Geoffrey Hinton: why his extinction-risk estimate remains 10–20%

‘When people come up with these probabilities, they're all based on gut feeling’

EXCERPT

HINTON: "Let me first direct my first question to Dr. Geoffrey Hinton. Uh, back in, in 2024, you told The Guardian that there was a, quote, "10 to 20% chance that AI would lead to human distinct- extinction in the next three decades," correct? That is correct. Obviously, nothing in the way of safety regulations has happened, uh, in the last two years since you made that quote. Um, and so at this point, AI has continued to advance, uh, rapidly. I'd assume that you'd now say there's a greater than 20% chance that AI will lead to human extinction in the next three decades? I should say something about how people come up with these probabilities. Um, we've never been here before. We don't have a sort of track record of making things smarter than ourselves and seeing what happened, where we can do statistics. So when people come up with these probabilities, they're all based on gut feeling. It's fairly clear to almost all the researchers that there's some probability that it will happen. So 1%'s probably a low estimate. Um, it's fairly clear that we're still very ingenious. We're still designing the things. We're still in control, so 99% seems like too much. Um, then people just estimate some number in between that seems plausible to them. So this isn't- Your gut- This isn't a- What's your gut instinct? ... scientific process. What's your gut instinct today? My gut instinct is still between 10 and 20%, partly because I think if we focus on how do we make advanced superintelligences not want to get rid of us, if we focus on the other aspects of these beings other than just their intelligence, and we train them so that they like people more than they like themselves, then we do stand a good chance."

Source: Congressman Raja Krishnamoorthi · Sep 27, 2026 · 107s clip