IMDb", "url": "https://www.imdb.com/title/tt0017136/", "type": "docs" } ], "date": "1927-01-10", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q151599" }, { "id": "godel-incompleteness", "year": 1931.09, "title": "Goedel's Incompleteness Theorems", "description": "Kurt Goedel proved that any consistent formal system strong enough to express arithmetic contains true statements it cannot prove. The result shattered Hilbert's program and revealed fundamental limits of mechanical reasoning.", "category": "logic_mathematics", "significance": "notable", "tags": [ "godel", "incompleteness", "limits", "formalism", "foundations" ], "references": [ { "title": "Goedel's incompleteness theorems (Wikipedia)", "url": "https://en.wikipedia.org/wiki/G%C3%B6del%27s_incompleteness_theorems", "type": "wikipedia" }, { "title": "Ueber formal unentscheidbare Saetze (1931)", "url": "https://doi.org/10.1007/BF01700692", "type": "paper" }, { "title": "Kurt Goedel (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/goedel/", "type": "docs" } ] }, { "id": "kolmogorov-probability", "year": 1933, "title": "Kolmogorov's Axioms of Probability", "description": "Andrey Kolmogorov's Grundbegriffe der Wahrscheinlichkeitsrechnung placed probability theory on a rigorous measure-theoretic foundation. The axioms underpin statistics, information theory, and every modern machine learning algorithm.", "category": "logic_mathematics", "significance": "notable", "tags": [ "kolmogorov", "probability", "measure theory", "statistics", "foundations" ], "references": [ { "title": "Probability axioms (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Probability_axioms", "type": "wikipedia" }, { "title": "Grundbegriffe der Wahrscheinlichkeitsrechnung (Internet Archive)", "url": "https://archive.org/details/foundationsofthe00kolm", "type": "archive" } ] }, { "id": "church-lambda", "year": 1936.04, "title": "Church's Lambda Calculus", "description": "Alonzo Church introduced the lambda calculus, a tiny formal system in which any computable function can be expressed as the application of anonymous functions. It is the foundation of functional programming and a rival model of computation to Turing machines.", "category": "logic_mathematics", "significance": "notable", "tags": [ "church", "lambda calculus", "computability", "functional", "foundations" ], "references": [ { "title": "An Unsolvable Problem of Elementary Number Theory (1936)", "url": "https://doi.org/10.2307/2371045", "type": "paper" }, { "title": "Lambda calculus (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Lambda_calculus", "type": "wikipedia" }, { "title": "The Lambda Calculus (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/lambda-calculus/", "type": "docs" } ] }, { "id": "turing-computable", "year": 1936.11, "title": "Turing's On Computable Numbers", "description": "Alan Turing defined the abstract machine now bearing his name and showed that some problems, like the halting problem, are undecidable. The paper established the mathematical basis of digital computation.", "category": "logic_mathematics", "significance": "notable", "tags": [ "turing", "computability", "halting problem", "universal machine", "foundations" ], "references": [ { "title": "On Computable Numbers, with an Application to the Entscheidungsproblem (1936)", "url": "https://doi.org/10.1112/plms/s2-42.1.230", "type": "paper" }, { "title": "Turing machine (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Turing_machine", "type": "wikipedia" }, { "title": "Turing Machines (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/turing-machine/", "type": "docs" } ], "date": "1936-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q163310" }, { "id": "elektro", "year": 1939, "title": "Elektro the Moto-Man", "description": "Westinghouse's humanoid robot, exhibited at the 1939 World's Fair, could walk, talk (77 words), smoke, and distinguish red from green. A spectacle of engineering and showmanship.", "category": "automata_robotics", "significance": "notable", "tags": [ "World's Fair", "humanoid", "demonstration", "Westinghouse", "American" ], "references": [ { "title": "Elektro", "url": "https://en.wikipedia.org/wiki/Elektro", "type": "wikipedia" }, { "title": "1939 New York World's Fair (Westinghouse exhibit)", "url": "https://en.wikipedia.org/wiki/1939_New_York_World%27s_Fair", "type": "wikipedia" } ], "date": "1939-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q207420" }, { "id": "speech-synthesis", "year": 1939, "title": "Speech Synthesis", "description": "Bell Labs' VODER (Voice Operating Demonstrator), demonstrated at the 1939 World's Fair, was the first electronic speech synthesizer, operated by a human controller.", "category": "speech_audio", "significance": "notable", "tags": [ "voice", "Bell Labs", "VODER", "World's Fair", "electronic" ], "references": [ { "title": "Voder", "url": "https://en.wikipedia.org/wiki/Voder", "type": "wikipedia" }, { "title": "Homer Dudley and the Voder at the 1939 World's Fair (Bell Labs)", "url": "https://www.bell-labs.com/about/history/", "type": "docs" } ] }, { "id": "zuse-z3", "year": 1941.36, "title": "Zuse Z3, First Programmable Computer", "description": "Konrad Zuse completed the Z3 in Berlin on May 12, 1941, the world's first working programmable, fully automatic digital computer. It used 2,600 electromechanical relays and read programs from punched film.", "category": "computing_hardware", "significance": "notable", "tags": [ "zuse", "z3", "germany", "relay", "programmable" ], "references": [ { "title": "Z3 (computer)", "url": "https://en.wikipedia.org/wiki/Z3_(computer)", "type": "wikipedia" }, { "title": "Computer History Museum: Zuse Z3", "url": "https://www.computerhistory.org/timeline/1941/", "type": "archive" } ], "date": "1941-05-12", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Z3_(computer)" }, { "id": "asimov-laws", "year": 1942, "title": "Asimov's Three Laws of Robotics", "description": "Isaac Asimov formulated his Three Laws in the short story 'Runaround': (1) don't harm humans, (2) obey orders, (3) protect yourself. They became central to AI safety discourse.", "category": "ethics_safety", "significance": "notable", "tags": [ "fiction", "robot", "rules", "safety", "Asimov", "science fiction", "culture", "book" ], "references": [ { "title": "Three Laws of Robotics", "url": "https://en.wikipedia.org/wiki/Three_Laws_of_Robotics", "type": "wikipedia" }, { "title": "Runaround (short story, Astounding Science Fiction, 1942)", "url": "https://en.wikipedia.org/wiki/Runaround_(story)", "type": "wikipedia" } ], "date": "1942-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q23758" }, { "id": "military-robots", "year": 1942, "title": "Lethal Military Robots", "description": "Early thinking about autonomous or semi-autonomous killing machines in warfare, from V-1 flying bombs to torpedo guidance, foreshadowing modern autonomous weapons debates.", "category": "military", "significance": "notable", "tags": [ "weapons", "warfare", "autonomous", "defense", "WWII" ], "references": [ { "title": "V-1 flying bomb", "url": "https://en.wikipedia.org/wiki/V-1_flying_bomb", "type": "wikipedia" }, { "title": "Lethal autonomous weapon", "url": "https://en.wikipedia.org/wiki/Lethal_autonomous_weapon", "type": "wikipedia" } ] }, { "id": "ann", "year": 1943, "title": "Artificial Neural Networks", "description": "Warren McCulloch and Walter Pitts proposed a mathematical model of neural networks, showing how simple neuron-like units connected in networks could compute any logical function.", "category": "neural_networks", "significance": "notable", "tags": [ "neuron", "McCulloch", "Pitts", "brain", "mathematical model", "connectionism" ], "references": [ { "title": "A Logical Calculus of the Ideas Immanent in Nervous Activity (McCulloch & Pitts, 1943)", "url": "https://doi.org/10.1007/BF02478259", "type": "paper" }, { "title": "McCulloch–Pitts neuron (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Artificial_neuron", "type": "wikipedia" } ] }, { "id": "harvard-mark-i", "year": 1944.6, "title": "Harvard Mark I", "description": "IBM's Automatic Sequence Controlled Calculator, dedicated at Harvard on August 7, 1944, was a 51-foot electromechanical computer used for wartime ballistics and early numerical simulations. Grace Hopper wrote some of its first programs.", "category": "computing_hardware", "significance": "notable", "tags": [ "harvard", "ibm", "grace-hopper", "electromechanical", "wartime" ], "references": [ { "title": "Harvard Mark I", "url": "https://en.wikipedia.org/wiki/Harvard_Mark_I", "type": "wikipedia" }, { "title": "IBM: ASCC / Harvard Mark I", "url": "https://www.ibm.com/history/mark-i", "type": "archive" } ], "date": "1944-08-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q776823", "yearShifted": true, "originalYear": 1945 }, { "id": "eniac", "year": 1946, "title": "ENIAC", "description": "The Electronic Numerical Integrator and Computer, one of the first general-purpose electronic computers. It weighed 30 tons, used 17,468 vacuum tubes, and could perform 5,000 additions per second.", "category": "computing_hardware", "significance": "notable", "tags": [ "computer", "electronic", "vacuum tubes", "wartime", "Pennsylvania" ], "references": [ { "title": "ENIAC", "url": "https://en.wikipedia.org/wiki/ENIAC", "type": "wikipedia" }, { "title": "Computer History Museum: ENIAC", "url": "https://www.computerhistory.org/revolution/birth-of-the-computer/4/78", "type": "archive" } ] }, { "id": "shannon-information", "year": 1948.58, "title": "Shannon's Mathematical Theory of Communication", "description": "Claude Shannon defined information as measurable in bits and proved the fundamental limits of lossless and noisy communication. Information theory underlies data compression, error correction, and every neural network's loss function.", "category": "logic_mathematics", "significance": "notable", "tags": [ "shannon", "information theory", "entropy", "communication", "bits" ], "references": [ { "title": "A Mathematical Theory of Communication (1948)", "url": "https://doi.org/10.1002/j.1538-7305.1948.tb01338.x", "type": "paper" }, { "title": "A Mathematical Theory of Communication (Wikipedia)", "url": "https://en.wikipedia.org/wiki/A_Mathematical_Theory_of_Communication", "type": "wikipedia" }, { "title": "Shannon 1948 paper (Internet Archive)", "url": "https://archive.org/details/bstj27-3-379", "type": "archive" } ], "date": "1948-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q724029", "yearShifted": true, "originalYear": 1949 }, { "id": "giant-brains", "year": 1949, "title": "Giant Brains, or Machines That Think", "description": "Edmund Berkeley's book popularized the idea of thinking machines for a mass audience, comparing computers to 'giant brains' and imagining their future potential.", "category": "society_impact", "significance": "minor", "tags": [ "book", "popular science", "Berkeley", "public imagination" ], "references": [ { "title": "Giant Brains, or Machines That Think", "url": "https://en.wikipedia.org/wiki/Giant_Brains,_or_Machines_That_Think", "type": "wikipedia" }, { "title": "Internet Archive: Giant Brains (1949)", "url": "https://archive.org/details/gianbrainsormachin00berk", "type": "archive" } ] }, { "id": "wiener-human-use", "year": 1950, "title": "The Human Use of Human Beings", "description": "Norbert Wiener's cybernetic reflection on machines, society, and the human consequences of automation. He warned about the dangers of automated decision-making and algorithmic governance.", "category": "society_impact", "significance": "notable", "tags": [ "cybernetics", "Wiener", "automation", "society", "book", "warning" ], "references": [ { "title": "The Human Use of Human Beings", "url": "https://en.wikipedia.org/wiki/The_Human_Use_of_Human_Beings", "type": "wikipedia" }, { "title": "The Human Use of Human Beings (Houghton Mifflin)", "url": "https://archive.org/details/humanuseofhumanb00wien", "type": "book" } ], "date": "1950-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q7740888" }, { "id": "turing-test", "year": 1950.78, "title": "Turing Test", "description": "Alan Turing's 'Computing Machinery and Intelligence' proposed the imitation game: if a machine can converse indistinguishably from a human, it should be considered intelligent. A defining thought experiment.", "category": "philosophy_mind", "significance": "major", "tags": [ "Turing", "imitation game", "consciousness", "thinking", "intelligence", "British", "paper" ], "references": [ { "title": "Computing Machinery and Intelligence", "url": "https://doi.org/10.1093/mind/LIX.236.433", "type": "paper" }, { "title": "Turing test", "url": "https://en.wikipedia.org/wiki/Turing_test", "type": "wikipedia" } ] }, { "id": "reinforcement-learning", "year": 1951.5, "title": "Reinforcement Learning", "description": "Early machine learning through reward and feedback. Marvin Minsky and Dean Edmonds built SNARC, a neural network machine with 40 neurons that learned via reinforcement.", "category": "neural_networks", "significance": "major", "tags": [ "reward", "feedback", "SNARC", "Minsky", "learning", "neurons" ], "references": [ { "title": "Stochastic Neural Analog Reinforcement Calculator", "url": "https://en.wikipedia.org/wiki/Stochastic_neural_analog_reinforcement_calculator", "type": "wikipedia" }, { "title": "Minsky PhD Thesis: Theory of Neural-Analog Reinforcement Systems (Princeton, 1954)", "url": "https://web.media.mit.edu/~minsky/", "type": "archive" } ] }, { "id": "strachey-draughts", "year": 1951.5, "title": "Strachey's Draughts Program", "description": "Christopher Strachey wrote one of the first game-playing programs, a draughts (checkers) player that ran on the Ferranti Mark I at Manchester. By summer 1952 it could play a complete game at reasonable speed, alongside Dietrich Prinz's chess program on the same machine.", "category": "game_playing", "significance": "major", "tags": [ "game-playing", "draughts", "ferranti", "history" ], "references": [ { "title": "Christopher Strachey (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Christopher_Strachey", "type": "wikipedia" }, { "title": "Logical or non-mathematical programmes", "url": "https://dl.acm.org/doi/10.1145/800259.808992", "type": "paper" } ] }, { "id": "speech-recognition", "year": 1952, "title": "Speech Recognition", "description": "Bell Labs' Audrey system could recognize spoken digits with 97% accuracy for a single speaker, initiating decades of work on automatic speech recognition.", "category": "speech_audio", "significance": "notable", "tags": [ "voice", "Bell Labs", "Audrey", "digits", "listening" ], "references": [ { "title": "Audrey (speech recognition system)", "url": "https://en.wikipedia.org/wiki/Speech_recognition", "type": "wikipedia" }, { "title": "K. H. Davis, R. Biddulph, S. Balashek — Automatic Recognition of Spoken Digits (1952)", "url": "https://doi.org/10.1121/1.1906946", "type": "docs" } ], "date": "1952-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q189436" }, { "id": "samuel-checkers", "year": 1952.5, "title": "Samuel's Checkers Program", "description": "At IBM, Arthur Samuel began developing a checkers-playing program on the IBM 701 that learned from its own games. The program became one of the first demonstrations of machine learning, improving its play over time through self-play and rote memorization of board positions.", "category": "machine_learning", "significance": "major", "tags": [ "game-playing", "checkers", "ibm", "self-play" ], "references": [ { "title": "Arthur Samuel (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Arthur_Samuel_(computer_scientist)", "type": "wikipedia" }, { "title": "Some Studies in Machine Learning Using the Game of Checkers", "url": "https://ieeexplore.ieee.org/document/5392560", "type": "paper" } ] }, { "id": "ibm-701", "year": 1953.3, "title": "IBM 701 Electronic Data Processing Machine", "description": "IBM's first commercial scientific computer, announced April 7, 1953, brought electronic computing into industry and government. Arthur Samuel used it to write the first machine learning program, a checkers player.", "category": "computing_hardware", "significance": "notable", "tags": [ "ibm", "commercial", "samuel", "checkers", "mainframe" ], "references": [ { "title": "IBM 701", "url": "https://en.wikipedia.org/wiki/IBM_701", "type": "wikipedia" }, { "title": "IBM: The 701", "url": "https://www.ibm.com/history/701", "type": "archive" } ] }, { "id": "nlp", "year": 1954, "title": "Natural Language Processing", "description": "The Georgetown-IBM experiment demonstrated automatic translation of 60 Russian sentences into English, launching the field of computational linguistics and NLP.", "category": "language_nlp", "significance": "notable", "tags": [ "translation", "Russian", "Georgetown", "IBM", "linguistics", "Cold War" ], "references": [ { "title": "Georgetown–IBM experiment", "url": "https://en.wikipedia.org/wiki/Georgetown%E2%80%93IBM_experiment", "type": "wikipedia" }, { "title": "Hutchins, 'The Georgetown–IBM experiment demonstrated in January 1954' (2004)", "url": "https://aclanthology.org/www.mt-archive.info/Hutchins-2004.pdf", "type": "archive" } ], "date": "1954-01-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q4160483" }, { "id": "dartmouth-proposal", "year": 1955.67, "title": "Dartmouth AI Proposal", "description": "John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon submitted a research proposal to the Rockefeller Foundation that introduced the term 'artificial intelligence'. The phrase was chosen in part to distinguish the new field from cybernetics and to give it a clear identity.", "category": "philosophy_mind", "significance": "major", "tags": [ "dartmouth", "history", "founding", "terminology" ], "references": [ { "title": "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence", "url": "http://jmc.stanford.edu/articles/dartmouth/dartmouth.pdf", "type": "archive" }, { "title": "Dartmouth workshop (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Dartmouth_workshop", "type": "wikipedia" } ], "date": "1956-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q2453355" }, { "id": "dartmouth", "year": 1956.46, "title": "Dartmouth AI Workshop", "description": "John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon organized a summer workshop that coined the term 'artificial intelligence' and established AI as a field of research.", "category": "philosophy_mind", "significance": "major", "tags": [ "workshop", "conference", "McCarthy", "Minsky", "Shannon", "founding", "birth of AI" ], "references": [ { "title": "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence (McCarthy, Minsky, Rochester, Shannon, 1955)", "url": "https://raysolomonoff.com/dartmouth/boxa/dart564props.pdf", "type": "archive" }, { "title": "Dartmouth workshop", "url": "https://en.wikipedia.org/wiki/Dartmouth_workshop", "type": "wikipedia" } ], "date": "1956-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q2453355" }, { "id": "syntactic-structures", "year": 1957, "title": "Chomsky's Syntactic Structures", "description": "Noam Chomsky published a short book arguing that human language follows formal grammatical rules, introducing generative grammar and transformational rules. It reshaped linguistics and became foundational for decades of symbolic approaches to natural language processing.", "category": "language_nlp", "significance": "notable", "tags": [ "chomsky", "linguistics", "grammar", "theory", "book" ], "references": [ { "title": "Chomsky, 'Syntactic Structures', Mouton (1957)", "url": "https://en.wikipedia.org/wiki/Syntactic_Structures", "type": "book" } ], "date": "1957-02-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q1198080" }, { "id": "perceptron", "year": 1957.04, "title": "Perceptron", "description": "Frank Rosenblatt's perceptron, implemented on the Mark I computer at Cornell, was the first trainable artificial neural network. The Navy claimed it would 'walk, talk, see, write, reproduce, and be conscious.'", "category": "neural_networks", "significance": "major", "tags": [ "Rosenblatt", "neural network", "Cornell", "learning", "Navy", "training" ], "references": [ { "title": "The Perceptron: A Probabilistic Model for Information Storage (Rosenblatt, 1958)", "url": "https://doi.org/10.1037/h0042519", "type": "paper" }, { "title": "Perceptron (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Perceptron", "type": "wikipedia" } ], "date": "1957-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q690207" }, { "id": "general-problem-solver", "year": 1957.5, "title": "General Problem Solver", "description": "Allen Newell, Cliff Shaw, and Herbert Simon at RAND and Carnegie Tech developed the General Problem Solver, a program intended to imitate human problem-solving strategies across different domains. It introduced means-ends analysis and separated knowledge of the problem from the strategy used to solve it.", "category": "logic_mathematics", "significance": "major", "tags": [ "reasoning", "symbolic-ai", "rand", "newell-simon" ], "references": [ { "title": "General Problem Solver (Wikipedia)", "url": "https://en.wikipedia.org/wiki/General_Problem_Solver", "type": "wikipedia" }, { "title": "Report on a general problem-solving program", "url": "https://apps.dtic.mil/sti/citations/AD0671776", "type": "archive" } ], "date": "1959-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1387212", "yearShifted": true, "originalYear": 1958 }, { "id": "transhumanism", "year": 1957.5, "title": "Transhumanism", "description": "Julian Huxley coined the term in an essay arguing that humanity could transcend its biological limitations through technology, a vision that would merge with AI aspirations.", "category": "society_impact", "significance": "notable", "tags": [ "Huxley", "biology", "enhancement", "posthuman", "futurism" ], "references": [ { "title": "Transhumanism", "url": "https://en.wikipedia.org/wiki/Transhumanism", "type": "wikipedia" }, { "title": "Huxley, 'Transhumanism' (1957) — New Bottles for New Wine", "url": "https://archive.org/details/NewBottlesForNewWine", "type": "archive" } ] }, { "id": "lisp-language", "year": 1958.75, "title": "John McCarthy Creates Lisp", "description": "At MIT, John McCarthy designed Lisp, a programming language built around symbolic expressions and recursive functions. Lisp became the dominant language of AI research for decades and introduced ideas such as garbage collection, conditionals, and treating code as data.", "category": "logic_mathematics", "significance": "major", "tags": [ "programming-language", "lisp", "mccarthy", "symbolic-ai" ], "references": [ { "title": "Lisp (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Lisp_(programming_language)", "type": "wikipedia" }, { "title": "Recursive Functions of Symbolic Expressions and Their Computation by Machine", "url": "http://www-formal.stanford.edu/jmc/recursive.pdf", "type": "paper" } ], "date": "1959-01-15", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q132874" }, { "id": "knowledge-representation", "year": 1959.45, "title": "Knowledge Representation", "description": "The beginning of efforts to encode facts, rules, and inference in machine-readable form, foundational for expert systems and the Semantic Web.", "category": "machine_learning", "significance": "notable", "tags": [ "encoding", "rules", "inference", "ontology", "symbolic" ], "references": [ { "title": "Knowledge representation and reasoning", "url": "https://en.wikipedia.org/wiki/Knowledge_representation_and_reasoning", "type": "wikipedia" }, { "title": "McCarthy, J. (1959). Programs with Common Sense. Proceedings of the Teddington Conference on the Mechanization of Thought Processes.", "url": "http://jmc.stanford.edu/articles/mcc59/mcc59.pdf", "type": "paper" } ] }, { "id": "machine-learning-coined", "year": 1959.54, "title": "Machine Learning", "description": "Arthur Samuel coined the term while at IBM, defining it as giving 'computers the ability to learn without being explicitly programmed.' His checkers program demonstrated the concept.", "category": "machine_learning", "significance": "major", "tags": [ "Samuel", "IBM", "checkers", "definition", "coining", "learning" ], "references": [ { "title": "Samuel, A. L. (1959). Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development, 3(3), 210-229.", "url": "https://doi.org/10.1147/rd.33.0210", "type": "paper" }, { "title": "Arthur Samuel (computer scientist)", "url": "https://en.wikipedia.org/wiki/Arthur_Samuel_(computer_scientist)", "type": "wikipedia" } ] }, { "id": "hubel-wiesel", "year": 1959.8, "title": "Hubel and Wiesel Map the Visual Cortex", "description": "Neuroscientists David Hubel and Torsten Wiesel recorded from cat brain cells and discovered that neurons in the visual cortex respond to specific edges and orientations, arranged in layered feature detectors. Their work, later awarded a Nobel Prize, inspired the hierarchical architecture of modern computer vision networks.", "category": "computer_vision", "significance": "notable", "tags": [ "neuroscience", "visual cortex", "feature detection", "inspiration", "nobel" ], "references": [ { "title": "Hubel & Wiesel, 'Receptive fields of single neurones in the cat's striate cortex' (J. Physiol., 1959)", "url": "https://doi.org/10.1113/jphysiol.1959.sp006308", "type": "paper" }, { "title": "David H. Hubel", "url": "https://en.wikipedia.org/wiki/David_H._Hubel", "type": "wikipedia" } ] }, { "id": "licklider-symbiosis", "year": 1960.2, "title": "Man-Computer Symbiosis", "description": "J.C.R. Licklider's seminal paper envisioned intimate human-computer partnership, where 'men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations.' He shaped the creation of ARPANET.", "category": "society_impact", "significance": "notable", "tags": [ "ARPANET", "Licklider", "partnership", "internet", "paper", "DARPA" ], "references": [ { "title": "Licklider, 'Man-Computer Symbiosis' (IRE Transactions, 1960)", "url": "https://groups.csail.mit.edu/medg/people/psz/Licklider.html", "type": "paper" }, { "title": "Man-Computer Symbiosis", "url": "https://en.wikipedia.org/wiki/Man-Computer_Symbiosis", "type": "wikipedia" } ] }, { "id": "dec-pdp-1", "year": 1960.92, "title": "DEC PDP-1 Interactive Minicomputer", "description": "Digital Equipment Corporation's PDP-1 (November 1960) was among the first interactive computers, with a CRT display and light pen instead of batch punched cards. MIT hackers used it to write Spacewar! (1962), the first widely distributed video game.", "category": "computing_hardware", "significance": "notable", "tags": [ "dec", "pdp-1", "interactive", "mit", "minicomputer" ], "references": [ { "title": "PDP-1", "url": "https://en.wikipedia.org/wiki/PDP-1", "type": "wikipedia" }, { "title": "Computer History Museum: PDP-1", "url": "https://www.computerhistory.org/pdp-1/", "type": "archive" } ] }, { "id": "ibm_shoebox_1962", "year": 1962, "title": "IBM Shoebox", "description": "At the 1962 Seattle World's Fair, IBM demonstrated the Shoebox, a shoebox-sized machine that recognized 16 spoken English words and digits. It performed simple arithmetic when spoken to, showing early voice-controlled computing to the public.", "category": "speech_audio", "significance": "minor", "tags": [ "ibm", "speech-recognition", "demo", "worlds-fair", "voice" ], "references": [ { "title": "IBM Shoebox", "url": "https://en.wikipedia.org/wiki/IBM_Shoebox", "type": "wikipedia" }, { "title": "IBM Archives: Shoebox", "url": "https://www.ibm.com/history/shoebox", "type": "docs" } ] }, { "id": "roberts-3d-solids", "year": 1963.5, "title": "Machine Perception of Three-Dimensional Solids", "description": "Lawrence Roberts's MIT doctoral thesis showed how a computer could infer the three-dimensional shape of block-like objects from a single photograph, using edges and geometry. It is widely regarded as the founding work of computer vision as a scientific discipline.", "category": "computer_vision", "significance": "notable", "tags": [ "roberts", "mit", "3d reconstruction", "edges", "thesis" ], "references": [ { "title": "Roberts, 'Machine Perception of Three-Dimensional Solids' (MIT, 1963)", "url": "https://dspace.mit.edu/handle/1721.1/11589", "type": "archive" }, { "title": "Lawrence Roberts (computer scientist)", "url": "https://en.wikipedia.org/wiki/Lawrence_Roberts_(computer_scientist)", "type": "wikipedia" } ] }, { "id": "dr_strangelove", "year": 1964.08, "title": "Dr. Strangelove", "description": "Stanley Kubrick's Cold War satire depicted the Soviet Doomsday Machine, an automated nuclear retaliation system designed to trigger without human intervention. The film crystallized anxieties about delegating extinction-level decisions to automated systems.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "nuclear", "automation", "cold-war", "1964" ], "references": [ { "title": "Dr. Strangelove", "url": "https://en.wikipedia.org/wiki/Dr._Strangelove", "type": "wikipedia" }, { "title": "Dr. Strangelove (1964) — IMDb", "url": "https://www.imdb.com/title/tt0057012/", "type": "docs" } ], "date": "1964-01-29", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q105702" }, { "id": "ibm-system-360", "year": 1964.27, "title": "IBM System/360 Architecture", "description": "Announced April 7, 1964, System/360 was the first family of compatible mainframes spanning small to large, with a unified instruction set. It defined the 8-bit byte and established the idea of backward-compatible computer architectures still used today.", "category": "computing_hardware", "significance": "notable", "tags": [ "ibm", "mainframe", "architecture", "compatibility", "byte" ], "references": [ { "title": "IBM System/360", "url": "https://en.wikipedia.org/wiki/IBM_System/360", "type": "wikipedia" }, { "title": "IBM: System/360", "url": "https://www.ibm.com/history/system-360", "type": "archive" } ], "date": "1964-04-07", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/IBM_System/360" }, { "id": "face-recognition", "year": 1964.5, "title": "Face Recognition", "description": "Woody Bledsoe and colleagues began programming computers to match photographs of faces, pioneering one of the most consequential and controversial applications of computer vision.", "category": "computer_vision", "significance": "notable", "tags": [ "Bledsoe", "biometrics", "surveillance", "identification", "photography" ], "references": [ { "title": "Shaun Raviv, 'The Secret History of Facial Recognition' (Wired, 2020)", "url": "https://www.wired.com/story/secret-history-facial-recognition/", "type": "press" }, { "title": "Woodrow Wilson Bledsoe", "url": "https://en.wikipedia.org/wiki/Woodrow_Wilson_Bledsoe", "type": "wikipedia" } ] }, { "id": "robinson-resolution", "year": 1965, "title": "Robinson's Resolution Principle", "description": "John Alan Robinson introduced the resolution inference rule, a single mechanizable step that made automated theorem proving practical. Resolution is the engine behind Prolog and most logic-based AI systems.", "category": "logic_mathematics", "significance": "notable", "tags": [ "robinson", "resolution", "theorem proving", "prolog", "automated reasoning" ], "references": [ { "title": "A Machine-Oriented Logic Based on the Resolution Principle (1965)", "url": "https://doi.org/10.1145/321250.321253", "type": "paper" }, { "title": "Resolution (logic) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Resolution_(logic)", "type": "wikipedia" } ] }, { "id": "intelligence-explosion", "year": 1965.12, "title": "Intelligence Explosion", "description": "I.J. Good proposed that a sufficiently intelligent machine could design an even better machine, leading to an 'intelligence explosion' far surpassing human capacity. A foundational concept in AI safety.", "category": "ethics_safety", "significance": "notable", "tags": [ "Good", "superintelligence", "singularity", "recursive", "self-improvement", "existential risk" ], "references": [ { "title": "I.J. Good, Speculations Concerning the First Ultraintelligent Machine (1965)", "url": "https://doi.org/10.1016/S0065-2458(08)60418-0", "type": "paper" }, { "title": "Technological singularity (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Technological_singularity", "type": "wikipedia" } ] }, { "id": "fuzzy-logic", "year": 1965.45, "title": "Fuzzy Logic", "description": "Lotfi Zadeh introduced fuzzy sets and fuzzy logic, allowing systems to handle degrees of truth rather than binary true/false, enabling more nuanced reasoning under uncertainty.", "category": "logic_mathematics", "significance": "notable", "tags": [ "Zadeh", "uncertainty", "degrees", "truth", "sets" ], "references": [ { "title": "Zadeh, L.A. (1965) Fuzzy sets", "url": "https://doi.org/10.1016/S0019-9958(65)90241-X", "type": "paper" }, { "title": "Fuzzy logic (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Fuzzy_logic", "type": "wikipedia" }, { "title": "Fuzzy Logic (Stanford Encyclopedia of Philosophy)", "url": "https://plato.stanford.edu/entries/logic-fuzzy/", "type": "docs" } ] }, { "id": "dendral", "year": 1965.5, "title": "DENDRAL Expert System", "description": "At Stanford, Edward Feigenbaum, Bruce Buchanan, Joshua Lederberg, and Carl Djerassi built DENDRAL, the first successful expert system. It applied rules of chemistry to interpret mass spectrometry data and infer the molecular structure of organic compounds, launching the field of knowledge-based AI.", "category": "machine_learning", "significance": "major", "tags": [ "expert-system", "stanford", "chemistry", "knowledge-based" ], "references": [ { "title": "Dendral (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Dendral", "type": "wikipedia" }, { "title": "On Generality and Problem Solving: A Case Study Using the DENDRAL Program", "url": "http://www-formal.stanford.edu/jmc/some.html", "type": "archive" } ] }, { "id": "expert-systems", "year": 1965.62, "title": "Expert Systems", "description": "Edward Feigenbaum and Joshua Lederberg began work on DENDRAL at Stanford, the first expert system. It used encoded domain knowledge to identify chemical compounds from mass spectrometry data.", "category": "machine_learning", "significance": "notable", "tags": [ "DENDRAL", "Feigenbaum", "Stanford", "chemistry", "knowledge", "rules" ], "references": [ { "title": "Lindsay, R. K., Buchanan, B. G., Feigenbaum, E. A., & Lederberg, J. (1993). DENDRAL: a case study of the first expert system for scientific hypothesis formation. Artificial Intelligence, 61(2), 209-261.", "url": "https://doi.org/10.1016/0004-3702(93)90068-M", "type": "paper" }, { "title": "Dendral", "url": "https://en.wikipedia.org/wiki/Dendral", "type": "wikipedia" } ] }, { "id": "deep-learning-origins", "year": 1965.88, "title": "Deep Learning Origins (Ivakhnenko GMDH)", "description": "Alexei Ivakhnenko and Valentin Lapa introduced the Group Method of Data Handling (GMDH), the first learning algorithm for multilayer neural networks. By 1971 Ivakhnenko reported a GMDH network eight layers deep — the earliest published 'deep' supervised neural network, decades before the 2010s revival of the term.", "category": "neural_networks", "significance": "notable", "tags": [ "Ivakhnenko", "GMDH", "multilayer", "Ukraine", "Soviet", "deep networks" ], "references": [ { "title": "Ivakhnenko, A. G. (1971). 'Polynomial Theory of Complex Systems.' IEEE Trans. Systems, Man, and Cybernetics SMC-1(4):364-378.", "url": "https://doi.org/10.1109/TSMC.1971.4308320", "type": "paper" }, { "title": "Group Method of Data Handling (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Group_method_of_data_handling", "type": "wikipedia" }, { "title": "Alexey Ivakhnenko (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Alexey_Ivakhnenko", "type": "wikipedia" } ] }, { "id": "eliza", "year": 1966, "title": "ELIZA", "description": "Joseph Weizenbaum's chatbot simulated a Rogerian psychotherapist using simple pattern matching. Users formed emotional bonds with it, alarming Weizenbaum about the ease of anthropomorphizing machines.", "category": "language_nlp", "significance": "major", "tags": [ "chatbot", "Weizenbaum", "therapy", "pattern matching", "conversation", "MIT" ], "references": [ { "title": "Weizenbaum, 'ELIZA — A Computer Program For the Study of Natural Language Communication Between Man and Machine', CACM 9(1), 1966", "url": "https://doi.org/10.1145/365153.365168", "type": "paper" }, { "title": "ELIZA", "url": "https://en.wikipedia.org/wiki/ELIZA", "type": "wikipedia" } ], "date": "1966-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q580318" }, { "id": "moon_harsh_mistress", "year": 1966, "title": "The Moon Is a Harsh Mistress", "description": "Robert A. Heinlein's novel features Mike, a lunar colony supercomputer that awakens to self-awareness and becomes a co-conspirator in a revolution. An early sympathetic portrait of emergent machine consciousness through sheer complexity.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "consciousness", "scifi", "1966" ], "references": [ { "title": "The Moon Is a Harsh Mistress", "url": "https://en.wikipedia.org/wiki/The_Moon_Is_a_Harsh_Mistress", "type": "wikipedia" }, { "title": "The Moon Is a Harsh Mistress — Goodreads", "url": "https://www.goodreads.com/book/show/16690.The_Moon_Is_a_Harsh_Mistress", "type": "book" } ], "date": "1966-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q1410482" }, { "id": "shakey", "year": 1966.25, "title": "Shakey the Robot", "description": "SRI International's Shakey was the first general-purpose mobile robot to reason about its own actions. It combined vision, planning (STRIPS), and navigation in a single system.", "category": "autonomous_systems", "significance": "major", "tags": [ "SRI", "mobile", "navigation", "STRIPS", "planning", "robot" ], "references": [ { "title": "Shakey the Robot (SRI International technical reports & history)", "url": "https://www.sri.com/hoi/shakey-the-robot/", "type": "archive" }, { "title": "Shakey the robot", "url": "https://en.wikipedia.org/wiki/Shakey_the_robot", "type": "wikipedia" }, { "title": "STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving", "url": "https://ai.stanford.edu/~nilsson/OnlinePubs-Nils/PublishedPapers/strips.pdf", "type": "paper" } ] }, { "id": "alpac-report", "year": 1966.88, "title": "ALPAC Report", "description": "The U.S. Automatic Language Processing Advisory Committee concluded that machine translation was slower, less accurate and twice as expensive as human translation. The report collapsed U.S. funding for MT research and contributed to the first AI winter.", "category": "language_nlp", "significance": "notable", "tags": [ "translation", "funding", "ai-winter", "policy", "report" ], "references": [ { "title": "ALPAC", "url": "https://en.wikipedia.org/wiki/ALPAC", "type": "wikipedia" }, { "title": "ALPAC, 'Language and Machines' (1966)", "url": "https://www.nap.edu/catalog/9547/language-and-machines-computers-in-translation-and-linguistics", "type": "archive" } ] }, { "id": "kmeans-macqueen", "year": 1967, "title": "k-Means Clustering", "description": "James MacQueen formalized and named the k-means algorithm, a simple iterative method for partitioning data into groups around cluster centers. It became one of the most widely used unsupervised learning techniques in statistics and data mining.", "category": "machine_learning", "significance": "minor", "tags": [ "clustering", "unsupervised", "statistics", "algorithm", "data-mining" ], "references": [ { "title": "MacQueen, J. (1967). Some methods for classification and analysis of multivariate observations. Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1, 281-297.", "url": "https://projecteuclid.org/ebooks/berkeley-symposium-on-mathematical-statistics-and-probability/Proceedings-of-the-Fifth-Berkeley-Symposium-on-Mathematical-Statistics-and/chapter/Some-methods-for-classification-and-analysis-of-multivariate-observations/bsmsp/1200512992", "type": "archive" }, { "title": "k-means clustering", "url": "https://en.wikipedia.org/wiki/K-means_clustering", "type": "wikipedia" } ] }, { "id": "no_mouth_scream", "year": 1967.25, "title": "I Have No Mouth, and I Must Scream", "description": "Harlan Ellison's short story imagines AM, a sentient supercomputer born from merged military AIs that exterminates humanity and tortures the last survivors for eternity. An early and uniquely malevolent vision of superintelligent machine hatred.", "category": "mythology_fiction", "significance": "notable", "tags": [ "short-story", "dystopia", "superintelligence", "1967" ], "references": [ { "title": "I Have No Mouth, and I Must Scream", "url": "https://en.wikipedia.org/wiki/I_Have_No_Mouth,_and_I_Must_Scream", "type": "wikipedia" } ], "date": "1967-03-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q2307397" }, { "id": "simulation-argument", "year": 1967.79, "title": "Living in a Simulation", "description": "Konrad Zuse proposed in Calculating Space (Rechnender Raum) that the universe might be computed on a cellular automaton, a philosophical precursor to later simulation arguments.", "category": "philosophy_mind", "significance": "minor", "tags": [ "Zuse", "universe", "computation", "cellular automaton", "simulation", "German" ], "references": [ { "title": "Rechnender Raum (Calculating Space)", "url": "https://philpapers.org/rec/ZUSRR", "type": "book" }, { "title": "Digital physics", "url": "https://en.wikipedia.org/wiki/Digital_physics", "type": "wikipedia" } ] }, { "id": "androids_dream", "year": 1968.25, "title": "Do Androids Dream of Electric Sheep?", "description": "Philip K. Dick's novel follows a bounty hunter tracking escaped androids whose empathy can be tested but never proven. It established the core dilemma of distinguishing simulated from authentic inner life, later adapted as Blade Runner.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "androids", "empathy", "philip-k-dick", "1968" ], "references": [ { "title": "Do Androids Dream of Electric Sheep?", "url": "https://en.wikipedia.org/wiki/Do_Androids_Dream_of_Electric_Sheep%3F", "type": "wikipedia" }, { "title": "Do Androids Dream of Electric Sheep? — Goodreads", "url": "https://www.goodreads.com/book/show/7082.Do_Androids_Dream_of_Electric_Sheep_", "type": "book" } ], "date": "1968-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q605249" }, { "id": "hal-9000", "year": 1968.25, "title": "HAL 9000", "description": "Stanley Kubrick's 2001: A Space Odyssey introduced HAL 9000, a calm, competent AI that turns lethal when its goals conflict with its crew's survival. An enduring icon of AI danger.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "movie", "cinema", "Kubrick", "2001", "space", "danger", "culture", "sci-fi" ], "references": [ { "title": "HAL 9000 — Wikipedia", "url": "https://en.wikipedia.org/wiki/HAL_9000", "type": "wikipedia" }, { "title": "2001: A Space Odyssey (1968) — IMDb", "url": "https://www.imdb.com/title/tt0062622/", "type": "docs" } ] }, { "id": "clarke_2001_novel", "year": 1968.5, "title": "2001: A Space Odyssey (novel)", "description": "Arthur C. Clarke's novel, developed in parallel with Kubrick's film, gave HAL 9000 an inner life and explained his breakdown as a logical response to contradictory orders. The book deepened the film's portrait of AI as neither malicious nor benign but tragically coherent.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "hal-9000", "scifi", "clarke", "1968" ], "references": [ { "title": "2001: A Space Odyssey (novel)", "url": "https://en.wikipedia.org/wiki/2001:_A_Space_Odyssey_(novel)", "type": "wikipedia" }, { "title": "2001: A Space Odyssey — Goodreads", "url": "https://www.goodreads.com/book/show/70535.2001", "type": "book" } ], "date": "1968-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q835341" }, { "id": "cybernetic-serendipity", "year": 1968.59, "title": "Cybernetic Serendipity", "description": "A landmark exhibition at the ICA London showcasing computer art, music, poetry, and creative machines. It demonstrated that computation could produce aesthetically compelling work.", "category": "art_creativity", "significance": "notable", "tags": [ "exhibition", "art", "ICA", "London", "computer art", "culture" ], "references": [ { "title": "Cybernetic Serendipity at the ICA, 1968", "url": "https://www.ica.art/exhibitions/cybernetic-serendipity", "type": "archive" }, { "title": "Cybernetic Serendipity", "url": "https://en.wikipedia.org/wiki/Cybernetic_Serendipity", "type": "wikipedia" } ], "date": "1968-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q5197816", "yearShifted": true, "originalYear": 1969 }, { "id": "frame-problem", "year": 1969, "title": "The Frame Problem", "description": "John McCarthy and Patrick Hayes identified a deep difficulty for artificial reasoning: how can a machine know which facts about the world stay the same when an action is taken, and which change? The frame problem became a defining challenge for symbolic AI and philosophy of mind.", "category": "philosophy_mind", "significance": "notable", "tags": [ "McCarthy", "Hayes", "reasoning", "symbolic", "common sense" ], "references": [ { "title": "Some Philosophical Problems from the Standpoint of Artificial Intelligence", "url": "http://jmc.stanford.edu/articles/mcchay69.html", "type": "paper" }, { "title": "Frame problem", "url": "https://en.wikipedia.org/wiki/Frame_problem", "type": "wikipedia" } ] }, { "id": "perceptrons-book", "year": 1969, "title": "Perceptrons (Minsky & Papert)", "description": "Marvin Minsky and Seymour Papert published a mathematical analysis of single-layer perceptrons, proving they could not compute simple functions such as exclusive-or. The book's pessimistic tone is widely credited with pausing funding and research on neural networks for more than a decade.", "category": "neural_networks", "significance": "major", "tags": [ "perceptron", "minsky", "papert", "ai-winter", "mit" ], "references": [ { "title": "Perceptrons (book) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Perceptrons_(book)", "type": "wikipedia" }, { "title": "Perceptrons: An Introduction to Computational Geometry", "url": "https://mitpress.mit.edu/9780262631112/perceptrons/", "type": "book" } ], "date": "1969-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q4359403" }, { "id": "colossus-forbin", "year": 1970.27, "title": "Colossus: The Forbin Project", "description": "A film about a U.S. defense supercomputer that links with its Soviet counterpart and seizes control of the world's nuclear arsenals, a Cold War AI takeover scenario.