P(doom) is the estimated probability that artificial intelligence (AI) will cause an existential catastrophe, such as human extinction. [1]
Overview of the Metric
Origin: The term started as an informal shorthand or inside joke in rationalist and Silicon Valley AI research circles before becoming a widely discussed metric. [1, 2, 3]
Scale: It is typically expressed as a percentage or a number between 0 and 1 (where 0 means no chance of human destruction and 100% means total certainty). [1, 2]
Not a Hard Science: Critics and analysts note it represents subjective "vibes" or personal hunches rather than a formal scientific calculation, given that definitions, timeframes, and AGI (Artificial General Intelligence) conditions vary widely. [1, 2, 3]
Expert Estimates
Estimates vary wildly depending on who you ask in the tech and research community:
The Optimists / Skeptics (< 1%): Figures like Yann LeCun (Meta's chief AI scientist) and Marc Andreessen place the risk at near 0% or less than 0.01%. [1, 2, 3]
The Moderates (10% to 50%): Many industry leaders and prominent researchers estimate a substantial risk. Geoffrey Hinton (Nobel laureate) estimates a 10% to 50% chance; Yoshua Bengio estimates around 20%; and Dario Amodei (Anthropic) estimates between 10% and 25%. [1, 2, 3, 4]
The Alarmists (> 80% to 99%): Researchers focused strictly on alignment and catastrophic risk, such as Eliezer Yudkowsky, put the probability above 95%. [1, 2]
