p(doom) estimates the probability of an AI-related catastrophe, including human extinction. The estimate is uncertain: superintelligence is presented as potentially beneficial or catastrophic depending on its goals, and catastrophe is not treated as inevitable. section 9.2.6
Illustrative estimates
In a 2011 interview, Legg suggested a 5% or 50% chance of human extinction within a year, while emphasizing that no reliable estimate exists. He later estimated a 70% chance of resolving alignment by 2028. section 9.2.6