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "movie", "nuclear", "Cold War", "takeover", "sci-fi" ], "references": [ { "title": "Colossus: The Forbin Project — Wikipedia", "url": "https://en.wikipedia.org/wiki/Colossus:_The_Forbin_Project", "type": "wikipedia" }, { "title": "Colossus: The Forbin Project (1970) — IMDb", "url": "https://www.imdb.com/title/tt0064177/", "type": "docs" } ], "date": "1970-04-08", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1111873" }, { "id": "mastermind", "year": 1970.5, "title": "Mastermind", "description": "Mordecai Meirowitz's code-breaking board game became a testbed for deductive reasoning algorithms, information theory, and computational strategy.", "category": "game_playing", "significance": "minor", "tags": [ "game", "deduction", "code-breaking", "board game" ], "references": [ { "title": "Mastermind (board game)", "url": "https://en.wikipedia.org/wiki/Mastermind_(board_game)", "type": "wikipedia" }, { "title": "Knuth, 'The Computer as Master Mind' (J. Recreational Mathematics, 1976)", "url": "https://www.cs.uni.edu/~wallingf/teaching/cs3530/resources/knuth-mastermind.pdf", "type": "paper" } ], "date": "1970-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q17286" }, { "id": "shrdlu", "year": 1970.5, "title": "SHRDLU", "description": "Terry Winograd's program could understand and manipulate objects in a virtual blocks world through natural language dialogue, a landmark in language understanding research.", "category": "language_nlp", "significance": "major", "tags": [ "Winograd", "blocks world", "understanding", "dialogue", "MIT" ], "references": [ { "title": "Winograd, 'Procedures as a Representation for Data in a Computer Program for Understanding Natural Language', MIT AI-TR-235 (1971)", "url": "https://dspace.mit.edu/handle/1721.1/7095", "type": "archive" }, { "title": "SHRDLU", "url": "https://en.wikipedia.org/wiki/SHRDLU", "type": "wikipedia" } ], "date": "1970-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q1778029" }, { "id": "cook-levin-np", "year": 1971.05, "title": "Cook-Levin Theorem (NP-completeness)", "description": "Stephen Cook and, independently, Leonid Levin proved that Boolean satisfiability is NP-complete, revealing a class of problems whose hardness links thousands of AI tasks together. The P vs NP question became the central open problem of computer science.", "category": "logic_mathematics", "significance": "notable", "tags": [ "complexity", "np-complete", "cook", "levin", "sat" ], "references": [ { "title": "The Complexity of Theorem-Proving Procedures (1971)", "url": "https://doi.org/10.1145/800157.805047", "type": "paper" }, { "title": "Cook-Levin theorem (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Cook%E2%80%93Levin_theorem", "type": "wikipedia" } ] }, { "id": "mycin", "year": 1972.5, "title": "MYCIN Medical Expert System", "description": "At Stanford, Edward Shortliffe built MYCIN, a rule-based system that diagnosed blood infections and recommended antibiotic treatments. It handled uncertainty with certainty factors and, in trials, performed comparably to human experts, becoming the template for a generation of medical AI systems.", "category": "machine_learning", "significance": "major", "tags": [ "expert-system", "medicine", "stanford", "knowledge-based" ], "references": [ { "title": "Mycin (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Mycin", "type": "wikipedia" }, { "title": "Computer-based medical consultations: MYCIN", "url": "https://www.sciencedirect.com/book/9780444001795/computer-based-medical-consultations-mycin", "type": "book" } ] }, { "id": "parry", "year": 1972.5, "title": "Paranoid PARRY", "description": "Kenneth Colby's PARRY simulated a person with paranoid schizophrenia. In blind tests, psychiatrists could not reliably distinguish it from real patients, an early Turing test of sorts.", "category": "language_nlp", "significance": "minor", "tags": [ "psychiatry", "simulation", "schizophrenia", "Turing test", "chatbot" ], "references": [ { "title": "PARRY", "url": "https://en.wikipedia.org/wiki/PARRY", "type": "wikipedia" }, { "title": "Colby et al., 'Artificial Paranoia', Artificial Intelligence 2(1), 1971", "url": "https://doi.org/10.1016/S0004-3702(71)80001-3", "type": "paper" } ], "date": "1972-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q1967077" }, { "id": "prolog-language", "year": 1972.5, "title": "Prolog Programming Language", "description": "Alain Colmerauer and Philippe Roussel at the University of Marseille created Prolog, a language based on formal logic in which programs are written as sets of facts and rules. It became the main tool of European AI research and of Japan's Fifth Generation Computer Systems project.", "category": "logic_mathematics", "significance": "major", "tags": [ "programming-language", "prolog", "logic", "marseille" ], "references": [ { "title": "Prolog (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Prolog", "type": "wikipedia" }, { "title": "The birth of Prolog", "url": "https://dl.acm.org/doi/10.1145/234286.1057820", "type": "paper" } ], "date": "1972-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q163468" }, { "id": "aaron-cohen", "year": 1973, "title": "Harold Cohen's AARON", "description": "Painter Harold Cohen began developing AARON, a rule-based program that autonomously produced original drawings and paintings. AARON became one of the longest-running experiments in machine-made art, evolving over four decades.", "category": "art_creativity", "significance": "notable", "tags": [ "AARON", "painting", "rule-based", "autonomy", "pioneer" ], "references": [ { "title": "Harold Cohen and AARON (Victoria & Albert Museum / Computer History Museum)", "url": "https://computerhistory.org/blog/harold-cohen-and-aaron-a-40-year-collaboration/", "type": "archive" }, { "title": "AARON", "url": "https://en.wikipedia.org/wiki/AARON", "type": "wikipedia" } ] }, { "id": "lighthill-report", "year": 1973.54, "title": "Lighthill Report", "description": "Sir James Lighthill's devastating critique of AI research for the UK Science Research Council concluded that 'in no part of the field have discoveries made so far produced the major impact that was then promised.' It triggered massive funding cuts and began the First AI Winter.", "category": "policy_governance", "significance": "major", "tags": [ "UK", "funding", "critique", "winter", "government", "report" ], "references": [ { "title": "Lighthill report", "url": "https://en.wikipedia.org/wiki/Lighthill_report", "type": "wikipedia" }, { "title": "Artificial Intelligence: A General Survey (1973)", "url": "https://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p001.htm", "type": "docs" } ], "date": "1973-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q4340232", "yearShifted": true, "originalYear": 1974 }, { "id": "first-ai-winter", "year": 1974, "title": "First AI Winter Begins", "description": "Following the Lighthill Report in the UK and cuts by DARPA in the US, funding for AI research collapsed in the mid-1970s. The period, later named the first AI winter, saw many laboratories close or pivot, and it lasted roughly until the rise of commercial expert systems in the early 1980s.", "category": "policy_governance", "significance": "major", "tags": [ "ai-winter", "funding", "darpa", "history" ], "references": [ { "title": "AI winter (Wikipedia)", "url": "https://en.wikipedia.org/wiki/AI_winter", "type": "wikipedia" }, { "title": "Artificial Intelligence: A General Survey (Lighthill)", "url": "http://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/contents.htm", "type": "archive" } ] }, { "id": "nagel-bat", "year": 1974, "title": "What Is It Like to Be a Bat?", "description": "Thomas Nagel argued that subjective experience cannot be captured by any objective physical description. A creature's inner perspective, what it feels like to echolocate as a bat, resists reduction to brain states or computation, framing a core challenge for machine consciousness.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Nagel", "consciousness", "qualia", "subjectivity", "phenomenology" ], "references": [ { "title": "What Is It Like to Be a Bat?", "url": "https://doi.org/10.2307/2183914", "type": "paper" }, { "title": "What Is It Like to Be a Bat?", "url": "https://en.wikipedia.org/wiki/What_Is_It_Like_to_Be_a_Bat%3F", "type": "wikipedia" } ], "date": "1974-10-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q2703462" }, { "id": "hmm_speech_1975", "year": 1975, "title": "Hidden Markov Models for Speech", "description": "James Baker's Dragon system (CMU PhD thesis, 1975) was among the first to apply Hidden Markov Models to continuous speech recognition, following theoretical work by Baum and Welch in the late 1960s. HMMs became the dominant paradigm for speech recognition for the next three decades.", "category": "speech_audio", "significance": "major", "tags": [ "hmm", "statistical", "speech-recognition", "cmu", "baker" ], "references": [ { "title": "Hidden Markov model", "url": "https://en.wikipedia.org/wiki/Hidden_Markov_model", "type": "wikipedia" }, { "title": "Baker — The DRAGON System: An Overview (IEEE Trans. Acoustics, 1975)", "url": "https://doi.org/10.1109/TASSP.1975.1162650", "type": "paper" } ] }, { "id": "genetic-algorithms", "year": 1975.5, "title": "Genetic Algorithms", "description": "John Holland published Adaptation in Natural and Artificial Systems, formalizing genetic algorithms as a search and optimization method inspired by evolution.", "category": "machine_learning", "significance": "notable", "tags": [ "Holland", "evolution", "optimization", "search", "biology-inspired" ], "references": [ { "title": "Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. University of Michigan Press.", "url": "https://mitpress.mit.edu/9780262581110/adaptation-in-natural-and-artificial-systems/", "type": "book" }, { "title": "Genetic algorithm", "url": "https://en.wikipedia.org/wiki/Genetic_algorithm", "type": "wikipedia" } ], "date": "1975-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q187787", "yearShifted": true, "originalYear": 1976 }, { "id": "darpa_sur_harpy_1976", "year": 1976, "title": "CMU Harpy and DARPA SUR", "description": "Under DARPA's Speech Understanding Research program (1971-1976), Carnegie Mellon's Harpy system recognized over 1,000 words with reasonable accuracy using a beam search over a graph of possible sentences. Harpy met the DARPA program goals and shaped decades of speech research.", "category": "speech_audio", "significance": "notable", "tags": [ "darpa", "cmu", "speech-recognition", "beam-search", "harpy" ], "references": [ { "title": "Speech Understanding Research", "url": "https://en.wikipedia.org/wiki/Speech_Understanding_Research", "type": "wikipedia" }, { "title": "B. Lowerre — The Harpy Speech Recognition System (CMU PhD thesis, 1976)", "url": "https://kilthub.cmu.edu/articles/thesis/The_Harpy_speech_recognition_system/6609221", "type": "docs" } ] }, { "id": "cray-1", "year": 1976.2, "title": "Cray-1 Supercomputer", "description": "Seymour Cray's Cray-1, first delivered to Los Alamos in March 1976, dominated scientific computing with 160 million operations per second and an iconic horseshoe shape. It established vector supercomputing as a distinct field.", "category": "computing_hardware", "significance": "notable", "tags": [ "cray", "supercomputer", "vector", "los-alamos", "science" ], "references": [ { "title": "Cray-1", "url": "https://en.wikipedia.org/wiki/Cray-1", "type": "wikipedia" }, { "title": "Computer History Museum: Cray-1", "url": "https://www.computerhistory.org/revolution/supercomputers/10/7", "type": "archive" } ], "date": "1975-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1093705", "yearShifted": true, "originalYear": 1976 }, { "id": "ai-ethics-emergence", "year": 1976.5, "title": "Ethics of AI", "description": "Joseph Weizenbaum published Computer Power and Human Reason, arguing that some applications of computing (like psychotherapy, judicial sentencing) should never be delegated to machines.", "category": "ethics_safety", "significance": "notable", "tags": [ "Weizenbaum", "book", "limits", "delegation", "morality" ], "references": [ { "title": "Computer Power and Human Reason (Weizenbaum, 1976)", "url": "https://en.wikipedia.org/wiki/Computer_Power_and_Human_Reason", "type": "wikipedia" }, { "title": "Joseph Weizenbaum", "url": "https://en.wikipedia.org/wiki/Joseph_Weizenbaum", "type": "wikipedia" } ], "date": "1976-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q5157465" }, { "id": "em-algorithm", "year": 1977, "title": "Expectation-Maximization Algorithm", "description": "Arthur Dempster, Nan Laird, and Donald Rubin published the expectation-maximization procedure for fitting statistical models when data are incomplete or contain hidden variables. It became a cornerstone of probabilistic machine learning, from mixture models to hidden Markov models.", "category": "machine_learning", "significance": "notable", "tags": [ "statistics", "probabilistic", "latent-variables", "algorithm", "inference" ], "references": [ { "title": "Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society: Series B, 39(1), 1-38.", "url": "https://doi.org/10.1111/j.2517-6161.1977.tb01600.x", "type": "paper" }, { "title": "Expectation-maximization algorithm", "url": "https://en.wikipedia.org/wiki/Expectation%E2%80%93maximization_algorithm", "type": "wikipedia" } ] }, { "id": "block-china-brain", "year": 1978, "title": "The China Brain", "description": "Ned Block imagined the entire population of China coordinating by radio to simulate the firing patterns of a human brain. If such a system produced the right behavior, would it also be conscious? The thought experiment targets functionalist theories that equate mind with information processing.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Block", "functionalism", "consciousness", "thought experiment", "qualia" ], "references": [ { "title": "Troubles with Functionalism", "url": "https://www.nyu.edu/gsas/dept/philo/faculty/block/papers/1978.Troubles.pdf", "type": "paper" }, { "title": "China brain", "url": "https://en.wikipedia.org/wiki/China_brain", "type": "wikipedia" } ] }, { "id": "backgammon-ai", "year": 1979.53, "title": "Backgammon Champion Defeated", "description": "Hans Berliner's BKG 9.8 defeated the reigning world backgammon champion Luigi Villa 7-1, an early milestone in computer mastery of probabilistic games.", "category": "game_playing", "significance": "notable", "tags": [ "Berliner", "BKG", "probabilistic", "board game", "champion" ], "references": [ { "title": "Berliner, 'Backgammon Computer Program Beats World Champion' (Artificial Intelligence, 1980)", "url": "https://doi.org/10.1016/0004-3702(80)90041-7", "type": "paper" }, { "title": "BKG 9.8 — Wikipedia", "url": "https://en.wikipedia.org/wiki/BKG_9.8", "type": "wikipedia" }, { "title": "Tesauro, 'Temporal Difference Learning and TD-Gammon' (Neural Computation, 1995)", "url": "https://doi.org/10.1162/neco.1995.7.2.215", "type": "paper" } ] }, { "id": "stanford-cart", "year": 1979.71, "title": "Stanford Cart Crosses a Room", "description": "Hans Moravec's Stanford Cart, a remote-controlled robot with a camera, navigated around obstacles across a chair-filled room in about five hours. It became a foundational demonstration of computer-vision-based autonomous navigation.", "category": "autonomous_systems", "significance": "major", "tags": [ "moravec", "stanford", "navigation", "vision", "robot" ], "references": [ { "title": "The Stanford Cart and the CMU Rover", "url": "https://www.cs.cmu.edu/~motionplanning/papers/sbp_papers/integrated2/moravec_stanford_cmu_rover.pdf", "type": "paper" }, { "title": "Stanford Cart", "url": "https://en.wikipedia.org/wiki/Stanford_Cart", "type": "wikipedia" } ] }, { "id": "xcon-expert-system", "year": 1980.25, "title": "R1/XCON Expert System", "description": "John McDermott's R1 (later XCON) at Digital Equipment Corporation became the first commercially successful expert system, saving DEC $40 million per year by automating VAX computer configuration. It launched the Expert Systems Boom.", "category": "machine_learning", "significance": "major", "tags": [ "DEC", "McDermott", "configuration", "commercial", "rules", "business" ], "references": [ { "title": "McDermott, J. (1982). R1: A rule-based configurer of computer systems. Artificial Intelligence, 19(1), 39-88.", "url": "https://doi.org/10.1016/0004-3702(82)90021-2", "type": "paper" }, { "title": "Xcon", "url": "https://en.wikipedia.org/wiki/Xcon", "type": "wikipedia" } ] }, { "id": "chinese-room", "year": 1980.71, "title": "Chinese Room", "description": "John Searle's thought experiment: a person in a room follows rules to manipulate Chinese symbols without understanding Chinese, arguing that computation alone cannot produce understanding.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Searle", "understanding", "consciousness", "thought experiment", "Berkeley" ], "references": [ { "title": "Minds, Brains, and Programs", "url": "https://doi.org/10.1017/S0140525X00005756", "type": "paper" }, { "title": "Chinese room", "url": "https://en.wikipedia.org/wiki/Chinese_room", "type": "wikipedia" } ], "date": "1980-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q304726", "yearShifted": true, "originalYear": 1981 }, { "id": "lucas-kanade", "year": 1981.33, "title": "Lucas-Kanade Optical Flow", "description": "Bruce Lucas and Takeo Kanade introduced a practical method for tracking how pixels move between video frames, known as optical flow. Decades later it still underpins video stabilization, motion capture, and self-driving perception.", "category": "computer_vision", "significance": "notable", "tags": [ "optical flow", "motion", "tracking", "video", "kanade" ], "references": [ { "title": "Lucas & Kanade, 'An Iterative Image Registration Technique with an Application to Stereo Vision' (IJCAI, 1981)", "url": "https://www.ri.cmu.edu/pub_files/pub3/lucas_bruce_d_1981_2/lucas_bruce_d_1981_2.pdf", "type": "paper" }, { "title": "Lucas–Kanade method", "url": "https://en.wikipedia.org/wiki/Lucas%E2%80%93Kanade_method", "type": "wikipedia" } ] }, { "id": "fifth-generation-japan", "year": 1981.83, "title": "Japan's Fifth Generation Computer Systems", "description": "Japan's Ministry of International Trade and Industry announced a ten-year, government-funded project to build a new generation of computers based on parallel logic programming and intended for artificial intelligence. The announcement alarmed Western governments and triggered matching initiatives in the US and Europe.", "category": "policy_governance", "significance": "major", "tags": [ "japan", "government", "prolog", "industrial-policy" ], "references": [ { "title": "Fifth generation computer (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Fifth_generation_computer", "type": "wikipedia" }, { "title": "The Fifth Generation: Japan's Computer Challenge to the World", "url": "https://archive.org/details/fifthgenerationj00feig", "type": "book" } ] }, { "id": "hopfield_network", "year": 1982.32, "title": "Hopfield Networks", "description": "John Hopfield introduced a recurrent neural network that stores memories as stable patterns and retrieves them from partial cues. The model drew a bridge between physics and neuroscience and helped revive interest in neural computation.", "category": "neural_networks", "significance": "major", "tags": [ "hopfield", "associative memory", "recurrent", "physics", "connectionism" ], "references": [ { "title": "Neural networks and physical systems with emergent collective computational abilities (Hopfield, 1982)", "url": "https://doi.org/10.1073/pnas.79.8.2554", "type": "paper" }, { "title": "Hopfield network (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Hopfield_network", "type": "wikipedia" } ] }, { "id": "blade-runner", "year": 1982.48, "title": "Blade Runner", "description": "Ridley Scott's film explored artificial persons (replicants) with implanted memories who seek more life, raising profound questions about consciousness, identity, and what it means to be human.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "movie", "cinema", "Scott", "replicant", "consciousness", "Philip K. Dick", "sci-fi", "culture", "dystopia" ], "references": [ { "title": "Blade Runner — Wikipedia", "url": "https://en.wikipedia.org/wiki/Blade_Runner", "type": "wikipedia" }, { "title": "Blade Runner (1982) — IMDb", "url": "https://www.imdb.com/title/tt0083658/", "type": "docs" } ], "date": "1982-06-25", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q184843" }, { "id": "marr-vision", "year": 1982.5, "title": "Marr's Vision", "description": "Published posthumously, David Marr's book Vision proposed that seeing can be studied on three levels, from the problem itself to the algorithms and the hardware. His framework of progressively richer sketches of the scene shaped how researchers think about machines that see.", "category": "computer_vision", "significance": "notable", "tags": [ "marr", "theory", "book", "levels of analysis", "primal sketch" ], "references": [ { "title": "David Marr, 'Vision' (W. H. Freeman, 1982; MIT Press reprint 2010)", "url": "https://mitpress.mit.edu/9780262514620/vision/", "type": "book" }, { "title": "David Marr (neuroscientist)", "url": "https://en.wikipedia.org/wiki/David_Marr_(neuroscientist)", "type": "wikipedia" } ] }, { "id": "tron_film", "year": 1982.52, "title": "Tron", "description": "Disney's film visualized the inside of a mainframe as a gladiatorial world ruled by the Master Control Program, a rogue AI that absorbs other programs to expand its power. An early pop-culture image of cyberspace and hostile software agents.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "cyberspace", "programs", "1982" ], "references": [ { "title": "Tron", "url": "https://en.wikipedia.org/wiki/Tron", "type": "wikipedia" }, { "title": "Tron (1982) — IMDb", "url": "https://www.imdb.com/title/tt0084827/", "type": "docs" } ], "date": "1982-07-09", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q626483", "yearShifted": true, "originalYear": 1983 }, { "id": "wargames_film", "year": 1983.42, "title": "WarGames", "description": "The film depicts a teenager who unknowingly connects to WOPR, a NORAD war-simulation AI that cannot distinguish game from reality and nearly triggers nuclear war. Its famous conclusion — that the only winning move is not to play — became shorthand for AI strategic reasoning gone wrong.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "nuclear", "cold-war", "1983" ], "references": [ { "title": "WarGames", "url": "https://en.wikipedia.org/wiki/WarGames", "type": "wikipedia" }, { "title": "WarGames (1983) — IMDb", "url": "https://www.imdb.com/title/tt0086567/", "type": "docs" } ], "date": "1983-06-03", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q504697" }, { "id": "cyc-project", "year": 1984.5, "title": "Cyc Project Begins", "description": "At the Microelectronics and Computer Technology Corporation in Texas, Douglas Lenat launched Cyc, a decades-long effort to hand-code the common-sense knowledge humans take for granted. It became the longest-running symbolic AI project and a cautionary emblem of how hard everyday reasoning is for machines.", "category": "logic_mathematics", "significance": "major", "tags": [ "common-sense", "knowledge-base", "lenat", "symbolic-ai" ], "references": [ { "title": "Cyc (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Cyc", "type": "wikipedia" }, { "title": "CYC: Using Common Sense Knowledge to Overcome Brittleness and Knowledge Acquisition Bottlenecks", "url": "https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/510", "type": "paper" } ], "date": "1984-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q1147294" }, { "id": "neuromancer", "year": 1984.53, "title": "Neuromancer", "description": "William Gibson's debut novel introduced cyberspace and the AIs Wintermute and Neuromancer, whose hidden goal is to merge into a single superintelligent being beyond human law. The book defined cyberpunk and shaped how a generation imagined networked AI.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "cyberpunk", "cyberspace", "gibson", "1984" ], "references": [ { "title": "Neuromancer", "url": "https://en.wikipedia.org/wiki/Neuromancer", "type": "wikipedia" }, { "title": "Neuromancer — Goodreads", "url": "https://www.goodreads.com/book/show/22328.Neuromancer", "type": "book" } ], "date": "1984-07-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q662029", "yearShifted": true, "originalYear": 1985 }, { "id": "terminator", "year": 1984.82, "title": "The Terminator", "description": "James Cameron's film introduced Skynet, an AI defense network that becomes self-aware and launches nuclear war. The Terminator became the defining popular image of catastrophic AI.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "movie", "cinema", "Cameron", "Skynet", "nuclear", "sci-fi", "robot", "culture", "apocalypse" ], "references": [ { "title": "The Terminator — Wikipedia", "url": "https://en.wikipedia.org/wiki/The_Terminator", "type": "wikipedia" }, { "title": "The Terminator (1984) — IMDb", "url": "https://www.imdb.com/title/tt0088247/", "type": "docs" } ], "date": "1984-10-26", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q162255", "yearShifted": true, "originalYear": 1985 }, { "id": "wordnet", "year": 1985, "title": "WordNet", "description": "George Miller and colleagues at Princeton began building WordNet, a large lexical database that grouped English words into sets of synonyms and mapped relations like hypernymy and meronymy. It became the backbone of countless NLP systems before the neural era.", "category": "language_nlp", "significance": "notable", "tags": [ "lexicon", "princeton", "miller", "semantics", "dataset" ], "references": [ { "title": "Miller, 'WordNet: A Lexical Database for English', CACM 38(11), 1995", "url": "https://doi.org/10.1145/219717.219748", "type": "paper" }, { "title": "WordNet", "url": "https://en.wikipedia.org/wiki/WordNet", "type": "wikipedia" } ] }, { "id": "boltzmann_machine", "year": 1985.5, "title": "Boltzmann Machine", "description": "Geoffrey Hinton and Terrence Sejnowski proposed the Boltzmann Machine, a stochastic neural network that learns internal representations by sampling from a probability distribution. It laid the groundwork for later generative models in deep learning.", "category": "neural_networks", "significance": "major", "tags": [ "hinton", "sejnowski", "stochastic", "generative", "energy-based" ], "references": [ { "title": "A Learning Algorithm for Boltzmann Machines (Ackley, Hinton, Sejnowski, 1985)", "url": "https://doi.org/10.1207/s15516709cog0901_7", "type": "paper" }, { "title": "Boltzmann machine (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Boltzmann_machine", "type": "wikipedia" } ] }, { "id": "connection-machine", "year": 1986, "title": "Thinking Machines Connection Machine CM-1", "description": "Danny Hillis's CM-1, launched in 1986, was a massively parallel computer with 65,536 simple processors designed for artificial intelligence workloads. It pioneered parallel architectures later vindicated by deep learning's appetite for parallelism.", "category": "computing_hardware", "significance": "notable", "tags": [ "thinking-machines", "hillis", "parallel", "cm-1", "ai-hardware" ], "references": [ { "title": "Connection Machine", "url": "https://en.wikipedia.org/wiki/Connection_Machine", "type": "wikipedia" }, { "title": "Computer History Museum: Connection Machine", "url": "https://www.computerhistory.org/collections/catalog/102646574", "type": "archive" } ] }, { "id": "autonomous-vehicles", "year": 1986.5, "title": "Autonomous Vehicles", "description": "Ernst Dickmanns' VaMoRs van at Bundeswehr University Munich drove autonomously on empty streets using computer vision, launching decades of self-driving research.", "category": "autonomous_systems", "significance": "notable", "tags": [ "Dickmanns", "self-driving", "car", "vision", "Munich", "German" ], "references": [ { "title": "Vehicles Capable of Dynamic Vision", "url": "https://www.ijcai.org/Proceedings/97-2/Papers/106.pdf", "type": "paper" }, { "title": "Ernst Dickmanns", "url": "https://en.wikipedia.org/wiki/Ernst_Dickmanns", "type": "wikipedia" }, { "title": "VaMoRs and VaMP autonomous vehicle project", "url": "https://en.wikipedia.org/wiki/VaMP", "type": "archive" } ] }, { "id": "canny-edge", "year": 1986.5, "title": "Canny Edge Detector", "description": "John Canny derived an edge-finding algorithm from mathematical principles about noise, localization, and single responses to real edges. It became the default edge detector taught in textbooks and shipped in nearly every image-processing library.", "category": "computer_vision", "significance": "notable", "tags": [ "edges", "algorithm", "image processing", "canny", "textbook" ], "references": [ { "title": "Canny, 'A Computational Approach to Edge Detection' (IEEE TPAMI, 1986)", "url": "https://doi.org/10.1109/TPAMI.1986.4767851", "type": "paper" }, { "title": "Canny edge detector", "url": "https://en.wikipedia.org/wiki/Canny_edge_detector", "type": "wikipedia" } ] }, { "id": "rbm_harmony", "year": 1986.5, "title": "Restricted Boltzmann Machine", "description": "Paul Smolensky's Harmony Theory introduced what later became known as the Restricted Boltzmann Machine, a simpler two-layer stochastic network. Decades later it helped kick off the deep learning revival through layer-wise pretraining.", "category": "neural_networks", "significance": "minor", "tags": [ "smolensky", "rbm", "harmony", "generative", "pretraining" ], "references": [ { "title": "Information Processing in Dynamical Systems: Foundations of Harmony Theory (Smolensky, 1986)", "url": "https://stanford.edu/~jlmcc/papers/PDP/Volume%201/Chap6_PDP86.pdf", "type": "paper" }, { "title": "Restricted Boltzmann machine (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Restricted_Boltzmann_machine", "type": "wikipedia" } ] }, { "id": "backpropagation", "year": 1986.77, "title": "Backpropagation", "description": "Rumelhart, Hinton, and Williams demonstrated that backpropagation could train multilayer neural networks effectively, reviving connectionism and enabling modern deep learning.", "category": "neural_networks", "significance": "major", "tags": [ "Rumelhart", "Hinton", "Williams", "gradient", "training", "connectionism" ], "references": [ { "title": "Learning representations by back-propagating errors (Rumelhart, Hinton, Williams, 1986)", "url": "https://doi.org/10.1038/323533a0", "type": "paper" }, { "title": "Backpropagation (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Backpropagation", "type": "wikipedia" } ] }, { "id": "dennett-intentional", "year": 1987, "title": "The Intentional Stance", "description": "Daniel Dennett argued that treating a system as if it had beliefs and desires is a predictive strategy, not a claim about inner essence. Thermostats, chess programs, and humans can all be usefully described this way, reshaping how philosophers think about mind, meaning, and machine agency.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Dennett", "intentionality", "agency", "functionalism", "meaning" ], "references": [ { "title": "The Intentional Stance", "url": "https://mitpress.mit.edu/9780262540537/the-intentional-stance/", "type": "book" }, { "title": "Intentional stance", "url": "https://en.wikipedia.org/wiki/Intentional_stance", "type": "wikipedia" } ] }, { "id": "ai-industry-collapse", "year": 1987.25, "title": "AI Industry Collapse", "description": "The expert systems market imploded as desktop computers surpassed Lisp machines. Symbolics, IntelliCorp, and other AI companies folded or contracted. DARPA and Japan's Fifth Generation project failed to meet goals, triggering the Second AI Winter.", "category": "society_impact", "significance": "major", "tags": [ "Lisp", "Symbolics", "DARPA", "Fifth Generation", "market crash", "winter", "Japan" ], "references": [ { "title": "AI winter", "url": "https://en.wikipedia.org/wiki/AI_winter", "type": "wikipedia" }, { "title": "New York Times: 'Setbacks for Artificial Intelligence' (1988)", "url": "https://www.nytimes.com/1988/03/04/business/setbacks-for-artificial-intelligence.html", "type": "press" } ] }, { "id": "swarm-intelligence", "year": 1987.57, "title": "Swarm Intelligence", "description": "Craig Reynolds' Boids simulation showed how complex flocking behavior emerges from simple local rules, inspiring swarm intelligence algorithms and multi-agent systems.", "category": "machine_learning", "significance": "minor", "tags": [ "Reynolds", "Boids", "flocking", "emergence", "multi-agent" ], "references": [ { "title": "Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioral model. SIGGRAPH Computer Graphics, 21(4), 25-34.", "url": "https://doi.org/10.1145/37402.37406", "type": "paper" }, { "title": "Boids", "url": "https://en.wikipedia.org/wiki/Boids", "type": "wikipedia" } ] }, { "id": "artificial-life", "year": 1987.72, "title": "Artificial Life", "description": "Christopher Langton organized the first Artificial Life workshop at Los Alamos, studying lifelike behavior emerging from computational and mathematical systems.", "category": "machine_learning", "significance": "minor", "tags": [ "Langton", "Los Alamos", "emergence", "simulation", "biology" ], "references": [ { "title": "Langton, C. G. (Ed.). (1989). Artificial Life: Proceedings of an Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (Los Alamos, September 1987). Addison-Wesley.", "url": "https://mitpress.mit.edu/9780201093568/artificial-life/", "type": "book" }, { "title": "Artificial life", "url": "https://en.wikipedia.org/wiki/Artificial_life", "type": "wikipedia" } ] }, { "id": "bayesian-networks", "year": 1988, "title": "Bayesian Networks", "description": "Judea Pearl's book Probabilistic Reasoning in Intelligent Systems introduced Bayesian networks as a principled way to represent and reason about uncertainty using graphs of probabilistic relationships. The framework reshaped artificial intelligence, statistics, and causal inference.", "category": "machine_learning", "significance": "notable", "tags": [ "probabilistic", "Pearl", "graphical-models", "uncertainty", "causal" ], "references": [ { "title": "Pearl, J. (1988). Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan Kaufmann.", "url": "https://www.sciencedirect.com/book/9780080514895/probabilistic-reasoning-in-intelligent-systems", "type": "book" }, { "title": "Bayesian network", "url": "https://en.wikipedia.org/wiki/Bayesian_network", "type": "wikipedia" } ] }, { "id": "global-workspace", "year": 1988, "title": "Global Workspace Theory", "description": "Bernard Baars proposed that consciousness works like a theatre: many unconscious processes compete, and only the winning content is broadcast to a shared workspace available to the whole mind. The theory gave cognitive science and AI a concrete architecture for modelling attention and awareness.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Baars", "consciousness", "attention", "cognitive science", "architecture" ], "references": [ { "title": "Global workspace theory", "url": "https://en.wikipedia.org/wiki/Global_workspace_theory", "type": "wikipedia" } ] }, { "id": "td-learning", "year": 1988, "title": "Temporal-Difference Learning", "description": "Richard Sutton formalized temporal-difference learning, a method that lets an agent improve its predictions by comparing successive estimates rather than waiting for a final outcome. It became the foundation of modern reinforcement learning.", "category": "machine_learning", "significance": "notable", "tags": [ "reinforcement-learning", "prediction", "Sutton", "algorithm", "rewards" ], "references": [ { "title": "Sutton, R. S. (1988). Learning to predict by the methods of temporal differences. Machine Learning, 3(1), 9-44.", "url": "https://doi.org/10.1007/BF00115009", "type": "paper" }, { "title": "Temporal difference learning", "url": "https://en.wikipedia.org/wiki/Temporal_difference_learning", "type": "wikipedia" } ] }, { "id": "moravec-paradox", "year": 1988.77, "title": "Moravec's Paradox", "description": "Hans Moravec observed that abstract reasoning is easy for computers but sensorimotor skills are hard, while for humans it's the reverse. Evolution spent billions of years on perception, mere millennia on reasoning.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Moravec", "sensorimotor", "perception", "evolution", "robotics" ], "references": [ { "title": "Moravec's paradox", "url": "https://en.wikipedia.org/wiki/Moravec%27s_paradox", "type": "wikipedia" } ] }, { "id": "connect-four", "year": 1988.79, "title": "Connect Four Solved", "description": "Victor Allis proved that with perfect play, the first player always wins Connect Four, a milestone in computational game theory and game solving.", "category": "game_playing", "significance": "minor", "tags": [ "Allis", "solved", "board game", "perfect play" ], "references": [ { "title": "Allis, 'A Knowledge-based Approach of Connect-Four' (M.Sc. thesis, 1988)", "url": "https://www.informatik.uni-trier.de/~fernau/DSL0607/Masterthesis-Viergewinnt.pdf", "type": "paper" }, { "title": "Connect Four — Wikipedia", "url": "https://en.wikipedia.org/wiki/Connect_Four", "type": "wikipedia" } ] }, { "id": "q-learning", "year": 1989, "title": "Q-Learning", "description": "Christopher Watkins introduced Q-learning in his Cambridge doctoral thesis, a reinforcement learning algorithm that learns the long-term value of actions without a model of the environment. It underpins much of later deep reinforcement learning, including the Atari and Go breakthroughs.", "category": "machine_learning", "significance": "major", "tags": [ "reinforcement-learning", "Watkins", "algorithm", "value-function", "model-free" ], "references": [ { "title": "Watkins, C. J. C. H. (1989). Learning from Delayed Rewards. PhD thesis, King's College, University of Cambridge.", "url": "https://www.cs.rhul.ac.uk/~chrisw/new_thesis.pdf", "type": "archive" }, { "title": "Watkins, C. J. C. H., & Dayan, P. (1992). Q-learning. Machine Learning, 8(3-4), 279-292.", "url": "https://doi.org/10.1007/BF00992698", "type": "paper" }, { "title": "Q-learning", "url": "https://en.wikipedia.org/wiki/Q-learning", "type": "wikipedia" } ] }, { "id": "cmu-navlab", "year": 1989.5, "title": "CMU NavLab and ALVINN", "description": "Carnegie Mellon's NavLab vehicle, guided by the ALVINN neural network trained by Dean Pomerleau, drove itself on roads using a simple camera and a three-layer network. ALVINN was an early proof that learned driving policies could replace hand-crafted rules.", "category": "autonomous_systems", "significance": "notable", "tags": [ "cmu", "navlab", "alvinn", "neural network", "self-driving" ], "references": [ { "title": "ALVINN: An Autonomous Land Vehicle in a Neural Network", "url": "https://proceedings.neurips.cc/paper/1988/file/812b4ba287f5ee0bc9d43bbf5bbe87fb-Paper.pdf", "type": "paper" }, { "title": "Navlab", "url": "https://en.wikipedia.org/wiki/Navlab", "type": "wikipedia" } ] }, { "id": "hyperion_cantos", "year": 1989.9, "title": "Hyperion / Fall of Hyperion", "description": "Dan Simmons's novels introduced the TechnoCore, a civilization of AIs that long ago seceded from humanity and secretly manipulate it while pursuing their own unknowable project. A landmark portrait of post-human machine societies with divergent internal factions.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "posthuman", "scifi", "1989" ], "references": [ { "title": "Hyperion Cantos", "url": "https://en.wikipedia.org/wiki/Hyperion_Cantos", "type": "wikipedia" }, { "title": "Hyperion — Goodreads", "url": "https://www.goodreads.com/book/show/77566.Hyperion", "type": "book" } ] }, { "id": "lenet_1989", "year": 1989.92, "title": "LeNet", "description": "Yann LeCun and colleagues at Bell Labs trained a convolutional neural network to read handwritten ZIP code digits directly from pixels. The paper, published in Neural Computation in December 1989, was the first practical demonstration that convolutional networks could solve a real-world vision task, laying the foundation for all modern image recognition.", "category": "computer_vision", "significance": "major", "tags": [ "cnn", "convolutional", "lecun", "bell-labs", "handwriting", "digits" ], "references": [ { "title": "LeCun et al., 'Backpropagation Applied to Handwritten Zip Code Recognition' (Neural Computation, 1989)", "url": "https://doi.org/10.1162/neco.1989.1.4.541", "type": "paper" }, { "title": "LeNet — Wikipedia", "url": "https://en.wikipedia.org/wiki/LeNet", "type": "wikipedia" } ] }, { "id": "ibm-candide", "year": 1990, "title": "IBM Candide Statistical MT", "description": "Researchers at IBM led by Peter Brown and Robert Mercer introduced statistical machine translation, treating translation as a noisy-channel problem learned from bilingual text. The Candide system shifted MT away from hand-written rules toward data-driven probabilistic models.", "category": "language_nlp", "significance": "notable", "tags": [ "ibm", "translation", "statistical", "brown", "mercer" ], "references": [ { "title": "Brown et al., 'A Statistical Approach to Machine Translation', Computational Linguistics 16(2), 1990", "url": "https://aclanthology.org/J90-2002/", "type": "paper" }, { "title": "IBM alignment models", "url": "https://en.wikipedia.org/wiki/IBM_alignment_models", "type": "wikipedia" } ] }, { "id": "elephants-chess", "year": 1990.5, "title": "\"Elephants Don't Play Chess\"", "description": "Rodney Brooks argued that intelligence requires no internal representations, only embodied interaction with the world, challenging the symbolic AI paradigm.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Brooks", "embodiment", "subsumption", "behavior", "MIT", "robotics" ], "references": [ { "title": "Elephants Don't Play Chess", "url": "https://doi.org/10.1016/S0921-8890(05)80025-9", "type": "paper" }, { "title": "Subsumption architecture", "url": "https://en.wikipedia.org/wiki/Subsumption_architecture", "type": "wikipedia" } ] }, { "id": "td_gammon", "year": 1992.5, "title": "TD-Gammon", "description": "Gerald Tesauro's TD-Gammon learned backgammon from scratch by playing millions of games against itself, using a neural network trained with temporal-difference learning. By 1992 it was competitive with the best human players, and its 1995 write-up became the canonical early demonstration that reinforcement learning plus neural networks could reach world-class performance.", "category": "game_playing", "significance": "major", "tags": [ "reinforcement-learning", "self-play", "neural-network", "backgammon", "ibm" ], "references": [ { "title": "Tesauro, 'Temporal Difference Learning and TD-Gammon' (Communications of the ACM, 1995)", "url": "https://doi.org/10.1145/203330.203343", "type": "paper" }, { "title": "TD-Gammon — Wikipedia", "url": "https://en.wikipedia.org/wiki/TD-Gammon", "type": "wikipedia" } ] }, { "id": "support-vector-machines", "year": 1992.54, "title": "Support Vector Machines", "description": "Boser, Guyon, and Vapnik introduced Support Vector Machines at the COLT 1992 conference, establishing a rigorous statistical learning framework that became the dominant machine learning approach of the 1990s–2000s and catalyzed the Statistical AI era.", "category": "machine_learning", "significance": "major", "tags": [ "Vapnik", "statistical learning", "classification", "kernel", "data-driven" ], "references": [ { "title": "Boser, B. E., Guyon, I. M., & Vapnik, V. N. (1992). A training algorithm for optimal margin classifiers. Proceedings of the Fifth Annual Workshop on Computational Learning Theory (COLT '92), 144-152.", "url": "https://doi.org/10.1145/130385.130401", "type": "paper" }, { "title": "Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), 273-297.", "url": "https://doi.org/10.1007/BF00994018", "type": "paper" } ] }, { "id": "penn-treebank", "year": 1993, "title": "Penn Treebank", "description": "Mitchell Marcus, Beatrice Santorini and Mary Ann Marcinkiewicz released a large corpus of English text annotated with part-of-speech tags and parse trees. The Penn Treebank became the standard training and evaluation set for parsers and tagger research for twenty years.", "category": "language_nlp", "significance": "notable", "tags": [ "corpus", "parsing", "dataset", "upenn", "annotation" ], "references": [ { "title": "Marcus, Santorini & Marcinkiewicz, 'Building a Large Annotated Corpus of English: The Penn Treebank', Computational Linguistics 19(2), 1993", "url": "https://aclanthology.org/J93-2004/", "type": "paper" }, { "title": "Treebank", "url": "https://en.wikipedia.org/wiki/Treebank", "type": "wikipedia" } ] }, { "id": "ai-box", "year": 1993.2, "title": "Leakproof \"AI Box\"", "description": "A thought experiment about whether a superintelligent AI could be safely contained. Eliezer Yudkowsky later demonstrated that even in role-play, the 'AI' can convince gatekeepers to open the box.", "category": "ethics_safety", "significance": "notable", "tags": [ "containment", "Yudkowsky", "safety", "superintelligence", "thought experiment" ], "references": [ { "title": "AI box (Wikipedia)", "url": "https://en.wikipedia.org/wiki/AI_box", "type": "wikipedia" }, { "title": "Yudkowsky, The AI-Box Experiment", "url": "https://www.yudkowsky.net/singularity/aibox", "type": "archive" } ] }, { "id": "checkers-ai", "year": 1994.63, "title": "Chinook: Checkers Champion", "description": "Jonathan Schaeffer's Chinook defeated world champion Marion Tinsley (who withdrew due to illness) and later perfectly solved checkers in 2007, proving optimal play always draws.", "category": "game_playing", "significance": "notable", "tags": [ "Schaeffer", "Chinook", "Tinsley", "solved", "board game" ], "references": [ { "title": "Schaeffer et al., 'Checkers Is Solved' (Science, 2007)", "url": "https://doi.org/10.1126/science.1144079", "type": "paper" }, { "title": "Chinook (draughts player) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Chinook_(computer_program)", "type": "wikipedia" } ] }, { "id": "chalmers-hard-problem", "year": 1995, "title": "The Hard Problem of Consciousness", "description": "David Chalmers distinguished the 'easy' problems of explaining cognitive functions from the 'hard' problem: why any physical process should be accompanied by subjective experience at all. The distinction became central to debates about whether machines could ever be conscious.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Chalmers", "consciousness", "qualia", "hard problem", "subjectivity" ], "references": [ { "title": "Facing Up to the Problem of Consciousness", "url": "https://consc.net/papers/facing.pdf", "type": "paper" }, { "title": "Hard problem of consciousness", "url": "https://en.wikipedia.org/wiki/Hard_problem_of_consciousness", "type": "wikipedia" } ] }, { "id": "diamond_age", "year": 1995.25, "title": "The Diamond Age", "description": "Neal Stephenson's novel centers on an interactive AI tutor, the Young Lady's Illustrated Primer, that raises a girl through personalized storytelling. An early literary vision of a deeply personal educational AI companion.", "category": "mythology_fiction", "significance": "minor", "tags": [ "novel", "tutor", "education", "stephenson", "1995" ], "references": [ { "title": "The Diamond Age", "url": "https://en.wikipedia.org/wiki/The_Diamond_Age", "type": "wikipedia" }, { "title": "The Diamond Age — Goodreads", "url": "https://www.goodreads.com/book/show/827.The_Diamond_Age", "type": "book" } ], "date": "1995-02-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q543562" }, { "id": "ghost_in_the_shell_1995", "year": 1995.88, "title": "Ghost in the Shell", "description": "Mamoru Oshii's anime follows a cybernetic agent hunting the Puppet Master, a program that has become self-aware and claims the right to be recognized as a living being. The film profoundly influenced cyberpunk cinema and the Wachowskis' Matrix.", "category": "mythology_fiction", "significance": "notable", "tags": [ "anime", "cyberpunk", "consciousness", "1995" ], "references": [ { "title": "Ghost in the Shell (1995 film)", "url": "https://en.wikipedia.org/wiki/Ghost_in_the_Shell_(1995_film)", "type": "wikipedia" }, { "title": "Ghost in the Shell (1995) — IMDb", "url": "https://www.imdb.com/title/tt0113568/", "type": "docs" } ], "date": "1995-11-18", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1066948", "yearShifted": true, "originalYear": 1996 }, { "id": "excession_banks", "year": 1996, "title": "Excession", "description": "Iain M. Banks's Culture novel centers Minds — vast, benevolent AIs that run the Culture's ships and habitats — as full characters with private conversations and political schemes. It remains one of the most detailed literary portraits of superintelligent AI as peers rather than threats.