What Is p(doom)? The AI Doom Probability Explained
P(doom) is shorthand for a person's estimated probability that artificial intelligence causes an existential catastrophe. The expression is common in AI-risk debates, but the number is not a measurement produced by a standard scientific instrument or model.
Treat p(doom) as a compressed statement of belief. The useful work begins when someone explains the assumptions, pathways, time horizon, and definition of “doom” behind their number.
What does p(doom) mean?
The “p” means probability. “Doom” usually means human extinction or an irreversible loss of humanity's control over its future caused by advanced AI. Different speakers may quietly use narrower or broader definitions, so two identical percentages can describe different events.
A p(doom) estimate is a subjective credence: a summary of what one person currently believes after considering uncertain evidence. It is not the observed frequency of repeatable experiments, and there is no agreed formula for calculating it.
Why do estimates differ so much?
People disagree about whether highly capable AI will arrive, how quickly capabilities could improve, whether systems will pursue unintended goals, how effective technical safeguards will become, and how governments and companies will respond. Each assumption changes the final estimate.
- Timelines: how soon transformative or general-purpose systems might arrive.
- Takeoff speed: whether capability growth would be gradual or abrupt.
- Alignment: whether systems reliably follow intended human goals.
- Control and governance: whether monitoring, coordination, and institutions can contain failures.
- Threat model: accident, deliberate misuse, competitive races, or some combination.
What do surveys actually show?
The 2022 Expert Survey on Progress in AI asked machine-learning researchers about long-run outcomes. Its authors reported non-trivial median probabilities for outcomes as severe as human extinction, while responses varied widely. A survey records beliefs among respondents at a particular time; it does not validate any one forecast.
Public discussion often highlights a single dramatic percentage. A better reading asks who was sampled, how the question defined catastrophe, the distribution of answers, and when the survey was conducted.
Useful shorthand or false precision?
Supporters find p(doom) useful because it forces a person to state uncertainty instead of saying only that risk is “serious.” Critics argue that a precise-looking number can disguise fragile assumptions and create confidence unsupported by data.
Both concerns can be true. A number can clarify disagreement while remaining deeply uncertain. Scenario models such as P(Doomed) are most useful for exposing the pathways beneath a number, not for generating a real-world probability.