According to Sina Finance, OpenAI Foundation board member Paul Christiano said on a technology podcast aired Monday that a catastrophe within six months was a reasonable possibility, potentially causing mass deaths or the overthrow of a government. He said the pace of AI development would affect whether such an event occurs and that AI models could improve themselves through recursive self-improvement, potentially reaching a dangerous threshold within a year.
Christiano, who leads the Berkeley-based nonprofit Alignment Research Center, said the risk of AI taking control from humans within a year was now “closer to 10% than 1%.” When he joined the OpenAI Foundation board in September, he estimated in a public essay that the risk of a catastrophic event resulting from humans losing control of AI within a year was about 4%. He said the risk estimate would rise significantly if companies did not take further safeguards. Christiano emphasized that his comments reflected his personal views, not those of the OpenAI board.