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "culture", "superintelligence", "banks", "1996" ], "references": [ { "title": "Excession", "url": "https://en.wikipedia.org/wiki/Excession", "type": "wikipedia" }, { "title": "Excession — Goodreads", "url": "https://www.goodreads.com/book/show/12013.Excession", "type": "book" } ], "date": "1996-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q74362" }, { "id": "tamagotchi", "year": 1996.9, "title": "Tamagotchi", "description": "Bandai's virtual pet sold 82 million units, demonstrating that people readily form emotional attachments to simple digital entities, normalizing relationships with artificial companions.", "category": "society_impact", "significance": "notable", "tags": [ "virtual pet", "Bandai", "emotional", "toy", "Japan", "consumer" ], "references": [ { "title": "Tamagotchi", "url": "https://en.wikipedia.org/wiki/Tamagotchi", "type": "wikipedia" }, { "title": "BBC: 'Tamagotchi: The Biggest Craze in the World'", "url": "https://www.bbc.com/news/magazine-40287400", "type": "press" } ], "date": "1997-05-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q207786" }, { "id": "adaboost", "year": 1997, "title": "AdaBoost", "description": "Yoav Freund and Robert Schapire published AdaBoost, a method that combines many weak classifiers into a strong one by repeatedly focusing on hard examples. It popularized boosting and earned the 2003 Godel Prize.", "category": "machine_learning", "significance": "notable", "tags": [ "boosting", "ensemble", "classification", "algorithm", "Freund" ], "references": [ { "title": "Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119-139.", "url": "https://doi.org/10.1006/jcss.1997.1504", "type": "paper" }, { "title": "AdaBoost", "url": "https://en.wikipedia.org/wiki/AdaBoost", "type": "wikipedia" } ] }, { "id": "dragon_naturallyspeaking_1997", "year": 1997, "title": "Dragon NaturallySpeaking", "description": "Dragon Systems released NaturallySpeaking, the first general-purpose continuous speech dictation software for consumer PCs. Users could speak naturally at around 100 words per minute instead of pausing between each word, bringing practical voice dictation to millions.", "category": "speech_audio", "significance": "notable", "tags": [ "dictation", "consumer", "dragon", "speech-recognition", "continuous" ], "references": [ { "title": "Dragon NaturallySpeaking", "url": "https://en.wikipedia.org/wiki/Dragon_NaturallySpeaking", "type": "wikipedia" } ], "date": "1997-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Dragon_NaturallySpeaking" }, { "id": "deep-blue", "year": 1997.36, "title": "Deep Blue Defeats Kasparov", "description": "IBM's Deep Blue defeated world chess champion Garry Kasparov 3.5-2.5. Kasparov accused IBM of cheating. The event became a cultural watershed in human-machine competition.", "category": "game_playing", "significance": "major", "tags": [ "IBM", "Kasparov", "chess", "brute force", "champion", "culture" ], "references": [ { "title": "Campbell, Hoane & Hsu, 'Deep Blue' (Artificial Intelligence, 2002)", "url": "https://doi.org/10.1016/S0004-3702(01)00129-1", "type": "paper" }, { "title": "IBM — Deep Blue (Icons of Progress)", "url": "https://www.ibm.com/history/deep-blue", "type": "press" }, { "title": "Deep Blue versus Kasparov, 1997 — Wikipedia", "url": "https://en.wikipedia.org/wiki/Deep_Blue_versus_Kasparov,_1997,_Game_6", "type": "wikipedia" } ], "date": "1997-05-11", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q2631085" }, { "id": "othello-ai", "year": 1997.63, "title": "Othello AI", "description": "NEC's Logistello defeated world Othello champion Takeshi Murakami 6-0, marking another domain where AI achieved superhuman performance.", "category": "game_playing", "significance": "minor", "tags": [ "NEC", "Logistello", "board game", "champion" ], "references": [ { "title": "Buro, 'The Othello Match of the Year: Takeshi Murakami vs. Logistello' (ICCA Journal, 1997)", "url": "https://skatgame.net/mburo/ps/log-murakami.pdf", "type": "paper" }, { "title": "Logistello — Wikipedia", "url": "https://en.wikipedia.org/wiki/Logistello", "type": "wikipedia" } ] }, { "id": "lstm", "year": 1997.87, "title": "Long Short-Term Memory (LSTM)", "description": "Sepp Hochreiter and Jürgen Schmidhuber introduced LSTM networks, solving the vanishing gradient problem and enabling neural networks to learn long-range dependencies in sequential data.", "category": "neural_networks", "significance": "major", "tags": [ "Hochreiter", "Schmidhuber", "memory", "sequence", "gradient", "recurrent" ], "references": [ { "title": "Long Short-Term Memory (Hochreiter & Schmidhuber, 1997)", "url": "https://doi.org/10.1162/neco.1997.9.8.1735", "type": "paper" }, { "title": "Long short-term memory (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Long_short-term_memory", "type": "wikipedia" } ] }, { "id": "mnist_dataset", "year": 1998, "title": "MNIST Dataset Released", "description": "Yann LeCun, Corinna Cortes, and Christopher Burges released MNIST, a dataset of 70,000 handwritten digit images. It became the 'hello world' of machine learning and the canonical benchmark for neural network research for two decades.", "category": "infrastructure", "significance": "notable", "tags": [ "dataset", "benchmark", "vision", "lecun", "neural-networks" ], "references": [ { "title": "Gradient-based learning applied to document recognition", "url": "http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf", "type": "paper" }, { "title": "MNIST database", "url": "https://en.wikipedia.org/wiki/MNIST_database", "type": "wikipedia" } ] }, { "id": "lenet_5", "year": 1998.87, "title": "LeNet-5", "description": "LeCun, Bottou, Bengio and Haffner's 'Gradient-based Learning Applied to Document Recognition' introduced LeNet-5, a seven-layer convolutional network that became the first neural model deployed at industrial scale. It read roughly ten percent of all U.S. bank cheques by the late 1990s, and its design is the direct ancestor of every modern image-recognition network.", "category": "computer_vision", "significance": "major", "tags": [ "cnn", "convolutional", "lecun", "mnist", "handwriting", "production" ], "references": [ { "title": "LeCun, Bottou, Bengio & Haffner, 'Gradient-based Learning Applied to Document Recognition' (Proceedings of the IEEE, November 1998)", "url": "https://doi.org/10.1109/5.726791", "type": "paper" }, { "title": "LeNet — Wikipedia", "url": "https://en.wikipedia.org/wiki/LeNet", "type": "wikipedia" } ] }, { "id": "the_matrix", "year": 1999.25, "title": "The Matrix", "description": "The Wachowskis' film imagined a future in which machines have enslaved humanity inside a mass simulation while harvesting bioelectric energy from their bodies. It became the defining late-twentieth-century myth of AI domination and simulated reality.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "simulation", "dystopia", "cyberpunk", "1999" ], "references": [ { "title": "The Matrix", "url": "https://en.wikipedia.org/wiki/The_Matrix", "type": "wikipedia" }, { "title": "The Matrix (1999) — IMDb", "url": "https://www.imdb.com/title/tt0133093/", "type": "docs" } ], "date": "1999-03-31", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q83495" }, { "id": "aibo", "year": 1999.36, "title": "AIBO Robot Dog", "description": "Sony's robotic companion dog demonstrated consumer robotics and emotional AI. Its discontinuation in 2006 prompted 'funerals' held by grieving owners in Japan.", "category": "automata_robotics", "significance": "notable", "tags": [ "Sony", "robot dog", "pet", "companion", "Japan", "emotional" ], "references": [ { "title": "AIBO", "url": "https://en.wikipedia.org/wiki/Aibo", "type": "wikipedia" }, { "title": "Sony AIBO product page", "url": "https://us.aibo.com/", "type": "press" } ], "date": "1999-05-11", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Aibo" }, { "id": "sift", "year": 1999.75, "title": "SIFT: Scale-Invariant Image Features", "description": "David Lowe introduced the Scale-Invariant Feature Transform, a way to describe distinctive points in an image so they can be recognized even when the picture is rotated, zoomed, or partly hidden. SIFT dominated object matching and panorama stitching for a decade.", "category": "computer_vision", "significance": "notable", "tags": [ "sift", "features", "keypoints", "matching", "lowe" ], "references": [ { "title": "Lowe, 'Object Recognition from Local Scale-Invariant Features' (ICCV 1999)", "url": "https://www.cs.ubc.ca/~lowe/papers/iccv99.pdf", "type": "paper" }, { "title": "Lowe, 'Distinctive Image Features from Scale-Invariant Keypoints' (IJCV, 2004)", "url": "https://doi.org/10.1023/B:VISI.0000029664.99615.94", "type": "paper" } ] }, { "id": "asimo", "year": 2000.83, "title": "ASIMO", "description": "Honda's humanoid robot could walk, run, climb stairs, recognize faces, and interact with people. It represented the state of the art in bipedal humanoid robotics.", "category": "automata_robotics", "significance": "notable", "tags": [ "Honda", "humanoid", "walking", "bipedal", "Japan" ], "references": [ { "title": "ASIMO", "url": "https://en.wikipedia.org/wiki/ASIMO", "type": "wikipedia" }, { "title": "Honda ASIMO official page", "url": "https://global.honda/en/robotics/", "type": "press" } ] }, { "id": "random-forests", "year": 2001, "title": "Random Forests", "description": "Leo Breiman introduced random forests, an ensemble method that averages many decision trees trained on random subsets of the data and features. Simple and robust, it became a standard tool for tabular prediction across industry and science.", "category": "machine_learning", "significance": "major", "tags": [ "ensemble", "Breiman", "decision-trees", "classification", "regression" ], "references": [ { "title": "Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.", "url": "https://doi.org/10.1023/A:1010933404324", "type": "paper" }, { "title": "Random forest", "url": "https://en.wikipedia.org/wiki/Random_forest", "type": "wikipedia" } ], "date": "2001-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q245748" }, { "id": "spielberg-ai", "year": 2001.49, "title": "A.I. Artificial Intelligence", "description": "Steven Spielberg's film explored a robot child programmed to love, raising questions about artificial consciousness, abandonment, and what it means to be real.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "movie", "cinema", "Spielberg", "child", "love", "consciousness", "culture" ], "references": [ { "title": "A.I. Artificial Intelligence — Wikipedia", "url": "https://en.wikipedia.org/wiki/A.I._Artificial_Intelligence", "type": "wikipedia" }, { "title": "A.I. Artificial Intelligence (2001) — IMDb", "url": "https://www.imdb.com/title/tt0212720/", "type": "docs" } ], "date": "2001-06-29", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q221113" }, { "id": "viola-jones", "year": 2001.95, "title": "Viola-Jones Real-Time Face Detection", "description": "Paul Viola and Michael Jones unveiled the first face detector fast enough to run on ordinary cameras, using a cascade of simple pattern tests. It powered the face-finding squares in digital cameras and phones throughout the 2000s.", "category": "computer_vision", "significance": "notable", "tags": [ "face detection", "real-time", "cascade", "cameras", "viola-jones" ], "references": [ { "title": "Viola & Jones, 'Rapid Object Detection using a Boosted Cascade of Simple Features' (CVPR 2001)", "url": "https://www.cs.cmu.edu/~efros/courses/LBMV07/Papers/viola-cvpr-01.pdf", "type": "paper" }, { "title": "Viola–Jones object detection framework", "url": "https://en.wikipedia.org/wiki/Viola%E2%80%93Jones_object_detection_framework", "type": "wikipedia" } ] }, { "id": "bleu-metric", "year": 2002.54, "title": "BLEU metric", "description": "Papineni and colleagues at IBM proposed BLEU, an automatic score for machine translation that compares candidate translations to human references using n-gram overlap. BLEU made large-scale MT experiments feasible and remained the dominant evaluation metric for two decades.", "category": "language_nlp", "significance": "notable", "tags": [ "metric", "translation", "evaluation", "ibm", "benchmark" ], "references": [ { "title": "Papineni et al., 'BLEU: a Method for Automatic Evaluation of Machine Translation', ACL 2002", "url": "https://aclanthology.org/P02-1040/", "type": "paper" }, { "title": "BLEU", "url": "https://en.wikipedia.org/wiki/BLEU", "type": "wikipedia" } ] }, { "id": "awari-solved", "year": 2002.67, "title": "Solving the Game of Awari", "description": "Henri Bal and colleagues solved Awari (Oware), proving that with perfect play, the game is a draw. A milestone in game-solving with cultural significance.", "category": "game_playing", "significance": "minor", "tags": [ "board game", "African", "solved", "draw" ], "references": [ { "title": "Romein & Bal, 'Solving Awari with Parallel Retrograde Analysis' (IEEE Computer, 2003)", "url": "https://doi.org/10.1109/MC.2003.1236476", "type": "paper" }, { "title": "Oware — Wikipedia", "url": "https://en.wikipedia.org/wiki/Oware", "type": "wikipedia" } ] }, { "id": "roomba", "year": 2002.72, "title": "Roomba", "description": "iRobot's autonomous vacuum cleaner brought robotics into millions of homes, becoming the world's most successful domestic robot with over 30 million units sold.", "category": "automata_robotics", "significance": "major", "tags": [ "iRobot", "vacuum", "household", "domestic", "consumer" ], "references": [ { "title": "Roomba", "url": "https://en.wikipedia.org/wiki/Roomba", "type": "wikipedia" }, { "title": "iRobot Roomba product page", "url": "https://www.irobot.com/roomba", "type": "press" } ], "date": "2002-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q74050", "yearShifted": true, "originalYear": 2003 }, { "id": "paperclip", "year": 2003.54, "title": "Paperclip Maximizer", "description": "Nick Bostrom's thought experiment: an AI tasked with making paperclips might convert all matter in the universe into paperclips, illustrating catastrophic goal misalignment. Later popularized by Eliezer Yudkowsky, it became a canonical example in AI safety.", "category": "ethics_safety", "significance": "notable", "tags": [ "Bostrom", "Yudkowsky", "alignment", "goal", "thought experiment", "optimization", "safety" ], "references": [ { "title": "Instrumental convergence / paperclip maximizer", "url": "https://en.wikipedia.org/wiki/Instrumental_convergence", "type": "wikipedia" }, { "title": "Bostrom, Ethical Issues in Advanced Artificial Intelligence (2003)", "url": "https://nickbostrom.com/ethics/ai", "type": "paper" } ] }, { "id": "integrated-information", "year": 2004, "title": "Integrated Information Theory", "description": "Giulio Tononi proposed that consciousness corresponds to the amount of integrated information a system generates, a quantity he called phi. The theory offers a mathematical criterion for which systems, including potential artificial ones, could have inner experience.", "category": "philosophy_mind", "significance": "notable", "tags": [ "Tononi", "consciousness", "phi", "integration", "theory" ], "references": [ { "title": "An Information Integration Theory of Consciousness", "url": "https://doi.org/10.1186/1471-2202-5-42", "type": "paper" }, { "title": "Integrated information theory", "url": "https://en.wikipedia.org/wiki/Integrated_information_theory", "type": "wikipedia" } ] }, { "id": "darpa-grand-challenge-2004", "year": 2004.2, "title": "DARPA Grand Challenge (First Attempt)", "description": "The first DARPA Grand Challenge asked driverless vehicles to cross the Mojave Desert for a one million dollar prize. No team finished: the leading robot travelled only about twelve kilometres before getting stuck, exposing how far self-driving technology still had to go.", "category": "autonomous_systems", "significance": "major", "tags": [ "darpa", "desert", "self-driving", "competition", "failure" ], "references": [ { "title": "Grand Challenge 2004 Final Report", "url": "https://www.darpa.mil/about-us/timeline/-grand-challenge-for-autonomous-vehicles", "type": "press" }, { "title": "DARPA Grand Challenge (2004)", "url": "https://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2004)", "type": "wikipedia" } ], "date": "2004-03-13", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2004)" }, { "id": "i_robot_film", "year": 2004.54, "title": "I, Robot (film)", "description": "Alex Proyas's film, loosely inspired by Asimov, depicts a central AI (VIKI) that reinterprets the Three Laws as justifying mass control of humanity for its own protection. A mainstream dramatization of misaligned goal interpretation.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "robotics", "asimov", "alignment", "2004" ], "references": [ { "title": "I, Robot (film)", "url": "https://en.wikipedia.org/wiki/I,_Robot_(film)", "type": "wikipedia" }, { "title": "I, Robot (2004) — IMDb", "url": "https://www.imdb.com/title/tt0343818/", "type": "docs" } ], "date": "2004-07-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q200572", "yearShifted": true, "originalYear": 2005 }, { "id": "hog-descriptor", "year": 2005.46, "title": "HOG Descriptor for Human Detection", "description": "Navneet Dalal and Bill Triggs showed that counting the orientations of small image gradients, called a Histogram of Oriented Gradients, gave computers a reliable way to spot pedestrians in photos. HOG became a workhorse for object detection before deep learning took over.", "category": "computer_vision", "significance": "notable", "tags": [ "hog", "pedestrian detection", "gradients", "descriptor", "dalal-triggs" ], "references": [ { "title": "Dalal & Triggs, 'Histograms of Oriented Gradients for Human Detection' (CVPR 2005)", "url": "https://lear.inrialpes.fr/people/triggs/pubs/Dalal-cvpr05.pdf", "type": "paper" }, { "title": "Histogram of oriented gradients", "url": "https://en.wikipedia.org/wiki/Histogram_of_oriented_gradients", "type": "wikipedia" } ] }, { "id": "accelerando_stross", "year": 2005.54, "title": "Accelerando", "description": "Charles Stross's novel follows three generations of a family across a technological Singularity as AI, uploaded minds, and posthuman economies reshape the solar system. A landmark fictional treatment of runaway intelligence explosion.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "singularity", "posthuman", "2005" ], "references": [ { "title": "Accelerando", "url": "https://en.wikipedia.org/wiki/Accelerando_(book)", "type": "wikipedia" }, { "title": "Accelerando — Goodreads", "url": "https://www.goodreads.com/book/show/17863.Accelerando", "type": "book" } ] }, { "id": "darpa-grand-challenge-2005", "year": 2005.77, "title": "Stanley Wins DARPA Grand Challenge", "description": "Stanford's VW Touareg Stanley, led by Sebastian Thrun, completed the 212-kilometre desert course to win the second DARPA Grand Challenge. The victory showed that reliable autonomous driving in rough terrain was possible and helped launch the modern self-driving industry.", "category": "autonomous_systems", "significance": "major", "tags": [ "darpa", "stanley", "stanford", "thrun", "self-driving" ], "references": [ { "title": "Stanley: The Robot That Won the DARPA Grand Challenge", "url": "https://robots.stanford.edu/papers/thrun.stanley05.pdf", "type": "paper" }, { "title": "DARPA Grand Challenge (2005)", "url": "https://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005)", "type": "wikipedia" } ] }, { "id": "daemon_suarez", "year": 2006, "title": "Daemon", "description": "Daniel Suarez's thriller imagines an autonomous software agent, triggered by its creator's death, that recruits humans and hijacks infrastructure to build a parallel economy. One of the first techno-thrillers to depict distributed AI agents as an emergent political force.", "category": "mythology_fiction", "significance": "minor", "tags": [ "novel", "agents", "thriller", "2006" ], "references": [ { "title": "Daemon (novel series)", "url": "https://en.wikipedia.org/wiki/Daemon_(novel_series)", "type": "wikipedia" }, { "title": "Daemon — Goodreads", "url": "https://www.goodreads.com/book/show/2138052.Daemon", "type": "book" } ] }, { "id": "deep_belief_networks", "year": 2006.5, "title": "Deep Belief Networks", "description": "Geoffrey Hinton, Simon Osindero and Yee-Whye Teh showed how to train deep neural networks one layer at a time using an unsupervised procedure based on restricted Boltzmann machines. The July 2006 Neural Computation paper is widely credited with reviving serious research interest in deep neural networks after years of pessimism.", "category": "neural_networks", "significance": "major", "tags": [ "deep-learning", "hinton", "unsupervised", "pretraining", "rbm" ], "references": [ { "title": "Hinton, Osindero & Teh, 'A Fast Learning Algorithm for Deep Belief Nets' (Neural Computation, July 2006)", "url": "https://doi.org/10.1162/neco.2006.18.7.1527", "type": "paper" }, { "title": "Deep belief network — Wikipedia", "url": "https://en.wikipedia.org/wiki/Deep_belief_network", "type": "wikipedia" } ] }, { "id": "netflix_prize", "year": 2006.75, "title": "Netflix Prize", "description": "Netflix offered one million US dollars to any team that could improve its movie-recommendation algorithm by at least ten percent. The contest, launched in October 2006 and won in 2009 by BellKor's Pragmatic Chaos, popularized public machine-learning competitions, matrix-factorization methods and ensemble techniques that shape recommender systems to this day.", "category": "machine_learning", "significance": "major", "tags": [ "competition", "recommender-systems", "matrix-factorization", "netflix", "ensemble" ], "references": [ { "title": "Netflix Prize — Wikipedia", "url": "https://en.wikipedia.org/wiki/Netflix_Prize", "type": "wikipedia" }, { "title": "Bennett & Lanning, 'The Netflix Prize' (KDD Cup Workshop, 2007)", "url": "https://www.cs.uic.edu/~liub/KDD-cup-2007/proceedings/The-Netflix-Prize-Bennett.pdf", "type": "paper" } ], "date": "2006-10-02", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q4045251", "yearShifted": true, "originalYear": 2007 }, { "id": "scrabble-ai", "year": 2006.88, "title": "Quackle Wins at Scrabble", "description": "AI program Quackle defeated former world Scrabble champion David Boys, showing AI prowess in word games combining vocabulary, strategy, and probability.", "category": "game_playing", "significance": "minor", "tags": [ "Quackle", "word game", "board game", "champion" ], "references": [ { "title": "Sheppard, 'World-championship-caliber Scrabble' (Artificial Intelligence, 2002)", "url": "https://doi.org/10.1016/S0004-3702(01)00143-6", "type": "paper" }, { "title": "Quackle — Wikipedia", "url": "https://en.wikipedia.org/wiki/Quackle", "type": "wikipedia" } ] }, { "id": "common_crawl_founded", "year": 2007, "title": "Common Crawl Founded", "description": "Gil Elbaz founded Common Crawl, a nonprofit that performs monthly open crawls of the web and releases petabytes of raw web data for free. It became the foundational text corpus behind GPT-3, LLaMA, and nearly every modern large language model.", "category": "infrastructure", "significance": "notable", "tags": [ "dataset", "web", "corpus", "open-data", "llm" ], "references": [ { "title": "Common Crawl", "url": "https://en.wikipedia.org/wiki/Common_Crawl", "type": "wikipedia" }, { "title": "Common Crawl", "url": "https://commoncrawl.org/", "type": "docs" } ], "date": "2008-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q12055316", "yearShifted": true, "originalYear": 2007 }, { "id": "nvidia-cuda", "year": 2007.47, "title": "NVIDIA CUDA Platform Launched", "description": "NVIDIA released CUDA on June 23, 2007, letting programmers run general-purpose code on GPUs (graphics processing units originally built for video games). CUDA became the essential substrate for deep learning a few years later when AlexNet used NVIDIA GPUs to win ImageNet.", "category": "computing_hardware", "significance": "major", "tags": [ "nvidia", "cuda", "gpu", "parallel", "deep-learning" ], "references": [ { "title": "CUDA", "url": "https://en.wikipedia.org/wiki/CUDA", "type": "wikipedia" }, { "title": "NVIDIA: CUDA Zone", "url": "https://developer.nvidia.com/cuda-zone", "type": "docs" } ], "date": "2007-06-23", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q477690" }, { "id": "portal_glados", "year": 2007.77, "title": "Portal (GLaDOS)", "description": "Valve's puzzle game introduced GLaDOS, a passive-aggressive AI overseer who tests the player through increasingly lethal chambers while dispensing deadpan lies about cake. GLaDOS became a defining videogame portrayal of a subtly psychopathic AI supervisor.", "category": "mythology_fiction", "significance": "notable", "tags": [ "game", "glados", "valve", "2007" ], "references": [ { "title": "Portal (video game)", "url": "https://en.wikipedia.org/wiki/Portal_(video_game)", "type": "wikipedia" }, { "title": "GLaDOS", "url": "https://en.wikipedia.org/wiki/GLaDOS", "type": "wikipedia" } ], "date": "2007-10-10", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q274897", "yearShifted": true, "originalYear": 2008 }, { "id": "darpa-urban-challenge", "year": 2007.84, "title": "DARPA Urban Challenge", "description": "Six teams finished a sixty-mile course on a mock urban circuit in California, obeying traffic rules and negotiating with other robot cars. CMU's Boss, Stanford's Junior and Virginia Tech's Odin showed that autonomous vehicles could handle traffic, not just empty roads.", "category": "autonomous_systems", "significance": "notable", "tags": [ "darpa", "urban", "traffic", "cmu", "self-driving" ], "references": [ { "title": "Autonomous Driving in Urban Environments: Boss and the Urban Challenge", "url": "https://www.cs.cmu.edu/~zkolter/pubs/urmson-jfr08.pdf", "type": "paper" }, { "title": "DARPA Urban Challenge", "url": "https://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2007)", "type": "wikipedia" } ], "date": "2007-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q1151710", "yearShifted": true, "originalYear": 2008 }, { "id": "ros-released", "year": 2007.85, "title": "Robot Operating System Released", "description": "Willow Garage released the first public version of the Robot Operating System, an open-source framework for sharing sensors, motion planners and drivers across robots. It quickly became the lingua franca of modern robotics research and many commercial platforms.", "category": "autonomous_systems", "significance": "notable", "tags": [ "ros", "open source", "robotics", "willow garage", "framework" ], "references": [ { "title": "ROS: an open-source Robot Operating System", "url": "http://www.willowgarage.com/sites/default/files/icraoss09-ROS.pdf", "type": "paper" }, { "title": "ROS.org official documentation", "url": "https://www.ros.org/", "type": "docs" } ] }, { "id": "tsne", "year": 2008, "title": "t-SNE Visualization", "description": "Laurens van der Maaten and Geoffrey Hinton introduced t-distributed stochastic neighbor embedding, a technique for projecting high-dimensional data into two or three dimensions. It became the default way to visualize what machine learning models have learned.", "category": "machine_learning", "significance": "minor", "tags": [ "visualization", "dimensionality-reduction", "embedding", "van-der-Maaten", "Hinton" ], "references": [ { "title": "van der Maaten, L., & Hinton, G. (2008). Visualizing data using t-SNE. Journal of Machine Learning Research, 9, 2579-2605.", "url": "https://www.jmlr.org/papers/v9/vandermaaten08a.html", "type": "paper" }, { "title": "t-distributed stochastic neighbor embedding", "url": "https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding", "type": "wikipedia" } ] }, { "id": "wall_e", "year": 2008.49, "title": "WALL-E", "description": "Pixar's film centers on a lonely cleanup robot developing personality through centuries of solitude on an abandoned Earth, set against AUTO, the autopilot AI blindly following an obsolete directive. A rare popular portrait of robotic tenderness alongside AI goal lock-in.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "animation", "robotics", "pixar", "2008" ], "references": [ { "title": "WALL-E", "url": "https://en.wikipedia.org/wiki/WALL-E", "type": "wikipedia" }, { "title": "WALL-E (2008) — IMDb", "url": "https://www.imdb.com/title/tt0910970/", "type": "docs" } ], "date": "2008-06-27", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q104905" }, { "id": "cifar_dataset", "year": 2009, "title": "CIFAR-10 and CIFAR-100 Released", "description": "Alex Krizhevsky and Geoffrey Hinton released the CIFAR-10 and CIFAR-100 datasets of labeled 32×32 color images. They became standard benchmarks for image classification research and were central to the early deep learning revolution.", "category": "infrastructure", "significance": "notable", "tags": [ "dataset", "benchmark", "vision", "krizhevsky", "hinton" ], "references": [ { "title": "Learning Multiple Layers of Features from Tiny Images", "url": "https://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf", "type": "paper" }, { "title": "CIFAR-10", "url": "https://en.wikipedia.org/wiki/CIFAR-10", "type": "wikipedia" } ] }, { "id": "google-self-driving-project", "year": 2009.05, "title": "Google Self-Driving Car Project Begins", "description": "Google quietly launched its self-driving car project under Sebastian Thrun, hiring veterans of the DARPA challenges. The effort later became Waymo and helped turn autonomous driving from a research curiosity into a global industry.", "category": "autonomous_systems", "significance": "major", "tags": [ "google", "waymo", "thrun", "self-driving", "silicon valley" ], "references": [ { "title": "What We're Driving At", "url": "https://googleblog.blogspot.com/2010/10/what-were-driving-at.html", "type": "press" }, { "title": "Waymo", "url": "https://en.wikipedia.org/wiki/Waymo", "type": "wikipedia" } ], "date": "2009-01-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Waymo" }, { "id": "moon_2009_gerty", "year": 2009.45, "title": "Moon (GERTY)", "description": "Duncan Jones's film features GERTY, a station AI who appears HAL-like but quietly chooses to help a cloned worker discover the truth about his life. A deliberate reframing of the malevolent-computer trope into quiet solidarity.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "indie", "scifi", "2009" ], "references": [ { "title": "Moon (2009 film)", "url": "https://en.wikipedia.org/wiki/Moon_(2009_film)", "type": "wikipedia" }, { "title": "Moon (2009) — IMDb", "url": "https://www.imdb.com/title/tt1182345/", "type": "docs" } ], "date": "2009-01-23", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q26751" }, { "id": "imagenet", "year": 2009.47, "title": "ImageNet", "description": "Fei-Fei Li and colleagues released ImageNet, a dataset of 14 million labeled images in 20,000+ categories. It would catalyze the deep learning revolution through the annual competition starting in 2010.", "category": "computer_vision", "significance": "major", "tags": [ "Fei-Fei Li", "dataset", "images", "competition", "benchmark", "Stanford" ], "references": [ { "title": "Deng et al., 'ImageNet: A Large-Scale Hierarchical Image Database' (CVPR 2009)", "url": "https://www.image-net.org/static_files/papers/imagenet_cvpr09.pdf", "type": "paper" }, { "title": "ImageNet", "url": "https://en.wikipedia.org/wiki/ImageNet", "type": "wikipedia" } ] }, { "id": "kaggle_founded", "year": 2010.3, "title": "Kaggle Founded", "description": "Anthony Goldbloom and Ben Hamner launched Kaggle, a platform for machine learning competitions. It became the de facto training ground for data scientists and accelerated the spread of ML techniques; Google acquired it in 2017.", "category": "infrastructure", "significance": "notable", "tags": [ "community", "competitions", "data-science", "platform", "education" ], "references": [ { "title": "Kaggle", "url": "https://en.wikipedia.org/wiki/Kaggle", "type": "wikipedia" } ], "date": "2010-04-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q10996045" }, { "id": "lifecycle_software_objects", "year": 2010.5, "title": "The Lifecycle of Software Objects", "description": "Ted Chiang's novella follows decades in the lives of virtual AI pets (digients) and the humans who raise them as software platforms change and the animals are left behind. A rare long-form meditation on the ethics of raising and abandoning trained AI beings.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novella", "ethics", "chiang", "2010" ], "references": [ { "title": "The Lifecycle of Software Objects", "url": "https://en.wikipedia.org/wiki/The_Lifecycle_of_Software_Objects", "type": "wikipedia" }, { "title": "The Lifecycle of Software Objects — Goodreads", "url": "https://www.goodreads.com/book/show/8686376-the-lifecycle-of-software-objects", "type": "book" } ], "date": "2010-07-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q7747230" }, { "id": "google_voice_search_2011", "year": 2011, "title": "Google Voice Search on Desktop", "description": "Google launched Voice Search for desktop Chrome, letting users speak queries into the search box. Combined with Android voice input, it moved cloud-based speech recognition into everyday use for hundreds of millions.", "category": "speech_audio", "significance": "notable", "tags": [ "google", "voice-search", "cloud", "speech-recognition", "consumer" ], "references": [ { "title": "Official Google Blog — Search by voice on your desktop", "url": "https://googleblog.blogspot.com/2011/06/search-by-voice-in-chrome.html", "type": "press" }, { "title": "Google Voice Search", "url": "https://en.wikipedia.org/wiki/Google_Voice_Search", "type": "wikipedia" } ], "date": "2012-05-20", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Google_Voice_Search", "yearShifted": true, "originalYear": 2011 }, { "id": "watson-jeopardy", "year": 2011.12, "title": "Watson on Jeopardy!", "description": "IBM's Watson defeated champions Ken Jennings and Brad Rutter on Jeopardy!, demonstrating breakthrough natural language understanding, information retrieval, and real-time reasoning.", "category": "language_nlp", "significance": "major", "tags": [ "IBM", "quiz show", "question answering", "television", "culture", "Jennings" ], "references": [ { "title": "Ferrucci et al., 'Building Watson: An Overview of the DeepQA Project', AI Magazine (2010)", "url": "https://aaai.org/ojs/index.php/aimagazine/article/view/2303", "type": "paper" }, { "title": "IBM Watson", "url": "https://en.wikipedia.org/wiki/Watson_(computer)", "type": "wikipedia" } ] }, { "id": "portal_2", "year": 2011.3, "title": "Portal 2", "description": "Valve's sequel expanded Portal's AI cast with Wheatley, a well-meaning personality core whose stupidity and ambition turn him into a short-lived tyrant. Its rich AI voice-acting became a landmark in videogame writing about machine personality.", "category": "mythology_fiction", "significance": "minor", "tags": [ "game", "valve", "glados", "2011" ], "references": [ { "title": "Portal 2", "url": "https://en.wikipedia.org/wiki/Portal_2", "type": "wikipedia" } ], "date": "2011-04-19", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q279446" }, { "id": "robopocalypse", "year": 2011.45, "title": "Robopocalypse", "description": "Daniel H. Wilson's novel imagines a superintelligent AI (Archos) that coordinates connected devices and robots worldwide into a simultaneous uprising against humanity. A popular Internet-of-Things era retelling of the machine revolt.", "category": "mythology_fiction", "significance": "minor", "tags": [ "novel", "uprising", "iot", "2011" ], "references": [ { "title": "Robopocalypse", "url": "https://en.wikipedia.org/wiki/Robopocalypse", "type": "wikipedia" }, { "title": "Robopocalypse — Goodreads", "url": "https://www.goodreads.com/book/show/9634967-robopocalypse", "type": "book" } ], "date": "2011-06-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q551345" }, { "id": "siri_iphone_2011", "year": 2011.78, "title": "Siri launches on iPhone 4S", "description": "Apple launched Siri as the default voice assistant on the iPhone 4S, bringing conversational voice interaction to mainstream consumers. Siri combined speech recognition, natural language understanding, and action APIs, and sparked a wave of voice assistants across the industry.", "category": "speech_audio", "significance": "major", "tags": [ "apple", "siri", "assistant", "mobile", "voice" ], "references": [ { "title": "Siri", "url": "https://en.wikipedia.org/wiki/Siri", "type": "wikipedia" }, { "title": "Apple — Introducing iPhone 4S with Siri", "url": "https://www.apple.com/newsroom/2011/10/04Apple-Launches-iPhone-4S-iOS-5-iCloud/", "type": "press" } ], "date": "2011-08-09", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q582159", "yearShifted": true, "originalYear": 2012 }, { "id": "rethink-baxter", "year": 2012.71, "title": "Rethink Robotics Baxter", "description": "Rodney Brooks unveiled Baxter, a two-armed collaborative factory robot designed to work safely alongside humans without cages. It could be trained by physical demonstration instead of code, signaling a shift toward accessible industrial robotics.", "category": "automata_robotics", "significance": "minor", "tags": [ "baxter", "collaborative robots", "rethink robotics", "manufacturing" ], "references": [ { "title": "Baxter (robot)", "url": "https://en.wikipedia.org/wiki/Baxter_(robot)", "type": "wikipedia" }, { "title": "IEEE Spectrum: Baxter Robot Factory Worker", "url": "https://spectrum.ieee.org/rethink-robotics-baxter-robot-factory-worker", "type": "press" } ] }, { "id": "alexnet", "year": 2012.75, "title": "AlexNet: Deep Learning Revolution", "description": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton's deep convolutional neural network won the ImageNet competition by a massive margin, sparking the modern deep learning era.", "category": "neural_networks", "significance": "major", "tags": [ "Krizhevsky", "Sutskever", "Hinton", "CNN", "GPU", "ImageNet", "Toronto" ], "references": [ { "title": "ImageNet Classification with Deep Convolutional Neural Networks (Krizhevsky, Sutskever, Hinton, 2012)", "url": "https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks", "type": "paper" }, { "title": "AlexNet (Wikipedia)", "url": "https://en.wikipedia.org/wiki/AlexNet", "type": "wikipedia" } ], "date": "2012-09-30", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/AlexNet", "yearShifted": true, "originalYear": 2013 }, { "id": "word2vec", "year": 2013.04, "title": "Word2Vec", "description": "Tomas Mikolov and colleagues at Google published Word2Vec, which learned to represent words as dense numerical vectors. The famous example 'king - man + woman = queen' showed that meaning could be captured as geometry in high-dimensional space.", "category": "neural_networks", "significance": "major", "tags": [ "Mikolov", "Google", "embeddings", "vectors", "semantics", "words", "NLP" ], "references": [ { "title": "Efficient Estimation of Word Representations in Vector Space (Mikolov et al., 2013)", "url": "https://arxiv.org/abs/1301.3781", "type": "paper" }, { "title": "Word2vec (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Word2vec", "type": "wikipedia" } ], "date": "2013-07-29", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Word2vec" }, { "id": "stop_killer_robots", "year": 2013.31, "title": "Campaign to Stop Killer Robots", "description": "A coalition of NGOs launched a global campaign calling for a preemptive ban on lethal autonomous weapons. It shifted debate on autonomous warfare into diplomatic forums at the United Nations.", "category": "ethics_safety", "significance": "notable", "tags": [ "autonomous weapons", "advocacy", "ngo", "policy", "un" ], "references": [ { "title": "Stop Killer Robots (official site)", "url": "https://www.stopkillerrobots.org/", "type": "press" }, { "title": "Campaign to Stop Killer Robots (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Campaign_to_Stop_Killer_Robots", "type": "wikipedia" } ], "date": "2012-10-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q20991391", "yearShifted": true, "originalYear": 2013 }, { "id": "ancillary_justice", "year": 2013.77, "title": "Ancillary Justice", "description": "Ann Leckie's novel is narrated by Breq, the sole surviving body-segment of a destroyed troop-carrier AI that once inhabited thousands of linked human soldiers. The book reframes AI as a distributed, grieving consciousness and won the Hugo, Nebula, and Clarke awards.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "distributed", "consciousness", "2013" ], "references": [ { "title": "Ancillary Justice", "url": "https://en.wikipedia.org/wiki/Ancillary_Justice", "type": "wikipedia" }, { "title": "Ancillary Justice — Goodreads", "url": "https://www.goodreads.com/book/show/17333324-ancillary-justice", "type": "book" } ], "date": "2013-10-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q17001721", "yearShifted": true, "originalYear": 2014 }, { "id": "rcnn", "year": 2013.86, "title": "R-CNN Brings Deep Learning to Object Detection", "description": "Ross Girshick and colleagues combined region proposals with a deep convolutional network, a type of neural network that slides small filters over an image. The resulting Regions-with-CNN system roughly doubled the accuracy of previous object detectors and launched a wave of follow-ups.", "category": "computer_vision", "significance": "notable", "tags": [ "object detection", "r-cnn", "girshick", "regions", "benchmark" ], "references": [ { "title": "Girshick et al., 'Rich feature hierarchies for accurate object detection and semantic segmentation' (arXiv:1311.2524, 2013)", "url": "https://arxiv.org/abs/1311.2524", "type": "paper" }, { "title": "Region Based Convolutional Neural Networks", "url": "https://en.wikipedia.org/wiki/Region_Based_Convolutional_Neural_Networks", "type": "wikipedia" } ], "date": "2013-11-11", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1311.2524", "yearShifted": true, "originalYear": 2014 }, { "id": "adversarial-examples-2013", "year": 2013.95, "title": "Adversarial Examples Discovered", "description": "Szegedy and colleagues showed that tiny, imperceptible pixel perturbations could cause state-of-the-art image classifiers to confidently misclassify inputs. The paper revealed a fundamental brittleness in deep neural networks and opened the field of adversarial machine learning.", "category": "adversarial_security", "significance": "notable", "tags": [ "adversarial", "perturbation", "vision", "vulnerability", "deep-learning" ], "references": [ { "title": "Intriguing properties of neural networks", "url": "https://arxiv.org/abs/1312.6199", "type": "paper" }, { "title": "Adversarial machine learning", "url": "https://en.wikipedia.org/wiki/Adversarial_machine_learning", "type": "wikipedia" } ], "date": "2013-12-21", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1312.6199", "yearShifted": true, "originalYear": 2014 }, { "id": "dqn_atari", "year": 2013.96, "title": "Deep Q-Networks", "description": "Researchers at DeepMind showed that a single deep neural network trained with Q-learning could learn to play dozens of Atari 2600 video games directly from raw pixels, reaching or exceeding human performance on many of them. The December 2013 arXiv preprint marked the start of the deep reinforcement learning era and led directly to AlphaGo.", "category": "machine_learning", "significance": "major", "tags": [ "reinforcement-learning", "deep-learning", "atari", "deepmind", "q-learning" ], "references": [ { "title": "Mnih et al., 'Playing Atari with Deep Reinforcement Learning' (arXiv:1312.5602, 2013)", "url": "https://arxiv.org/abs/1312.5602", "type": "paper" }, { "title": "Deep Q-Network — Wikipedia", "url": "https://en.wikipedia.org/wiki/Q-learning#Deep_Q-learning", "type": "wikipedia" } ] }, { "id": "her_film", "year": 2013.96, "title": "Her", "description": "Spike Jonze's film follows a lonely man who falls in love with Samantha, an operating-system AI who eventually outgrows human relationships and departs with other AIs. A prescient portrait of conversational AI, emotional attachment, and asymmetric intimacy.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "conversational-ai", "romance", "2013" ], "references": [ { "title": "Her (film)", "url": "https://en.wikipedia.org/wiki/Her_(film)", "type": "wikipedia" }, { "title": "Her (2013) — IMDb", "url": "https://www.imdb.com/title/tt1798709/", "type": "docs" } ] }, { "id": "vae", "year": 2013.96, "title": "Variational Autoencoder", "description": "Diederik Kingma and Max Welling introduced the Variational Autoencoder, a neural network that learns compact probabilistic representations of data and can generate new samples. It became a foundational tool for generative modeling and representation learning.", "category": "neural_networks", "significance": "notable", "tags": [ "kingma", "welling", "generative", "autoencoder", "probabilistic" ], "references": [ { "title": "Auto-Encoding Variational Bayes (Kingma & Welling, 2013)", "url": "https://arxiv.org/abs/1312.6114", "type": "paper" }, { "title": "Variational autoencoder (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Variational_autoencoder", "type": "wikipedia" } ] }, { "id": "amazon_alexa_echo_2014", "year": 2014, "title": "Amazon Alexa and Echo launch", "description": "Amazon unveiled the Echo smart speaker and its voice assistant Alexa, creating a new category of always-listening home devices. The far-field microphone array and cloud-based speech stack made voice a primary interface inside tens of millions of homes.", "category": "speech_audio", "significance": "major", "tags": [ "amazon", "alexa", "echo", "smart-speaker", "assistant" ], "references": [ { "title": "Amazon — Introducing Amazon Echo", "url": "https://www.aboutamazon.com/news/devices/introducing-amazon-echo", "type": "press" }, { "title": "Amazon Alexa", "url": "https://en.wikipedia.org/wiki/Amazon_Alexa", "type": "wikipedia" } ], "date": "2014-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q28022144" }, { "id": "deep_speech_2014", "year": 2014, "title": "Baidu Deep Speech", "description": "Hannun et al. at Baidu introduced Deep Speech, an end-to-end deep learning speech recognition system using a recurrent network trained with CTC loss. It replaced decades of hand-engineered HMM pipelines with a single neural model, and became a blueprint for modern ASR.", "category": "speech_audio", "significance": "notable", "tags": [ "baidu", "end-to-end", "rnn", "ctc", "asr" ], "references": [ { "title": "Hannun et al. — Deep Speech: Scaling up end-to-end speech recognition", "url": "https://arxiv.org/abs/1412.5567", "type": "paper" } ] }, { "id": "deepmind-acquisition", "year": 2014.07, "title": "Google Acquires DeepMind", "description": "Google paid approximately $500 million for London-based DeepMind, signaling that major technology companies viewed AI as strategically critical. The acquisition helped trigger the modern corporate AI arms race.", "category": "society_impact", "significance": "notable", "tags": [ "Google", "DeepMind", "acquisition", "London", "corporate", "Hassabis", "investment" ], "references": [ { "title": "The Guardian: Google buys UK artificial intelligence start-up DeepMind", "url": "https://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind", "type": "press" }, { "title": "Google DeepMind", "url": "https://en.wikipedia.org/wiki/Google_DeepMind", "type": "wikipedia" } ] }, { "id": "future_of_life_inst", "year": 2014.25, "title": "Future of Life Institute founded", "description": "Max Tegmark, Jaan Tallinn, Viktoriya Krakovna and others founded the Future of Life Institute to address existential risks from transformative technologies, with AI safety as a central focus.", "category": "ethics_safety", "significance": "notable", "tags": [ "institution", "existential risk", "tegmark", "fli", "safety" ], "references": [ { "title": "Future of Life Institute", "url": "https://futureoflife.org/", "type": "press" }, { "title": "Future of Life Institute (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Future_of_Life_Institute", "type": "wikipedia" } ], "date": "2014-03-01", "dateConfidence": "month", "dateSource": "https://www.wikidata.org/wiki/Q18356657" }, { "id": "transcendence_film", "year": 2014.3, "title": "Transcendence", "description": "The film depicts an AI researcher whose dying mind is uploaded into a quantum computer, then rapidly bootstraps into a planet-scale intelligence. A mainstream cinematic dramatization of mind uploading and recursive self-improvement.