Scenarios · takeoff · p(doom)
How It Could Go
Where the AI story might end, mapped three ways: a field of possible outcomes arranged by how much control humanity keeps and how much wellbeing results; the speed of the transition to transformatively powerful AI; and what the researchers closest to the field believe the odds of catastrophe actually are.
Outcome scenarios
The Space of Futures
26 ways humans and AI might coexist — from egalitarian utopia to cosmic coalition — with 8 routes to an AI takeover you can layer on top. Each is positioned by how much control humans retain and how much wellbeing results. The clustering reveals something: most scenarios involve surrendering control.
Each dot is a plausible AI future, positioned by two axes: how much control humans retain (horizontal) and how well humans fare (vertical). Click any scenario — or focus it with the keyboard and press Enter — to read its narrative and the story it echoes.
Recap
- What you saw
A space of 34 plausible AI futures plotted by human control and human wellbeing.
- What it means
Most scenarios involve surrendering control — and the downside scenarios are the ones that cannot be reversed.
- What to watch
Which scenarios industry leadership publicly endorses, and which remain unspoken.
Takeoff speed
The Speed of the Fall
How fast do we get from narrow AI to general intelligence — and does speed leave room for course correction? 17 scenarios arranged from safest to most dangerous, mapped onto the three stages of artificial intelligence.
The "takeoff speed" debate: how quickly might AI transition from human-level to superintelligent capability? The dial reflects the range of expert opinion — from a gradual transition over decades to an abrupt leap in days or weeks.
Recap
- What you saw
17 takeoff scenarios from a decades-long transition to an abrupt leap in days.
- What it means
Speed is the variable that decides whether mistakes can be corrected or only regretted.
- What to watch
Forecasting updates from Metaculus and Manifold on AGI arrival distributions.
p(doom) — expert estimates
Expert Estimates, Wide Uncertainty
p(doom) is informal shorthand for the probability someone assigns to advanced AI ending badly — and it is not one quantity. The people below are answering at least four different questions: will AI kill everyone, will it end or permanently subjugate us, will we lose control of it, or will things simply go very badly. The marker shape says which. Every value plotted is one the person actually stated; figures we could not trace to their own words were removed rather than estimated, which is why several prominent voices appear without a dot. For context, the band beneath shows the safety thresholds we set for other technologies. Most are yearly or per-use rates, so below the chart we convert the forecasts to the same clock.
Four bands: aggregate surveys, individual estimates, people on record without a number, and the safety thresholds we require of other technologies. A dashed arrow marks an estimate its author has revised; a hollow marker is a second figure from the same source. Hover any row for the wording, the date and the citation.
Same clock, like for like
1 − Pby 2100 = (1 − pper year)75
A chance P of disaster by 2100 is the same as facing a yearly risk p for 75 years in a row and having to come through every one of them.
- Superforecasters, AI extinction by 2100: 0.38%about 1 in 20,000 a year
- AI experts, same tournament: 3%about 1 in 2,500 a year
- What we tolerate for one existing dam1 in 10,000 a year
Year for year, the most cautious group’s figure comes within a factor of two of the limit we set for a single dam, and the experts’ is four times above it. A dam failure is a local disaster. This one would fall on everyone. The researchers’ 10% cannot be converted: their survey sets no date.
Survey data: Grace, K. et al. (2026), “Advanced AI according to 1,580 researchers”, AI Impacts — the fourth Expert Survey on Progress in AI, run December 2024; and Grace, K. et al. (2024), “Thousands of AI Authors on the Future of AI”. Superforecaster and AI-expert figures from the Forecasting Research Institute’s Existential Risk Persuasion Tournament (2022). The per-year conversion follows the Lancet Commission on 21st century global threats to health (Murray, Kanem et al., 2026), which used the same tournament answers. Individual estimates are cited per entry in the tooltip; see also PauseAI and Wikipedia for compiled references. Industry thresholds from regulatory standards.
p(doom) is informal shorthand, not a rigorous statistical measure, and the figures are not strictly comparable to each other — they differ in what counts as doom, over what horizon, and under what conditions, which is what the marker shapes encode. Ranges reflect stated uncertainty. Nothing here is inferred from the strength of someone’s language: a person who has never given a number appears in the third band, with their words, rather than on the axis. Industry thresholds measure per-event or per-year risk while p(doom) estimates a one-time civilizational probability, so on the axis the comparison is illustrative, not equivalent; the per-year conversion above puts the dated forecasts on the same clock as the per-year thresholds.
Recap
- What you saw
Estimates spanning from zero to near-certain among serious people — and four different questions being answered as if they were one.
- What it means
The median AI researcher’s figure doubled between the 2023 and 2024 surveys. Put on a yearly clock, even the superforecasters’ low figure sits close to the limit we accept for a single dam or reactor, where the harm is local; here it would be global.
- What to watch
Whether the frontier-lab researchers who went public in September 2026 are followed by their colleagues — and whether the fifth expert survey shows the median moving again.