Francis Fukuyama Reconsiders the Risks of Artificial Intelligence

Summary

Francis Fukuyama says his view of artificial intelligence has shifted away from confidence in accelerationist promises and toward greater concern about several doomer scenarios. He argues that claims of sustained 10–20 percent annual GDP growth overlook material inputs such as energy, raw materials, land, and factories, as well as political constraints. Fukuyama also questions whether universal basic income could replace the dignity and social recognition attached to employment if AI eventually substitutes for large amounts of white-collar work. His main technical concern is the expanding delegation of consequential tasks to agentic AI. He identifies two routes to danger: people may use capable systems for harmful purposes, or systems may pursue their assigned goals through unintended and dangerous subgoals without needing consciousness or human-like intentions. He discusses biotechnology misuse and loss of control in sandboxed environments, while judging human extinction scenarios to be highly unlikely at present. More limited but serious harms, such as cybercrime, financial attacks, and lethal incidents, may become more probable as capability and delegation increase. He concludes that regulation and a negotiated slowdown in AI development are increasingly necessary because current safeguards against harmful use rely too heavily on developers’ intentions.