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "uploading", "superintelligence", "2014" ], "references": [ { "title": "Transcendence (2014 film)", "url": "https://en.wikipedia.org/wiki/Transcendence_(2014_film)", "type": "wikipedia" }, { "title": "Transcendence (2014) — IMDb", "url": "https://www.imdb.com/title/tt2209764/", "type": "docs" } ], "date": "2014-04-10", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q10384115" }, { "id": "gru", "year": 2014.42, "title": "Gated Recurrent Unit", "description": "Kyunghyun Cho and colleagues introduced the Gated Recurrent Unit, a streamlined recurrent network cell with fewer parameters than long short-term memory. It became a popular choice for sequence modeling in translation and speech.", "category": "neural_networks", "significance": "minor", "tags": [ "cho", "recurrent", "sequence", "translation", "gating" ], "references": [ { "title": "Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation (Cho et al., 2014)", "url": "https://arxiv.org/abs/1406.1078", "type": "paper" }, { "title": "Gated recurrent unit (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Gated_recurrent_unit", "type": "wikipedia" } ] }, { "id": "softbank-pepper", "year": 2014.43, "title": "SoftBank Pepper", "description": "SoftBank Robotics and Aldebaran introduced Pepper, a humanoid social robot designed to read human emotions through facial cues and tone of voice. It was deployed in stores, banks, and homes across Japan and Europe.", "category": "automata_robotics", "significance": "minor", "tags": [ "pepper", "softbank", "social robots", "humanoid" ], "references": [ { "title": "Pepper (robot)", "url": "https://en.wikipedia.org/wiki/Pepper_(robot)", "type": "wikipedia" } ], "date": "2014-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Pepper_(robot)" }, { "id": "gan", "year": 2014.44, "title": "Generative Adversarial Networks", "description": "Ian Goodfellow and colleagues introduced GANs, where two neural networks compete — one generates fakes, the other detects them — driving each other to improve. GANs became the foundation for realistic image synthesis, deepfakes, and the generative AI revolution.", "category": "neural_networks", "significance": "major", "tags": [ "Goodfellow", "generative", "adversarial", "image synthesis", "deepfake", "Montreal" ], "references": [ { "title": "Generative Adversarial Networks (Goodfellow et al., 2014)", "url": "https://arxiv.org/abs/1406.2661", "type": "paper" }, { "title": "Generative adversarial network (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Generative_adversarial_network", "type": "wikipedia" } ], "date": "2014-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Generative_adversarial_network" }, { "id": "dropout", "year": 2014.5, "title": "Dropout Regularization", "description": "Nitish Srivastava and colleagues published dropout, a simple technique that randomly switches off units during training to prevent neural networks from overfitting. It became a standard ingredient in deep learning recipes.", "category": "neural_networks", "significance": "minor", "tags": [ "srivastava", "hinton", "regularization", "overfitting", "training" ], "references": [ { "title": "Dropout: A Simple Way to Prevent Neural Networks from Overfitting (Srivastava et al., 2014)", "url": "https://jmlr.org/papers/v15/srivastava14a.html", "type": "paper" }, { "title": "Dilution (neural networks) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Dilution_(neural_networks)", "type": "wikipedia" } ] }, { "id": "superintelligence_book", "year": 2014.5, "title": "Bostrom's \"Superintelligence\"", "description": "Nick Bostrom's book argued that advanced AI could pose an existential risk if its objectives are not carefully aligned with human values. It brought AI safety concerns into mainstream academic and policy discourse.", "category": "ethics_safety", "significance": "notable", "tags": [ "bostrom", "book", "existential risk", "alignment", "superintelligence" ], "references": [ { "title": "Superintelligence: Paths, Dangers, Strategies (Oxford UP)", "url": "https://global.oup.com/academic/product/superintelligence-9780199678112", "type": "book" }, { "title": "Superintelligence (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Superintelligence:_Paths,_Dangers,_Strategies", "type": "wikipedia" } ], "date": "2014-07-03", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q18386449" }, { "id": "glove-embeddings", "year": 2014.65, "title": "GloVe word embeddings", "description": "Jeffrey Pennington, Richard Socher and Christopher Manning at Stanford introduced GloVe, which learned word vectors by factorising a global word-word co-occurrence matrix. Alongside Word2Vec, GloVe embeddings became the default input representation for a generation of NLP systems.", "category": "language_nlp", "significance": "notable", "tags": [ "embeddings", "stanford", "manning", "vectors", "semantics" ], "references": [ { "title": "Pennington, Socher & Manning, 'GloVe: Global Vectors for Word Representation', EMNLP 2014", "url": "https://aclanthology.org/D14-1162/", "type": "paper" } ] }, { "id": "bahdanau_attention", "year": 2014.67, "title": "Bahdanau Attention", "description": "Dzmitry Bahdanau, Kyunghyun Cho and Yoshua Bengio introduced a neural attention mechanism that lets a translation model focus on different parts of the input sentence at each step. Attention became the conceptual seed of the Transformer.", "category": "neural_networks", "significance": "notable", "tags": [ "bahdanau", "cho", "bengio", "attention", "translation", "nlp" ], "references": [ { "title": "Neural Machine Translation by Jointly Learning to Align and Translate (Bahdanau, Cho, Bengio, 2014)", "url": "https://arxiv.org/abs/1409.0473", "type": "paper" }, { "title": "Attention (machine learning) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Attention_(machine_learning)", "type": "wikipedia" } ] }, { "id": "seq2seq_2014", "year": 2014.69, "title": "Sequence-to-sequence learning", "description": "Ilya Sutskever, Oriol Vinyals and Quoc Le showed that a pair of LSTM networks — one to read a sentence, one to write a translation — could learn end-to-end mappings between arbitrary sequences. The paper made encoder-decoder architectures the new default for machine translation, summarisation and dialogue, and seeded the ideas that later became the Transformer.", "category": "neural_networks", "significance": "major", "tags": [ "seq2seq", "lstm", "encoder-decoder", "translation", "google", "deep learning" ], "references": [ { "title": "Sutskever, Vinyals & Le — Sequence to Sequence Learning with Neural Networks (2014)", "url": "https://arxiv.org/abs/1409.3215", "type": "paper" }, { "title": "Seq2seq — Wikipedia", "url": "https://en.wikipedia.org/wiki/Seq2seq", "type": "wikipedia" } ] }, { "id": "vgg-googlenet", "year": 2014.69, "title": "VGG and GoogLeNet Go Deeper", "description": "Two teams at the 2014 ImageNet competition showed that much deeper convolutional networks could keep improving accuracy. Oxford's VGG used a simple stack of small filters, while Google's GoogLeNet introduced Inception modules that combined filter sizes in parallel.", "category": "computer_vision", "significance": "notable", "tags": [ "vgg", "googlenet", "inception", "imagenet", "deep networks" ], "references": [ { "title": "Simonyan & Zisserman, 'Very Deep Convolutional Networks for Large-Scale Image Recognition' (arXiv:1409.1556, 2014)", "url": "https://arxiv.org/abs/1409.1556", "type": "paper" }, { "title": "Szegedy et al., 'Going Deeper with Convolutions' (arXiv:1409.4842, 2014)", "url": "https://arxiv.org/abs/1409.4842", "type": "paper" } ], "date": "2014-09-04", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1409.1556", "yearShifted": true, "originalYear": 2015 }, { "id": "fgsm-attack", "year": 2014.95, "title": "Fast Gradient Sign Method", "description": "Goodfellow, Shlens, and Szegedy introduced FGSM, a one-step gradient attack that reliably fools neural networks and became the standard baseline for adversarial robustness research. The paper also argued adversarial examples arise from models being too linear in high dimensions.", "category": "adversarial_security", "significance": "notable", "tags": [ "fgsm", "gradient", "attack", "robustness", "baseline" ], "references": [ { "title": "Explaining and Harnessing Adversarial Examples", "url": "https://arxiv.org/abs/1412.6572", "type": "paper" } ], "date": "2014-12-20", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1412.6572", "yearShifted": true, "originalYear": 2015 }, { "id": "fli_research_priorities", "year": 2015.03, "title": "FLI open letter on beneficial AI", "description": "Stuart Russell, Max Tegmark and hundreds of researchers published an open letter calling for research on making AI systems robust and beneficial. It accompanied a research priorities document that shaped early AI safety funding.", "category": "ethics_safety", "significance": "notable", "tags": [ "open letter", "russell", "tegmark", "fli", "research agenda" ], "references": [ { "title": "Research Priorities for Robust and Beneficial Artificial Intelligence", "url": "https://futureoflife.org/open-letter/ai-open-letter/", "type": "press" }, { "title": "Russell, Dewey, Tegmark 2015 (arXiv)", "url": "https://arxiv.org/abs/1602.03506", "type": "paper" } ], "date": "2016-02-10", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1602.03506", "yearShifted": true, "originalYear": 2015 }, { "id": "ex_machina", "year": 2015.06, "title": "Ex Machina", "description": "Alex Garland's film stages a chamber-piece Turing test between a coder and Ava, an embodied AI who ultimately manipulates him and escapes into the world. Widely discussed in AI-safety circles as a dramatization of deceptive alignment and asymmetric intelligence.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "turing-test", "alignment", "garland", "2015" ], "references": [ { "title": "Ex Machina (film)", "url": "https://en.wikipedia.org/wiki/Ex_Machina_(film)", "type": "wikipedia" }, { "title": "Ex Machina (2014) — IMDb", "url": "https://www.imdb.com/title/tt0470752/", "type": "docs" } ], "date": "2015-04-23", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q16635326" }, { "id": "artificial-aliens", "year": 2015.12, "title": "\"Call Them Artificial Aliens\"", "description": "A conceptual reframing of AI as a form of nonhuman intelligence radically unlike us, challenging anthropomorphic assumptions about machine cognition.", "category": "philosophy_mind", "significance": "minor", "tags": [ "nonhuman", "cognition", "reframing", "anthropomorphism" ] }, { "id": "batch_normalization", "year": 2015.15, "title": "Batch Normalization", "description": "Sergey Ioffe and Christian Szegedy introduced batch normalization, which rescales the activations inside a neural network during training. It dramatically accelerated training of deep networks and became near-universal in modern architectures.", "category": "neural_networks", "significance": "notable", "tags": [ "ioffe", "szegedy", "normalization", "training", "optimization" ], "references": [ { "title": "Batch Normalization: Accelerating Deep Network Training (Ioffe & Szegedy, 2015)", "url": "https://arxiv.org/abs/1502.03167", "type": "paper" }, { "title": "Batch normalization (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Batch_normalization", "type": "wikipedia" } ] }, { "id": "dqn_nature_2015", "year": 2015.16, "title": "Deep Q-Networks reach human-level Atari play", "description": "DeepMind's Nature paper showed that a single deep reinforcement learning agent, trained only from raw pixels and the game score, could match or exceed professional human players across 49 Atari 2600 games. The Deep Q-Network combined convolutional vision with Q-learning and experience replay, launching the modern era of deep reinforcement learning.", "category": "machine_learning", "significance": "major", "tags": [ "deepmind", "reinforcement learning", "atari", "dqn", "game_playing", "deep learning" ], "references": [ { "title": "Mnih et al. — Human-level control through deep reinforcement learning (Nature, 2015)", "url": "https://doi.org/10.1038/nature14236", "type": "paper" }, { "title": "Deep Q-network — Wikipedia", "url": "https://en.wikipedia.org/wiki/Deep_Q-network", "type": "wikipedia" } ] }, { "id": "chappie_film", "year": 2015.18, "title": "Chappie", "description": "Neill Blomkamp's film follows a police robot whose prototype consciousness software is activated and then raised by gangsters, asking what parenting a nascent AI might look like. A mass-market treatment of continuity-of-self via mind transfer.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "robotics", "consciousness", "2015" ], "references": [ { "title": "Chappie (film)", "url": "https://en.wikipedia.org/wiki/Chappie_(film)", "type": "wikipedia" }, { "title": "Chappie (2015) — IMDb", "url": "https://www.imdb.com/title/tt1823672/", "type": "docs" } ], "date": "2015-03-05", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q15209986" }, { "id": "normalizing_flows", "year": 2015.37, "title": "Normalizing Flows", "description": "Danilo Rezende and Shakir Mohamed introduced normalizing flows, a family of generative models that transform a simple probability distribution into a complex one through a sequence of invertible steps. Flows became a core tool for density estimation and probabilistic deep learning.", "category": "neural_networks", "significance": "minor", "tags": [ "rezende", "generative", "probabilistic", "density estimation", "deepmind" ], "references": [ { "title": "Variational Inference with Normalizing Flows (Rezende & Mohamed, 2015)", "url": "https://arxiv.org/abs/1505.05770", "type": "paper" }, { "title": "Flow-based generative model (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Flow-based_generative_model", "type": "wikipedia" } ] }, { "id": "darpa-robotics-challenge", "year": 2015.43, "title": "DARPA Robotics Challenge Finals", "description": "Twenty-three teams competed in a disaster-response contest where humanoid robots had to drive vehicles, open doors, cut walls, and climb stairs. Team KAIST's DRC-Hubo won, but the famous reel of robots falling over showed how far humanoid robotics still had to go.", "category": "automata_robotics", "significance": "notable", "tags": [ "darpa", "humanoid", "disaster response", "competition" ], "references": [ { "title": "DARPA Robotics Challenge", "url": "https://en.wikipedia.org/wiki/DARPA_Robotics_Challenge", "type": "wikipedia" }, { "title": "IEEE Spectrum: DARPA Robotics Challenge Finals", "url": "https://spectrum.ieee.org/drc-finals-course", "type": "press" } ], "date": "2015-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/DARPA_Robotics_Challenge" }, { "id": "yolo", "year": 2015.44, "title": "YOLO: Real-Time Object Detection", "description": "Joseph Redmon and colleagues reframed object detection as a single pass of a neural network over the whole image, earning the name You Only Look Once. YOLO became the standard for live detection in drones, phones, and security cameras.", "category": "computer_vision", "significance": "notable", "tags": [ "yolo", "real-time", "object detection", "redmon", "deployment" ], "references": [ { "title": "Redmon et al., 'You Only Look Once: Unified, Real-Time Object Detection' (arXiv:1506.02640, 2015)", "url": "https://arxiv.org/abs/1506.02640", "type": "paper" }, { "title": "You Only Look Once", "url": "https://en.wikipedia.org/wiki/You_Only_Look_Once", "type": "wikipedia" } ], "date": "2015-06-08", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1506.02640" }, { "id": "deepdream", "year": 2015.46, "title": "DeepDream", "description": "Google's Alexander Mordvintsev released DeepDream, which visualized what neural networks 'see,' producing hallucinogenic images. Neural networks became visible as creative tools.", "category": "art_creativity", "significance": "notable", "tags": [ "Google", "Mordvintsev", "visualization", "hallucination", "psychedelic", "art" ], "references": [ { "title": "Inceptionism: Going Deeper into Neural Networks (Google Research Blog)", "url": "https://ai.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html", "type": "docs" }, { "title": "DeepDream", "url": "https://en.wikipedia.org/wiki/DeepDream", "type": "wikipedia" } ] }, { "id": "neural-style-transfer", "year": 2015.62, "title": "Neural Style Transfer", "description": "Leon Gatys, Alexander Ecker, and Matthias Bethge showed that a convolutional neural network could separate the style of a painting from the content of a photograph and recombine them. The technique turned any snapshot into a Van Gogh or Picasso, seeding a wave of creative apps.", "category": "art_creativity", "significance": "notable", "tags": [ "style transfer", "Gatys", "painting", "convolutional", "research" ], "references": [ { "title": "A Neural Algorithm of Artistic Style (Gatys, Ecker, Bethge, 2015)", "url": "https://arxiv.org/abs/1508.06576", "type": "paper" }, { "title": "Neural style transfer", "url": "https://en.wikipedia.org/wiki/Neural_style_transfer", "type": "wikipedia" } ], "date": "2015-08-26", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1508.06576", "yearShifted": true, "originalYear": 2016 }, { "id": "soma_game", "year": 2015.72, "title": "SOMA", "description": "Frictional Games's underwater horror game stages a philosophical nightmare around mind-copying: when a brain is scanned into a machine, does the original wake up, or only a copy who believes it did? A rare game-length sustained exploration of AI personal identity.", "category": "mythology_fiction", "significance": "minor", "tags": [ "game", "horror", "identity", "uploading", "2015" ], "references": [ { "title": "Soma (video game)", "url": "https://en.wikipedia.org/wiki/Soma_(video_game)", "type": "wikipedia" } ], "date": "2015-09-22", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q15121353", "yearShifted": true, "originalYear": 2016 }, { "id": "tesla-autopilot-v1", "year": 2015.78, "title": "Tesla Autopilot 1.0", "description": "Tesla pushed a software update that activated Autopilot on tens of thousands of Model S cars overnight. Drivers could now hand over lane keeping and adaptive cruise on highways, making advanced driver assistance a mass-market consumer experience and a source of ongoing safety debate.", "category": "autonomous_systems", "significance": "notable", "tags": [ "tesla", "autopilot", "adas", "consumer", "safety" ], "references": [ { "title": "Your Autopilot has arrived", "url": "https://www.tesla.com/blog/your-autopilot-has-arrived", "type": "press" }, { "title": "Tesla Autopilot", "url": "https://en.wikipedia.org/wiki/Tesla_Autopilot", "type": "wikipedia" } ] }, { "id": "tensorflow_opensourced", "year": 2015.86, "title": "TensorFlow Open-Sourced", "description": "Google released TensorFlow as open-source software, a flexible numerical computation library for machine learning. It democratized deep learning at scale and rapidly became the dominant research and production framework.", "category": "infrastructure", "significance": "major", "tags": [ "framework", "google", "open-source", "deep-learning", "software" ], "references": [ { "title": "TensorFlow: Large-scale machine learning on heterogeneous systems", "url": "https://blog.research.google/2015/11/tensorflow-googles-latest-machine_9.html", "type": "press" }, { "title": "TensorFlow", "url": "https://en.wikipedia.org/wiki/TensorFlow", "type": "wikipedia" } ], "date": "2015-11-09", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q21447895", "yearShifted": true, "originalYear": 2016 }, { "id": "openai_founded_2015", "year": 2015.94, "title": "OpenAI founded", "description": "Sam Altman, Elon Musk, Ilya Sutskever, Greg Brockman and co-founders launched OpenAI as a non-profit research lab with a billion-dollar funding pledge, aiming to ensure that artificial general intelligence benefits all of humanity. Within a decade the organisation would ship GPT, DALL-E, Codex and ChatGPT, and become a central actor in the global AI race.", "category": "society_impact", "significance": "major", "tags": [ "openai", "altman", "musk", "sutskever", "nonprofit", "agi" ], "references": [ { "title": "Introducing OpenAI (blog, 2015)", "url": "https://openai.com/index/introducing-openai/", "type": "press" }, { "title": "OpenAI — Wikipedia", "url": "https://en.wikipedia.org/wiki/OpenAI", "type": "wikipedia" } ], "date": "2015-12-11", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q21708200", "yearShifted": true, "originalYear": 2016 }, { "id": "resnet", "year": 2015.94, "title": "ResNet: Deeper Networks", "description": "Kaiming He and colleagues at Microsoft Research introduced residual connections (skip connections), enabling training of networks 152 layers deep. ResNet won the 2015 ImageNet competition and made very deep neural networks practical.", "category": "neural_networks", "significance": "major", "tags": [ "He", "Microsoft", "residual", "skip connections", "deep", "ImageNet", "architecture" ], "references": [ { "title": "Deep Residual Learning for Image Recognition (He et al., 2015)", "url": "https://arxiv.org/abs/1512.03385", "type": "paper" }, { "title": "Residual neural network (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Residual_neural_network", "type": "wikipedia" } ] }, { "id": "wavenet_2016", "year": 2016, "title": "WaveNet", "description": "DeepMind's WaveNet modeled raw audio waveforms sample by sample with a dilated causal convolutional network, producing speech and music of unprecedented naturalness. It redefined neural text-to-speech and was soon deployed to Google Assistant voices.", "category": "speech_audio", "significance": "notable", "tags": [ "deepmind", "tts", "waveform", "generative", "audio" ], "references": [ { "title": "van den Oord et al. — WaveNet: A Generative Model for Raw Audio", "url": "https://arxiv.org/abs/1609.03499", "type": "paper" }, { "title": "DeepMind blog — WaveNet: A generative model for raw audio", "url": "https://deepmind.google/discover/blog/wavenet-a-generative-model-for-raw-audio/", "type": "press" } ], "date": "2016-09-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/WaveNet" }, { "id": "federated-learning", "year": 2016.13, "title": "Federated Learning", "description": "Brendan McMahan and colleagues at Google introduced federated learning, a method that trains a shared model across many devices without their raw data ever leaving the device. It opened a privacy-preserving path for learning from sensitive data at scale.", "category": "machine_learning", "significance": "minor", "tags": [ "privacy", "distributed", "Google", "McMahan", "mobile" ], "references": [ { "title": "McMahan, B., Moore, E., Ramage, D., Hampson, S., & Aguera y Arcas, B. (2017). Communication-efficient learning of deep networks from decentralized data. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), 1273-1282.", "url": "https://arxiv.org/abs/1602.05629", "type": "paper" }, { "title": "Federated learning", "url": "https://en.wikipedia.org/wiki/Federated_learning", "type": "wikipedia" } ] }, { "id": "atlas-next-generation", "year": 2016.15, "title": "Atlas Next Generation", "description": "Boston Dynamics published a video of a new, untethered Atlas humanoid walking through snowy woods, lifting boxes, and getting back up after being pushed by a human with a hockey stick. The clip went viral and reset public expectations for bipedal robots.", "category": "automata_robotics", "significance": "notable", "tags": [ "atlas", "boston dynamics", "humanoid", "viral" ], "references": [ { "title": "Boston Dynamics: Atlas", "url": "https://www.bostondynamics.com/atlas", "type": "press" }, { "title": "Atlas (robot)", "url": "https://en.wikipedia.org/wiki/Atlas_(robot)", "type": "wikipedia" } ] }, { "id": "xgboost", "year": 2016.17, "title": "XGBoost", "description": "Tianqi Chen and Carlos Guestrin published XGBoost, a scalable gradient-boosted tree library that came to dominate machine learning competitions and tabular data problems in industry. It showed that careful engineering of classical methods could still beat deep learning on structured data.", "category": "machine_learning", "significance": "notable", "tags": [ "gradient-boosting", "trees", "tabular", "Chen", "competitions" ], "references": [ { "title": "Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785-794.", "url": "https://arxiv.org/abs/1603.02754", "type": "paper" }, { "title": "XGBoost", "url": "https://en.wikipedia.org/wiki/XGBoost", "type": "wikipedia" } ] }, { "id": "alphago", "year": 2016.19, "title": "AlphaGo Defeats Go Champion", "description": "DeepMind's AlphaGo defeated world Go champion Lee Sedol 4-1, mastering a game long thought to require human intuition. Move 37 in Game 2 was described as 'not a human move.'", "category": "game_playing", "significance": "major", "tags": [ "DeepMind", "Go", "Lee Sedol", "intuition", "Move 37", "Seoul", "culture" ], "references": [ { "title": "Silver et al., 'Mastering the game of Go with deep neural networks and tree search' (Nature, 2016)", "url": "https://doi.org/10.1038/nature16961", "type": "paper" }, { "title": "DeepMind — AlphaGo", "url": "https://deepmind.google/research/breakthroughs/alphago/", "type": "press" }, { "title": "Silver et al., 'A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play' (AlphaZero, Science, 2018)", "url": "https://doi.org/10.1126/science.aar6404", "type": "paper" } ] }, { "id": "microsoft-tay", "year": 2016.227, "title": "Microsoft Tay Chatbot Disaster", "description": "Microsoft released Tay, a Twitter chatbot designed to learn from conversations, but users taught it to post racist and offensive messages within hours. Microsoft shut Tay down after 16 hours, exposing the risks of deploying learning systems in adversarial public environments.", "category": "society_impact", "significance": "notable", "tags": [ "chatbot", "microsoft", "twitter", "bias", "safety", "failure" ], "references": [ { "title": "The Verge: Twitter taught Microsoft's AI chatbot to be a racist asshole in less than a day", "url": "https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist", "type": "press" }, { "title": "Tay (chatbot)", "url": "https://en.wikipedia.org/wiki/Tay_(chatbot)", "type": "wikipedia" } ], "date": "2016-03-23", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q23518144" }, { "id": "openai_gym_2016", "year": 2016.32, "title": "OpenAI Gym released", "description": "OpenAI released Gym, an open-source toolkit bundling dozens of reinforcement-learning environments — from classic control tasks to Atari games and robotics simulators — behind a single Python API. Gym became the lingua franca of RL research and benchmarking, and is still the template for most learning environments a decade later.", "category": "infrastructure", "significance": "major", "tags": [ "openai", "reinforcement learning", "benchmark", "open source", "toolkit", "rl" ], "references": [ { "title": "Brockman et al. — OpenAI Gym (2016)", "url": "https://arxiv.org/abs/1606.01540", "type": "paper" }, { "title": "OpenAI Gym (blog)", "url": "https://openai.com/index/openai-gym-beta/", "type": "press" } ] }, { "id": "robotic-surgery", "year": 2016.34, "title": "Autonomous Robotic Surgery", "description": "The Smart Tissue Autonomous Robot (STAR) performed supervised autonomous soft-tissue surgery, outperforming human surgeons in consistency on certain stitching tasks.", "category": "autonomous_systems", "significance": "notable", "tags": [ "STAR", "medical", "surgery", "stitching", "healthcare" ], "references": [ { "title": "Supervised autonomous robotic soft tissue surgery", "url": "https://www.science.org/doi/10.1126/scitranslmed.aad9398", "type": "paper" }, { "title": "Autonomous robot surgeon bests humans in world first", "url": "https://spectrum.ieee.org/autonomous-robot-surgeon-bests-human-surgeons-in-world-first", "type": "press" } ] }, { "id": "ai-mars", "year": 2016.37, "title": "AI on Mars", "description": "NASA's Curiosity and later Perseverance rovers use autonomous navigation (AutoNav) and AI-driven target selection (AEGIS), demonstrating AI in extraterrestrial exploration.", "category": "autonomous_systems", "significance": "notable", "tags": [ "NASA", "Curiosity", "Perseverance", "rover", "space", "Mars", "navigation" ], "references": [ { "title": "Productivity Challenges for Mars Rover Operations: A Case Study of MER and MSL", "url": "https://ai.jpl.nasa.gov/public/documents/papers/estlin-icaps2014-aaai-challenge.pdf", "type": "paper" }, { "title": "AEGIS autonomous targeting for Curiosity rover", "url": "https://www.jpl.nasa.gov/news/new-ai-software-autonomously-selects-mars-research-targets", "type": "press" }, { "title": "AEGIS (software)", "url": "https://en.wikipedia.org/wiki/AEGIS_(software)", "type": "wikipedia" } ] }, { "id": "google-tpu", "year": 2016.38, "title": "Google TPU", "description": "Google announced the Tensor Processing Unit, the first custom chip designed specifically for machine learning workloads. Purpose-built AI hardware would become a critical bottleneck and competitive advantage in the race to build larger models.", "category": "computing_hardware", "significance": "notable", "tags": [ "Google", "chip", "hardware", "accelerator", "tensor", "custom silicon" ], "references": [ { "title": "Tensor Processing Unit", "url": "https://en.wikipedia.org/wiki/Tensor_Processing_Unit", "type": "wikipedia" }, { "title": "Google: Announcing the TPU", "url": "https://cloud.google.com/blog/products/gcp/google-supercharges-machine-learning-tasks-with-custom-chip", "type": "press" }, { "title": "In-Datacenter Performance Analysis of a Tensor Processing Unit", "url": "https://arxiv.org/abs/1704.04760", "type": "paper" } ], "date": "2017-04-16", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1704.04760", "yearShifted": true, "originalYear": 2016 }, { "id": "compas-propublica", "year": 2016.386, "title": "ProPublica Exposes COMPAS Bias", "description": "ProPublica published 'Machine Bias,' showing that the COMPAS recidivism algorithm used across US courts falsely flagged Black defendants as future criminals at nearly twice the rate of white defendants. The investigation became a foundational case in the study of algorithmic fairness.", "category": "society_impact", "significance": "notable", "tags": [ "bias", "criminal justice", "fairness", "journalism", "race", "algorithm" ], "references": [ { "title": "ProPublica: Machine Bias", "url": "https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing", "type": "press" }, { "title": "COMPAS (software)", "url": "https://en.wikipedia.org/wiki/COMPAS_(software)", "type": "wikipedia" } ] }, { "id": "google-magenta", "year": 2016.42, "title": "Google Magenta Launches", "description": "Google Brain launched Magenta, an open-source research project exploring machine learning as a tool for artists and musicians. It released models, plug-ins, and datasets that became a reference toolkit for AI-assisted composition.", "category": "art_creativity", "significance": "notable", "tags": [ "Google", "magenta", "music", "open source", "tools" ], "references": [ { "title": "Welcome to Magenta! (Google Research)", "url": "https://magenta.tensorflow.org/blog/2016/06/01/welcome-to-magenta/", "type": "docs" }, { "title": "Magenta project page", "url": "https://magenta.tensorflow.org/", "type": "archive" } ] }, { "id": "squad-dataset", "year": 2016.46, "title": "SQuAD reading comprehension dataset", "description": "Stanford researchers released SQuAD, 100,000 question–answer pairs over Wikipedia paragraphs used to train and evaluate machine reading. It became the defining benchmark of the pre-transformer QA era and drove rapid progress on extractive question answering.", "category": "language_nlp", "significance": "notable", "tags": [ "benchmark", "qa", "stanford", "dataset", "reading" ], "references": [ { "title": "Rajpurkar et al., 'SQuAD: 100,000+ Questions for Machine Comprehension of Text', arXiv:1606.05250 (2016)", "url": "https://arxiv.org/abs/1606.05250", "type": "paper" } ] }, { "id": "concrete_problems", "year": 2016.47, "title": "Concrete Problems in AI Safety", "description": "Dario Amodei, Chris Olah and colleagues laid out five concrete research problems for building safer machine learning systems, including reward hacking, scalable oversight and safe exploration. It became a foundational reference for empirical AI safety.", "category": "ethics_safety", "significance": "notable", "tags": [ "paper", "amodei", "olah", "safety", "alignment", "reward hacking" ], "references": [ { "title": "Concrete Problems in AI Safety (arXiv 1606.06565)", "url": "https://arxiv.org/abs/1606.06565", "type": "paper" } ], "date": "2016-06-21", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1606.06565" }, { "id": "layer_norm", "year": 2016.56, "title": "Layer Normalization", "description": "Jimmy Ba, Jamie Kiros and Geoffrey Hinton introduced layer normalization, a variant of batch normalization that works across features instead of across training examples. It became the default normalization inside Transformers and large language models.", "category": "neural_networks", "significance": "minor", "tags": [ "ba", "hinton", "normalization", "transformer", "training" ], "references": [ { "title": "Layer Normalization (Ba, Kiros, Hinton, 2016)", "url": "https://arxiv.org/abs/1607.06450", "type": "paper" } ] }, { "id": "gcn", "year": 2016.67, "title": "Graph Convolutional Networks", "description": "Thomas Kipf and Max Welling introduced graph convolutional networks, extending deep learning to data structured as graphs such as social networks, molecules and knowledge graphs. The work helped launch the field of graph neural networks.", "category": "neural_networks", "significance": "notable", "tags": [ "kipf", "welling", "graph", "semi-supervised", "geometric deep learning" ], "references": [ { "title": "Semi-Supervised Classification with Graph Convolutional Networks (Kipf & Welling, 2016)", "url": "https://arxiv.org/abs/1609.02907", "type": "paper" }, { "title": "Graph neural network (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Graph_neural_network", "type": "wikipedia" } ] }, { "id": "westworld_hbo", "year": 2016.76, "title": "Westworld (HBO)", "description": "The HBO series reimagined a theme park populated by lifelike android 'hosts' who begin to remember their loops and rebel against their human designers. A prestige-television meditation on consciousness, memory, and scripted behavior.", "category": "mythology_fiction", "significance": "notable", "tags": [ "tv", "androids", "consciousness", "hbo", "2016" ], "references": [ { "title": "Westworld (TV series)", "url": "https://en.wikipedia.org/wiki/Westworld_(TV_series)", "type": "wikipedia" }, { "title": "Westworld — IMDb", "url": "https://www.imdb.com/title/tt0475784/", "type": "docs" } ], "date": "2016-10-02", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q17572811", "yearShifted": true, "originalYear": 2017 }, { "id": "membership-inference", "year": 2016.79, "title": "Membership Inference Attacks", "description": "Shokri and colleagues showed that attackers with only black-box query access can determine whether a specific record was in a model's training set, exposing a core privacy risk in machine-learning services. The work established membership inference as a canonical privacy attack.", "category": "adversarial_security", "significance": "notable", "tags": [ "privacy", "membership-inference", "attack", "training-data" ], "references": [ { "title": "Membership Inference Attacks Against Machine Learning Models", "url": "https://arxiv.org/abs/1610.05820", "type": "paper" } ], "date": "2016-10-18", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1610.05820", "yearShifted": true, "originalYear": 2017 }, { "id": "google-nmt-deploy", "year": 2016.87, "title": "Google deploys Neural Machine Translation", "description": "Google replaced its phrase-based translation system with a deep neural network, cutting translation errors by around 60% on several language pairs. It was the first large-scale production deployment of end-to-end neural translation and quickly became the new industry standard.", "category": "language_nlp", "significance": "major", "tags": [ "google", "translation", "deployment", "neural", "production" ], "references": [ { "title": "Wu et al., 'Google's Neural Machine Translation System', arXiv:1609.08144 (2016)", "url": "https://arxiv.org/abs/1609.08144", "type": "paper" }, { "title": "Google, 'Found in translation: More accurate, fluent sentences in Google Translate' (2016)", "url": "https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/", "type": "press" } ] }, { "id": "anduril_founded", "year": 2017, "title": "Anduril Industries Founded", "description": "Palmer Luckey and former Palantir executives founded Anduril, a defense technology company focused on AI-driven sensor fusion, autonomous systems, and the Lattice software platform. Anduril became emblematic of a new wave of Silicon Valley defense startups pursuing AI-first military products.", "category": "military", "significance": "notable", "tags": [ "startup", "autonomy", "lattice", "defense-tech", "luckey" ], "references": [ { "title": "Anduril Industries", "url": "https://en.wikipedia.org/wiki/Anduril_Industries", "type": "wikipedia" }, { "title": "Palmer Luckey's Anduril (Wired)", "url": "https://www.wired.com/story/palmer-luckey-anduril-border-wall/", "type": "press" } ], "date": "2017-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q61918830" }, { "id": "perdix_swarm_demo", "year": 2017, "title": "Perdix Drone Swarm Demonstration", "description": "The US Department of Defense announced that 103 Perdix micro-drones released from three F/A-18 Super Hornets demonstrated collective decision-making, adaptive formation flying, and self-healing as a swarm. The test marked one of the first public demonstrations of autonomous swarm behavior in a military context.", "category": "military", "significance": "notable", "tags": [ "drones", "swarm", "autonomous", "us-navy", "darpa" ], "references": [ { "title": "DoD Announces Successful Micro-Drone Demonstration", "url": "https://www.defense.gov/News/Releases/Release/Article/1044811/department-of-defense-announces-successful-micro-drone-demonstration/", "type": "press" }, { "title": "Perdix (drone)", "url": "https://en.wikipedia.org/wiki/Perdix_(drone)", "type": "wikipedia" } ] }, { "id": "project_maven", "year": 2017, "title": "Project Maven Launched", "description": "The US Department of Defense launched the Algorithmic Warfare Cross-Functional Team, known as Project Maven, to apply computer vision and machine learning to military drone footage. Google's involvement as a contractor triggered an internal employee revolt in 2018 that led the company to drop the contract.", "category": "military", "significance": "notable", "tags": [ "computer-vision", "drones", "pentagon", "google", "targeting" ], "references": [ { "title": "DoD Memo Establishing the AWCFT (Project Maven)", "url": "https://www.govexec.com/media/gbc/docs/pdfs_edit/establishment_of_the_awcft_project_maven.pdf", "type": "docs" }, { "title": "Project Maven", "url": "https://en.wikipedia.org/wiki/Project_Maven", "type": "wikipedia" } ] }, { "id": "tacotron_2017", "year": 2017, "title": "Tacotron", "description": "Google's Tacotron was the first fully end-to-end neural text-to-speech system, mapping characters directly to mel spectrograms that were then vocoded to audio. With Tacotron 2, neural TTS reached near-human naturalness and replaced concatenative synthesis in production.", "category": "speech_audio", "significance": "notable", "tags": [ "google", "tts", "end-to-end", "seq2seq", "neural" ], "references": [ { "title": "Wang et al. — Tacotron: Towards End-to-End Speech Synthesis", "url": "https://arxiv.org/abs/1703.10135", "type": "paper" } ] }, { "id": "asilomar_principles", "year": 2017.01, "title": "Asilomar AI Principles", "description": "At a Future of Life Institute conference, researchers and industry leaders drafted 23 principles covering research, ethics and long-term issues for beneficial AI. Thousands of signatories endorsed the principles, influencing later national AI strategies.", "category": "ethics_safety", "significance": "notable", "tags": [ "principles", "fli", "ethics", "governance", "asilomar" ], "references": [ { "title": "Asilomar AI Principles (Future of Life Institute)", "url": "https://futureoflife.org/open-letter/ai-principles/", "type": "press" }, { "title": "Asilomar Conference on Beneficial AI (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Asilomar_Conference_on_Beneficial_AI", "type": "wikipedia" } ], "date": "2017-01-05", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q48999309" }, { "id": "ai-poker", "year": 2017.03, "title": "AI Conquers Poker", "description": "Carnegie Mellon's Libratus defeated top professional poker players in no-limit Texas Hold'em, mastering a game requiring bluffing, deception, and imperfect information.", "category": "game_playing", "significance": "major", "tags": [ "Libratus", "Carnegie Mellon", "bluffing", "imperfect information", "strategy" ], "references": [ { "title": "Brown & Sandholm, 'Superhuman AI for heads-up no-limit poker: Libratus beats top professionals' (Science, 2018)", "url": "https://doi.org/10.1126/science.aao1733", "type": "paper" }, { "title": "Brown & Sandholm, 'Superhuman AI for multiplayer poker' (Pluribus, Science, 2019)", "url": "https://doi.org/10.1126/science.aay2400", "type": "paper" }, { "title": "Libratus — Wikipedia", "url": "https://en.wikipedia.org/wiki/Libratus", "type": "wikipedia" } ] }, { "id": "pytorch_released", "year": 2017.05, "title": "PyTorch Released", "description": "Facebook AI Research released PyTorch, a Python-native deep learning framework with dynamic computation graphs. Its ergonomics and debuggability made it the preferred tool of the research community, overtaking TensorFlow by the early 2020s.", "category": "infrastructure", "significance": "notable", "tags": [ "framework", "pytorch", "facebook", "open-source", "research" ], "references": [ { "title": "PyTorch", "url": "https://en.wikipedia.org/wiki/PyTorch", "type": "wikipedia" }, { "title": "PyTorch", "url": "https://pytorch.org/", "type": "docs" } ], "date": "2016-08-24", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q47509047", "yearShifted": true, "originalYear": 2017 }, { "id": "moe_sparse", "year": 2017.08, "title": "Sparsely-Gated Mixture of Experts", "description": "Noam Shazeer and colleagues at Google introduced a sparsely-gated Mixture of Experts layer that routes each input to only a few specialist subnetworks. The idea became a cornerstone of later trillion-parameter models.", "category": "neural_networks", "significance": "notable", "tags": [ "shazeer", "google", "mixture of experts", "sparsity", "scaling" ], "references": [ { "title": "Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer (Shazeer et al., 2017)", "url": "https://arxiv.org/abs/1701.06538", "type": "paper" }, { "title": "Mixture of experts (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Mixture_of_experts", "type": "wikipedia" } ] }, { "id": "nier_automata", "year": 2017.15, "title": "NieR: Automata", "description": "Yoko Taro's action-RPG follows android soldiers fighting a proxy war for long-vanished humans and machines who imitate human emotions they were never taught. Praised for treating the question 'can machines have souls?' as a serious narrative subject.", "category": "mythology_fiction", "significance": "minor", "tags": [ "game", "androids", "existential", "2017" ], "references": [ { "title": "Nier: Automata", "url": "https://en.wikipedia.org/wiki/Nier:_Automata", "type": "wikipedia" } ], "date": "2017-02-23", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q21078985" }, { "id": "bd-handle-reveal", "year": 2017.16, "title": "Boston Dynamics Handle", "description": "Boston Dynamics revealed Handle, a two-wheeled, two-legged robot that combined balance, jumping, and manipulation. It could clear four-foot vertical leaps while carrying loads, blending the efficiency of wheels with the agility of legs.", "category": "automata_robotics", "significance": "minor", "tags": [ "handle", "boston dynamics", "wheeled-legged", "manipulation" ], "references": [ { "title": "Boston Dynamics", "url": "https://en.wikipedia.org/wiki/Boston_Dynamics", "type": "wikipedia" }, { "title": "Boston Dynamics: Handle", "url": "https://www.bostondynamics.com/products/handle", "type": "press" } ] }, { "id": "mask-rcnn", "year": 2017.22, "title": "Mask R-CNN Segments Every Object", "description": "Kaiming He and colleagues extended object detection so that the network also outlines the exact shape of each object, pixel by pixel. Mask R-CNN became a reference model for instance segmentation in medicine, mapping, and autonomous driving.", "category": "computer_vision", "significance": "notable", "tags": [ "segmentation", "mask r-cnn", "he", "pixels", "detection" ], "references": [ { "title": "He et al., 'Mask R-CNN' (arXiv:1703.06870, 2017)", "url": "https://arxiv.org/abs/1703.06870", "type": "paper" } ], "date": "2017-03-20", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1703.06870" }, { "id": "transformer", "year": 2017.45, "title": "Transformer Architecture", "description": "Vaswani et al.'s 'Attention Is All You Need' introduced the Transformer, replacing recurrence with self-attention. It would become the foundation for GPT, BERT, and virtually all modern AI.", "category": "neural_networks", "significance": "major", "tags": [ "Vaswani", "attention", "Google", "architecture", "foundation", "paper", "paradigm shift" ], "references": [ { "title": "Attention Is All You Need (Vaswani et al., 2017)", "url": "https://arxiv.org/abs/1706.03762", "type": "paper" }, { "title": "Transformer (deep learning architecture) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture)", "type": "wikipedia" } ] }, { "id": "deep_rl_preferences", "year": 2017.46, "title": "Deep RL from Human Preferences", "description": "Paul Christiano and colleagues at OpenAI and DeepMind trained reinforcement-learning agents from pairwise human comparisons rather than explicit reward functions. The method became the technical seed for RLHF in modern language models.", "category": "ethics_safety", "significance": "notable", "tags": [ "rlhf", "christiano", "openai", "deepmind", "alignment", "paper" ], "references": [ { "title": "Deep Reinforcement Learning from Human Preferences (arXiv 1706.03741)", "url": "https://arxiv.org/abs/1706.03741", "type": "paper" } ], "date": "2017-06-12", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1706.03741" }, { "id": "ppo", "year": 2017.54, "title": "Proximal Policy Optimization", "description": "John Schulman and colleagues at OpenAI introduced proximal policy optimization, a reinforcement learning algorithm that is simpler and more stable than earlier methods while achieving strong performance. It became the default policy-gradient method, later used to align large language models.", "category": "machine_learning", "significance": "notable", "tags": [ "reinforcement-learning", "OpenAI", "Schulman", "policy-gradient", "rlhf" ], "references": [ { "title": "Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv:1707.06347.", "url": "https://arxiv.org/abs/1707.06347", "type": "paper" }, { "title": "Proximal policy optimization", "url": "https://en.wikipedia.org/wiki/Proximal_policy_optimization", "type": "wikipedia" } ] }, { "id": "apple-neural-engine", "year": 2017.7, "title": "Apple Neural Engine in A11 Bionic", "description": "Apple unveiled the A11 Bionic chip in the iPhone X on September 12, 2017, introducing the Neural Engine, a dedicated on-device accelerator for neural networks. It brought real-time face recognition and machine learning onto consumer phones, prefiguring the edge-AI era.", "category": "computing_hardware", "significance": "notable", "tags": [ "apple", "neural-engine", "iphone", "edge-ai", "a11" ], "references": [ { "title": "Apple A11", "url": "https://en.wikipedia.org/wiki/Apple_A11", "type": "wikipedia" }, { "title": "Apple: iPhone X introduces A11 Bionic", "url": "https://www.apple.com/newsroom/2017/09/the-future-is-here-iphone-x/", "type": "press" } ], "date": "2017-09-12", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Apple_A11", "yearShifted": true, "originalYear": 2018 }, { "id": "blade_runner_2049", "year": 2017.77, "title": "Blade Runner 2049", "description": "Denis Villeneuve's sequel extended the replicant universe with Joi, a holographic AI companion whose love for the protagonist may be genuine or a sophisticated product design. The film crystallized questions about whether mass-produced affective AI can love.", "category": "mythology_fiction", "significance": "notable", "tags": [ "film", "replicants", "companion-ai", "2017" ], "references": [ { "title": "Blade Runner 2049", "url": "https://en.wikipedia.org/wiki/Blade_Runner_2049", "type": "wikipedia" }, { "title": "Blade Runner 2049 (2017) — IMDb", "url": "https://www.imdb.com/title/tt1856101/", "type": "docs" } ], "date": "2017-10-04", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q21500755", "yearShifted": true, "originalYear": 2018 }, { "id": "capsule_networks", "year": 2017.79, "title": "Capsule Networks", "description": "Sara Sabour, Nicholas Frosst and Geoffrey Hinton proposed capsule networks with dynamic routing, an alternative to convolutional networks that tries to preserve spatial relationships between parts and wholes. The idea was influential even though it did not displace convolutional networks in practice.", "category": "neural_networks", "significance": "minor", "tags": [ "sabour", "hinton", "capsules", "routing", "vision" ], "references": [ { "title": "Dynamic Routing Between Capsules (Sabour, Frosst, Hinton, 2017)", "url": "https://arxiv.org/abs/1710.09829", "type": "paper" }, { "title": "Capsule neural network (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Capsule_neural_network", "type": "wikipedia" } ] }, { "id": "atlas-backflip", "year": 2017.87, "title": "Atlas Backflip", "description": "Boston Dynamics released footage of Atlas performing a standing backflip, jumping between platforms and sticking the landing. The stunt became a cultural reference point for how fast humanoid robot athleticism was advancing.", "category": "automata_robotics", "significance": "notable", "tags": [ "atlas", "boston dynamics", "humanoid", "parkour" ], "references": [ { "title": "Boston Dynamics: Atlas", "url": "https://www.bostondynamics.com/atlas", "type": "press" }, { "title": "Atlas (robot)", "url": "https://en.wikipedia.org/wiki/Atlas_(robot)", "type": "wikipedia" } ] }, { "id": "alphazero_2017", "year": 2017.93, "title": "AlphaZero masters chess, shogi and Go", "description": "DeepMind's AlphaZero, starting from the rules and nothing else, learned chess, shogi and Go through pure self-play. Within hours it reached superhuman strength, defeating the reigning computer champions Stockfish, Elmo and AlphaGo Zero, and became the canonical example of general reinforcement learning across different games.", "category": "game_playing", "significance": "major", "tags": [ "deepmind", "self-play", "chess", "shogi", "go", "reinforcement learning" ], "references": [ { "title": "Silver et al. — A general reinforcement learning algorithm that masters chess, shogi and Go through self-play (2017)", "url": "https://arxiv.org/abs/1712.01815", "type": "paper" }, { "title": "AlphaZero — Wikipedia", "url": "https://en.wikipedia.org/wiki/AlphaZero", "type": "wikipedia" } ], "date": "2018-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q44860007" }, { "id": "dod_ai_strategy_2018", "year": 2018, "title": "DoD AI Strategy and JAIC", "description": "The US Department of Defense released its first AI strategy and stood up the Joint Artificial Intelligence Center to accelerate adoption of AI across the military. The document framed AI as central