Исследователи ИИ указывают разные вероятности экзистенциального риска на фоне споров о методологии

A number of AI industry insiders have recently shared personal estimates of the likelihood that artificial intelligence could lead to human extinction within the next decade. Anthropic alignment researcher Evan Hubinger posted on X that he believes the probability exceeds 10%, while Anthropic co-founder Dario Amodei has said his estimate ranges from 10% to 25%. Former OpenAI researcher Daniel Kokotajlo quoted a 70% chance, and Hubinger previously cited an 80% figure for existential risk.

The estimates, often referred to as “p(doom),” are presented as either ranges or single values, but experts acknowledge that no established mathematical model exists to calculate such odds. OpenAI CEO Sam Altman has said he sees a non-zero risk but cannot assign a precise number.

Critics argue that the forecasts may serve public-relations purposes, especially as major AI firms prepare for initial public offerings. Georgetown computer scientist Cal Newport described the discourse as “doom trolling” that could harm public mental health. In response, Anthropic AI-safety researcher Drake Thomas said the warnings stem from genuine fear rather than marketing.

The debate highlights tension between portraying AI as an inevitable technological advance and the uncertainty surrounding its long-term societal impacts. While some executives suggest that only their companies can mitigate the risk, others caution that overstated probabilities may erode public trust.