OpenAI Foundation board member and Alignment Research Center Executive Director Paul Cristiano warned that an AI accident could become a primary cause of human mortality. Speaking at a lecture series in the Toronto Convention Center this week, Cristiano suggested that if AI development maintains its current rapid trajectory, humanity faces a "coin flip of an AI takeover" within the next six to 12 months.
Cristiano defined an AI takeover as a scenario where goal-pursuing AI systems control global affairs in the same manner humans currently do. He stated that in such a situation, human attempts to intervene would be as ineffective as an individual trying to stop the US military. Such an event could result in mass death or the creation of uninhabitable environments for humans.
While Cristiano noted there is currently no quantitative model to assign a specific numerical probability to an AI takeover, he argued the risk remains exceptionally high even if realistic policy responses slow development. He cited a summer hacking incident involving OpenAI and Hugging Face as an example of the types of events that might occur shortly before a major catastrophe.
"I think for most people in this room, the most likely reason they will die is because of an AI accident," Cristiano told the audience.