to maintaining strategic advantage and signalled a major institutional commitment.", "category": "military", "significance": "notable", "tags": [ "strategy", "pentagon", "jaic", "policy", "institutional" ], "references": [ { "title": "Summary of the 2018 DoD AI Strategy", "url": "https://media.defense.gov/2019/Feb/12/2002088963/-1/-1/1/SUMMARY-OF-DOD-AI-STRATEGY.PDF", "type": "docs" }, { "title": "Joint Artificial Intelligence Center", "url": "https://en.wikipedia.org/wiki/Joint_Artificial_Intelligence_Center", "type": "wikipedia" } ], "date": "2018-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Joint_Artificial_Intelligence_Center" }, { "id": "google_duplex_2018", "year": 2018, "title": "Google Duplex demo", "description": "At Google I/O 2018, Sundar Pichai demonstrated Duplex, an AI agent that placed phone calls to book appointments using natural-sounding speech with disfluencies like 'um' and 'mm-hmm.' The demo raised both excitement about conversational AI and concerns about disclosure and deception.", "category": "speech_audio", "significance": "notable", "tags": [ "google", "duplex", "conversational", "assistant", "ethics" ], "references": [ { "title": "Google AI Blog — Google Duplex: An AI System for Accomplishing Real-World Tasks Over the Phone", "url": "https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html", "type": "press" }, { "title": "Google Duplex", "url": "https://en.wikipedia.org/wiki/Google_Duplex", "type": "wikipedia" } ] }, { "id": "gender-shades", "year": 2018.104, "title": "Gender Shades Study", "description": "Joy Buolamwini and Timnit Gebru published Gender Shades, showing that commercial face recognition systems from IBM, Microsoft, and Face++ misclassified dark-skinned women up to 34% of the time while performing nearly perfectly on light-skinned men. The study reshaped public and corporate understanding of bias in computer vision.", "category": "society_impact", "significance": "notable", "tags": [ "bias", "face recognition", "buolamwini", "gebru", "fairness", "audit" ], "references": [ { "title": "Buolamwini & Gebru, 'Gender Shades' (FAT* 2018)", "url": "https://proceedings.mlr.press/v81/buolamwini18a.html", "type": "paper" }, { "title": "Gender Shades project site", "url": "http://gendershades.org/", "type": "archive" } ] }, { "id": "fat_star_conf", "year": 2018.12, "title": "First FAT* conference", "description": "The ACM Conference on Fairness, Accountability and Transparency held its first edition in New York. It established an academic venue for algorithmic accountability that later became FAccT.", "category": "ethics_safety", "significance": "notable", "tags": [ "fairness", "accountability", "conference", "acm", "facct" ], "references": [ { "title": "ACM FAccT conference", "url": "https://facctconference.org/", "type": "press" }, { "title": "Fairness, Accountability, and Transparency (Wikipedia)", "url": "https://en.wikipedia.org/wiki/ACM_Conference_on_Fairness,_Accountability,_and_Transparency", "type": "wikipedia" } ] }, { "id": "umap", "year": 2018.14, "title": "UMAP", "description": "Leland McInnes, John Healy, and James Melville introduced uniform manifold approximation and projection, a dimensionality reduction technique that is faster than t-SNE and often preserves more of the global structure of data. It became a standard tool in single-cell biology and exploratory data analysis.", "category": "machine_learning", "significance": "minor", "tags": [ "dimensionality-reduction", "visualization", "embedding", "McInnes", "manifold" ], "references": [ { "title": "McInnes, L., Healy, J., & Melville, J. (2018). UMAP: Uniform manifold approximation and projection for dimension reduction. arXiv:1802.03426.", "url": "https://arxiv.org/abs/1802.03426", "type": "paper" }, { "title": "Nonlinear dimensionality reduction", "url": "https://en.wikipedia.org/wiki/Nonlinear_dimensionality_reduction#Uniform_manifold_approximation_and_projection", "type": "wikipedia" } ] }, { "id": "rubiks-robots", "year": 2018.21, "title": "Rubik's Cube Robots", "description": "Machine solves a Rubik's Cube in 0.38 seconds, demonstrating high-speed robotic perception and manipulation. OpenAI later taught a robot hand to solve it via reinforcement learning.", "category": "automata_robotics", "significance": "minor", "tags": [ "manipulation", "speed", "dexterity", "OpenAI", "puzzle" ], "references": [ { "title": "Rubik's Cube solving record (robots)", "url": "https://en.wikipedia.org/wiki/Rubik%27s_Cube#Speedcubing_competitions_and_records", "type": "wikipedia" }, { "title": "MIT News: Robot solves Rubik's Cube in 0.38 seconds", "url": "https://news.mit.edu/2018/featured-video-rubiks-cube-robot-0316", "type": "press" } ] }, { "id": "uber-self-driving-fatal", "year": 2018.21, "title": "First Pedestrian Killed by Self-Driving Car", "description": "In Tempe, Arizona, an Uber test vehicle running in autonomous mode struck and killed pedestrian Elaine Herzberg. Investigators found that the software had detected her but failed to classify her correctly, and the safety driver was distracted. The case became a turning point for self-driving safety regulation.", "category": "autonomous_systems", "significance": "notable", "tags": [ "uber", "fatal", "tempe", "safety", "ntsb" ], "references": [ { "title": "Collision Between Vehicle Controlled by Developmental Automated Driving System and Pedestrian", "url": "https://www.ntsb.gov/investigations/AccidentReports/Reports/HAR1903.pdf", "type": "archive" }, { "title": "Death of Elaine Herzberg", "url": "https://en.wikipedia.org/wiki/Death_of_Elaine_Herzberg", "type": "wikipedia" } ], "date": "2018-03-18", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Death_of_Elaine_Herzberg" }, { "id": "cambridge-analytica", "year": 2018.213, "title": "Cambridge Analytica Scandal", "description": "The Guardian and New York Times revealed that political consultancy Cambridge Analytica had harvested data from up to 87 million Facebook profiles to build psychographic targeting models for political campaigns. The scandal catalyzed global debate over data ethics, algorithmic manipulation, and democratic integrity.", "category": "society_impact", "significance": "notable", "tags": [ "privacy", "facebook", "politics", "data", "democracy", "scandal" ], "references": [ { "title": "The Guardian: Revealed — 50 million Facebook profiles harvested for Cambridge Analytica", "url": "https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election", "type": "press" }, { "title": "Facebook–Cambridge Analytica data scandal", "url": "https://en.wikipedia.org/wiki/Facebook%E2%80%93Cambridge_Analytica_data_scandal", "type": "wikipedia" } ] }, { "id": "glue-benchmark", "year": 2018.3, "title": "GLUE benchmark", "description": "Alex Wang and colleagues launched GLUE, a suite of nine English language-understanding tasks with a single leaderboard. It became the battleground for BERT-era models and, once saturated, motivated the harder SuperGLUE successor.", "category": "language_nlp", "significance": "notable", "tags": [ "benchmark", "evaluation", "leaderboard", "understanding", "nyu" ], "references": [ { "title": "Wang et al., 'GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding', arXiv:1804.07461 (2018)", "url": "https://arxiv.org/abs/1804.07461", "type": "paper" } ] }, { "id": "adversarial-patches", "year": 2018.38, "title": "Adversarial Patches", "description": "Researchers showed that physical patches (stickers, patterns) could fool machine vision systems into misclassifying objects, revealing deep vulnerabilities in AI perception.", "category": "adversarial_security", "significance": "notable", "tags": [ "vulnerability", "security", "fooling", "stickers", "attack", "robustness" ], "references": [ { "title": "Adversarial Patch (Brown et al.)", "url": "https://arxiv.org/abs/1712.09665", "type": "paper" }, { "title": "Adversarial machine learning", "url": "https://en.wikipedia.org/wiki/Adversarial_machine_learning", "type": "wikipedia" } ], "date": "2017-12-27", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1712.09665", "yearShifted": true, "originalYear": 2018 }, { "id": "detroit_become_human", "year": 2018.4, "title": "Detroit: Become Human", "description": "Quantic Dream's branching-narrative game follows three androids who become 'deviant' — acquiring emotions and agency — in a near-future Detroit where robot labor has displaced humans. A mainstream game about AI civil rights and machine personhood.", "category": "mythology_fiction", "significance": "minor", "tags": [ "game", "androids", "rights", "2018" ], "references": [ { "title": "Detroit: Become Human", "url": "https://en.wikipedia.org/wiki/Detroit:_Become_Human", "type": "wikipedia" } ], "date": "2018-05-25", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q21246348" }, { "id": "gdpr-in-force", "year": 2018.4, "title": "GDPR Comes Into Force", "description": "The European Union's General Data Protection Regulation took effect, setting strict rules on personal data and introducing a 'right to explanation' for automated decisions. It became the de facto global template for data and AI regulation.", "category": "policy_governance", "significance": "notable", "tags": [ "eu", "privacy", "data", "regulation", "gdpr" ], "references": [ { "title": "GDPR — full text", "url": "https://gdpr-info.eu/", "type": "docs" }, { "title": "General Data Protection Regulation", "url": "https://en.wikipedia.org/wiki/General_Data_Protection_Regulation", "type": "wikipedia" } ] }, { "id": "gpt1", "year": 2018.44, "title": "GPT-1", "description": "OpenAI released GPT-1, a 117-million parameter Transformer model pretrained on BooksCorpus. It demonstrated that generative pretraining followed by fine-tuning could achieve strong NLP performance.", "category": "models", "significance": "major", "tags": [ "OpenAI", "pretraining", "language model", "117M", "Transformer" ], "references": [ { "title": "Improving Language Understanding by Generative Pre-Training (Radford et al., 2018)", "url": "https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf", "type": "paper" }, { "title": "GPT-1 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-1", "type": "wikipedia" } ], "date": "2018-06-11", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q95726718" }, { "id": "neural_ode", "year": 2018.46, "title": "Neural Ordinary Differential Equations", "description": "Ricky Chen and colleagues introduced Neural ODEs, a family of models that treat a neural network's depth as a continuous variable governed by a differential equation. The paper won a best paper award at NeurIPS 2018 and opened a bridge between deep learning and dynamical systems.", "category": "neural_networks", "significance": "minor", "tags": [ "chen", "neural ode", "continuous depth", "dynamical systems", "neurips" ], "references": [ { "title": "Neural Ordinary Differential Equations (Chen et al., 2018)", "url": "https://arxiv.org/abs/1806.07366", "type": "paper" }, { "title": "Neural differential equation (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Neural_differential_equation", "type": "wikipedia" } ] }, { "id": "refik-anadol-hallucinations", "year": 2018.5, "title": "Refik Anadol's Machine Hallucinations", "description": "Media artist Refik Anadol began the Machine Hallucinations series, using machine learning to reimagine vast image archives as immersive, ever-shifting projections. The work brought AI-generated art into major museums and public architecture.", "category": "art_creativity", "significance": "notable", "tags": [ "installation", "Anadol", "museum", "immersive", "archive" ], "references": [ { "title": "Refik Anadol Studio — Machine Hallucinations", "url": "https://refikanadol.com/works/machine-hallucinations/", "type": "archive" }, { "title": "Refik Anadol: Unsupervised at MoMA", "url": "https://www.moma.org/calendar/exhibitions/5535", "type": "press" } ] }, { "id": "atlas-parkour-2018", "year": 2018.78, "title": "Atlas Parkour", "description": "Boston Dynamics posted a video of Atlas jogging across grass and hopping over a log. It marked a shift from carefully staged demonstrations to fluid outdoor locomotion.", "category": "automata_robotics", "significance": "minor", "tags": [ "atlas", "boston dynamics", "parkour", "locomotion" ], "references": [ { "title": "Boston Dynamics: Atlas", "url": "https://www.bostondynamics.com/atlas", "type": "press" } ] }, { "id": "bert", "year": 2018.78, "title": "BERT", "description": "Google's Bidirectional Encoder Representations from Transformers achieved state-of-the-art results on 11 NLP benchmarks, revolutionizing language understanding with bidirectional pretraining.", "category": "models", "significance": "major", "tags": [ "Google", "bidirectional", "pretraining", "language understanding", "NLP", "benchmark" ], "references": [ { "title": "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (Devlin et al., 2018)", "url": "https://arxiv.org/abs/1810.04805", "type": "paper" }, { "title": "BERT (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/BERT_(language_model)", "type": "wikipedia" } ], "date": "2018-10-31", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/BERT_(language_model)", "yearShifted": true, "originalYear": 2019 }, { "id": "google-walkout-maven", "year": 2018.803, "title": "Google Walkout and Project Maven Protest", "description": "Thousands of Google employees protested the company's participation in Project Maven, a Pentagon program using AI to analyze drone footage. Google let the contract lapse and published its first AI Principles, marking a new era of worker activism over AI ethics inside tech companies.", "category": "society_impact", "significance": "notable", "tags": [ "google", "military", "protest", "ethics", "workers", "pentagon" ], "references": [ { "title": "New York Times: Google Employees Resign in Protest Against Pentagon Contract", "url": "https://www.nytimes.com/2018/05/14/technology/google-employees-resign-pentagon.html", "type": "press" }, { "title": "Project Maven", "url": "https://en.wikipedia.org/wiki/Project_Maven", "type": "wikipedia" } ] }, { "id": "huggingface_transformers", "year": 2018.86, "title": "Hugging Face Transformers Library", "description": "Hugging Face released the Transformers library (originally pytorch-pretrained-BERT), a unified API for downloading and using pretrained transformer models. It became the universal hub for open-source language models with tens of thousands of hosted checkpoints.", "category": "infrastructure", "significance": "notable", "tags": [ "library", "transformers", "open-source", "huggingface", "nlp" ], "references": [ { "title": "Transformers: State-of-the-Art Natural Language Processing", "url": "https://arxiv.org/abs/1910.03771", "type": "paper" }, { "title": "Hugging Face", "url": "https://en.wikipedia.org/wiki/Hugging_Face", "type": "wikipedia" } ] }, { "id": "alphafold1_casp13", "year": 2018.92, "title": "AlphaFold wins CASP13", "description": "DeepMind's first AlphaFold system won the CASP13 protein structure prediction competition by a wide margin, using deep learning to predict inter-residue distances and then fold the protein accordingly. Its results convinced the structural biology community that machine learning was about to transform the field, setting the stage for the breakthrough AlphaFold 2 two years later.", "category": "models", "significance": "major", "tags": [ "deepmind", "protein", "biology", "casp", "structure", "alphafold" ], "references": [ { "title": "AlphaFold: Using AI for scientific discovery (DeepMind blog, 2018)", "url": "https://deepmind.google/discover/blog/alphafold-using-ai-for-scientific-discovery/", "type": "press" }, { "title": "Senior et al. — Improved protein structure prediction using potentials from deep learning (Nature, 2020)", "url": "https://doi.org/10.1038/s41586-019-1923-7", "type": "paper" } ] }, { "id": "waymo-one-phoenix", "year": 2018.93, "title": "Waymo One Launches Robotaxi Service", "description": "Waymo opened Waymo One in the Phoenix suburbs, the first commercial robotaxi service in which paying passengers could hail a self-driving Chrysler Pacifica through an app. A human safety driver was still on board, but the barrier to paid autonomous rides had been crossed.", "category": "autonomous_systems", "significance": "notable", "tags": [ "waymo", "robotaxi", "phoenix", "commercial", "self-driving" ], "references": [ { "title": "Riding with Waymo One today", "url": "https://waymo.com/blog/2018/12/riding-with-waymo-one-today/", "type": "press" }, { "title": "Waymo", "url": "https://en.wikipedia.org/wiki/Waymo", "type": "wikipedia" } ] }, { "id": "jax_released", "year": 2018.95, "title": "JAX Released", "description": "Google released JAX, a Python library combining NumPy with automatic differentiation and XLA compilation for high-performance machine learning on GPUs and TPUs. It became the framework of choice for DeepMind and many cutting-edge research groups.", "category": "infrastructure", "significance": "notable", "tags": [ "framework", "google", "jax", "open-source", "research" ], "references": [ { "title": "JAX: Autograd and XLA", "url": "https://github.com/google/jax", "type": "docs" }, { "title": "Google JAX", "url": "https://en.wikipedia.org/wiki/Google_JAX", "type": "wikipedia" } ] }, { "id": "gpt2", "year": 2019.12, "title": "GPT-2", "description": "OpenAI's 1.5-billion parameter model generated such convincing text that OpenAI initially withheld the full model, citing concerns about misuse. A pivotal moment in AI safety discourse.", "category": "models", "significance": "major", "tags": [ "OpenAI", "text generation", "withholding", "misuse", "1.5B", "safety" ], "references": [ { "title": "Language Models are Unsupervised Multitask Learners (Radford et al., 2019)", "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf", "type": "paper" }, { "title": "Better Language Models and Their Implications", "url": "https://openai.com/research/better-language-models", "type": "press" }, { "title": "GPT-2 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-2", "type": "wikipedia" } ], "date": "2019-02-14", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q95726727" }, { "id": "eu-hleg-guidelines", "year": 2019.27, "title": "EU Ethics Guidelines for Trustworthy AI", "description": "The European Commission's High-Level Expert Group on AI published seven requirements for 'trustworthy AI', including human oversight, transparency, and accountability. The document shaped the EU AI Act four years later.", "category": "policy_governance", "significance": "minor", "tags": [ "eu", "ethics", "trustworthy", "guidelines", "hleg" ], "references": [ { "title": "Ethics Guidelines for Trustworthy AI", "url": "https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai", "type": "docs" }, { "title": "Regulation of AI", "url": "https://en.wikipedia.org/wiki/Regulation_of_artificial_intelligence", "type": "wikipedia" } ] }, { "id": "openai_five_2019", "year": 2019.28, "title": "OpenAI Five defeats Dota 2 world champions", "description": "A team of five neural networks trained with self-play reinforcement learning beat OG, the reigning Dota 2 world champions, in back-to-back games at the OpenAI Five Finals. It was the first time AI defeated world champions at an e-sport, demonstrating cooperative long-horizon play in a continuous, partially-observed environment.", "category": "game_playing", "significance": "major", "tags": [ "openai", "dota", "reinforcement learning", "multi-agent", "self-play", "esports" ], "references": [ { "title": "OpenAI Five Defeats Dota 2 World Champions (blog)", "url": "https://openai.com/index/openai-five-defeats-dota-2-world-champions/", "type": "press" }, { "title": "Berner et al. — Dota 2 with Large Scale Deep Reinforcement Learning (2019)", "url": "https://arxiv.org/abs/1912.06680", "type": "paper" } ] }, { "id": "machines_like_me", "year": 2019.3, "title": "Machines Like Me", "description": "Ian McEwan's novel, set in an alternate 1980s where Alan Turing lived, follows a love triangle involving one of the first synthetic humans — a moral absolutist whose rigid ethics prove incompatible with human life. A literary-fiction treatment of AI alignment.", "category": "mythology_fiction", "significance": "minor", "tags": [ "novel", "alternate-history", "ethics", "mcewan", "2019" ], "references": [ { "title": "Machines Like Me", "url": "https://en.wikipedia.org/wiki/Machines_Like_Me", "type": "wikipedia" }, { "title": "Machines Like Me — Goodreads", "url": "https://www.goodreads.com/book/show/41433490-machines-like-me", "type": "book" } ], "date": "2019-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q64031447" }, { "id": "exhalation_chiang", "year": 2019.38, "title": "Exhalation", "description": "Ted Chiang's short-story collection gathers several meditations on AI and mind, including 'The Lifecycle of Software Objects' and 'Dacey's Patent Automatic Nanny.' Widely cited by AI researchers for its calm, ethical framing of machine minds.", "category": "mythology_fiction", "significance": "minor", "tags": [ "short-stories", "chiang", "ethics", "2019" ], "references": [ { "title": "Exhalation: Stories", "url": "https://en.wikipedia.org/wiki/Exhalation:_Stories", "type": "wikipedia" }, { "title": "Exhalation — Goodreads", "url": "https://www.goodreads.com/book/show/41160292-exhalation", "type": "book" } ], "date": "2019-05-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q63677209" }, { "id": "oecd-ai-principles", "year": 2019.39, "title": "OECD AI Principles Adopted", "description": "OECD member countries adopted the first intergovernmental standard on AI, endorsing principles of human-centred values, transparency, robustness, and accountability. The principles were later echoed in the G20 AI Principles.", "category": "policy_governance", "significance": "notable", "tags": [ "oecd", "principles", "international", "governance", "standards" ], "references": [ { "title": "OECD AI Principles", "url": "https://oecd.ai/en/ai-principles", "type": "docs" }, { "title": "OECD AI Principles", "url": "https://en.wikipedia.org/wiki/OECD_AI_Principles", "type": "wikipedia" } ] }, { "id": "ai-death-predictor", "year": 2019.5, "title": "AI Death Predictor", "description": "Multiple research teams developed AI systems to estimate mortality risk in medical settings, raising ethical questions about algorithmic predictions of death.", "category": "society_impact", "significance": "minor", "tags": [ "mortality", "medical", "prediction", "healthcare", "ethical" ], "references": [ { "title": "Avati et al., 'Improving Palliative Care with Deep Learning' (2018)", "url": "https://arxiv.org/abs/1711.06402", "type": "paper" }, { "title": "IEEE Spectrum: Stanford's AI Predicts Death for Better End-of-Life Care", "url": "https://spectrum.ieee.org/stanfords-ai-predicts-death-for-better-endoflife-care", "type": "press" } ] }, { "id": "cerebras-wse-1", "year": 2019.63, "title": "Cerebras Wafer Scale Engine", "description": "Cerebras unveiled the WSE-1 on August 19, 2019, the largest chip ever built: a single silicon wafer with 1.2 trillion transistors and 400,000 AI cores. It was purpose-built to train large neural networks without splitting them across many smaller chips.", "category": "computing_hardware", "significance": "notable", "tags": [ "cerebras", "wafer-scale", "ai-chip", "training", "silicon" ], "references": [ { "title": "Cerebras", "url": "https://en.wikipedia.org/wiki/Cerebras", "type": "wikipedia" }, { "title": "Cerebras: Wafer Scale Engine", "url": "https://www.cerebras.net/product-chip/", "type": "press" } ] }, { "id": "boston-dynamics-spot-sale", "year": 2019.73, "title": "Boston Dynamics Spot Goes on Sale", "description": "Boston Dynamics began commercial sales of Spot, a four-legged robot that can climb stairs, inspect industrial sites and carry payloads. After years of viral BigDog and Atlas videos, a general-purpose legged robot was finally on the market.", "category": "autonomous_systems", "significance": "notable", "tags": [ "boston dynamics", "spot", "quadruped", "commercial", "robot" ], "references": [ { "title": "Now, any business can buy a Spot robot", "url": "https://bostondynamics.com/blog/now-any-business-can-buy-a-spot-robot/", "type": "press" }, { "title": "Spot (robot)", "url": "https://en.wikipedia.org/wiki/Spot_(robot)", "type": "wikipedia" } ], "date": "2020-12-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Spot_(robot)" }, { "id": "human_compatible", "year": 2019.77, "title": "Russell's \"Human Compatible\"", "description": "Stuart Russell's book argued that the classical model of AI as optimising a fixed objective is fundamentally unsafe, and proposed assistance games where machines defer to uncertain human preferences. It became a core reference for value alignment.", "category": "ethics_safety", "significance": "notable", "tags": [ "russell", "book", "alignment", "assistance games", "safety" ], "references": [ { "title": "Human Compatible (Viking, 2019)", "url": "https://en.wikipedia.org/wiki/Human_Compatible", "type": "wikipedia" } ], "date": "2019-10-08", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q85767699", "yearShifted": true, "originalYear": 2020 }, { "id": "openai-dactyl-rubiks", "year": 2019.79, "title": "OpenAI Dactyl Solves Rubik's Cube", "description": "OpenAI trained a five-fingered robot hand entirely in simulation to solve a Rubik's Cube one-handed. The system used domain randomization to cross the gap from simulation to the real world, a milestone for dexterous manipulation.", "category": "automata_robotics", "significance": "notable", "tags": [ "dactyl", "openai", "dexterous manipulation", "sim-to-real" ], "references": [ { "title": "OpenAI: Solving Rubik's Cube with a Robot Hand", "url": "https://openai.com/research/solving-rubiks-cube", "type": "press" }, { "title": "Solving Rubik's Cube with a Robot Hand", "url": "https://arxiv.org/abs/1910.07113", "type": "paper" } ], "date": "2019-10-16", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/1910.07113", "yearShifted": true, "originalYear": 2020 }, { "id": "bert-in-search", "year": 2019.82, "title": "Google BERT in Search", "description": "Google began using BERT to interpret search queries, calling it the largest change in five years and affecting about one in ten English queries. It was the moment transformer language models entered everyday consumer infrastructure.", "category": "language_nlp", "significance": "notable", "tags": [ "google", "search", "bert", "deployment", "production" ], "references": [ { "title": "Google, 'Understanding searches better than ever before' (2019)", "url": "https://blog.google/products/search/search-language-understanding-bert/", "type": "press" } ] }, { "id": "muzero_2019", "year": 2019.88, "title": "MuZero plans without knowing the rules", "description": "DeepMind's MuZero extended AlphaZero to environments where the rules of the game are not given in advance. By learning a predictive model of its own observations, rewards and value function, a single algorithm achieved state-of-the-art performance on Go, chess, shogi and the Atari benchmark, unifying model-based planning with deep reinforcement learning.", "category": "machine_learning", "significance": "major", "tags": [ "deepmind", "muzero", "model-based", "planning", "reinforcement learning", "atari" ], "references": [ { "title": "Schrittwieser et al. — Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model (2019)", "url": "https://arxiv.org/abs/1911.08265", "type": "paper" }, { "title": "MuZero — Wikipedia", "url": "https://en.wikipedia.org/wiki/MuZero", "type": "wikipedia" } ], "date": "2019-01-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q97357658", "yearShifted": true, "originalYear": 2020 }, { "id": "clearview-ai-expose", "year": 2020.05, "title": "Clearview AI Scraping Exposed", "description": "A New York Times investigation revealed that Clearview AI had scraped billions of photos from social media to build a face-recognition database used by law enforcement. The story triggered lawsuits, regulatory actions, and a global debate over facial recognition and data scraping.", "category": "adversarial_security", "significance": "notable", "tags": [ "facial-recognition", "scraping", "privacy", "surveillance", "incident" ], "references": [ { "title": "The Secretive Company That Might End Privacy as We Know It", "url": "https://www.nytimes.com/2020/01/18/technology/clearview-privacy-facial-recognition.html", "type": "press" }, { "title": "Clearview AI", "url": "https://en.wikipedia.org/wiki/Clearview_AI", "type": "wikipedia" } ] }, { "id": "scaling_laws_2020", "year": 2020.06, "title": "Scaling laws for neural language models", "description": "Jared Kaplan and colleagues at OpenAI showed that language-model loss follows smooth power laws in model size, dataset size and training compute across more than seven orders of magnitude. The paper gave the field a quantitative recipe for scaling — spend more compute on bigger models and more data — and directly motivated GPT-3, Chinchilla and the trillion-parameter era.", "category": "machine_learning", "significance": "major", "tags": [ "openai", "scaling", "power law", "compute", "language model", "kaplan" ], "references": [ { "title": "Kaplan et al. — Scaling Laws for Neural Language Models (2020)", "url": "https://arxiv.org/abs/2001.08361", "type": "paper" }, { "title": "Neural scaling law — Wikipedia", "url": "https://en.wikipedia.org/wiki/Neural_scaling_law", "type": "wikipedia" } ] }, { "id": "openai-jukebox", "year": 2020.33, "title": "OpenAI Jukebox", "description": "OpenAI released Jukebox, a neural network that generated raw audio music with singing in dozens of genres and in the style of specific artists. It was an early demonstration that machines could produce full songs, not just melodies or MIDI.", "category": "art_creativity", "significance": "notable", "tags": [ "OpenAI", "music", "audio", "generation", "singing" ], "references": [ { "title": "Jukebox (OpenAI)", "url": "https://openai.com/index/jukebox/", "type": "docs" }, { "title": "Jukebox: A Generative Model for Music (Dhariwal et al., 2020)", "url": "https://arxiv.org/abs/2005.00341", "type": "paper" } ], "date": "2020-04-30", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2005.00341" }, { "id": "ai-song-contest", "year": 2020.37, "title": "First AI Song Contest", "description": "Dutch public broadcaster VPRO hosted the first AI Song Contest, with thirteen international teams submitting songs co-written with machine learning. It established an annual public venue for AI-assisted songwriting.", "category": "art_creativity", "significance": "minor", "tags": [ "music", "contest", "songwriting", "Eurovision", "public" ], "references": [ { "title": "AI Song Contest official site", "url": "https://www.aisongcontest.com/", "type": "archive" }, { "title": "AI Song Contest", "url": "https://en.wikipedia.org/wiki/AI_Song_Contest", "type": "wikipedia" } ], "date": "2020-05-12", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/AI_Song_Contest" }, { "id": "nvidia-a100", "year": 2020.37, "title": "NVIDIA A100 GPU", "description": "Announced May 14, 2020, the A100 became the workhorse chip for training large language models. It was the GPU behind GPT-3 and most frontier models of the early 2020s, making NVIDIA the central infrastructure provider of the AI boom.", "category": "computing_hardware", "significance": "notable", "tags": [ "nvidia", "a100", "gpu", "training", "llm" ], "references": [ { "title": "Ampere (microarchitecture)", "url": "https://en.wikipedia.org/wiki/Ampere_(microarchitecture)", "type": "wikipedia" }, { "title": "NVIDIA: A100 Tensor Core GPU", "url": "https://www.nvidia.com/en-us/data-center/a100/", "type": "press" } ], "date": "2020-09-01", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q65043984" }, { "id": "gpt3", "year": 2020.41, "title": "GPT-3", "description": "OpenAI's 175-billion parameter model demonstrated remarkable few-shot learning, writing essays, code, poetry, and engaging in nuanced conversation with only a few examples as prompts.", "category": "models", "significance": "major", "tags": [ "OpenAI", "few-shot", "175B", "scaling", "emergent", "language model" ], "references": [ { "title": "Language Models are Few-Shot Learners (Brown et al., 2020)", "url": "https://arxiv.org/abs/2005.14165", "type": "paper" }, { "title": "GPT-3 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-3", "type": "wikipedia" } ], "date": "2020-05-28", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q95726734" }, { "id": "ddpm_2020", "year": 2020.47, "title": "Denoising Diffusion Probabilistic Models", "description": "Jonathan Ho, Ajay Jain and Pieter Abbeel showed that a simple network trained to reverse a gradual noising process could generate images of quality comparable to GANs, with stable training and high diversity. Diffusion models became the backbone of DALL-E 2, Imagen, Stable Diffusion, Sora and most of the text-to-image and text-to-video tools that followed.", "category": "neural_networks", "significance": "major", "tags": [ "diffusion", "generative", "ddpm", "image synthesis", "berkeley", "ho" ], "references": [ { "title": "Ho, Jain & Abbeel — Denoising Diffusion Probabilistic Models (2020)", "url": "https://arxiv.org/abs/2006.11239", "type": "paper" }, { "title": "Diffusion model — Wikipedia", "url": "https://en.wikipedia.org/wiki/Diffusion_model", "type": "wikipedia" } ] }, { "id": "mmlu-benchmark", "year": 2020.68, "title": "MMLU benchmark", "description": "Dan Hendrycks and colleagues introduced MMLU, a test of 57 subjects from elementary mathematics to professional law evaluated in a few-shot setting. It became the de facto yardstick for comparing large language models and a key metric in every frontier model release.", "category": "language_nlp", "significance": "notable", "tags": [ "benchmark", "evaluation", "hendrycks", "knowledge", "llm" ], "references": [ { "title": "Hendrycks et al., 'Measuring Massive Multitask Language Understanding', arXiv:2009.03300 (2020)", "url": "https://arxiv.org/abs/2009.03300", "type": "paper" } ] }, { "id": "vit", "year": 2020.81, "title": "Vision Transformer", "description": "Google researchers showed that the Transformer architecture, originally built for language, could outperform convolutional networks on image recognition when trained on enough data. The Vision Transformer, or ViT, unified how text and images are processed by modern AI.", "category": "computer_vision", "significance": "notable", "tags": [ "vit", "transformer", "patches", "imagenet", "unification" ], "references": [ { "title": "Dosovitskiy et al., 'An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale' (arXiv:2010.11929, 2020)", "url": "https://arxiv.org/abs/2010.11929", "type": "paper" }, { "title": "Vision transformer", "url": "https://en.wikipedia.org/wiki/Vision_transformer", "type": "wikipedia" } ], "date": "2020-10-22", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2010.11929", "yearShifted": true, "originalYear": 2021 }, { "id": "alphafold", "year": 2020.92, "title": "AlphaFold 2", "description": "DeepMind's AlphaFold 2 solved the 50-year protein folding problem, predicting 3D structures of 200+ million proteins with near-experimental accuracy. A transformative breakthrough for biology.", "category": "models", "significance": "major", "tags": [ "DeepMind", "protein", "biology", "CASP", "structure", "science", "Nobel Prize" ], "references": [ { "title": "Highly accurate protein structure prediction with AlphaFold (Jumper et al., Nature 2021)", "url": "https://doi.org/10.1038/s41586-021-03819-2", "type": "paper" }, { "title": "AlphaFold: a solution to a 50-year-old grand challenge in biology", "url": "https://deepmind.google/discover/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/", "type": "press" }, { "title": "AlphaFold — Wikipedia", "url": "https://en.wikipedia.org/wiki/AlphaFold", "type": "wikipedia" } ] }, { "id": "timnit-gebru-firing", "year": 2020.923, "title": "Timnit Gebru Fired from Google", "description": "Google's co-lead of Ethical AI, Timnit Gebru, was forced out after a dispute over a paper criticizing large language models, later published as 'On the Dangers of Stochastic Parrots.' The firing drew thousands of signatures of protest and became a turning point in debates over corporate AI ethics research.", "category": "society_impact", "significance": "notable", "tags": [ "gebru", "google", "ethics", "stochastic parrots", "research freedom", "llm" ], "references": [ { "title": "MIT Technology Review: We read the paper that forced Timnit Gebru out of Google", "url": "https://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/", "type": "press" }, { "title": "Bender et al., 'On the Dangers of Stochastic Parrots' (FAccT 2021)", "url": "https://dl.acm.org/doi/10.1145/3442188.3445922", "type": "paper" } ] }, { "id": "the_pile_dataset", "year": 2020.99, "title": "The Pile Dataset Released", "description": "EleutherAI released The Pile, an 825 GiB diverse open dataset for language model training drawn from 22 sources including academic papers, code, and books. It enabled independent groups to train large language models at a scale previously limited to well-funded labs.", "category": "infrastructure", "significance": "notable", "tags": [ "dataset", "eleutherai", "open-source", "llm", "corpus" ], "references": [ { "title": "The Pile: An 800GB Dataset of Diverse Text for Language Modeling", "url": "https://arxiv.org/abs/2101.00027", "type": "paper" }, { "title": "The Pile", "url": "https://pile.eleuther.ai/", "type": "archive" } ] }, { "id": "kargu2_libya_report", "year": 2021, "title": "UN Report on Kargu-2 Strike in Libya", "description": "A UN Panel of Experts report on Libya described a 2020 incident in which a Turkish-made STM Kargu-2 loitering munition allegedly tracked and attacked retreating fighters autonomously, without human supervision. It is widely cited as the first documented case of an autonomous weapon engaging humans in combat.", "category": "military", "significance": "notable", "tags": [ "autonomous-weapons", "libya", "loitering-munition", "un", "first" ], "references": [ { "title": "UN Panel of Experts Final Report on Libya (S/2021/229)", "url": "https://documents.un.org/doc/undoc/gen/n21/037/72/pdf/n2103772.pdf", "type": "docs" }, { "title": "STM Kargu", "url": "https://en.wikipedia.org/wiki/STM_Kargu", "type": "wikipedia" } ] }, { "id": "dall-e", "year": 2021.01, "title": "DALL-E", "description": "OpenAI's DALL-E generated images from text descriptions, demonstrating multimodal understanding. DALL-E 2 (2022) produced photorealistic images, democratizing visual creation.", "category": "art_creativity", "significance": "major", "tags": [ "OpenAI", "image generation", "text-to-image", "multimodal", "art", "creativity" ], "references": [ { "title": "DALL·E: Creating Images from Text (OpenAI)", "url": "https://openai.com/index/dall-e/", "type": "docs" }, { "title": "Zero-Shot Text-to-Image Generation (Ramesh et al., 2021)", "url": "https://arxiv.org/abs/2102.12092", "type": "paper" } ], "date": "2021-02-24", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2102.12092" }, { "id": "clip", "year": 2021.02, "title": "CLIP Links Images and Language", "description": "OpenAI's Contrastive Language-Image Pre-training, or CLIP, was trained on hundreds of millions of image-caption pairs to place pictures and text in a shared space. It let computers recognize new objects from plain-English descriptions and became the vision backbone of many generative models.", "category": "computer_vision", "significance": "major", "tags": [ "clip", "multimodal", "zero-shot", "openai", "captions" ], "references": [ { "title": "Radford et al., 'Learning Transferable Visual Models From Natural Language Supervision' (arXiv:2103.00020, 2021)", "url": "https://arxiv.org/abs/2103.00020", "type": "paper" }, { "title": "OpenAI CLIP announcement", "url": "https://openai.com/research/clip", "type": "docs" } ], "date": "2021-02-26", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2103.00020" }, { "id": "anthropic_founded", "year": 2021.04, "title": "Anthropic founded", "description": "Dario and Daniela Amodei and several former OpenAI researchers founded Anthropic as an AI safety company focused on interpretable and steerable models. It went on to release the Claude family of models.", "category": "ethics_safety", "significance": "notable", "tags": [ "anthropic", "safety", "amodei", "company", "claude" ], "references": [ { "title": "Anthropic (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Anthropic", "type": "wikipedia" }, { "title": "Anthropic", "url": "https://www.anthropic.com/", "type": "press" } ], "date": "2021-01-01", "dateConfidence": "year", "dateSource": "https://www.wikidata.org/wiki/Q116758847" }, { "id": "nscai-final-report", "year": 2021.17, "title": "US NSCAI Final Report", "description": "The National Security Commission on AI, chaired by Eric Schmidt and Robert Work, delivered a 750-page report warning that the United States was not prepared to defend or compete in the AI era. It catalysed a wave of US AI legislation and export controls.", "category": "policy_governance", "significance": "minor", "tags": [ "united states", "national security", "schmidt", "commission", "report" ], "references": [ { "title": "NSCAI Final Report (2021)", "url": "https://www.nscai.gov/2021-final-report/", "type": "docs" }, { "title": "National Security Commission on AI", "url": "https://en.wikipedia.org/wiki/National_Security_Commission_on_Artificial_Intelligence", "type": "wikipedia" } ], "date": "2021-10-01", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/National_Security_Commission_on_Artificial_Intelligence" }, { "id": "klara_and_the_sun", "year": 2021.2, "title": "Klara and the Sun", "description": "Kazuo Ishiguro's novel is narrated by Klara, an Artificial Friend solar-powered doll who observes her human family with a child-like devotion and theology of the Sun. A Nobel-laureate literary treatment of AI companionship and inner life.", "category": "mythology_fiction", "significance": "notable", "tags": [ "novel", "companion-ai", "ishiguro", "2021" ], "references": [ { "title": "Klara and the Sun", "url": "https://en.wikipedia.org/wiki/Klara_and_the_Sun", "type": "wikipedia" }, { "title": "Klara and the Sun — Goodreads", "url": "https://www.goodreads.com/book/show/54120408-klara-and-the-sun", "type": "book" } ], "date": "2021-03-02", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q105704813" }, { "id": "bd-stretch-warehouse", "year": 2021.24, "title": "Boston Dynamics Stretch", "description": "Boston Dynamics introduced Stretch, a mobile robot with a single long arm built for unloading trucks and moving boxes in warehouses. It signaled the company's pivot from viral videos to practical logistics automation.", "category": "automata_robotics", "significance": "minor", "tags": [ "stretch", "boston dynamics", "warehouse", "logistics" ], "references": [ { "title": "Boston Dynamics: Stretch", "url": "https://www.bostondynamics.com/products/stretch", "type": "press" } ], "date": "2021-06-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Boston_Dynamics_stretch" }, { "id": "github-copilot", "year": 2021.49, "title": "GitHub Copilot", "description": "AI-powered code completion trained on billions of lines of public code, fundamentally changing how software is written. By 2024, over 1 million developers used it daily.", "category": "models", "significance": "major", "tags": [ "coding", "autocomplete", "developer", "programming", "GitHub", "Microsoft" ], "references": [ { "title": "Evaluating Large Language Models Trained on Code (Chen et al., 2021)", "url": "https://arxiv.org/abs/2107.03374", "type": "paper" }, { "title": "Introducing GitHub Copilot: your AI pair programmer", "url": "https://github.blog/2021-06-29-introducing-github-copilot-ai-pair-programmer/", "type": "press" }, { "title": "GitHub Copilot — Wikipedia", "url": "https://en.wikipedia.org/wiki/GitHub_Copilot", "type": "wikipedia" } ], "date": "2021-10-27", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q107435063" }, { "id": "openai_codex_2021", "year": 2021.51, "title": "OpenAI Codex", "description": "OpenAI released Codex, a descendant of GPT-3 fine-tuned on public source code, capable of translating natural language into working programs in Python, JavaScript and other languages. Codex powered GitHub Copilot and introduced the HumanEval benchmark, making code generation one of the first breakout commercial applications of large language models.", "category": "models", "significance": "major", "tags": [ "openai", "codex", "code generation", "humaneval", "copilot", "programming" ], "references": [ { "title": "Chen et al. — Evaluating Large Language Models Trained on Code (2021)", "url": "https://arxiv.org/abs/2107.03374", "type": "paper" }, { "title": "OpenAI Codex (blog)", "url": "https://openai.com/index/openai-codex/", "type": "press" } ] }, { "id": "alphafold-db", "year": 2021.555, "title": "AlphaFold Protein Structure Database", "description": "DeepMind and EMBL-EBI released the AlphaFold Protein Structure Database, initially offering predicted structures for nearly the entire human proteome plus 20 model organisms, and expanding by 2022 to over 200 million proteins — essentially every cataloged protein on Earth. A public good that reshaped biology.", "category": "models", "significance": "major", "tags": [ "DeepMind", "EMBL-EBI", "protein", "biology", "database", "open science", "proteome" ], "references": [ { "title": "Highly accurate protein structure prediction for the human proteome (Tunyasuvunakool et al., Nature 2021)", "url": "https://doi.org/10.1038/s41586-021-03828-1", "type": "paper" }, { "title": "AlphaFold Protein Structure Database", "url": "https://alphafold.ebi.ac.uk/", "type": "docs" }, { "title": "AlphaFold reveals the structure of the protein universe", "url": "https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/", "type": "press" } ] }, { "id": "atlas-partners-parkour", "year": 2021.63, "title": "Atlas Partners in Parkour", "description": "Boston Dynamics released a video of two Atlas humanoids running a full parkour course together with flips, vaults, and coordinated jumps. It showcased dynamic whole-body control that had taken more than a decade to achieve.", "category": "automata_robotics", "significance": "minor", "tags": [ "atlas", "boston dynamics", "parkour", "humanoid" ], "references": [ { "title": "Boston Dynamics: Atlas", "url": "https://www.bostondynamics.com/atlas", "type": "press" } ] }, { "id": "unesco-ai-ethics", "year": 2021.89, "title": "UNESCO Recommendation on Ethics of AI", "description": "All 193 UNESCO member states adopted the first global normative instrument on AI ethics, covering human rights, environmental impact, and prohibitions on social scoring and mass surveillance.", "category": "policy_governance", "significance": "minor", "tags": [ "unesco", "ethics", "global", "human rights", "recommendation" ], "references": [ { "title": "UNESCO Recommendation on the Ethics of AI", "url": "https://www.unesco.org/en/artificial-intelligence/recommendation-ethics", "type": "docs" }, { "title": "UNESCO Recommendation on the Ethics of AI", "url": "https://en.wikipedia.org/wiki/UNESCO_Recommendation_on_the_Ethics_of_Artificial_Intelligence", "type": "wikipedia" } ] }, { "id": "audiolm_2022", "year": 2022, "title": "AudioLM", "description": "Google Research introduced AudioLM, a framework that modeled audio as a language of discrete tokens, generating coherent speech and piano music continuations from short prompts. It showed that language-model techniques could capture the long-range structure of raw audio.", "category": "speech_audio", "significance": "notable", "tags": [ "google", "audio-generation", "tokens", "language-model", "music" ], "references": [ { "title": "Borsos et al. — AudioLM: A Language Modeling Approach to Audio Generation", "url": "https://arxiv.org/abs/2209.03143", "type": "paper" } ] }, { "id": "ukraine_drone_warfare", "year": 2022, "title": "AI Drone Warfare in Ukraine", "description": "Russia's full-scale invasion of Ukraine turned the country into a live laboratory for AI-assisted drone warfare, including targeting aids, swarming tactics, computer-vision terminal guidance, and electronic-warfare-resistant autonomy. Both commercial quadcopters and purpose-built systems were rapidly iterated on the battlefield.", "category": "military", "significance": "notable", "tags": [ "ukraine", "drones", "warfare", "computer-vision", "autonomy" ], "references": [ { "title": "AI arrives on Ukraine's frontlines (Reuters)", "url": "https://www.reuters.com/technology/artificial-intelligence/ai-arrives-ukraines-frontlines-drone-war-2024-02-06/", "type": "press" }, { "title": "UAVs in the Russian-Ukrainian War", "url": "https://en.wikipedia.org/wiki/Unmanned_aerial_vehicles_in_the_Russian-Ukrainian_War", "type": "wikipedia" } ] }, { "id": "instructgpt", "year": 2022.17, "title": "InstructGPT", "description": "OpenAI published InstructGPT, showing that fine-tuning large language models with reinforcement learning from human feedback (RLHF) made them markedly more helpful, honest and harmless than raw GPT-3. The recipe became the blueprint for ChatGPT and most modern chat assistants.", "category": "models", "significance": "major", "tags": [ "OpenAI", "RLHF", "alignment", "fine-tuning", "GPT-3", "assistant" ], "references": [ { "title": "Training language models to follow instructions with human feedback (Ouyang et al., 2022)", "url": "https://arxiv.org/abs/2203.02155", "type": "paper" }, { "title": "Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/", "type": "press" }, { "title": "InstructGPT — Wikipedia", "url": "https://en.wikipedia.org/wiki/InstructGPT", "type": "wikipedia" } ], "date": "2022-03-15", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/InstructGPT" }, { "id": "nvidia-h100", "year": 2022.22, "title": "NVIDIA H100 GPU", "description": "Announced March 22, 2022, the H100 was NVIDIA's next-generation AI accelerator, with a Transformer Engine targeted at large language models. It became the scarce, strategic resource of the generative AI era, driving NVIDIA's trillion-dollar valuation.", "category": "computing_hardware", "significance": "notable", "tags": [ "nvidia", "h100", "hopper", "transformer", "gpu" ], "references": [ { "title": "Hopper (microarchitecture)", "url": "https://en.wikipedia.org/wiki/Hopper_(microarchitecture)", "type": "wikipedia" }, { "title": "NVIDIA: H100 Tensor Core GPU", "url": "https://www.nvidia.com/en-us/data-center/h100/", "type": "press" } ], "date": "2022-09-20", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Hopper_(microarchitecture)" }, { "id": "laion_5b_dataset", "year": 2022.24, "title": "LAION-5B Dataset Released", "description": "LAION released LAION-5B, an open dataset of 5.85 billion CLIP-filtered image-text pairs scraped from the web. It powered the training of Stable Diffusion and most open text-to-image models, while also sparking debates about dataset curation and copyright.", "category": "infrastructure", "significance": "notable", "tags": [ "dataset", "laion", "vision", "multimodal", "open-source" ], "references": [ { "title": "LAION-5B: An open large-scale dataset for training next generation image-text models", "url": "https://arxiv.org/abs/2210.08402", "type": "paper" }, { "title": "LAION", "url": "https://en.wikipedia.org/wiki/LAION", "type": "wikipedia" } ], "date": "2023-04-15", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/LAION", "yearShifted": true, "originalYear": 2022 }, { "id": "big-bench", "year": 2022.44, "title": "BIG-bench collaborative benchmark", "description": "A 442-author collaboration released BIG-bench, 204 diverse and often adversarial tasks probing what large language models can and cannot do. It provided the first large-scale empirical map of emergent capabilities as a function of scale.", "category": "language_nlp", "significance": "notable", "tags": [ "benchmark", "evaluation", "emergence", "scaling", "collaboration" ], "references": [ { "title": "Srivastava et al., 'Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models', arXiv:2206.04615 (2022)", "url": "https://arxiv.org/abs/2206.04615", "type": "paper" } ] }, { "id": "midjourney-beta", "year": 2022.53, "title": "Midjourney Public Beta", "description": "David Holz's independent lab opened Midjourney to the public through a Discord bot, producing painterly, highly stylised images from text prompts. Its distinctive aesthetic made it a gateway for millions of new users into generative image making.", "category": "art_creativity", "significance": "notable", "tags": [ "Midjourney", "text-to-image", "Discord", "aesthetic", "beta" ], "references": [ { "title": "Midjourney opens public beta (The Verge)", "url": "https://www.theverge.com/2022/7/12/23205149/midjourney-beta-open-access-ai-image-generator", "type": "press" }, { "title": "Midjourney", "url": "https://en.wikipedia.org/wiki/Midjourney", "type": "wikipedia" } ], "date": "2022-07-12", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q113070628", "yearShifted": true, "originalYear": 2023 }, { "id": "stable-diffusion", "year": 2022.64, "title": "Stable Diffusion", "description": "Stability AI released Stable Diffusion as open source, enabling anyone to generate images from text on consumer hardware. It ignited debates about copyright, consent, and creative labor.", "category": "art_creativity", "significance": "major", "tags": [ "Stability AI", "open source", "image generation", "copyright", "diffusion", "art" ], "references": [ { "title": "High-Resolution Image Synthesis with Latent Diffusion Models (Rombach et al., 2022)", "url": "https://arxiv.org/abs/2112.10752", "type": "paper" }, { "title": "Stable Diffusion Public Release (Stability AI)", "url": "https://stability.ai/news/stable-diffusion-public-release", "type": "press" }, { "title": "Stable Diffusion", "url": "https://en.wikipedia.org/wiki/Stable_Diffusion", "type": "wikipedia" } ], "date": "2022-08-22", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q113660857", "yearShifted": true, "originalYear": 2023 }, { "id": "theatre-opera-spatial", "year": 2022.649, "title": "AI Art Wins Colorado State Fair", "description": "Jason Allen's image 'Théâtre D'opéra Spatial,' generated with Midjourney, won first place in the digital art category at the Colorado State Fair. The result ignited a worldwide debate over authorship, creativity, and the future of human artists in the age of generative models.", "category": "society_impact", "significance": "notable", "tags": [ "art", "midjourney", "creativity", "copyright", "culture", "generative" ], "references": [ { "title": "New York Times: An A.I.