p(doom) reads "probability of doom" and is written the way any probability is written on a blackboard: P(A), where A is the event doom. It is the number someone states, half seriously, to summarize how worried they are that AI will end badly: "my p(doom) is 10%." By its users' own account, it was circulating on LessWrong, the forum of the rationalists, around 2009–2010; it reached the press in 2023, after GPT-4, when the New York Times presented it as Silicon Valley's grim new metric. The p(doom) song turned it into a chorus: "I'm upping my P(doom)" (osmarks, n.d.).
The figures people state in public range from practically zero to near certainty. More informative is the survey by Grace et al. (2025) of 2,778 authors at top AI venues: between 38% and 51% gave at least a 10% chance to outcomes as bad as human extinction. So this is not the obsession of a fringe forum, even if the notation was born there.
For a philosopher, p(doom) is a textbook case of subjective probability: a credence about an unrepeatable event, with no frequencies behind it, supposedly updated by Bayes' theorem. That is where the criticism aims. First, vagueness: is doom extinction or loss of control? By when? Conditional on AGI existing? Even on LessWrong, someone titled a post calling it "a dumb meme." Second, Narayanan & Kapoor (2024): with no reference class and no reliable causal model, these numbers are guesses with decimals and should not guide public policy. Third, a sociological point: the number also works as a password, and stating one's p(doom) is a way of announcing which side one is on (doomer).