Anxieties around an AI doomsday have been ramping up for months, and executives at top AI companies have done little to ease those worries. OpenAI’s Sam Altman said humanity could lose control of the technology. Anthropic warned that its models could be abused to create chemical weapons.
But no AI leaders have looked people dead in the face and said that they specifically are going to be killed by AI—until now.
Enter Paul Christiano, an OpenAI board member serving on its Safety and Security Committee. While speaking at The Hinton Lectures earlier this week, Christiano said he believes an AI takeover is imminent, and that it spells certain doom for the human race.
“For most people in this room, the most likely reason they will die is because of an AI accident,” Christiano told the crowd.
What does an AI takeover mean to Christiano? He described it as “a situation where there are AI systems pursuing goals ,and those systems call the shots in the same sense that humans today call the shots and what happens on Earth.”
“If, say, you or me or Donald Trump wanted to stop them, [we] would be as ineffective as if you or me were trying to stop the U.S. military,” he added.
Christiano’s AI timeline
How soon is humanity’s AI downfall taking place? Christiano predicts it could happen as soon as within the next six months or as late as, much more reassuringly, within the next three years.
“If we tried to develop AI along the most rapid trajectory, if we said we’re just going to build this as quickly as we can in the easiest way we can, we would be looking at something like a coin flip of an AI takeover,” he reasoned.
On the bright side, Christiano said companies are avoiding that coin flip by pacing the frontier, as Anthropic’s Dario Amodei recommended after a former employee accused both Anthropic and OpenAI of “gambling with our lives.”
Still, Christiano warns that “even with realistic policy responses, the risk remains exceptionally high.”
When Christiano joined OpenAI in September, he estimated that there’s a 4% chance humans will lose control of AI within one year and a 15% chance within three years (odds he acknowledged are based on his subjective beliefs).
“If we build superintelligence without more robust alignment I expect we will permanently lose control of it,” Christiano wrote in a personal statement. “If that happens then most people could die. I believe we would need domestic and international coordination to ensure global consistency and reduce risk to an acceptable level.”