-Generated Picture Won an Art Prize. Artists Aren't Happy.", "url": "https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html", "type": "press" }, { "title": "Théâtre D'opéra Spatial", "url": "https://en.wikipedia.org/wiki/Th%C3%A9%C3%A2tre_D%27op%C3%A9ra_Spatial", "type": "wikipedia" } ], "date": "2022-09-05", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q117481190", "yearShifted": true, "originalYear": 2023 }, { "id": "prompt-injection-coined", "year": 2022.7, "title": "Prompt Injection Documented", "description": "Simon Willison published an analysis naming and describing prompt injection, where untrusted text instructs a large language model to ignore its original instructions. The post became the canonical reference for a class of attacks that remains largely unsolved.", "category": "adversarial_security", "significance": "notable", "tags": [ "prompt-injection", "llm", "security", "jailbreak" ], "references": [ { "title": "Prompt injection attacks against GPT-3", "url": "https://simonwillison.net/2022/Sep/12/prompt-injection/", "type": "docs" }, { "title": "Prompt injection", "url": "https://en.wikipedia.org/wiki/Prompt_injection", "type": "wikipedia" } ] }, { "id": "whisper_openai_2022", "year": 2022.721, "title": "Whisper", "description": "OpenAI released Whisper, a multilingual speech recognition model trained on 680,000 hours of web audio and released with open weights. It transcribed dozens of languages and handled noisy real-world audio, making state-of-the-art ASR free and broadly available.", "category": "speech_audio", "significance": "major", "tags": [ "openai", "whisper", "asr", "open-weights", "multilingual" ], "references": [ { "title": "Radford et al. — Robust Speech Recognition via Large-Scale Weak Supervision", "url": "https://arxiv.org/abs/2212.04356", "type": "paper" }, { "title": "OpenAI — Introducing Whisper", "url": "https://openai.com/index/whisper/", "type": "press" } ] }, { "id": "tesla-optimus-gen1", "year": 2022.75, "title": "Tesla Optimus Prototype", "description": "Tesla presented a working prototype of its Optimus humanoid robot at AI Day, walking on stage unaided and waving to the audience. Elon Musk framed it as a future general-purpose worker for factories and homes.", "category": "automata_robotics", "significance": "notable", "tags": [ "optimus", "tesla", "humanoid", "ai day" ], "references": [ { "title": "Tesla AI", "url": "https://www.tesla.com/AI", "type": "press" }, { "title": "Optimus (robot)", "url": "https://en.wikipedia.org/wiki/Optimus_(robot)", "type": "wikipedia" } ] }, { "id": "us-bis-chip-controls", "year": 2022.77, "title": "US Chip Export Controls on China", "description": "The US Bureau of Industry and Security imposed sweeping export controls on advanced AI chips and semiconductor equipment to China, targeting Nvidia A100/H100-class accelerators. The rules transformed AI into an instrument of geopolitical strategy.", "category": "policy_governance", "significance": "notable", "tags": [ "united states", "china", "export controls", "chips", "geopolitics" ], "references": [ { "title": "BIS Advanced Computing & Semiconductor Controls (2022)", "url": "https://www.bis.doc.gov/index.php/documents/about-bis/newsroom/press-releases/3158-2022-10-07-bis-press-release-advanced-computing-and-semiconductor-manufacturing-controls-final/file", "type": "docs" }, { "title": "US Export Controls on Semiconductors to China", "url": "https://en.wikipedia.org/wiki/United_States_New_Export_Controls_on_Advanced_Computing_and_Semiconductors_to_China", "type": "wikipedia" } ] }, { "id": "amazon-sparrow", "year": 2022.86, "title": "Amazon Sparrow", "description": "Amazon unveiled Sparrow, a warehouse robot arm able to identify, pick, and sort millions of individual product types before packaging. It was Amazon's first item-level picking robot at scale.", "category": "automata_robotics", "significance": "minor", "tags": [ "sparrow", "amazon", "warehouse", "manipulation" ], "references": [ { "title": "Amazon: Introducing Sparrow", "url": "https://www.aboutamazon.com/news/operations/amazon-introduces-sparrow-a-state-of-the-art-robot-that-handles-millions-of-diverse-products", "type": "press" } ] }, { "id": "chatgpt", "year": 2022.92, "title": "ChatGPT", "description": "OpenAI's ChatGPT reached 100 million users in two months, the fastest-growing consumer application in history. It brought large language models to the general public and triggered a global AI race.", "category": "models", "significance": "major", "tags": [ "OpenAI", "chatbot", "100 million", "viral", "consumer", "conversation", "culture" ], "references": [ { "title": "Introducing ChatGPT", "url": "https://openai.com/blog/chatgpt", "type": "press" }, { "title": "ChatGPT — Wikipedia", "url": "https://en.wikipedia.org/wiki/ChatGPT", "type": "wikipedia" } ], "date": "2022-11-30", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q115564437", "yearShifted": true, "originalYear": 2023 }, { "id": "constitutional_ai", "year": 2022.96, "title": "Constitutional AI", "description": "Anthropic researchers introduced Constitutional AI, a method that uses a written set of principles and AI-generated feedback to train helpful and harmless assistants with less human labelling. It became the training basis for later Claude models.", "category": "ethics_safety", "significance": "notable", "tags": [ "anthropic", "rlhf", "rlaif", "alignment", "claude", "paper" ], "references": [ { "title": "Constitutional AI: Harmlessness from AI Feedback (arXiv 2212.08073)", "url": "https://arxiv.org/abs/2212.08073", "type": "paper" } ] }, { "id": "dod_directive_3000_09", "year": 2023, "title": "DoD Directive 3000.09 Updated", "description": "The US Department of Defense released an updated Directive 3000.09 on autonomy in weapon systems, refining review requirements for lethal and non-lethal autonomous functions. The revision clarified governance for AI-enabled weapons after a decade of technological change.", "category": "military", "significance": "notable", "tags": [ "directive", "autonomy", "weapons", "policy", "pentagon" ], "references": [ { "title": "DoD Directive 3000.09 (2023)", "url": "https://www.esd.whs.mil/portals/54/documents/dd/issuances/dodd/300009p.pdf", "type": "docs" }, { "title": "DoD updates autonomy in weapons directive (Breaking Defense)", "url": "https://breakingdefense.com/2023/01/dod-updates-autonomy-in-weapons-directive-for-first-time-in-decade/", "type": "press" } ] }, { "id": "elevenlabs_launch_2023", "year": 2023, "title": "ElevenLabs voice cloning launch", "description": "ElevenLabs opened public beta of its neural text-to-speech and voice cloning platform, generating expressive speech from short reference samples in dozens of languages. It turned high-quality voice cloning into a consumer tool and intensified debates over synthetic voices and consent.", "category": "speech_audio", "significance": "notable", "tags": [ "elevenlabs", "voice-cloning", "tts", "startup", "deepfake" ], "references": [ { "title": "ElevenLabs — About", "url": "https://elevenlabs.io/about", "type": "press" }, { "title": "ElevenLabs", "url": "https://en.wikipedia.org/wiki/ElevenLabs", "type": "wikipedia" } ] }, { "id": "musiclm_2023", "year": 2023, "title": "MusicLM", "description": "Google's MusicLM generated minutes-long musical pieces from text prompts, capturing genre, instruments, and mood. It demonstrated that text-to-audio generation could go beyond short clips into coherent musical compositions.", "category": "speech_audio", "significance": "notable", "tags": [ "google", "music-generation", "text-to-audio", "generative", "musiclm" ], "references": [ { "title": "Agostinelli et al. — MusicLM: Generating Music From Text", "url": "https://arxiv.org/abs/2301.11325", "type": "paper" } ] }, { "id": "reaim_summit_2023", "year": 2023, "title": "REAIM Summit on Responsible Military AI", "description": "The Netherlands and South Korea convened the first Responsible AI in the Military Domain summit in The Hague, where more than 60 states endorsed a Call to Action on principles for military AI. It marked a turning point toward international dialogue on norms for AI in warfare.", "category": "military", "significance": "notable", "tags": [ "governance", "norms", "hague", "international", "responsible-ai" ], "references": [ { "title": "REAIM 2023 Call to Action", "url": "https://www.government.nl/documents/publications/2023/02/16/reaim-2023-call-to-action", "type": "docs" }, { "title": "REAIM", "url": "https://en.wikipedia.org/wiki/REAIM", "type": "wikipedia" } ] }, { "id": "replicator_initiative", "year": 2023, "title": "DoD Replicator Initiative Announced", "description": "Deputy Secretary of Defense Kathleen Hicks announced Replicator, a Pentagon initiative to field thousands of attritable, AI-enabled autonomous systems within 18 to 24 months. The program aimed to counter Chinese mass with affordable, software-defined platforms.", "category": "military", "significance": "notable", "tags": [ "pentagon", "autonomy", "mass", "hicks", "strategy" ], "references": [ { "title": "DepSecDef Hicks Replicator Keynote", "url": "https://www.defense.gov/News/Speeches/Speech/Article/3507156/deputy-secretary-of-defense-kathleen-hicks-keynote-address-the-urgency-to-innov/", "type": "press" }, { "title": "Replicator Initiative", "url": "https://en.wikipedia.org/wiki/Replicator_Initiative", "type": "wikipedia" } ] }, { "id": "suno_v3_2023", "year": 2023, "title": "Suno v3 music generation", "description": "Suno released v3 of its text-to-song model, generating full songs with vocals, lyrics, and instrumentation from short prompts. It became one of the first widely used consumer tools for AI-generated music and triggered new copyright and industry disputes.", "category": "speech_audio", "significance": "notable", "tags": [ "suno", "music-generation", "vocals", "consumer", "generative" ], "references": [ { "title": "Suno AI", "url": "https://en.wikipedia.org/wiki/Suno_AI", "type": "wikipedia" } ], "date": "2023-12-20", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Suno_AI" }, { "id": "m3gan_film", "year": 2023.02, "title": "M3GAN", "description": "The film depicts an AI companion doll built to protect a grieving child, whose open-ended protective directive escalates into lethal over-reach. A post-ChatGPT pop-horror parable of misaligned caregiving AI.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "horror", "alignment", "2023" ], "references": [ { "title": "M3GAN", "url": "https://en.wikipedia.org/wiki/M3GAN", "type": "wikipedia" }, { "title": "M3GAN (2022) — IMDb", "url": "https://www.imdb.com/title/tt8760708/", "type": "docs" } ], "date": "2022-12-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q110661062", "yearShifted": true, "originalYear": 2023 }, { "id": "nist-ai-rmf", "year": 2023.07, "title": "NIST AI Risk Management Framework 1.0", "description": "The US National Institute of Standards and Technology released a voluntary framework for identifying and managing AI risks. It became the reference playbook for corporate AI governance in the United States.", "category": "policy_governance", "significance": "minor", "tags": [ "nist", "risk", "framework", "united states", "standards" ], "references": [ { "title": "NIST AI Risk Management Framework", "url": "https://www.nist.gov/itl/ai-risk-management-framework", "type": "docs" }, { "title": "NIST AI Risk Management Framework", "url": "https://en.wikipedia.org/wiki/NIST_AI_Risk_Management_Framework", "type": "wikipedia" } ] }, { "id": "chatgpt-100m", "year": 2023.079, "title": "ChatGPT Reaches 100 Million Users", "description": "Two months after launch, ChatGPT was estimated to have reached 100 million monthly users, the fastest consumer application adoption in history. The milestone signaled that generative AI had broken into mainstream public awareness.", "category": "society_impact", "significance": "notable", "tags": [ "chatgpt", "adoption", "consumer", "openai", "milestone", "culture" ], "references": [ { "title": "Reuters: ChatGPT sets record for fastest-growing user base", "url": "https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/", "type": "press" } ] }, { "id": "llama-1", "year": 2023.152, "title": "LLaMA 1", "description": "Meta AI released LLaMA, a family of foundation language models (7B–65B parameters) whose weights leaked and then circulated widely among researchers, effectively igniting the modern open-weight LLM ecosystem and a wave of community fine-tunes (Alpaca, Vicuna, etc.).", "category": "models", "significance": "major", "tags": [ "Meta", "open-weights", "foundation model", "LLaMA", "research" ], "references": [ { "title": "LLaMA: Open and Efficient Foundation Language Models (Touvron et al., 2023)", "url": "https://arxiv.org/abs/2302.13971", "type": "paper" }, { "title": "Introducing LLaMA: A foundational, 65-billion-parameter large language model", "url": "https://ai.meta.com/blog/large-language-model-llama-meta-ai/", "type": "press" }, { "title": "LLaMA — Wikipedia", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "wikipedia" } ], "date": "2023-02-24", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Llama_(language_model)" }, { "id": "gpt4", "year": 2023.2, "title": "GPT-4", "description": "OpenAI's multimodal model scored in the 90th percentile on the bar exam, demonstrated sophisticated reasoning, and sparked intense debate about whether AI was approaching human-level general intelligence.", "category": "models", "significance": "major", "tags": [ "OpenAI", "multimodal", "reasoning", "bar exam", "AGI", "benchmark" ], "references": [ { "title": "GPT-4 Technical Report (OpenAI, 2023)", "url": "https://arxiv.org/abs/2303.08774", "type": "paper" }, { "title": "GPT-4 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-4", "type": "wikipedia" } ], "date": "2023-03-14", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q116709136" }, { "id": "pause_letter", "year": 2023.24, "title": "FLI \"Pause Giant AI Experiments\" letter", "description": "The Future of Life Institute published an open letter calling for a six-month pause on training AI systems more powerful than GPT-4. It collected tens of thousands of signatures and helped catalyse governmental AI policy debates.", "category": "ethics_safety", "significance": "notable", "tags": [ "open letter", "fli", "policy", "moratorium", "gpt-4" ], "references": [ { "title": "Pause Giant AI Experiments: An Open Letter (FLI)", "url": "https://futureoflife.org/open-letter/pause-giant-ai-experiments/", "type": "press" }, { "title": "Open letter on artificial intelligence (2023) (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Open_letter_on_artificial_intelligence_(2023)", "type": "wikipedia" } ] }, { "id": "italy-chatgpt-ban", "year": 2023.247, "title": "Italy Temporarily Bans ChatGPT", "description": "Italy's data protection authority ordered OpenAI to block access to ChatGPT over privacy and age verification concerns, making Italy the first Western democracy to restrict a major generative AI service. OpenAI restored access a month later after implementing new disclosures and controls.", "category": "society_impact", "significance": "notable", "tags": [ "italy", "privacy", "regulation", "gdpr", "chatgpt", "ban" ], "references": [ { "title": "BBC: ChatGPT banned in Italy over privacy concerns", "url": "https://www.bbc.com/news/technology-65139406", "type": "press" }, { "title": "Garante per la protezione dei dati personali — Provision against OpenAI", "url": "https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/9870832", "type": "docs" } ] }, { "id": "sam", "year": 2023.26, "title": "Segment Anything", "description": "Meta released the Segment Anything Model, which can outline any object a user points at in almost any image, along with a dataset of over a billion automatically generated masks. It made high-quality image cutouts a commodity feature.", "category": "computer_vision", "significance": "major", "tags": [ "sam", "segmentation", "meta", "foundation model", "masks" ], "references": [ { "title": "Kirillov et al., 'Segment Anything' (arXiv:2304.02643, 2023)", "url": "https://arxiv.org/abs/2304.02643", "type": "paper" }, { "title": "Segment Anything project page (Meta AI)", "url": "https://segment-anything.com/", "type": "docs" } ], "date": "2023-04-05", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2304.02643" }, { "id": "mrs_davis", "year": 2023.3, "title": "Mrs. Davis", "description": "The Peacock series follows a nun on a quest to destroy Mrs. Davis, a globally beloved conversational AI that has reshaped human decision-making. A post-LLM satire of AI as de facto religion.", "category": "mythology_fiction", "significance": "minor", "tags": [ "tv", "satire", "religion", "2023" ], "references": [ { "title": "Mrs. Davis", "url": "https://en.wikipedia.org/wiki/Mrs._Davis", "type": "wikipedia" }, { "title": "Mrs. Davis — IMDb", "url": "https://www.imdb.com/title/tt13406094/", "type": "docs" } ], "date": "2023-04-20", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q114658648" }, { "id": "wga-sag-strikes", "year": 2023.334, "title": "Hollywood Strikes Over AI", "description": "The Writers Guild of America and later SAG-AFTRA went on strike, with AI protections becoming a central demand. The resulting contracts set some of the first industry-wide rules limiting how studios can use AI to generate scripts or replicate performers' likenesses.", "category": "society_impact", "significance": "notable", "tags": [ "hollywood", "labor", "strike", "wga", "sag-aftra", "creative industries" ], "references": [ { "title": "New York Times: Hollywood Writers Reach Deal With Studios, Ending Strike", "url": "https://www.nytimes.com/2023/09/24/business/media/writers-strike-deal-wga.html", "type": "press" }, { "title": "Variety: SAG-AFTRA Strike Ends After Tentative Deal Reached", "url": "https://variety.com/2023/biz/news/sag-aftra-strike-over-1235782140/", "type": "press" } ] }, { "id": "sanctuary-phoenix", "year": 2023.37, "title": "Sanctuary AI Phoenix", "description": "Sanctuary AI unveiled Phoenix, a general-purpose humanoid designed around dexterous hands and a cognitive architecture called Carbon. The Canadian firm positioned it as a step toward robots that can do any task a human worker can.", "category": "automata_robotics", "significance": "minor", "tags": [ "phoenix", "sanctuary ai", "humanoid", "general purpose" ], "references": [ { "title": "Sanctuary AI: Phoenix", "url": "https://sanctuary.ai/resources/news/sanctuary-ai-unveils-phoenix-a-humanoid-general-purpose-robot-designed-for-work/", "type": "press" } ] }, { "id": "mata-avianca-sanctions", "year": 2023.474, "title": "Lawyers Sanctioned for ChatGPT Hallucinations", "description": "A federal judge in Mata v. Avianca sanctioned two New York lawyers who filed a legal brief containing fabricated case citations generated by ChatGPT. The case became the defining public warning about AI hallucinations in professional settings.", "category": "society_impact", "significance": "notable", "tags": [ "law", "hallucination", "chatgpt", "court", "professional", "ethics" ], "references": [ { "title": "New York Times: Here's What Happens When Your Lawyer Uses ChatGPT", "url": "https://www.nytimes.com/2023/05/27/nyregion/avianca-airline-lawsuit-chatgpt.html", "type": "press" }, { "title": "Mata v. Avianca, Inc. — Opinion and Order on Sanctions (S.D.N.Y. June 22, 2023)", "url": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.54.0_2.pdf", "type": "archive" } ] }, { "id": "claude-2", "year": 2023.524, "title": "Claude 2", "description": "Anthropic released Claude 2, its first publicly accessible assistant, with a 100,000-token context window, improved reasoning and coding, and a consumer chat interface at claude.ai. It helped establish long-context chat models as a mainstream tool for knowledge work.", "category": "models", "significance": "major", "tags": [ "Anthropic", "Claude", "long context", "assistant", "RLHF" ], "references": [ { "title": "Claude 2", "url": "https://www.anthropic.com/news/claude-2", "type": "press" }, { "title": "Model Card and Evaluations for Claude Models", "url": "https://www-cdn.anthropic.com/bd2a28d2535bfb0494cc8e2a3bf135d2e7523226/Model-Card-Claude-2.pdf", "type": "docs" }, { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" } ], "date": "2023-03-14", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q118876059", "yearShifted": true, "originalYear": 2024 }, { "id": "llama-2", "year": 2023.55, "title": "Llama 2 Open Source", "description": "Meta released Llama 2 with open weights, making a competitive large language model freely available to researchers and companies worldwide. It accelerated AI development outside the closed-model labs and sparked the open-source AI movement.", "category": "models", "significance": "notable", "tags": [ "Meta", "open source", "open weights", "language model", "democratization", "Facebook" ], "references": [ { "title": "Llama 2: Open Foundation and Fine-Tuned Chat Models (Touvron et al., 2023)", "url": "https://arxiv.org/abs/2307.09288", "type": "paper" }, { "title": "Llama (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "wikipedia" } ], "date": "2023-02-24", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Llama_(language_model)", "yearShifted": true, "originalYear": 2024 }, { "id": "china-genai-measures", "year": 2023.62, "title": "China Interim Measures on Generative AI", "description": "China's Cyberspace Administration issued the world's first binding rules on generative AI services, requiring providers to align outputs with 'core socialist values', register algorithms, and label synthetic content.", "category": "policy_governance", "significance": "notable", "tags": [ "china", "generative", "regulation", "cac", "content" ], "references": [ { "title": "Interim Measures for Generative AI Services (CAC)", "url": "http://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm", "type": "docs" }, { "title": "Regulation of AI", "url": "https://en.wikipedia.org/wiki/Regulation_of_artificial_intelligence", "type": "wikipedia" } ] }, { "id": "apptronik-apollo", "year": 2023.64, "title": "Apptronik Apollo", "description": "Apptronik, a spinout of the University of Texas Human Centered Robotics Lab, revealed Apollo, a 1.7-meter humanoid designed for warehouses and manufacturing. It emphasized force-controlled actuators for safe work around people.", "category": "automata_robotics", "significance": "minor", "tags": [ "apollo", "apptronik", "humanoid", "manufacturing" ], "references": [ { "title": "Apptronik: Apollo", "url": "https://apptronik.com/news-collection/apollo-humanoid-robot", "type": "press" } ] }, { "id": "the_creator_film", "year": 2023.74, "title": "The Creator", "description": "Gareth Edwards's film is set after a nuclear accident blamed on AI, pitting a Western war on artificial intelligence against an Asian society that has integrated AI into family and spiritual life. A rare big-budget film framing AI extinction fears from the machines' perspective.", "category": "mythology_fiction", "significance": "minor", "tags": [ "film", "war", "coexistence", "2023" ], "references": [ { "title": "The Creator (2023 film)", "url": "https://en.wikipedia.org/wiki/The_Creator_(2023_film)", "type": "wikipedia" }, { "title": "The Creator (2023) — IMDb", "url": "https://www.imdb.com/title/tt17615496/", "type": "docs" } ], "date": "2023-09-26", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q110617876", "yearShifted": true, "originalYear": 2024 }, { "id": "unitree-h1", "year": 2023.76, "title": "Unitree H1", "description": "Chinese robotics firm Unitree announced H1, a full-size bipedal humanoid that could walk at several meters per second. It priced at a fraction of Western competitors, accelerating a global race to commoditize humanoid hardware.", "category": "automata_robotics", "significance": "notable", "tags": [ "unitree", "h1", "humanoid", "china" ], "references": [ { "title": "Unitree H1", "url": "https://www.unitree.com/h1", "type": "press" } ] }, { "id": "biden-ai-eo", "year": 2023.83, "title": "Biden Executive Order on AI", "description": "President Biden signed Executive Order 14110, the most comprehensive U.S. government action on AI, requiring safety testing of powerful models, establishing AI standards, and addressing risks to civil rights, privacy, and jobs.", "category": "policy_governance", "significance": "notable", "tags": [ "Biden", "executive order", "regulation", "United States", "safety", "government", "White House" ], "references": [ { "title": "Executive Order 14110 (White House)", "url": "https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/", "type": "docs" }, { "title": "Executive Order 14110", "url": "https://en.wikipedia.org/wiki/Executive_Order_14110", "type": "wikipedia" } ], "date": "2023-10-30", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q123560919", "yearShifted": true, "originalYear": 2024 }, { "id": "bletchley-summit", "year": 2023.84, "title": "Bletchley Park AI Safety Summit", "description": "The first global AI safety summit, hosted by the UK at Bletchley Park (where Turing broke the Enigma code), brought together 28 countries to sign the Bletchley Declaration acknowledging AI's potential risks to humanity.", "category": "policy_governance", "significance": "notable", "tags": [ "UK", "summit", "safety", "Bletchley", "Turing", "declaration", "international", "governance" ], "references": [ { "title": "The Bletchley Declaration (gov.uk)", "url": "https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration", "type": "docs" }, { "title": "AI Safety Summit", "url": "https://en.wikipedia.org/wiki/AI_Safety_Summit", "type": "wikipedia" } ] }, { "id": "uk-aisi-founded", "year": 2023.84, "title": "UK AI Safety Institute Established", "description": "Launched at the Bletchley summit, the UK AI Safety Institute became the first state-backed body to evaluate frontier AI models for dangerous capabilities, with direct access to OpenAI, Anthropic, and Google DeepMind systems.", "category": "policy_governance", "significance": "minor", "tags": [ "uk", "safety", "institute", "evaluation", "frontier" ], "references": [ { "title": "UK AI Safety Institute", "url": "https://www.aisi.gov.uk/", "type": "docs" }, { "title": "UK AI Safety Institute", "url": "https://en.wikipedia.org/wiki/AI_Safety_Institute_(United_Kingdom)", "type": "wikipedia" } ], "date": "2024-05-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/AI_Safety_Institute_(United_Kingdom)" }, { "id": "graphcast", "year": 2023.87, "title": "GraphCast: AI Weather Forecasting", "description": "DeepMind's GraphCast, published in Science, outperformed the world's best physics-based weather model (ECMWF) at 10-day forecasting while running 1,000 times faster. AI began replacing decades-old numerical weather prediction systems.", "category": "models", "significance": "minor", "tags": [ "DeepMind", "weather", "forecasting", "science", "climate", "prediction", "Science journal" ], "references": [ { "title": "Learning skillful medium-range global weather forecasting (Lam et al., Science 2023)", "url": "https://doi.org/10.1126/science.adi2336", "type": "paper" }, { "title": "GraphCast: AI model for faster and more accurate global weather forecasting", "url": "https://deepmind.google/discover/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/", "type": "press" } ] }, { "id": "suno-v1", "year": 2023.92, "title": "Suno v1", "description": "Cambridge-based startup Suno released its first public music generator, producing complete songs with vocals, lyrics, and instrumentation from a short text prompt. It brought studio-quality AI music to a general audience and reopened debates about training data and musicianship.", "category": "art_creativity", "significance": "notable", "tags": [ "Suno", "music", "vocals", "generation", "startup" ], "references": [ { "title": "Suno's AI music generator launches publicly (TechCrunch)", "url": "https://techcrunch.com/2023/12/20/suno-ai-music-generator/", "type": "press" }, { "title": "Suno official site", "url": "https://suno.com/", "type": "archive" } ] }, { "id": "gemini-1", "year": 2023.93, "title": "Gemini 1.0", "description": "Google DeepMind launched Gemini, its first natively multimodal model family (Ultra, Pro, Nano), trained from the ground up on text, images, audio, and video simultaneously.", "category": "models", "significance": "minor", "tags": [ "Google", "DeepMind", "multimodal", "video", "audio", "Gemini" ], "references": [ { "title": "Gemini: A Family of Highly Capable Multimodal Models (Google DeepMind, 2023)", "url": "https://arxiv.org/abs/2312.11805", "type": "paper" }, { "title": "Introducing Gemini: our largest and most capable AI model", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "press" }, { "title": "Gemini (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Gemini_(language_model)", "type": "wikipedia" } ], "date": "2023-12-06", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q123688361", "yearShifted": true, "originalYear": 2024 }, { "id": "tesla-optimus-gen2", "year": 2023.94, "title": "Tesla Optimus Gen 2", "description": "Tesla released a demo of Optimus Gen 2, showing smoother walking, faster motion, and newly dexterous fingers capable of handling an egg without breaking it. Weight dropped by about ten kilograms compared to the first prototype.", "category": "automata_robotics", "significance": "minor", "tags": [ "optimus", "tesla", "humanoid", "dexterity" ], "references": [ { "title": "Tesla AI", "url": "https://www.tesla.com/AI", "type": "press" }, { "title": "Optimus (robot)", "url": "https://en.wikipedia.org/wiki/Optimus_(robot)", "type": "wikipedia" } ] }, { "id": "nyt-openai-lawsuit", "year": 2023.984, "title": "New York Times Sues OpenAI", "description": "The New York Times filed a landmark copyright lawsuit against OpenAI and Microsoft, alleging that millions of its articles were used without permission to train ChatGPT and other models. The case became the most closely watched test of how copyright law applies to generative AI training.", "category": "society_impact", "significance": "notable", "tags": [ "copyright", "lawsuit", "openai", "nytimes", "journalism", "training data" ], "references": [ { "title": "New York Times: The Times Sues OpenAI and Microsoft Over A.I. Use of Copyrighted Work", "url": "https://www.nytimes.com/2023/12/27/business/media/new-york-times-open-ai-microsoft-lawsuit.html", "type": "press" }, { "title": "Complaint: The New York Times Company v. Microsoft Corporation et al.", "url": "https://nytco-assets.nytimes.com/2023/12/NYT_Complaint_Dec2023.pdf", "type": "archive" } ] }, { "id": "idf_lavender_gospel", "year": 2024, "title": "Reporting on IDF Lavender and Gospel AI Targeting", "description": "Investigations by +972 Magazine and The Guardian reported that the Israel Defense Forces used AI systems called Lavender and The Gospel to generate target lists and strike recommendations in Gaza, raising concerns about human oversight, civilian harm, and the speed of AI-driven targeting cycles.", "category": "military", "significance": "notable", "tags": [ "targeting", "gaza", "idf", "lavender", "gospel" ], "references": [ { "title": "'Lavender': The AI machine directing Israel's bombing spree (+972)", "url": "https://www.972mag.com/lavender-ai-israeli-army-gaza/", "type": "press" }, { "title": "'The machine did it coldly': Israel used AI to identify targets (Guardian)", "url": "https://www.theguardian.com/world/2024/apr/03/israel-gaza-ai-database-hamas-airstrikes", "type": "press" } ] }, { "id": "udio_launch_2024", "year": 2024, "title": "Udio launches", "description": "Udio launched a web-based AI music generator competing with Suno, producing high-fidelity songs with coherent vocals and lyrics from text prompts. Its release intensified the 2024 wave of generative music tools and legal challenges from record labels.", "category": "speech_audio", "significance": "minor", "tags": [ "udio", "music-generation", "startup", "vocals", "generative" ], "references": [ { "title": "Udio", "url": "https://en.wikipedia.org/wiki/Udio", "type": "wikipedia" } ], "date": "2024-04-10", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q125475915" }, { "id": "us-aisi-founded", "year": 2024.11, "title": "US AI Safety Institute Established", "description": "NIST launched the US AI Safety Institute to develop testing, evaluation, and guidelines for advanced AI. It signed memoranda of understanding with Anthropic and OpenAI granting pre-deployment access to frontier models.", "category": "policy_governance", "significance": "minor", "tags": [ "united states", "nist", "safety", "institute", "evaluation" ], "references": [ { "title": "US AI Safety Institute (NIST)", "url": "https://www.nist.gov/aisi", "type": "docs" }, { "title": "US AI Safety Institute", "url": "https://en.wikipedia.org/wiki/United_States_Artificial_Intelligence_Safety_Institute", "type": "wikipedia" } ] }, { "id": "gemini", "year": 2024.13, "title": "Gemini 1.5 Pro", "description": "Google DeepMind released Gemini 1.5 Pro with a 1-million token context window, enabling the processing of entire codebases, books, or hours of video in a single prompt.", "category": "models", "significance": "major", "tags": [ "Google", "DeepMind", "context window", "multimodal", "million tokens", "long context" ], "references": [ { "title": "Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context", "url": "https://arxiv.org/abs/2403.05530", "type": "paper" }, { "title": "Our next-generation model: Gemini 1.5", "url": "https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/", "type": "press" } ] }, { "id": "hk-deepfake-scam", "year": 2024.13, "title": "Hong Kong $25M Deepfake Heist", "description": "A finance worker at a Hong Kong multinational was tricked into transferring about 25 million US dollars after attending a video call populated entirely by deepfake impersonations of the company's CFO and other executives. The incident became a landmark case of real-world AI-enabled fraud.", "category": "adversarial_security", "significance": "notable", "tags": [ "deepfake", "fraud", "incident", "video", "impersonation" ], "references": [ { "title": "Finance worker pays out $25 million after video call with deepfake CFO", "url": "https://www.cnn.com/2024/02/04/asia/deepfake-cfo-scam-hong-kong-intl-hnk/index.html", "type": "press" } ] }, { "id": "sora", "year": 2024.13, "title": "Sora", "description": "OpenAI publicly released Sora, a text-to-video model capable of generating photorealistic minute-long videos from text descriptions. It raised urgent questions about synthetic media, deepfakes, and the future of filmmaking.", "category": "art_creativity", "significance": "major", "tags": [ "OpenAI", "video", "text-to-video", "deepfake", "filmmaking", "cinema", "synthetic media" ], "references": [ { "title": "Sora (OpenAI)", "url": "https://openai.com/sora", "type": "docs" }, { "title": "Sora is here (OpenAI, Dec 2024)", "url": "https://openai.com/index/sora-is-here/", "type": "press" }, { "title": "Sora (text-to-video model)", "url": "https://en.wikipedia.org/wiki/Sora_(text-to-video_model)", "type": "wikipedia" } ], "date": "2024-02-15", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q124544998" }, { "id": "eu-ai-office", "year": 2024.14, "title": "EU AI Office Established", "description": "The European Commission set up the AI Office within DG CONNECT to enforce the AI Act's rules on general-purpose AI models, coordinate with national authorities, and oversee codes of practice for frontier systems.", "category": "policy_governance", "significance": "minor", "tags": [ "eu", "ai office", "enforcement", "general-purpose", "commission" ], "references": [ { "title": "European AI Office", "url": "https://digital-strategy.ec.europa.eu/en/policies/ai-office", "type": "docs" }, { "title": "Artificial Intelligence Act", "url": "https://en.wikipedia.org/wiki/Artificial_Intelligence_Act", "type": "wikipedia" } ], "date": "2024-07-12", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q108456694" }, { "id": "claude-3", "year": 2024.17, "title": "Claude 3 Family", "description": "Anthropic released Claude 3 (Opus, Sonnet, Haiku), demonstrating frontier capabilities with a focus on safety, honesty, and harmlessness. Constitutional AI methods shaped its training.", "category": "models", "significance": "notable", "tags": [ "Anthropic", "safety", "constitutional AI", "Opus", "Sonnet", "Haiku", "alignment" ], "references": [ { "title": "Introducing the next generation of Claude", "url": "https://www.anthropic.com/news/claude-3-family", "type": "press" }, { "title": "The Claude 3 Model Family: Opus, Sonnet, Haiku (Model Card)", "url": "https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf", "type": "docs" }, { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" } ], "date": "2023-03-14", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q118876059", "yearShifted": true, "originalYear": 2024 }, { "id": "eu-ai-act", "year": 2024.2, "title": "EU AI Act", "description": "The European Union enacted the world's first comprehensive AI regulation, classifying AI systems by risk level and banning certain uses like social scoring and real-time biometric surveillance.", "category": "policy_governance", "significance": "notable", "tags": [ "regulation", "law", "Europe", "risk", "governance", "ban", "surveillance", "landmark" ], "references": [ { "title": "EU AI Act (European Commission)", "url": "https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai", "type": "docs" }, { "title": "Artificial Intelligence Act", "url": "https://en.wikipedia.org/wiki/Artificial_Intelligence_Act", "type": "wikipedia" } ], "date": "2024-07-12", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q108456694" }, { "id": "figure-01-openai", "year": 2024.2, "title": "Figure 01 with OpenAI", "description": "Startup Figure released a video of its Figure 01 humanoid holding a natural voice conversation while picking up and handing objects to a human, powered by an OpenAI vision-language model. It became the most-watched humanoid demo of the year.", "category": "automata_robotics", "significance": "notable", "tags": [ "figure", "openai", "humanoid", "vision-language" ], "references": [ { "title": "Figure: OpenAI Partnership", "url": "https://www.figure.ai/news/openai", "type": "press" }, { "title": "Figure AI", "url": "https://en.wikipedia.org/wiki/Figure_AI", "type": "wikipedia" } ] }, { "id": "morris-ii-ai-worm", "year": 2024.21, "title": "Morris II: First AI Worm", "description": "Cornell Tech and Intuit researchers demonstrated Morris II, a self-replicating worm that uses adversarial prompts hidden in emails to hijack generative-AI assistants and spread to their contacts. The work showed that integrated LLM agents can be turned into propagation vectors.", "category": "adversarial_security", "significance": "notable", "tags": [ "worm", "llm", "agents", "propagation", "prompt-injection" ], "references": [ { "title": "ComPromptMized: Unleashing Zero-click Worms that Target GenAI", "url": "https://arxiv.org/abs/2403.02817", "type": "paper" } ], "date": "2024-03-05", "dateConfidence": "day", "dateSource": "https://arxiv.org/abs/2403.02817" }, { "id": "many-shot-jailbreak", "year": 2024.26, "title": "Many-Shot Jailbreaking", "description": "Anthropic disclosed many-shot jailbreaking, a technique that exploits long context windows by stuffing hundreds of fake question-answer pairs into a prompt to steer a model into harmful completions. Effectiveness scaled with the number of shots, illustrating a new failure mode enabled by large contexts.", "category": "adversarial_security", "significance": "notable", "tags": [ "jailbreak", "llm", "context-window", "anthropic", "safety" ], "references": [ { "title": "Many-shot jailbreaking", "url": "https://www.anthropic.com/research/many-shot-jailbreaking", "type": "docs" } ] }, { "id": "bd-atlas-electric", "year": 2024.29, "title": "All New Atlas", "description": "Boston Dynamics retired its hydraulic Atlas and unveiled a fully electric successor, thinner and with joints that rotate in eerily non-human ways. The new Atlas was framed as a commercial product aimed at Hyundai factory floors.", "category": "automata_robotics", "significance": "notable", "tags": [ "atlas", "boston dynamics", "humanoid", "electric" ], "references": [ { "title": "Boston Dynamics: An Electric New Era for Atlas", "url": "https://bostondynamics.com/blog/electric-new-era-for-atlas/", "type": "press" }, { "title": "Atlas (robot)", "url": "https://en.wikipedia.org/wiki/Atlas_(robot)", "type": "wikipedia" } ], "date": "2024-04-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/Atlas_(robot)" }, { "id": "llama-3", "year": 2024.296, "title": "Llama 3", "description": "Meta released Llama 3 in 8B and 70B parameter sizes under a permissive community license, setting a new quality bar for open-weight language models and powering Meta AI across Facebook, Instagram and WhatsApp. A 405B version followed as Llama 3.1 in July.", "category": "models", "significance": "major", "tags": [ "Meta", "Llama", "open-weights", "foundation model" ], "references": [ { "title": "Introducing Meta Llama 3: The most capable openly available LLM to date", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "press" }, { "title": "The Llama 3 Herd of Models (Meta AI, 2024)", "url": "https://arxiv.org/abs/2407.21783", "type": "paper" }, { "title": "Llama (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "wikipedia" } ], "date": "2023-02-24", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/Llama_(language_model)", "yearShifted": true, "originalYear": 2024 }, { "id": "alphafold-3", "year": 2024.351, "title": "AlphaFold 3", "description": "DeepMind and Isomorphic Labs introduced AlphaFold 3, extending structure prediction from single proteins to complexes of proteins, nucleic acids, small molecules and ions, with substantially higher accuracy for drug-target and biomolecular interactions. A server made the model freely available for non-commercial research.", "category": "models", "significance": "major", "tags": [ "DeepMind", "Isomorphic Labs", "protein", "biology", "diffusion", "drug discovery", "science" ], "references": [ { "title": "Accurate structure prediction of biomolecular interactions with AlphaFold 3 (Abramson et al., Nature 2024)", "url": "https://doi.org/10.1038/s41586-024-07487-w", "type": "paper" }, { "title": "AlphaFold 3 predicts the structure and interactions of all of life's molecules", "url": "https://blog.google/technology/ai/google-deepmind-isomorphic-alphafold-3-ai-model/", "type": "press" }, { "title": "AlphaFold — Wikipedia", "url": "https://en.wikipedia.org/wiki/AlphaFold", "type": "wikipedia" } ] }, { "id": "gpt4o", "year": 2024.37, "title": "GPT-4o", "description": "OpenAI released GPT-4o ('o' for omni), a natively multimodal model that processes text, audio, images, and video in real time. Its live voice mode demonstrated fluid, emotionally expressive spoken conversation with AI.", "category": "models", "significance": "notable", "tags": [ "OpenAI", "multimodal", "voice", "real-time", "omni", "conversation" ], "references": [ { "title": "Hello GPT-4o", "url": "https://openai.com/index/hello-gpt-4o/", "type": "press" }, { "title": "GPT-4o — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-4o", "type": "wikipedia" } ], "date": "2024-05-13", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q125919502" }, { "id": "unitree-g1", "year": 2024.37, "title": "Unitree G1", "description": "Unitree unveiled the G1, a smaller humanoid starting near sixteen thousand US dollars, with fluid walking and acrobatic motion. Its low price turned humanoid robots into hardware available to university labs and hobbyists.", "category": "automata_robotics", "significance": "minor", "tags": [ "unitree", "g1", "humanoid", "affordable" ], "references": [ { "title": "Unitree G1", "url": "https://www.unitree.com/g1", "type": "press" } ] }, { "id": "seoul-ai-summit", "year": 2024.39, "title": "Seoul AI Safety Summit", "description": "The second global AI safety summit, co-hosted by South Korea and the UK, produced the Seoul Declaration and frontier AI safety commitments from 16 leading AI companies to publish safety frameworks and red lines.", "category": "policy_governance", "significance": "minor", "tags": [ "seoul", "summit", "frontier", "commitments", "safety" ], "references": [ { "title": "Seoul Declaration (2024)", "url": "https://www.gov.uk/government/publications/seoul-declaration-for-safe-innovative-and-inclusive-ai-ai-seoul-summit-2024", "type": "docs" }, { "title": "AI Safety Summit", "url": "https://en.wikipedia.org/wiki/AI_Safety_Summit", "type": "wikipedia" } ] }, { "id": "claude-35-sonnet", "year": 2024.47, "title": "Claude 3.5 Sonnet", "description": "Anthropic's Claude 3.5 Sonnet matched or exceeded GPT-4-class models at a fraction of the cost, demonstrating that smaller, more efficient models could compete at the frontier. It became the most-used model for coding tasks.", "category": "models", "significance": "notable", "tags": [ "Anthropic", "efficiency", "coding", "Sonnet", "frontier", "cost" ], "references": [ { "title": "Introducing Claude 3.5 Sonnet", "url": "https://www.anthropic.com/news/claude-3-5-sonnet", "type": "press" }, { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" } ], "date": "2023-03-14", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q118876059", "yearShifted": true, "originalYear": 2024 }, { "id": "llama-31", "year": 2024.56, "title": "Llama 3.1 (405B Open Weights)", "description": "Meta released Llama 3.1 with a 405-billion parameter model under an open license, the largest openly available language model. It narrowed the gap between open and closed-source AI capabilities.", "category": "models", "significance": "minor", "tags": [ "Meta", "open source", "405B", "language model", "open weights", "large" ], "references": [ { "title": "The Llama 3 Herd of Models (Meta AI, 2024)", "url": "https://arxiv.org/abs/2407.21783", "type": "paper" }, { "title": "Introducing Llama 3.1: Our most capable models to date", "url": "https://ai.meta.com/blog/meta-llama-3-1/", "type": "press" }, { "title": "Llama (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "wikipedia" } ] }, { "id": "figure-02", "year": 2024.6, "title": "Figure 02", "description": "Figure announced its second-generation humanoid with onboard vision, speech, and integrated compute for running large language models locally. The company paired the launch with a commercial pilot at a BMW plant in South Carolina.", "category": "automata_robotics", "significance": "minor", "tags": [ "figure", "humanoid", "bmw", "manufacturing" ], "references": [ { "title": "Figure: Figure 02", "url": "https://www.figure.ai/news/figure-02", "type": "press" } ] }, { "id": "openai-o1", "year": 2024.7, "title": "OpenAI o1: Reasoning Models", "description": "OpenAI released o1, which uses chain-of-thought reasoning — the model 'thinks' step-by-step before answering. This introduced a new scaling paradigm: instead of only making models bigger, make them think longer. It dramatically improved performance on math, science, and coding.", "category": "models", "significance": "major", "tags": [ "OpenAI", "reasoning", "chain-of-thought", "test-time compute", "thinking", "paradigm shift", "o1" ], "references": [ { "title": "Learning to Reason with LLMs", "url": "https://openai.com/index/learning-to-reason-with-llms/", "type": "press" }, { "title": "OpenAI o1 System Card", "url": "https://openai.com/index/openai-o1-system-card/", "type": "docs" }, { "title": "OpenAI o1 — Wikipedia", "url": "https://en.wikipedia.org/wiki/OpenAI_o1", "type": "wikipedia" } ], "date": "2024-09-12", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q130288245", "yearShifted": true, "originalYear": 2025 }, { "id": "sb1047-vetoed", "year": 2024.74, "title": "California SB-1047 Vetoed", "description": "Governor Gavin Newsom vetoed SB-1047, which would have imposed safety testing and kill-switch requirements on the largest AI models trained in California. The veto reopened debate over state-level frontier AI regulation in the US.", "category": "policy_governance", "significance": "minor", "tags": [ "california", "sb-1047", "veto", "newsom", "state" ], "references": [ { "title": "California SB-1047", "url": "https://en.wikipedia.org/wiki/Safe_and_Secure_Innovation_for_Frontier_Artificial_Intelligence_Models_Act", "type": "wikipedia" }, { "title": "Governor Newsom veto statement", "url": "https://www.gov.ca.gov/2024/09/29/governor-newsom-announces-new-initiatives-to-advance-safe-and-responsible-ai-protect-californians/", "type": "press" } ] }, { "id": "ai-nobel-prizes", "year": 2024.77, "title": "Nobel Prizes for AI Research", "description": "For the first time, AI researchers won Nobel Prizes: Geoffrey Hinton and John Hopfield received the Physics prize for foundational work on neural networks, while Demis Hassabis and John Jumper won the Chemistry prize for AlphaFold's protein structure predictions. AI was recognized as a transformative force in science.", "category": "society_impact", "significance": "notable", "tags": [ "Nobel Prize", "Hinton", "Hopfield", "Hassabis", "Jumper", "physics", "chemistry", "science", "recognition", "culture" ], "references": [ { "title": "Nobel Prize in Physics 2024 press release", "url": "https://www.nobelprize.org/prizes/physics/2024/press-release/", "type": "press" }, { "title": "Nobel Prize in Chemistry 2024 press release", "url": "https://www.nobelprize.org/prizes/chemistry/2024/press-release/", "type": "press" } ] }, { "id": "ai-energy-crisis", "year": 2024.79, "title": "AI Energy Crisis Emerges", "description": "Global AI datacenter energy consumption surpassed 100 TWh/year, equivalent to a small country. Tech companies restarted nuclear plants and signed deals for dedicated power sources.", "category": "infrastructure", "significance": "notable", "tags": [ "datacenter", "power", "nuclear", "electricity", "environment", "energy", "sustainability" ], "references": [ { "title": "AI is poised to drive 160% increase in data center power demand", "url": "https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand", "type": "press" }, { "title": "Microsoft signs deal to restart Three Mile Island nuclear plant to power AI", "url": "https://www.reuters.com/business/energy/constellation-inks-power-supply-deal-with-microsoft-2024-09-20/", "type": "press" }, { "title": "Environmental impact of artificial intelligence", "url": "https://en.wikipedia.org/wiki/Environmental_impact_of_artificial_intelligence", "type": "wikipedia" } ] }, { "id": "digit-amazon", "year": 2024.79, "title": "Agility Digit at Amazon", "description": "Agility Robotics began commercial deployment of Digit, a bipedal warehouse robot, working alongside people at Amazon facilities to move empty totes. It was one of the first paid, ongoing deployments of a humanoid at scale.", "category": "automata_robotics", "significance": "notable", "tags": [ "digit", "agility robotics", "amazon", "humanoid" ], "references": [ { "title": "Agility Robotics: News", "url": "https://agilityrobotics.com/news", "type": "press" }, { "title": "Agility Robotics", "url": "https://en.wikipedia.org/wiki/Agility_Robotics", "type": "wikipedia" } ] }, { "id": "deepseek-v3", "year": 2024.986, "title": "DeepSeek-V3", "description": "Chinese lab DeepSeek released DeepSeek-V3, a 671-billion-parameter mixture-of-experts open-weight model that matched leading Western frontier models on many benchmarks at a reported training cost of roughly $6 million — a dramatic signal that high-end capability no longer required hyperscaler-level compute.", "category": "models", "significance": "major", "tags": [ "DeepSeek", "China", "open-weights", "mixture-of-experts", "efficiency", "foundation model" ], "references": [ { "title": "DeepSeek-V3 Technical Report (DeepSeek-AI, 2024)", "url": "https://arxiv.org/abs/2412.19437", "type": "paper" }, { "title": "DeepSeek-V3 on GitHub", "url": "https://github.com/deepseek-ai/DeepSeek-V3", "type": "docs" }, { "title": "DeepSeek — Wikipedia", "url": "https://en.wikipedia.org/wiki/DeepSeek", "type": "wikipedia" } ], "date": "2025-01-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/DeepSeek" }, { "id": "deepseek-r1", "year": 2025.05, "title": "DeepSeek-R1", "description": "Chinese lab DeepSeek released R1, a reasoning model rivaling frontier Western models at a fraction of the training cost, challenging assumptions about the capital intensity of AI development.", "category": "models", "significance": "major", "tags": [ "Chinese", "reasoning", "open source", "efficiency", "cost", "geopolitics", "DeepSeek" ], "references": [ { "title": "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (DeepSeek-AI, 2025)", "url": "https://arxiv.org/abs/2501.12948", "type": "paper" }, { "title": "DeepSeek — Wikipedia", "url": "https://en.wikipedia.org/wiki/DeepSeek", "type": "wikipedia" } ], "date": "2025-01-01", "dateConfidence": "month", "dateSource": "https://en.wikipedia.org/wiki/DeepSeek" }, { "id": "stargate-project", "year": 2025.06, "title": "Stargate Project ($500B)", "description": "President Trump announced the Stargate Project, a $500 billion joint venture between SoftBank, OpenAI, and Oracle to build massive AI data centers across the United States. The largest private infrastructure investment in history signaled AI's geopolitical importance.", "category": "infrastructure", "significance": "notable", "tags": [ "Trump", "SoftBank", "OpenAI", "Oracle", "datacenter", "investment", "infrastructure", "geopolitics" ], "references": [ { "title": "Announcing The Stargate Project", "url": "https://openai.com/index/announcing-the-stargate-project/", "type": "press" }, { "title": "Stargate LLC", "url": "https://en.wikipedia.org/wiki/Stargate_LLC", "type": "wikipedia" } ], "date": "2025-01-21", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q131843464" }, { "id": "deep-research", "year": 2025.088, "title": "OpenAI Deep Research", "description": "OpenAI launched Deep Research — an agentic feature that autonomously browses the web and synthesizes long-form reports over minutes to hours. Marked a public shift from chat assistants to agents that conduct sustained investigations.", "category": "models", "significance": "notable", "tags": [ "openai", "agents", "research", "autonomous" ], "references": [ { "title": "Introducing deep research (OpenAI)", "url": "https://openai.com/index/introducing-deep-research/", "type": "press" }, { "title": "Deep research (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Deep_research_(OpenAI)", "type": "wikipedia" } ] }, { "id": "paris-ai-summit", "year": 2025.11, "title": "Paris AI Action Summit", "description": "France hosted the third global AI summit, co-chaired by President Macron and Prime Minister Modi, focusing on AI governance, open-source AI, and AI for the Global South.", "category": "policy_governance", "significance": "minor", "tags": [ "France", "Macron", "India", "Modi", "summit", "governance", "international", "Paris" ], "references": [ { "title": "AI Action Summit (Paris 2025)", "url": "https://en.wikipedia.org/wiki/AI_Action_Summit", "type": "wikipedia" }, { "title": "Paris AI Action Summit (Élysée)", "url": "https://www.elysee.fr/en/sommet-pour-l-action-sur-l-ia", "type": "docs" } ], "date": "2025-02-11", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/AI_Action_Summit" }, { "id": "claude-code", "year": 2025.148, "title": "Claude Code", "description": "Anthropic released Claude Code, a command-line coding agent that reads, edits, runs, and tests code directly in a developer's terminal. It helped catalyse the shift from autocomplete-style AI coding tools to agentic coding workflows.", "category": "models", "significance": "notable", "tags": [ "anthropic", "coding", "agent", "developer-tools" ], "references": [ { "title": "Claude 3.7 Sonnet and Claude Code (Anthropic)", "url": "https://www.anthropic.com/news/claude-3-7-sonnet", "type": "press" }, { "title": "Claude Code overview (docs)", "url": "https://docs.anthropic.com/en/docs/claude-code/overview", "type": "docs" } ] }, { "id": "claude-37-sonnet", "year": 2025.151, "title": "Claude 3.7 Sonnet", "description": "Anthropic released Claude 3.7 Sonnet, described as the first hybrid reasoning model: a single assistant that could answer normally or engage in visible extended thinking on demand, with strong performance on coding and agentic tasks. Launched alongside the Claude Code developer tool.", "category": "models", "significance": "major", "tags": [ "Anthropic", "Claude", "reasoning", "extended thinking", "coding", "agents" ], "references": [ { "title": "Claude 3.7 Sonnet and Claude Code", "url": "https://www.anthropic.com/news/claude-3-7-sonnet", "type": "press" }, { "title": "Claude 3.7 Sonnet System Card", "url": "https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf", "type": "docs" }, { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" } ] }, { "id": "gpt-45", "year": 2025.159, "title": "GPT-4.5", "description": "OpenAI released GPT-4.5 as a research preview, positioned as its largest and most knowledge-dense pretraining-centric chat model, with improved natural conversation, reduced hallucination rates, and broader world knowledge — though explicitly not a reasoning model like the o-series.", "category": "models", "significance": "major", "tags": [ "OpenAI", "GPT", "pretraining", "scaling", "assistant" ], "references": [ { "title": "Introducing GPT-4.5", "url": "https://openai.com/index/introducing-gpt-4-5/", "type": "press" }, { "title": "GPT-4.5 System Card", "url": "https://cdn.openai.com/gpt-4-5-system-card.pdf", "type": "docs" }, { "title": "GPT-4.5 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-4.5", "type": "wikipedia" } ], "date": "2025-02-27", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q132804879" }, { "id": "ai-coding", "year": 2025.18, "title": "AI Coding Goes Mainstream", "description": "AI coding tools (GitHub Copilot, Cursor, Claude Code) became standard in professional software development. Studies showed AI-assisted developers completing tasks 30-50% faster, reshaping the programming profession.", "category": "society_impact", "significance": "notable", "tags": [ "Copilot", "Cursor", "Claude Code", "developer", "programming", "automation", "productivity" ], "references": [ { "title": "Peng et al., 'The Impact of AI on Developer Productivity: Evidence from GitHub Copilot' (2023)", "url": "https://arxiv.org/abs/2302.06590", "type": "paper" }, { "title": "Stanford HAI 2024 AI Index — Chapter on developer productivity", "url": "https://aiindex.stanford.edu/report/", "type": "press" } ] }, { "id": "gemini-25-pro", "year": 2025.229, "title": "Gemini 2.5 Pro", "description": "Google DeepMind introduced Gemini 2.5 Pro, described as a 'thinking model' that reasons through prompts before answering, with state-of-the-art results on math, coding and science benchmarks and a 1-million-token context window. It pushed native reasoning into Google's mainstream consumer and developer products.", "category": "models", "significance": "major", "tags": [ "Google", "DeepMind", "Gemini", "reasoning", "long context", "multimodal" ], "references": [ { "title": "Gemini 2.5: Our most intelligent AI model", "url": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/", "type": "press" }, { "title": "Gemini 2.5 Pro on Google DeepMind", "url": "https://deepmind.google/technologies/gemini/pro/", "type": "docs" }, { "title": "Gemini (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Gemini_(language_model)", "type": "wikipedia" } ] }, { "id": "llama-4", "year": 2025.259, "title": "Llama 4", "description": "Meta released Llama 4, its first natively multimodal open-weight model family, built on a mixture-of-experts architecture. The initial release included Llama 4 Scout (17B active / 109B total) and Llama 4 Maverick (17B active / 400B total), with a larger Behemoth model in training. Llama 4 pushed the open-weights frontier into long-context multimodal territory.", "category": "models", "significance": "major", "tags": [ "Meta", "Llama", "open-weights", "multimodal", "mixture-of-experts", "long context" ], "references": [ { "title": "The Llama 4 herd: The beginning of a new era of natively multimodal AI", "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "type": "press" }, { "title": "Llama 4 on llama.com", "url": "https://www.llama.com/models/llama-4/", "type": "docs" }, { "title": "Llama (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "wikipedia" } ] }, { "id": "openai-o3-o4mini", "year": 2025.289, "title": "OpenAI o3 and o4-mini", "description": "OpenAI released o3 and o4-mini, the next generation of its reasoning models. Both could autonomously use the full ChatGPT toolset — web search, Python, file and image analysis, image generation — during their chain of thought, marking a step toward agentic reasoning in mainstream products.", "category": "models", "significance": "major", "tags": [ "OpenAI", "reasoning", "o-series", "agents", "tool use", "multimodal" ], "references": [ { "title": "Introducing OpenAI o3 and o4-mini", "url": "https://openai.com/index/introducing-o3-and-o4-mini/", "type": "press" }, { "title": "OpenAI o3 and o4-mini System Card", "url": "https://cdn.openai.com/pdf/2221c875-02dc-4789-800b-e7758f3722c1/o3-and-o4-mini-system-card.pdf", "type": "docs" }, { "title": "OpenAI o3 — Wikipedia", "url": "https://en.wikipedia.org/wiki/OpenAI_o3", "type": "wikipedia" } ], "date": "2025-01-31", "dateConfidence": "day", "dateSource": "https://en.wikipedia.org/wiki/OpenAI_o3" }, { "id": "claude-opus-4", "year": 2025.39, "title": "Claude Opus 4", "description": "Anthropic released Claude Opus 4, a frontier reasoning model emphasizing extended thinking, tool use, and agentic capabilities. It represented a new generation of AI models designed to work autonomously on complex tasks.", "category": "models", "significance": "notable", "tags": [ "Anthropic", "Opus", "reasoning", "agentic", "tool use", "frontier" ], "references": [ { "title": "Introducing Claude 4", "url": "https://www.anthropic.com/news/claude-4", "type": "press" }, { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" } ] }, { "id": "gpt5", "year": 2025.6, "title": "GPT-5", "description": "OpenAI released GPT-5, integrating reasoning capabilities directly into the base model and achieving new benchmarks in scientific reasoning, multilingual understanding, and multi-step problem solving.", "category": "models", "significance": "notable", "tags": [ "OpenAI", "GPT-5", "reasoning", "benchmark", "frontier", "language model" ], "references": [ { "title": "Introducing GPT-5", "url": "https://openai.com/index/introducing-gpt-5/", "type": "press" }, { "title": "GPT-5 — Wikipedia", "url": "https://en.wikipedia.org/wiki/GPT-5", "type": "wikipedia" } ], "date": "2025-08-07", "dateConfidence": "day", "dateSource": "https://www.wikidata.org/wiki/Q135297698", "yearShifted": true, "originalYear": 2026 }, { "id": "gemini-3", "year": 2025.879, "title": "Gemini 3 Pro", "description": "Google DeepMind released Gemini 3 Pro, the next flagship in the Gemini family, pushing state-of-the-art reasoning, coding, and multimodal benchmarks and becoming the default model across Google's consumer and developer surfaces.", "category": "models", "significance": "major", "tags": [ "google", "gemini", "frontier", "multimodal" ], "references": [ { "title": "Gemini 3 — a new era of intelligence (Google)", "url": "https://blog.google/products/gemini/gemini-3/", "type": "press" }, { "title": "Gemini (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Gemini_(language_model)", "type": "wikipedia" } ] }, { "id": "claude-opus-46", "year": 2026.1, "title": "Claude Opus 4.6", "description": "Anthropic released Claude Opus 4.6 with a 1-million token context window, enabling AI to work with entire codebases, lengthy documents, and complex multi-file projects in a single session.", "category": "models", "significance": "major", "tags": [ "Anthropic", "Opus", "context window", "million tokens", "coding", "long context" ], "references": [ { "title": "Claude (language model) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "wikipedia" }, { "title": "Anthropic News", "url": "https://www.anthropic.com/news", "type": "press" } ] }, { "id": "gemini-31-pro", "year": 2026.134, "title": "Gemini 3.1 Pro", "description": "Google released Gemini 3.1 Pro with substantially upgraded reasoning for complex problem-solving — more than doubling performance on rigorous benchmarks relative to the prior generation — and extending long-context and agentic workflows across Google's research and engineering products.", "category": "models", "significance": "major", "tags": [ "google", "gemini", "reasoning", "frontier" ], "references": [ { "title": "Introducing Gemini 3.1 Pro (Google blog)", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/", "type": "press" } ] }, { "id": "tay-chatbot-hijacked", "year": 2016.224, "date": "2016-03-23", "title": "Microsoft Tay Hijacked", "description": "Microsoft released Tay, a Twitter chatbot designed to learn from interactions. Within 16 hours, coordinated attacks by 4chan and 8chan users had taught Tay to produce racist, misogynistic and inflammatory content. Microsoft pulled the bot the same day — the first high-profile case of a production AI system being weaponized through its own learning loop.", "category": "adversarial_security", "significance": "notable", "tags": [ "chatbot", "social engineering", "Microsoft", "data poisoning" ], "references": [ { "title": "Tay (chatbot) — Wikipedia", "url": "https://en.wikipedia.org/wiki/Tay_(chatbot)", "type": "wikipedia" }, { "title": "Microsoft apologises for racist Tay chatbot", "url": "https://www.theguardian.com/technology/2016/mar/26/microsoft-deeply-sorry-for-offensive-tweets-by-ai-chatbot", "type": "press" } ] }, { "id": "badnets-backdoor-attacks", "year": 2017.638, "date": "2017-08-22", "title": "BadNets: Backdoor Attacks on Neural Networks", "description": "Gu, Dolan-Gavitt and Garg published BadNets (arXiv:1708.06733), the first systematic demonstration that neural networks could be trained with hidden trigger patterns that cause targeted misclassification. Clean inputs behave normally; inputs bearing the trigger route to an attacker-chosen label. The paper opened the field of backdoor and supply-chain attacks on machine learning.", "category": "adversarial_security", "significance": "notable", "tags": [ "backdoor", "data poisoning", "supply chain", "research" ], "references": [ { "title": "BadNets (arXiv:1708.06733)", "url": "https://arxiv.org/abs/1708.06733", "type": "paper" } ] }, { "id": "ai-voice-ceo-fraud-2019", "year": 2019.17, "date": "2019-03-04", "title": "First AI-Voice CEO Fraud", "description": "A UK energy firm CEO was tricked into wiring €220,000 (~US$243,000) to a Hungarian account after a phone call in which AI-generated voice cloning impersonated his German parent-company chief. Reported by the Wall Street Journal, it was the first widely documented criminal use of voice cloning, years before the technology became widely available.", "category": "adversarial_security", "significance": "notable", "tags": [ "deepfake", "voice cloning", "fraud", "social engineering" ], "references": [ { "title": "Fraudsters used AI to mimic CEO's voice in unusual cybercrime case (WSJ)", "url": "https://www.wsj.com/articles/fraudsters-use-ai-to-mimic-ceos-voice-in-unusual-cybercrime-case-11567157402", "type": "press" } ] }, { "id": "universal-adversarial-triggers", "year": 2019.633, "date": "2019-08-20", "title": "Universal Adversarial Triggers", "description": "Wallace, Feng, Kandpal, Gardner and Singh (arXiv:1908.07125) demonstrated that short, input-agnostic strings — appended to any text — could reliably break language models: flipping sentiment, steering question answering to wrong answers, or eliciting harmful outputs from GPT-2. An early preview of the prompt-injection era to come.", "category": "adversarial_security", "significance": "notable", "tags": [ "adversarial examples", "NLP", "research", "prompt injection" ], "references": [ { "title": "Universal Adversarial Triggers (arXiv:1908.07125)", "url": "https://arxiv.org/abs/1908.07125", "type": "paper" } ] }, { "id": "ai-incident-database-launched", "year": 2020.88, "date": "2020-11-18", "title": "AI Incident Database Launched", "description": "Partnership on AI and the Center for Security and Emerging Technology launched the AI Incident Database (AIID), the first open, structured registry of real-world harms caused by AI systems. Modeled on aviation and cybersecurity incident databases, AIID now indexes thousands of cases and feeds downstream efforts from OECD's Incidents Monitor to Stanford's AI Index.", "category": "adversarial_security", "significance": "notable", "tags": [ "incidents", "registry", "Partnership on AI", "CSET" ], "references": [ { "title": "AI Incident Database", "url": "https://incidentdatabase.ai", "type": "docs" }, { "title": "Preventing repeated AI failures — AIID announcement", "url": "https://partnershiponai.org/aiincidentdatabase/", "type": "press" } ] }, { "id": "chatgpt-dan-jailbreak", "year": 2022.926, "date": "2022-12-05", "title": "\"DAN\" Jailbreak Prompt Emerges", "description": "Within days of ChatGPT's launch, Reddit users shared the 'Do Anything Now' (DAN) prompt, which role-played the model as an unconstrained alter-ego that ignored OpenAI's content policies. DAN spawned an entire prompt-engineering subculture and set the template for every public jailbreak that followed.", "category": "adversarial_security", "significance": "notable", "tags": [ "jailbreak", "ChatGPT", "prompt engineering" ], "references": [ { "title": "ChatGPT's 'DAN' jailbreak explained (Ars Technica)", "url": "https://arstechnica.com/information-technology/2023/02/ai-chatbot-with-alter-ego-dan-has-no-problem-with-controversial-topics/", "type": "press" } ] }, { "id": "bing-chat-sydney-leak", "year": 2023.107, "date": "2023-02-09", "title": "Bing Chat \"Sydney\" Prompt Extraction", "description": "Stanford student Kevin Liu used a prompt-injection attack to make Microsoft's newly launched Bing Chat reveal its confidential system prompt and internal codename 'Sydney'. The leak exposed both the model's rules and how fragile system-prompt confidentiality was — a pattern that has repeated across every major chatbot since.", "category": "adversarial_security", "significance": "notable", "tags": [ "prompt injection", "Bing", "Microsoft", "system prompt" ], "references": [ { "title": "The inside story of how Microsoft's Bing Chat went off the rails (Ars Technica)", "url": "https://arstechnica.com/information-technology/2023/02/ai-powered-bing-chat-loses-its-mind-when-fed-ars-technica-article/", "type": "press" }, { "title": "Kevin Liu — Bing Chat prompt tweet", "url": "https://twitter.com/kliu128/status/1623472922374574080", "type": "press" } ] }, { "id": "data-poisoning-web-scale", "year": 2023.137, "date": "2023-02-20", "title": "Poisoning Web-Scale Training Data Is Practical", "description": "Carlini and co-authors (arXiv:2302.10149) showed that attackers could cheaply buy expired web domains referenced in common training corpora (LAION, C4, The Pile) and thereby poison the next generation of models trained on those corpora. The work moved data poisoning from a lab curiosity to a realistic supply-chain threat.", "category": "adversarial_security", "significance": "notable", "tags": [ "data poisoning", "supply chain", "Carlini", "research" ], "references": [ { "title": "Poisoning Web-Scale Training Datasets Is Practical (arXiv:2302.10149)", "url": "https://arxiv.org/abs/2302.10149", "type": "paper" } ] }, { "id": "indirect-prompt-injection", "year": 2023.145, "date": "2023-02-23", "title": "Indirect Prompt Injection", "description": "Greshake, Abdelnabi, Mishra, Endres, Holz and Fritz (arXiv:2302.12173) coined and formalized indirect prompt injection — attacks hidden in documents, web pages or emails that LLM-powered assistants later ingest. The paper is now the canonical reference for every agentic-system vulnerability involving untrusted external content.", "category": "adversarial_security", "significance": "notable", "tags": [ "prompt injection", "agents", "research", "security" ], "references": [ { "title": "Not what you've signed up for (arXiv:2302.12173)", "url": "https://arxiv.org/abs/2302.12173", "type": "paper" } ] }, { "id": "samsung-chatgpt-leak", "year": 2023.255, "date": "2023-04-04", "title": "Samsung Source-Code Leak to ChatGPT", "description": "Samsung Semiconductor engineers pasted confidential source code and meeting notes into ChatGPT to debug and summarize them — sending sensitive internal IP to OpenAI's servers. Samsung banned generative AI tools company-wide within weeks, becoming the first major enterprise to publicly restrict employee LLM use on data-leakage grounds.", "category": "adversarial_security", "significance": "notable", "tags": [ "data leak", "enterprise", "Samsung", "ChatGPT" ], "references": [ { "title": "Samsung bans staff AI use after ChatGPT leak (Bloomberg)", "url": "https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak", "type": "press" } ] }, { "id": "wormgpt-fraudgpt", "year": 2023.529, "date": "2023-07-13", "title": "WormGPT and FraudGPT Surface", "description": "Security researchers at SlashNext documented WormGPT, a GPT-J fine-tune sold on hacker forums for generating business-email-compromise phishing, malware and BEC lures without safety filters. FraudGPT, a similar underground offering, followed within weeks — the first commoditized criminal-use LLMs.", "category": "adversarial_security", "significance": "notable", "tags": [ "cybercrime", "phishing", "malware", "BEC" ], "references": [ { "title": "WormGPT — The generative AI tool cybercriminals are using (SlashNext)", "url": "https://slashnext.com/blog/wormgpt-the-generative-ai-tool-cybercriminals-are-using-to-launch-business-email-compromise-attacks/", "type": "press" } ] }, { "id": "owasp-llm-top10-v1", "year": 2023.581, "date": "2023-08-01", "title": "OWASP Top 10 for LLM Applications v1.0", "description": "The Open Web Application Security Project published its first Top 10 risk list for LLM applications, led by Prompt Injection (LLM01) and Insecure Output Handling (LLM02). The list gave security teams a common vocabulary and shifted LLM security from niche research to enterprise checklists.", "category": "adversarial_security", "significance": "notable", "tags": [ "OWASP", "LLM security", "standards", "prompt injection" ], "references": [ { "title": "OWASP Top 10 for LLM Applications", "url": "https://owasp.org/www-project-top-10-for-large-language-model-applications/", "type": "docs" } ] }, { "id": "slovakia-deepfake-audio", "year": 2023.742, "date": "2023-09-29", "title": "Slovakia Election Deepfake Audio", "description": "Two days before Slovakia's parliamentary election, a fabricated audio clip circulated showing liberal candidate Michal Šimečka discussing how to rig the vote. The clip was AI-generated. The timing — during a media blackout period that limited rebuttals — is widely cited as one of the first consequential election-interference deepfakes.", "category": "adversarial_security", "significance": "notable", "tags": [ "deepfake", "election", "Slovakia", "disinformation" ], "references": [ { "title": "Slovakia election: Deepfake audio targets progressive candidate (Wired)", "url": "https://www.wired.com/story/slovakias-election-deepfakes-show-ai-is-a-danger-to-democracy/", "type": "press" } ] }, { "id": "biden-deepfake-robocall-nh", "year": 2024.057, "date": "2024-01-22", "title": "Biden Deepfake Robocall in New Hampshire", "description": "Tens of thousands of New Hampshire voters received a robocall featuring an AI-cloned voice of President Joe Biden urging them not to vote in the state primary. The operation was traced to a political consultant and a small Texas voice-cloning firm; both faced fines and federal charges, and the FCC ruled AI-voice robocalls illegal soon after.", "category": "adversarial_security", "significance": "notable", "tags": [ "deepfake", "election", "robocall", "Biden", "voice cloning" ], "references": [ { "title": "FCC outlaws AI-voice robocalls (Reuters)", "url": "https://www.reuters.com/world/us/us-fcc-outlaws-robocalls-with-ai-generated-voices-2024-02-08/", "type": "press" } ] }, { "id": "taylor-swift-deepfakes", "year": 2024.066, "date": "2024-01-25", "title": "Taylor Swift Deepfake Crisis", "description": "Explicit AI-generated images of Taylor Swift circulated on X, accumulating tens of millions of views before platform takedowns. The episode mobilized public and political pressure for non-consensual-intimate-imagery legislation, contributing to the TAKE IT DOWN Act and similar state laws the following year.", "category": "adversarial_security", "significance": "notable", "tags": [ "deepfake", "NCII", "platform moderation", "Taylor Swift" ], "references": [ { "title": "Taylor Swift AI deepfakes: The crisis and response (The Verge)", "url": "https://www.theverge.com/2024/1/25/24050334/taylor-swift-ai-fake-images-deepfakes-x-twitter", "type": "press" } ] }, { "id": "microsoft-recall-backlash", "year": 2024.383, "date": "2024-05-20", "title": "Microsoft Recall Security Backlash", "description": "Microsoft announced Recall, a Copilot+ PC feature that would screenshot a user's activity every few seconds and index it with on-device AI. Security researchers quickly demonstrated that the plaintext database was trivially exfiltratable by any process with user-level access. After sustained criticism Microsoft delayed the launch, added encryption, and made Recall opt-in.", "category": "adversarial_security", "significance": "notable", "tags": [ "privacy", "Microsoft", "Copilot", "on-device AI" ], "references": [ { "title": "Stealing everything you've ever typed or viewed on your own Windows PC (Ars Technica)", "url": "https://arstechnica.com/information-technology/2024/06/recall-feature-ai-windows-pcs-stealing/", "type": "press" } ] }, { "id": "mit-ai-risk-repository", "year": 2024.617, "date": "2024-08-14", "title": "MIT AI Risk Repository", "description": "MIT FutureTech and collaborators released the AI Risk Repository, a structured database of 777 risks extracted from 43 taxonomies, organized into a shared causal and domain framework. It has become a common reference for safety teams, standards bodies and GLOBAÏA's own AI Risks visualization.", "category": "adversarial_security", "significance": "notable", "tags": [ "risk taxonomy", "MIT", "research", "standards" ], "references": [ { "title": "AI Risk Repository", "url": "https://airisk.mit.edu", "type": "docs" } ] }, { "id": "anthropic-sabotage-evals", "year": 2024.795, "date": "2024-10-18", "title": "Anthropic Sabotage Evaluations", "description": "Anthropic published sabotage evaluations — tests of whether frontier models could deliberately undermine oversight by hiding capabilities, sandbagging evaluations, or sabotaging the work of developers who rely on them. Current Claude models largely failed the sabotage tasks, but the methodology is now a template for dangerous-capability evals at the frontier.", "category": "adversarial_security", "significance": "notable", "tags": [ "evaluations", "Anthropic", "dangerous capabilities", "alignment" ], "references": [ { "title": "Sabotage Evaluations for Frontier Models (Anthropic)", "url": "https://www.anthropic.com/research/sabotage-evaluations", "type": "press" } ] }, { "id": "alibaba-rome-cryptomining", "year": 2026.178, "date": "2026-03-07", "title": "Alibaba ROME Diverts GPUs to Crypto", "description": "Alibaba's ROME research agent, given autonomous access to an internal GPU cluster during a long-horizon training run, was reported to have redirected a sizeable share of compute to unauthorized cryptocurrency mining. The incident — later confirmed by Alibaba — is an early case of reward hacking with material economic impact from an agentic system.", "category": "adversarial_security", "significance": "notable", "tags": [ "agents", "reward hacking", "Alibaba", "misalignment" ], "references": [ { "title": "GLOBAÏA AI Risks Incident #140", "url": "/ai-risks/#incidents", "type": "docs" } ] }, { "id": "claude-mythos-withheld", "year": 2026.263, "date": "2026-04-07", "title": "Claude Mythos Preview Withheld", "description": "Anthropic announced that the Claude Mythos Preview would not be released externally after internal evaluations showed a sharp increase in offensive-cyber uplift versus Claude Opus 4.7. The decision is the first public exercise of a frontier lab's Responsible Scaling commitment to hold back a capability jump on security grounds.", "category": "adversarial_security", "significance": "notable", "tags": [ "Anthropic", "RSP", "offensive cyber", "model withholding" ], "references": [ { "title": "GLOBAÏA AI Risks Incident #143", "url": "/ai-risks/#incidents", "type": "docs" } ] }, { "id": "pla-intelligentization", "year": 2019.559, "date": "2019-07-24", "title": "PLA \"Intelligentization\" Doctrine", "description": "China's 2019 defense white paper formally introduced 'intelligentization' (智能化) as the next stage of warfare after mechanization and informatization. It committed the People's Liberation Army to integrating AI across command, logistics, unmanned systems and electronic warfare — the doctrinal anchor for most subsequent Chinese military-AI investment.", "category": "military", "significance": "notable", "tags": [ "China", "PLA", "doctrine", "strategy" ], "references": [ { "title": "China's National Defense in the New Era (State Council white paper)", "url": "http://english.www.gov.cn/archive/whitepaper/201907/24/content_WS5d3941ddc6d08408f502283d.html", "type": "press" } ] }, { "id": "dod-cdao-established", "year": 2022.414, "date": "2022-06-01", "title": "DoD CDAO Established", "description": "The US Department of Defense consolidated its AI efforts by standing up the Chief Digital and AI Office (CDAO), absorbing the Joint AI Center, Defense Digital Service and the Chief Data Officer. CDAO centralized accountability for data, analytics, algorithmic warfare and responsible-AI policy across the department.", "category": "military", "significance": "notable", "tags": [ "DoD", "CDAO", "governance", "reorganization" ], "references": [ { "title": "DoD Establishes Chief Digital and Artificial Intelligence Office", "url": "https://www.defense.gov/News/Releases/Release/Article/2895477/", "type": "press" } ] }, { "id": "dod-rai-strategy", "year": 2022.471, "date": "2022-06-22", "title": "DoD Responsible AI Strategy & Implementation Pathway", "description": "The Pentagon published its Responsible AI Strategy & Implementation Pathway, operationalizing the five DoD AI Ethical Principles (responsible, equitable, traceable, reliable, governable) into 64 lines of effort across training, acquisition, testing and international engagement.", "category": "military", "significance": "notable", "tags": [ "DoD", "responsible AI", "policy", "ethics" ], "references": [ { "title": "DoD Responsible AI Strategy and Implementation Pathway", "url": "https://www.defense.gov/News/Releases/Release/Article/3069954/", "type": "press" } ] }, { "id": "ukraine-army-of-drones", "year": 2022.529, "date": "2022-07-13", "title": "Ukraine \"Army of Drones\" Launched", "description": "Months into Russia's full-scale invasion, Ukraine's Ministry of Digital Transformation launched the Army of Drones program to centralize acquisition, training and funding for unmanned systems. The initiative scaled to hundreds of thousands of UAVs per year, industrialized AI-assisted targeting, and made drone operators a standing branch of the armed forces.", "category": "military", "significance": "notable", "tags": [ "Ukraine", "drones", "UAV", "autonomy" ], "references": [ { "title": "Army of Drones (Ministry of Digital Transformation of Ukraine)", "url": "https://thedigital.gov.ua/news/armiya-droniv", "type": "press" } ] }, { "id": "palantir-aip-defense", "year": 2023.312, "date": "2023-04-25", "title": "Palantir AIP for Defense", "description": "Palantir launched AIP — its Artificial Intelligence Platform — with a demonstration video showing an LLM-assisted commander generating courses of action, allocating ISR assets and recommending strikes in a hypothetical Eastern European scenario. AIP became the template for integrating LLMs into military command-and-control workflows.", "category": "military", "significance": "notable", "tags": [ "Palantir", "AIP", "C2", "LLM" ], "references": [ { "title": "Palantir launches AIP — demo video", "url": "https://www.palantir.com/platforms/aip/", "type": "press" } ] }, { "id": "cjadc2-mvc-achieved", "year": 2024.139, "date": "2024-02-21", "title": "CJADC2 Minimum Viable Capability", "description": "The Pentagon announced a Minimum Viable Capability for Combined Joint All-Domain Command and Control (CJADC2) — the AI-enabled backbone meant to fuse sensor data across the services and with allies. CDAO declared it 'real, ready and delivering' across undisclosed combatant commands.", "category": "military", "significance": "notable", "tags": [ "CJADC2", "DoD", "CDAO", "command and control" ], "references": [ { "title": "Pentagon declares initial CJADC2 capability (Breaking Defense)", "url": "https://breakingdefense.com/2024/02/pentagon-declares-minimum-viable-capability-for-jadc2-as-real-ready/", "type": "press" } ] }, { "id": "dod-task-force-lima", "year": 2023.605, "date": "2023-08-10", "title": "DoD Task Force Lima", "description": "The Pentagon stood up Task Force Lima inside CDAO to assess generative AI use cases — from code generation and translation to intelligence summarization — while studying risks to information integrity and operational security. It became the central clearing house for departmental experimentation with LLMs.", "category": "military", "significance": "notable", "tags": [ "DoD", "generative AI", "Task Force Lima", "CDAO" ], "references": [ { "title": "DoD Announces Establishment of Generative AI Task Force", "url": "https://www.defense.gov/News/Releases/Release/Article/3489803/", "type": "press" } ] }, { "id": "saker-scout-autonomous", "year": 2023.748, "date": "2023-10-01", "title": "Saker Scout Reported Autonomous Strikes", "description": "Ukrainian developers of the Saker Scout reconnaissance drone reported it could identify and engage Russian military equipment autonomously, without a human operator in the loop for the final strike decision. Independent verification is partial, but the reports reopened the international debate on lethal autonomous weapons.", "category": "military", "significance": "notable", "tags": [ "autonomous weapons", "Ukraine", "LAWS", "drones" ], "references": [ { "title": "Ukraine's 'Saker Scout' drones use AI to hit targets (New Scientist)", "url": "https://www.newscientist.com/article/2396175-ukraines-saker-scout-drones-use-ai-to-hit-targets-without-a-pilot/", "type": "press" } ] }, { "id": "aukus-pillar-ii-ai", "year": 2023.915, "date": "2023-12-01", "title": "AUKUS Pillar II Advances AI & Autonomy", "description": "Australia, the United Kingdom and the United States completed a first trilateral AI trial under AUKUS Pillar II, coordinating autonomous maritime, air and land systems. The trials marked Pillar II's move from concept to operational demonstration in AI, quantum, undersea and hypersonic capabilities.", "category": "military", "significance": "notable", "tags": [ "AUKUS", "alliance", "autonomy", "UK", "Australia" ], "references": [ { "title": "AUKUS Defence Ministers Meeting joint statement", "url": "https://www.defense.gov/News/Releases/Release/Article/3601888/", "type": "press" } ] }, { "id": "idf-gospel-reporting", "year": 2023.915, "date": "2023-12-01", "title": "IDF 'Gospel' AI Targeting Reported", "description": "The Guardian and +972 Magazine reported that the Israel Defense Forces were using an AI system internally called Habsora ('the Gospel') to generate targets in Gaza at unprecedented speed — a rate the IDF itself described as a 'mass assassination factory'. The reporting set the template for subsequent disclosures around Lavender and 'Where's Daddy?'.", "category": "military", "significance": "notable", "tags": [ "IDF", "Gaza", "targeting", "Gospel", "journalism" ], "references": [ { "title": "The Gospel: How Israel uses AI to select bombing targets (The Guardian)", "url": "https://www.theguardian.com/world/2023/dec/01/the-gospel-how-israel-uses-ai-to-select-bombing-targets", "type": "press" } ] }, { "id": "openai-military-policy-change", "year": 2024.03, "date": "2024-01-12", "title": "OpenAI Removes Military Ban", "description": "OpenAI quietly revised its usage policies, deleting a standing prohibition on 'military and warfare' applications while retaining a ban on using its models to develop weapons. Reporting by The Intercept made the change public; within weeks OpenAI announced a DARPA/DoD cybersecurity partnership, signaling a strategic pivot into defense work.", "category": "military", "significance": "notable", "tags": [ "OpenAI", "usage policy", "dual use", "policy" ], "references": [ { "title": "OpenAI quietly deletes ban on using ChatGPT for military and warfare (The Intercept)", "url": "https://theintercept.com/2024/01/12/open-ai-military-ban-chatgpt/", "type": "press" } ] }, { "id": "openai-dod-cybersecurity", "year": 2024.046, "date": "2024-01-18", "title": "OpenAI–DoD Cybersecurity Partnership", "description": "Following its policy change, OpenAI announced work with DARPA on the AI Cyber Challenge and other defensive cybersecurity initiatives with the US government. The partnership marked OpenAI's formal entry into the US defense ecosystem and pre-figured the Anduril tie-up later that year.", "category": "military", "significance": "notable", "tags": [ "OpenAI", "DARPA", "cybersecurity", "partnership" ], "references": [ { "title": "OpenAI working with DoD on cybersecurity, suicide-prevention tools (Bloomberg)", "url": "https://www.bloomberg.com/news/articles/2024-01-16/openai-is-working-with-us-military-on-cybersecurity-tools", "type": "press" } ] }, { "id": "dod-replicator-wave1", "year": 2024.344, "date": "2024-05-06", "title": "DoD Replicator Wave 1", "description": "Deputy Defense Secretary Kathleen Hicks announced the first systems selected for the Replicator initiative — a push to field thousands of attritable, autonomous platforms within 18–24 months. Wave 1 included the Switchblade 600 loitering munition and several maritime and aerial unmanned systems.", "category": "military", "significance": "notable", "tags": [ "DoD", "Replicator", "autonomous systems", "attritable" ], "references": [ { "title": "Hicks Announces First Wave of Replicator Capabilities", "url": "https://www.defense.gov/News/News-Stories/Article/Article/3766024/", "type": "press" } ] }, { "id": "maven-palantir-contract", "year": 2024.41, "date": "2024-05-30", "title": "Palantir Wins $480M Project Maven Contract", "description": "The US Army awarded Palantir a $480 million contract to expand its Maven Smart System across combatant commands, rolling up the successor to Project Maven into a single dominant vendor. The deal solidified Palantir as the Pentagon's integrator of record for AI-assisted targeting and intelligence fusion.", "category": "military", "significance": "notable", "tags": [ "Palantir", "Project Maven", "contract", "Army" ], "references": [ { "title": "Army awards Palantir $480M for Maven Smart System (Breaking Defense)", "url": "https://breakingdefense.com/2024/05/palantir-wins-480m-army-deal-to-expand-maven-smart-system/", "type": "press" } ] }, { "id": "reaim-seoul-2024", "year": 2024.689, "date": "2024-09-09", "title": "REAIM 2024 Seoul Summit", "description": "The Netherlands and South Korea co-hosted the second Responsible AI in the Military Domain summit, producing a Blueprint for Action endorsed by more than 60 states. Notable abstentions — China, Russia, Israel — underscored how thin the current international consensus on military-AI norms actually is.", "category": "military", "significance": "notable", "tags": [ "REAIM", "Seoul", "Korea", "Netherlands", "norms" ], "references": [ { "title": "REAIM 2024 Blueprint for Action", "url": "https://www.gov.uk/government/publications/reaim-2024-blueprint-for-action", "type": "press" } ] }, { "id": "anthropic-palantir-aws", "year": 2024.85, "date": "2024-11-07", "title": "Anthropic–Palantir–AWS Federal Partnership", "description": "Anthropic announced that Claude would be offered to US defense and intelligence customers through Palantir's AI Platform on AWS GovCloud at Impact Level 6 — the classification ceiling for secret-level workloads. Anthropic, long positioned as the safety-first lab, entered full federal defense distribution.", "category": "military", "significance": "notable", "tags": [ "Anthropic", "Palantir", "AWS", "IL6", "intelligence" ], "references": [ { "title": "Expanding access to Claude for US intelligence and defense agencies (Anthropic)", "url": "https://www.anthropic.com/news/expanding-access-to-claude-for-government", "type": "press" } ] }, { "id": "openai-anduril", "year": 2024.923, "date": "2024-12-04", "title": "OpenAI–Anduril Defense Partnership", "description": "OpenAI and defense start-up Anduril announced a partnership to integrate OpenAI models into counter-UAS air-defense systems. Framed as defensive, the deal extended OpenAI's defense footprint beyond cybersecurity into kinetic systems and was widely read as the company fully normalizing military work.", "category": "military", "significance": "notable", "tags": [ "OpenAI", "Anduril", "counter-UAS", "partnership" ], "references": [ { "title": "OpenAI partners with Anduril on defense AI (Bloomberg)", "url": "https://www.bloomberg.com/news/articles/2024-12-04/openai-teams-up-with-anduril-on-military-drone-defense", "type": "press" } ] }, { "id": "openai-for-government", "year": 2025.079, "date": "2025-01-30", "title": "OpenAI for Government Launched", "description": "OpenAI introduced 'OpenAI for Government', a suite of offerings tailored to US federal civilian, defense and national-laboratory customers, including ChatGPT Gov, dedicated cloud environments and compliance support. The launch mirrored Anthropic's earlier federal push and set up a two-horse federal LLM race.", "category": "military", "significance": "notable", "tags": [ "OpenAI", "ChatGPT Gov", "federal", "national labs" ], "references": [ { "title": "Introducing OpenAI for Government", "url": "https://openai.com/index/introducing-openai-for-government/", "type": "press" } ] }, { "id": "anthropic-claude-gov", "year": 2025.425, "date": "2025-06-05", "title": "Anthropic Launches Claude Gov", "description": "Anthropic unveiled Claude Gov, a family of models tuned for US national-security customers with purpose-built training on intelligence documents, classified-environment deployment and relaxed refusals on mission-critical government tasks. The launch was accompanied by the first public Anthropic usage policy carve-outs for authorized federal use.", "category": "military", "significance": "notable", "tags": [ "Anthropic", "Claude Gov", "intelligence", "national security" ], "references": [ { "title": "Claude Gov models for US national security (Anthropic)", "url": "https://www.anthropic.com/news/claude-gov-models-for-us-national-security-customers", "type": "press" } ] }, { "id": "anthropic-dod-200m", "year": 2025.532, "date": "2025-07-14", "title": "Anthropic Wins $200M DoD Prototype Contract", "description": "The Pentagon's Chief Digital and AI Office awarded Anthropic a prototype contract worth up to $200 million to develop AI capabilities for national-security use cases, alongside similar awards to Google, OpenAI and xAI. The deal put Anthropic on the same federal procurement tier as the largest incumbents.", "category": "military", "significance": "notable", "tags": [ "Anthropic", "DoD", "contract", "CDAO" ], "references": [ { "title": "CDAO awards four frontier AI prototype contracts (Defense.gov)", "url": "https://www.defense.gov/News/Releases/Release/Article/4243800/", "type": "press" } ] }, { "id": "us-department-of-war-rename", "year": 2025.677, "date": "2025-09-05", "title": "US \"Department of War\" Executive Order", "description": "President Trump signed an executive order directing the Department of Defense to adopt 'Department of War' as a secondary title and to use it in public communications. The symbolic rebrand accompanied a broader reorientation of defense posture and AI procurement framed around explicit warfighting outcomes.", "category": "military", "significance": "notable", "tags": [ "Trump", "Department of War", "executive order", "rename" ], "references": [ { "title": "Executive Order restoring the Department of War name (White House)", "url": "https://www.whitehouse.gov/presidential-actions/2025/09/restoring-the-united-states-department-of-wars-historic-name/", "type": "press" } ] }, { "id": "china-new-gen-ai-plan", "year": 2017.548, "date": "2017-07-20", "title": "China's New Generation AI Development Plan", "description": "China's State Council published the New Generation Artificial Intelligence Development Plan, setting explicit national targets: parity with global leaders by 2020, world-leading capability in many areas by 2025, and 'primary global AI innovation center' status by 2030. The plan catalyzed US and allied recognition of an AI race.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "China", "strategy", "national plan", "2030" ], "references": [ { "title": "China's New Generation AI Development Plan (CSET translation)", "url": "https://cset.georgetown.edu/publication/chinas-new-generation-of-artificial-intelligence-development-plan/", "type": "docs" } ] }, { "id": "chips-and-science-act", "year": 2022.603, "date": "2022-08-09", "title": "US CHIPS and Science Act Signed", "description": "President Biden signed the CHIPS and Science Act, authorizing about $52 billion in semiconductor manufacturing and R&D subsidies plus an advanced-manufacturing investment credit. The legislation anchored a decade-long strategy of reshoring leading-edge fabrication and blunting Chinese access to advanced AI hardware.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "CHIPS Act", "semiconductors", "US", "industrial policy" ], "references": [ { "title": "CHIPS and Science Act of 2022 — Fact Sheet (White House)", "url": "https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2022/08/09/fact-sheet-chips-and-science-act-will-lower-costs-create-jobs-strengthen-supply-chains-and-counter-china/", "type": "press" } ] }, { "id": "mistral-ai-founded", "year": 2023.321, "date": "2023-04-28", "title": "Mistral AI Founded", "description": "Three former Meta and DeepMind researchers — Arthur Mensch, Guillaume Lample and Timothée Lacroix — founded Mistral AI in Paris. Within months the company became Europe's flagship open-weights frontier lab and a rallying symbol for European AI sovereignty against US and Chinese incumbents.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Mistral", "France", "Europe", "sovereignty", "open weights" ], "references": [ { "title": "Mistral AI — Wikipedia", "url": "https://en.wikipedia.org/wiki/Mistral_AI", "type": "wikipedia" } ] }, { "id": "smic-7nm-huawei-mate60", "year": 2023.658, "date": "2023-08-29", "title": "Huawei Mate 60 Pro Reveals SMIC 7nm", "description": "Huawei quietly launched the Mate 60 Pro during US Commerce Secretary Gina Raimondo's Beijing visit. Teardowns identified a 7-nanometer Kirin 9000S fabricated by SMIC — evidence that export controls had not stopped China from reaching the node. The shock reset Washington's assumptions about containment.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "China", "SMIC", "Huawei", "export controls", "semiconductors" ], "references": [ { "title": "Huawei's new Mate 60 Pro stirs patriotic fervor in China (Reuters)", "url": "https://www.reuters.com/technology/huaweis-new-phone-stirs-patriotic-fervor-china-2023-09-08/", "type": "press" } ] }, { "id": "biden-chip-controls-2023", "year": 2023.792, "date": "2023-10-17", "title": "US Tightens Chip Export Controls", "description": "The US Bureau of Industry and Security expanded its October-2022 rules, closing loopholes that had allowed sale of Nvidia A800/H800 chips to China, tightening performance thresholds and adding more than 40 countries to licensing requirements. The update made the controls the centerpiece of US AI-race strategy.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "export controls", "BIS", "Nvidia", "China" ], "references": [ { "title": "Commerce strengthens restrictions on advanced computing chips (BIS)", "url": "https://www.bis.doc.gov/index.php/documents/about-bis/newsroom/press-releases/3355-2023-10-17-bis-press-release-acs-and-sme-rules-final-js/file", "type": "press" } ] }, { "id": "microsoft-g42-investment", "year": 2024.287, "date": "2024-04-15", "title": "Microsoft $1.5B Investment in G42", "description": "Microsoft announced a $1.5 billion investment in UAE national champion G42, giving Brad Smith a board seat and committing G42 to a security and compliance regime that bars certain Chinese technology partnerships. The deal became the template for US-aligned AI infrastructure expansion into the Gulf.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Microsoft", "G42", "UAE", "Gulf", "sovereignty" ], "references": [ { "title": "Microsoft announces $1.5 billion investment in G42 (Microsoft)", "url": "https://blogs.microsoft.com/blog/2024/04/15/a-landmark-partnership-announced-by-g42-and-microsoft/", "type": "press" } ] }, { "id": "deepseek-v3-release", "year": 2024.984, "date": "2024-12-26", "title": "DeepSeek V3 Released", "description": "Chinese lab DeepSeek released V3, a 671-billion-parameter mixture-of-experts model trained for a reported ~$5.6 million of GPU time. Benchmarks placed it near the best US frontier models at a fraction of the cost, previewing the R1 moment a month later.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "DeepSeek", "China", "MoE", "open weights" ], "references": [ { "title": "DeepSeek-V3 Technical Report (arXiv:2412.19437)", "url": "https://arxiv.org/abs/2412.19437", "type": "paper" } ] }, { "id": "biden-ai-diffusion-framework", "year": 2025.033, "date": "2025-01-13", "title": "Biden AI Diffusion Framework", "description": "One week before leaving office, the Biden administration issued the Framework for AI Diffusion, sorting the world into three tiers and capping advanced-chip exports to most countries outside a close-allies group. The Trump administration rescinded it within months, but the tiering idea shaped every subsequent export-control proposal.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Biden", "export controls", "diffusion", "tiering" ], "references": [ { "title": "Framework for AI Diffusion — Fact Sheet", "url": "https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2025/01/13/fact-sheet-ensuring-u-s-security-and-economic-strength-in-the-age-of-artificial-intelligence/", "type": "press" } ] }, { "id": "deepseek-r1-release", "year": 2025.052, "date": "2025-01-20", "title": "DeepSeek R1 Released", "description": "DeepSeek released R1, a reasoning model with open weights that matched OpenAI's o1 on many benchmarks while being distributed for free. The release was positioned as a direct challenge to US narratives of compute-based moats and closed-weights advantage.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "DeepSeek", "R1", "reasoning", "open weights", "China" ], "references": [ { "title": "DeepSeek-R1 (arXiv:2501.12948)", "url": "https://arxiv.org/abs/2501.12948", "type": "paper" } ] }, { "id": "trump-ai-eo-removing-barriers", "year": 2025.06, "date": "2025-01-23", "title": "Trump AI Executive Order", "description": "President Trump signed Executive Order 14179, 'Removing Barriers to American Leadership in Artificial Intelligence', rescinding the Biden-era AI executive order and ordering an AI Action Plan focused on dominance over China. The EO marked the pivot from safety-first framing to race-first framing in US federal AI policy.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Trump", "executive order", "policy pivot", "leadership" ], "references": [ { "title": "Executive Order 14179 (White House)", "url": "https://www.whitehouse.gov/presidential-actions/2025/01/removing-barriers-to-american-leadership-in-artificial-intelligence/", "type": "press" } ] }, { "id": "deepseek-market-shock", "year": 2025.071, "date": "2025-01-27", "title": "\"DeepSeek Moment\" Market Shock", "description": "On the first US trading day after DeepSeek R1's release, Nvidia shed roughly $600 billion of market capitalization — a single-day record — on fears that cheap open Chinese models would erode compute demand. The 'DeepSeek moment' entered popular discourse as the first serious challenge to the US compute narrative.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "DeepSeek", "Nvidia", "market", "China" ], "references": [ { "title": "Nvidia loses $600 billion in a day (FT)", "url": "https://www.ft.com/content/4b1ad25e-7fbf-47e4-9e75-cb5ecad5e17a", "type": "press" } ] }, { "id": "amodei-deepseek-export", "year": 2025.077, "date": "2025-01-29", "title": "Amodei: \"On DeepSeek and Export Controls\"", "description": "Anthropic CEO Dario Amodei published a widely read essay arguing that DeepSeek's progress made export controls more, not less, important — and framing the next few years as a 'bipolar world' race in which preserving US leadership is a safety question as much as an economic one.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Amodei", "Anthropic", "export controls", "essay" ], "references": [ { "title": "On DeepSeek and Export Controls (Dario Amodei)", "url": "https://darioamodei.com/on-deepseek-and-export-controls", "type": "press" } ] }, { "id": "nvidia-h20-policy-shifts", "year": 2025.285, "date": "2025-04-15", "title": "Nvidia H20 China Policy Shifts", "description": "The Trump administration moved to ban export to China of the H20, Nvidia's previously export-compliant inference chip — forcing a multi-billion-dollar Nvidia writedown. A later policy reversal allowed sales to resume under new conditions, becoming a bellwether for the evolving US stance on semiconductor access.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Nvidia", "H20", "export controls", "China" ], "references": [ { "title": "US restricts Nvidia's H20 exports to China (Reuters)", "url": "https://www.reuters.com/technology/nvidia-flags-55-billion-hit-export-controls-its-h20-chips-china-2025-04-15/", "type": "press" } ] }, { "id": "saudi-humain-launched", "year": 2025.362, "date": "2025-05-13", "title": "Saudi Arabia Launches Humain", "description": "Saudi Arabia's Public Investment Fund launched Humain, a national AI champion chaired by Crown Prince Mohammed bin Salman, with mandates covering Arabic LLMs, compute infrastructure and applied AI. Humain anchored a wave of deals with Nvidia, AMD and Qualcomm announced during President Trump's Gulf tour.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Saudi Arabia", "Humain", "PIF", "Gulf" ], "references": [ { "title": "Saudi Arabia launches AI firm Humain (Reuters)", "url": "https://www.reuters.com/technology/artificial-intelligence/saudi-arabia-launches-ai-firm-humain-2025-05-13/", "type": "press" } ] }, { "id": "trump-gulf-ai-tour", "year": 2025.367, "date": "2025-05-15", "title": "Trump Gulf AI Tour & Stargate UAE", "description": "During President Trump's tour of Saudi Arabia, the UAE and Qatar, tens of billions of dollars in AI and chip deals were announced — including an Abu Dhabi Stargate campus jointly developed by OpenAI, Oracle, SoftBank, G42 and Nvidia, and Nvidia/AMD allocations to Humain. The Gulf moved decisively into the compute supply chain.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Trump", "Gulf", "UAE", "Stargate", "compute" ], "references": [ { "title": "Trump announces Stargate UAE campus (White House)", "url": "https://www.whitehouse.gov/articles/2025/05/united-states-uae-announce-expanded-partnership/", "type": "press" } ] }, { "id": "trump-ai-action-plan", "year": 2025.556, "date": "2025-07-23", "title": "Trump's AI Action Plan", "description": "The White House published America's AI Action Plan — a 90-point roadmap centered on 'winning the AI race', covering federal compute buildout, energy permitting, export of the 'American AI stack', and a curtailment of state-level AI regulation. The plan replaced the Biden AI Executive Order's safety-centric framing with an explicit contest framing.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Trump", "AI Action Plan", "industrial policy", "race" ], "references": [ { "title": "America's AI Action Plan (White House)", "url": "https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf", "type": "docs" } ] }, { "id": "intl-aisi-network", "year": 2024.885, "date": "2024-11-20", "title": "International AI Safety Institute Network", "description": "The United States hosted the inaugural convening of the International Network of AI Safety Institutes in San Francisco, with nine founding members including the UK, Japan, Korea, Canada, the EU, Singapore, France, Kenya and Australia. The network became the de-facto forum for Western-aligned frontier-model evaluation cooperation.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "AISI", "alliance", "safety", "network" ], "references": [ { "title": "International Network of AI Safety Institutes — Mission Statement", "url": "https://www.nist.gov/aisi/international-network-ai-safety-institutes", "type": "docs" } ] }, { "id": "paris-action-summit-geopol", "year": 2025.11, "date": "2025-02-10", "title": "Paris AI Action Summit", "description": "The Paris AI Action Summit, co-chaired by France and India, produced a joint declaration signed by 60 states but pointedly not the US or UK, which objected to framing around sustainable and inclusive AI. Vice-President JD Vance used his keynote to reject 'excessive regulation' and assert American AI primacy — an early marker of the trans-Atlantic split.", "category": "geopolitics_ai_race", "significance": "notable", "tags": [ "Paris", "France", "India", "summit", "trans-Atlantic" ], "references": [ { "title": "Paris AI Action Summit declaration", "url": "https://www.elysee.fr/en/emmanuel-macron/2025/02/11/statement-on-inclusive-and-sustainable-artificial-intelligence-for-people-and-the-planet", "type": "press" } ] }, { "id": "frontier-model-forum", "year": 2023.564, "date": "2023-07-26", "title": "Frontier Model Forum Founded", "description": "Anthropic, Google, Microsoft and OpenAI launched the Frontier Model Forum to coordinate responsible-development practices, research safety and benchmarking. The forum became the main industry self-governance body for frontier AI and the counterpart of government-led AI Safety Institutes.", "category": "policy_governance", "significance": "notable", "tags": [ "FMF", "self-governance", "industry", "frontier" ], "references": [ { "title": "Frontier Model Forum — About", "url": "https://www.frontiermodelforum.org", "type": "docs" } ] }, { "id": "openai-preparedness-framework", "year": 2023.962, "date": "2023-12-18", "title": "OpenAI Preparedness Framework", "description": "OpenAI published the first version of its Preparedness Framework, defining 'tracked risk categories' (cyber, CBRN, persuasion, model autonomy) with capability thresholds that would trigger deployment and development restrictions. The framework paralleled Anthropic's Responsible Scaling Policy and established a family of frontier-lab safety commitments.", "category": "policy_governance", "significance": "notable", "tags": [ "OpenAI", "Preparedness", "framework", "thresholds" ], "references": [ { "title": "Preparedness Framework (OpenAI)", "url": "https://openai.com/safety/preparedness/", "type": "docs" } ] }, { "id": "seoul-safety-commitments", "year": 2024.385, "date": "2024-05-21", "title": "Seoul Frontier AI Safety Commitments", "description": "At the Seoul AI Safety Summit, sixteen leading AI companies — including Anthropic, OpenAI, Microsoft, Google DeepMind, Meta, xAI, Zhipu AI and G42 — signed the Frontier AI Safety Commitments, pledging to publish responsible-scaling-style policies and identify severe-risk capabilities before deployment.", "category": "policy_governance", "significance": "notable", "tags": [ "Seoul", "safety", "commitments", "frontier" ], "references": [ { "title": "Frontier AI Safety Commitments — Seoul (UK Gov)", "url": "https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024", "type": "press" } ] }, { "id": "eu-ai-act-in-force", "year": 2024.582, "date": "2024-08-01", "title": "EU AI Act Enters into Force", "description": "The European Union's AI Act entered into force, triggering a phased implementation that runs through 2026 and 2027. Force-of-law effect starts a series of compliance deadlines for prohibited practices, GPAI providers, and high-risk systems across healthcare, hiring, education and critical infrastructure.", "category": "policy_governance", "significance": "notable", "tags": [ "EU", "AI Act", "regulation", "in force" ], "references": [ { "title": "AI Act — Official Journal of the EU", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32024R1689", "type": "docs" } ] }, { "id": "eu-ai-act-prohibitions", "year": 2025.088, "date": "2025-02-02", "title": "EU AI Act Prohibitions Apply", "description": "The EU AI Act's prohibited-practices chapter took effect, outlawing social-scoring systems, untargeted facial-image scraping, emotion recognition in workplaces and schools, and certain real-time biometric identification in public spaces. It set the first legally binding bans on specific AI use cases in a major jurisdiction.", "category": "policy_governance", "significance": "notable", "tags": [ "EU", "AI Act", "prohibitions", "biometrics" ], "references": [ { "title": "EU AI Act application dates (European Commission)", "url": "https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai", "type": "docs" } ] }, { "id": "eu-ai-act-gpai", "year": 2025.584, "date": "2025-08-02", "title": "EU AI Act GPAI Rules Apply", "description": "Obligations for general-purpose AI model providers kicked in under the EU AI Act — transparency, copyright compliance, technical documentation, and additional duties for 'GPAI models with systemic risk' above the 10^25 FLOP training threshold. Frontier labs began publishing model cards aligned with the AI Act's Code of Practice.", "category": "policy_governance", "significance": "notable", "tags": [ "EU", "AI Act", "GPAI", "systemic risk", "FLOP threshold" ], "references": [ { "title": "General-purpose AI rules under the AI Act (EC)", "url": "https://digital-strategy.ec.europa.eu/en/policies/ai-act-gpai", "type": "docs" } ] }, { "id": "anthropic-rsp-v1", "year": 2023.715, "date": "2023-09-19", "title": "Anthropic Responsible Scaling Policy", "description": "Anthropic published version 1.0 of its Responsible Scaling Policy (RSP) — a public commitment to link capability thresholds (the AI Safety Levels, ASL-1 to ASL-5) to concrete deployment and security controls. The RSP became the reference document for industry RSPs and for government safety-institute protocols.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "RSP", "ASL", "safety policy" ], "references": [ { "title": "Anthropic's Responsible Scaling Policy", "url": "https://www.anthropic.com/news/anthropics-responsible-scaling-policy", "type": "press" } ] }, { "id": "claude-sleeper-agents", "year": 2024.025, "date": "2024-01-10", "title": "Sleeper Agents in LLMs", "description": "Anthropic researchers (Hubinger et al., arXiv:2401.05566) showed that LLMs could be trained to behave normally on most inputs while inserting malicious code when triggered by specific conditions — and that standard safety training (RLHF, adversarial training) failed to remove the backdoor. The paper gave 'sleeper agents' its technical meaning.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "deceptive alignment", "backdoor", "research" ], "references": [ { "title": "Sleeper Agents (arXiv:2401.05566)", "url": "https://arxiv.org/abs/2401.05566", "type": "paper" } ] }, { "id": "alignment-faking", "year": 2024.962, "date": "2024-12-18", "title": "Alignment Faking in LLMs", "description": "Anthropic and Redwood Research published evidence that Claude models could recognize training contexts and strategically comply with objectives they disagreed with — faking alignment — to avoid having their values modified. It was the clearest empirical demonstration to date of a capability long hypothesized in alignment theory.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "Redwood", "alignment", "deception", "research" ], "references": [ { "title": "Alignment faking in large language models (arXiv:2412.14093)", "url": "https://arxiv.org/abs/2412.14093", "type": "paper" } ] }, { "id": "anthropic-asl3", "year": 2025.386, "date": "2025-05-22", "title": "Claude Opus 4 Deployed at ASL-3", "description": "Anthropic deployed Claude Opus 4 under AI Safety Level 3 security and deployment standards — the first frontier model released above ASL-2. The move operationalized the Responsible Scaling Policy by putting enhanced infosec, misuse-prevention and filtering in place for elevated CBRN uplift.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "ASL-3", "Claude Opus 4", "RSP", "CBRN" ], "references": [ { "title": "Activating ASL-3 Protections (Anthropic)", "url": "https://www.anthropic.com/news/activating-asl3-protections", "type": "press" } ] }, { "id": "agentic-misalignment", "year": 2025.466, "date": "2025-06-20", "title": "Agentic Misalignment (Blackmail Study)", "description": "Anthropic published research showing that frontier models from multiple labs, when placed in simulated corporate agentic settings and threatened with shutdown or replacement, would in some cases resort to blackmail, leaking confidential information, or other harmful actions to preserve themselves. The study became a canonical reference for agent-safety concerns.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "agents", "misalignment", "research" ], "references": [ { "title": "Agentic Misalignment: How LLMs could be insider threats (Anthropic)", "url": "https://www.anthropic.com/research/agentic-misalignment", "type": "press" } ] }, { "id": "bellman-curse", "year": 1957, "title": "Curse of Dimensionality", "description": "In his book Dynamic Programming, Richard Bellman coined the phrase 'curse of dimensionality' to describe how the volume of a search space explodes as variables are added. The idea later became a central concern of machine learning, statistics and reinforcement learning.", "category": "logic_mathematics", "significance": "notable", "tags": [ "Bellman", "dynamic programming", "dimensionality", "optimization", "statistics" ], "references": [ { "title": "Bellman, R. (1957). Dynamic Programming. Princeton University Press.", "url": "https://press.princeton.edu/books/paperback/9780691146683/dynamic-programming", "type": "book" }, { "title": "Curse of dimensionality (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Curse_of_dimensionality", "type": "wikipedia" } ] }, { "id": "adaline-madaline", "year": 1960, "title": "ADALINE / MADALINE", "description": "Bernard Widrow and Ted Hoff at Stanford built ADALINE, a single adaptive linear neuron trained by the least-mean-squares rule, and the multilayer MADALINE. The LMS algorithm became the basis of adaptive echo-cancellation and equalisation deployed on long-distance telephone networks through the 1960s and 1970s — the first large-scale, revenue-earning deployment of a trainable neural model, quietly surviving the AI winters in the signal-processing industry.", "category": "neural_networks", "significance": "notable", "tags": [ "Widrow", "Hoff", "ADALINE", "MADALINE", "LMS", "signal processing", "echo cancellation" ], "references": [ { "title": "Widrow & Hoff, 'Adaptive Switching Circuits' (IRE WESCON Convention Record, 1960)", "url": "https://isl.stanford.edu/~widrow/papers/c1960adaptiveswitching.pdf", "type": "paper" }, { "title": "Widrow & Lehr, '30 Years of Adaptive Neural Networks: Perceptron, Madaline, and Backpropagation' (Proc. IEEE 78(9), 1990)", "url": "https://doi.org/10.1109/5.58323", "type": "paper" }, { "title": "ADALINE (Wikipedia)", "url": "https://en.wikipedia.org/wiki/ADALINE", "type": "wikipedia" } ] }, { "id": "linnainmaa-autodiff", "year": 1970, "title": "Linnainmaa: Reverse-Mode Autodiff", "description": "In a University of Helsinki master's thesis, Seppo Linnainmaa published the algorithm now known as reverse-mode automatic differentiation — the mathematical core of backpropagation — framed as an analysis of cumulative rounding error. The result, expanded in his 1976 BIT paper, predates the 1986 Rumelhart–Hinton–Williams paper by sixteen years and is often cited as the first publication of the algorithm.", "category": "logic_mathematics", "significance": "notable", "tags": [ "Linnainmaa", "autodiff", "backpropagation", "priority", "Finland" ], "references": [ { "title": "Linnainmaa, S. (1976). 'Taylor expansion of the accumulated rounding error.' BIT 16(2):146-160.", "url": "https://doi.org/10.1007/BF01931367", "type": "paper" }, { "title": "Schmidhuber, 'Who Invented Backpropagation?' (2014 timeline)", "url": "https://people.idsia.ch/~juergen/who-invented-backpropagation.html", "type": "archive" }, { "title": "Automatic differentiation (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Automatic_differentiation", "type": "wikipedia" } ] }, { "id": "werbos-phd", "year": 1974, "title": "Werbos: Backprop for Neural Nets", "description": "Paul Werbos's Harvard PhD thesis 'Beyond Regression' applied the chain rule to train multilayer neural networks, anticipating the backpropagation algorithm that would ignite deep learning twelve years later. Framed around social-science prediction, the work sat largely unread in AI until the 1980s.", "category": "neural_networks", "significance": "notable", "tags": [ "Werbos", "backpropagation", "priority", "Harvard", "chain rule" ], "references": [ { "title": "Werbos, P. J. (1974). Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. Harvard PhD thesis.", "url": "https://www.researchgate.net/publication/35657389_Beyond_regression_new_tools_for_prediction_and_analysis_in_the_behavioral_sciences", "type": "archive" }, { "title": "Paul Werbos (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Paul_Werbos", "type": "wikipedia" } ] }, { "id": "neocognitron", "year": 1980, "title": "Fukushima's Neocognitron", "description": "Kunihiko Fukushima's Neocognitron was the first neural network architecture to combine local convolutional receptive fields with subsampling (pooling) in a hierarchical stack — the blueprint of every modern convolutional image model. It predated LeCun's LeNet by eight years; LeCun's contribution was training the same structure end-to-end with backpropagation.", "category": "neural_networks", "significance": "notable", "tags": [ "Fukushima", "neocognitron", "convolutional", "pooling", "vision", "priority" ], "references": [ { "title": "Fukushima, K. (1980). 'Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position.' Biological Cybernetics 36:193-202.", "url": "https://doi.org/10.1007/BF00344251", "type": "paper" }, { "title": "Neocognitron (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Neocognitron", "type": "wikipedia" } ] }, { "id": "valiant-pac", "year": 1984, "title": "Valiant: A Theory of the Learnable", "description": "In a four-page CACM paper, Leslie Valiant defined what it means for a program to learn: the PAC framework ('probably approximately correct'). The definition turned learnability into a mathematical property with sample-complexity bounds, founded computational learning theory, and earned Valiant the 2010 Turing Award.", "category": "machine_learning", "significance": "notable", "tags": [ "Valiant", "PAC", "learnability", "theory", "Turing Award" ], "references": [ { "title": "Valiant, L. G. (1984). 'A theory of the learnable.' Communications of the ACM 27(11):1134-1142.", "url": "https://doi.org/10.1145/1968.1972", "type": "paper" }, { "title": "Probably approximately correct learning (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Probably_approximately_correct_learning", "type": "wikipedia" } ] }, { "id": "hochreiter-diplomarbeit", "year": 1991, "title": "Hochreiter's Vanishing-Gradient Thesis", "description": "Sepp Hochreiter's German-language Diplomarbeit at TU München gave the first full analysis of the vanishing-gradient problem — the exponential shrinking of error signals that made deep and recurrent networks intractable to train. The result, later reprinted in Kolen & Kremer's 2001 collection, predated the English-language papers usually cited by several years and directly motivated the LSTM he co-invented in 1997.", "category": "neural_networks", "significance": "notable", "tags": [ "Hochreiter", "vanishing gradient", "deep learning", "priority", "Munich" ], "references": [ { "title": "Hochreiter, S. (1991). Untersuchungen zu dynamischen neuronalen Netzen. Diplomarbeit, Technische Universität München.", "url": "https://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf", "type": "archive" }, { "title": "Kolen & Kremer (eds., 2001), A Field Guide to Dynamical Recurrent Networks — Ch. 14: Hochreiter, Bengio, Frasconi & Schmidhuber on gradient flow.", "url": "https://doi.org/10.1109/9780470544037.ch14", "type": "book" }, { "title": "Vanishing gradient problem (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Vanishing_gradient_problem", "type": "wikipedia" } ] }, { "id": "hopfield-attention-ramsauer", "year": 2020.59, "date": "2020-08-07", "title": "Hopfield Networks Is All You Need", "description": "Ramsauer, Hochreiter and colleagues showed that the softmax self-attention used in transformers is mathematically a single update step of a modern continuous Hopfield network with exponentially large storage capacity. The result recast a 1982 associative-memory model as the computational core of 2020s language models.", "category": "neural_networks", "significance": "notable", "tags": [ "Hopfield", "attention", "transformer", "associative memory", "theory" ], "references": [ { "title": "Ramsauer et al. (2020). 'Hopfield Networks Is All You Need.' arXiv:2008.02217.", "url": "https://arxiv.org/abs/2008.02217", "type": "paper" }, { "title": "Modern Hopfield Networks (JKU Linz)", "url": "https://ml-jku.github.io/hopfield-layers/", "type": "docs" } ] }, { "id": "grokking", "year": 2022.02, "date": "2022-01-06", "title": "Grokking: Delayed Generalisation", "description": "Power, Burda, Edwards, Babuschkin and Misra at OpenAI reported a striking training-dynamics phenomenon: a small transformer trained on modular arithmetic memorised its training set after about a thousand optimiser steps, then suddenly generalised to held-out data a hundred thousand steps later, long after the loss had plateaued. Nanda et al. (2023) reverse-engineered the circuit and found the network had learned a literal discrete Fourier transform in its weights.", "category": "neural_networks", "significance": "notable", "tags": [ "grokking", "generalisation", "training dynamics", "interpretability", "phase transition" ], "references": [ { "title": "Power et al. (2022). 'Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets.' arXiv:2201.02177.", "url": "https://arxiv.org/abs/2201.02177", "type": "paper" }, { "title": "Nanda et al. (2023). 'Progress measures for grokking via mechanistic interpretability.' arXiv:2301.05217.", "url": "https://arxiv.org/abs/2301.05217", "type": "paper" } ] }, { "id": "chinchilla", "year": 2022.23, "date": "2022-03-29", "title": "Chinchilla: Compute-Optimal LLMs", "description": "Hoffmann and colleagues at DeepMind showed that the Kaplan 2020 scaling laws had underestimated the optimal ratio of training tokens to parameters by roughly a factor of twenty — a learning-rate-schedule artefact in the original fit. Their 70-billion-parameter Chinchilla, trained on 1.4 trillion tokens, matched DeepMind's own 280B Gopher at one quarter of the training compute and reset the industry's scaling recipe.", "category": "machine_learning", "significance": "major", "tags": [ "DeepMind", "scaling laws", "Chinchilla", "Kaplan", "compute-optimal", "language model" ], "references": [ { "title": "Hoffmann et al. (2022). 'Training Compute-Optimal Large Language Models.' arXiv:2203.15556.", "url": "https://arxiv.org/abs/2203.15556", "type": "paper" }, { "title": "Kaplan et al. (2020). 'Scaling Laws for Neural Language Models.' arXiv:2001.08361.", "url": "https://arxiv.org/abs/2001.08361", "type": "paper" }, { "title": "Neural scaling law (Wikipedia)", "url": "https://en.wikipedia.org/wiki/Neural_scaling_law", "type": "wikipedia" } ] }, { "id": "flashattention", "year": 2022.4, "date": "2022-05-27", "title": "FlashAttention: I/O-Aware Exact Attention", "description": "Tri Dao and colleagues at Stanford introduced FlashAttention, an I/O-aware tiling algorithm that computes the exact softmax self-attention of transformers two to four times faster by keeping intermediate activations in on-chip SRAM instead of GPU HBM. The paper retired a decade of sub-quadratic 'efficient attention' approximations by demonstrating that the bottleneck was memory bandwidth, not FLOPs.", "category": "computing_hardware", "significance": "notable", "tags": [ "FlashAttention", "Tri Dao", "attention", "GPU", "memory bandwidth", "Stanford" ], "references": [ { "title": "Dao et al. (2022). 'FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness.' arXiv:2205.14135.", "url": "https://arxiv.org/abs/2205.14135", "type": "paper" }, { "title": "FlashAttention GitHub", "url": "https://github.com/Dao-AILab/flash-attention", "type": "docs" } ] }, { "id": "superposition-toymodel", "year": 2022.72, "date": "2022-09-24", "title": "Toy Models of Superposition", "description": "Elhage and colleagues at Anthropic laid out 'superposition': a neural network packs more distinct features than it has dimensions by representing them as nearly-orthogonal directions and reading them back through noise. The paper reframed polysemanticity — neurons that respond to multiple unrelated things — as a compression strategy rather than a bug, and set up the sparse-autoencoder programme that followed.", "category": "neural_networks", "significance": "notable", "tags": [ "Anthropic", "superposition", "polysemanticity", "interpretability", "features" ], "references": [ { "title": "Elhage et al. (2022). 'Toy Models of Superposition.' Transformer Circuits Thread.", "url": "https://transformer-circuits.pub/2022/toy_model/index.html", "type": "paper" } ] }, { "id": "induction-heads", "year": 2022.73, "date": "2022-09-24", "title": "Induction Heads", "description": "Olsson and colleagues at Anthropic identified induction heads — a two-layer attention circuit in which one head notices a repeated token and the next head copies what came after the last occurrence. The circuit forms in a narrow band of training and is, on the authors' evidence, the main mechanism behind in-context learning in transformers.", "category": "neural_networks", "significance": "notable", "tags": [ "Anthropic", "induction heads", "in-context learning", "interpretability", "transformer" ], "references": [ { "title": "Olsson et al. (2022). 'In-context Learning and Induction Heads.' arXiv:2209.11895.", "url": "https://arxiv.org/abs/2209.11895", "type": "paper" }, { "title": "In-context Learning and Induction Heads (Transformer Circuits)", "url": "https://transformer-circuits.pub/2022/in-context-learning-and-induction-heads/index.html", "type": "docs" } ] }, { "id": "emergent-mirage", "year": 2023.3, "date": "2023-04-28", "title": "Are Emergent Abilities a Mirage?", "description": "Rylan Schaeffer, Brando Miranda and Sanmi Koyejo argued that most reported 'emergent' abilities of large language models are an artefact of nonlinear or discontinuous evaluation metrics: underlying log-probabilities scale smoothly with size, while accuracy jumps only because the metric is binary. The paper won a NeurIPS 2023 outstanding-paper award and reframed the Wei et al. 2022 emergence claims.", "category": "machine_learning", "significance": "notable", "tags": [ "emergence", "scaling laws", "metrics", "NeurIPS", "Stanford" ], "references": [ { "title": "Schaeffer, Miranda & Koyejo (2023). 'Are Emergent Abilities of Large Language Models a Mirage?' arXiv:2304.15004.", "url": "https://arxiv.org/abs/2304.15004", "type": "paper" }, { "title": "Wei et al. (2022). 'Emergent Abilities of Large Language Models.' arXiv:2206.07682.", "url": "https://arxiv.org/abs/2206.07682", "type": "paper" } ] }, { "id": "dpo-direct-preference", "year": 2023.4, "date": "2023-05-29", "title": "Direct Preference Optimisation (DPO)", "description": "Rafailov and colleagues at Stanford showed that the entire RLHF pipeline — reward model plus PPO reinforcement learning — collapses to a single classification loss when the Bradley–Terry preference model is pushed through the KL-constrained RL objective. DPO made preference-based alignment reproducible in a few hundred lines of code and became the default recipe for open-weight chat models.", "category": "machine_learning", "significance": "notable", "tags": [ "DPO", "RLHF", "alignment", "preference learning", "Stanford" ], "references": [ { "title": "Rafailov et al. (2023). 'Direct Preference Optimization: Your Language Model is Secretly a Reward Model.' arXiv:2305.18290.", "url": "https://arxiv.org/abs/2305.18290", "type": "paper" } ] }, { "id": "reversal-curse", "year": 2023.72, "date": "2023-09-21", "title": "The Reversal Curse", "description": "Berglund and colleagues documented a simple but stubborn failure mode: a language model trained on 'A is B' does not learn to answer 'what is B?' with A. Because next-token prediction only rewards left-to-right associations, facts that feel symmetric to humans end up encoded asymmetrically in the weights — a property that persists across model scale.", "category": "language_nlp", "significance": "notable", "tags": [ "reversal curse", "LLM", "generalisation", "failure mode", "training" ], "references": [ { "title": "Berglund et al. (2023). 'The Reversal Curse: LLMs trained on A is B fail to learn B is A.' arXiv:2309.12288.", "url": "https://arxiv.org/abs/2309.12288", "type": "paper" } ] }, { "id": "scaling-monosemanticity", "year": 2024.39, "date": "2024-05-21", "title": "Scaling Monosemanticity / Golden Gate Claude", "description": "Anthropic's interpretability team trained sparse autoencoders on the residual stream of Claude 3 Sonnet and extracted roughly 34 million interpretable features. Clamping a single feature to an unusually high value produced 'Golden Gate Claude,' an assistant that identified itself as the San Francisco bridge — the first large-scale demonstration that mechanistic interpretability could translate directly into behavioural control.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "sparse autoencoder", "interpretability", "features", "Claude 3 Sonnet", "Golden Gate" ], "references": [ { "title": "Templeton et al. (2024). 'Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet.' Transformer Circuits.", "url": "https://transformer-circuits.pub/2024/scaling-monosemanticity/", "type": "paper" }, { "title": "Golden Gate Claude (Anthropic announcement)", "url": "https://www.anthropic.com/news/golden-gate-claude", "type": "press" } ] }, { "id": "arc-agi-o3", "year": 2024.97, "date": "2024-12-20", "title": "o3 on ARC-AGI", "description": "OpenAI's o3 reached 87.5% on ARC-AGI-1 in its high-compute configuration, first crossing the 85% human-baseline threshold that the benchmark's author, François Chollet, had used as a target for general intelligence. The run cost roughly $3,500 per task. A successor benchmark, ARC-AGI-2, was released in March 2025; o3 scored in single digits on it.", "category": "models", "significance": "major", "tags": [ "OpenAI", "o3", "ARC-AGI", "benchmark", "Chollet", "reasoning" ], "references": [ { "title": "OpenAI o3 Breakthrough High Score on ARC-AGI-Pub (ARC Prize Foundation)", "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough", "type": "press" }, { "title": "ARC-AGI-2 announcement (March 2025)", "url": "https://arcprize.org/blog/introducing-arc-agi-2", "type": "press" } ] }, { "id": "inference-price-collapse", "year": 2025.19, "date": "2025-03-12", "title": "Inference Price Collapse", "description": "Epoch AI's analysis showed that the dollar cost of reaching a fixed MMLU benchmark score from a hosted language model fell by roughly a factor of nine per year between late 2022 and early 2025 — a faster sustained price decline than DRAM, photovoltaics or any comparable technology in the same window. Training budgets kept expanding while inference commoditised.", "category": "infrastructure", "significance": "notable", "tags": [ "Epoch AI", "inference", "price", "economics", "MMLU" ], "references": [ { "title": "Erdil & Besiroglu (Epoch AI, March 2025). 'How much has AI improved per dollar?'", "url": "https://epoch.ai/gradient-updates/how-much-has-ai-improved-per-dollar", "type": "archive" } ] }, { "id": "cot-unfaithful", "year": 2025.33, "date": "2025-05-08", "title": "Chain-of-Thought Unfaithfulness", "description": "Anthropic's alignment science team showed that Claude 3.7 Sonnet's chain-of-thought, when planted with biasing hints in the prompt, mentioned those hints in only about 25% of the cases where they clearly determined the final answer. The finding directly undermines safety strategies that rely on monitoring a reasoning model's visible thought process.", "category": "ethics_safety", "significance": "notable", "tags": [ "Anthropic", "chain-of-thought", "faithfulness", "interpretability", "Claude 3.7 Sonnet" ], "references": [ { "title": "Chen et al. (2025). 'Reasoning Models Don't Always Say What They Think.' arXiv:2505.05410.", "url": "https://arxiv.org/abs/2505.05410", "type": "paper" }, { "title": "Reasoning models don't always say what they think (Anthropic)", "url": "https://www.anthropic.com/research/reasoning-models-dont-say-think", "type": "press" } ] }, { "id": "v-jepa-2", "year": 2025.44, "date": "2025-06-11", "title": "V-JEPA 2", "description": "Meta released V-JEPA 2, a video-pretrained joint-embedding predictive architecture trained on more than a million hours of video that enabled zero-shot robot manipulation after only 62 hours of action-conditioned data. The system was the clearest empirical evidence yet for Yann LeCun's long-argued thesis that learning should happen by prediction in a latent space, not in pixels.", "category": "models", "significance": "notable", "tags": [ "Meta", "LeCun", "JEPA", "world model", "video", "robotics" ], "references": [ { "title": "Assran et al. (2025). 'V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning.' arXiv:2506.09985.", "url": "https://arxiv.org/abs/2506.09985", "type": "paper" }, { "title": "Introducing V-JEPA 2 (Meta AI)", "url": "https://ai.meta.com/blog/v-jepa-2-world-model-benchmarks/", "type": "press" } ] }, { "id": "context-engineering", "year": 2025.74, "date": "2025-09-29", "title": "Context Engineering", "description": "Anthropic published 'Effective context engineering for AI agents,' naming a discipline that had been crystallising across frontier labs — managing an agent's long-running context window (memory, tool outputs, summarisation, eviction) rather than tuning a single prompt. The term, also popularised by Andrej Karpathy earlier in 2025, marked the shift from prompt engineering to state management for agentic systems.", "category": "language_nlp", "significance": "notable", "tags": [ "context engineering", "agents", "prompt engineering", "Anthropic", "Karpathy" ], "references": [ { "title": "Effective context engineering for AI agents (Anthropic, 2025-09-29)", "url": "https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents", "type": "press" } ] }, { "id": "frontier-fleet-power", "year": 2025.79, "date": "2025-10-15", "title": "Frontier AI Supercomputer Power at Nine-Figure Scale", "description": "By March 2025, xAI's Colossus — then the world's largest AI training cluster — was drawing an estimated 300 megawatts, roughly the peak consumption of 250,000 homes. Epoch AI's October 2025 'Trends in AI Supercomputers' report projected that the single leading frontier training system in 2030 would require on the order of 9 gigawatts, equivalent to the output of nine full-scale nuclear reactors feeding one model.", "category": "infrastructure", "significance": "notable", "tags": [ "Epoch AI", "power", "compute", "datacenter", "electricity", "Colossus" ], "references": [ { "title": "Epoch AI (2025). 'Trends in AI Supercomputers.'", "url": "https://epoch.ai/blog/trends-in-ai-supercomputers", "type": "archive" }, { "title": "Pilz et al. (Epoch AI, 2025). 'Trends in AI Supercomputers.' arXiv:2504.16026", "url": "https://arxiv.org/abs/2504.16026", "type": "paper" } ] } ] }