Abstract:
On September 11, according to "Business Insider", "AI Godfather" Geoffrey Hinton said,
It is difficult to estimate the probability that AI will wipe out all humans, but this probability is not zero.
Hinton told BBC Newsnight on Wednesday: "To me,10 per cent seems like a not outrageous estimate, but in reality no one knows how to give a reasonable estimate.
”
Hinton
"Newsnight" host Victoria Derbyshire expressed surprise because Hinton basically agreed with the judgment of Jacob Coxon, Anthropic's top AI security researcher. The researcher said,
In the next 10 years, AI "has a probability of more than 10% killing all humans"
."Wow," Derbyshire said, "Oh my God."
Hinton said it is difficult to estimate the specific probability because human society has never seen a technology that "soon may be smarter than us."
“It would be foolish to say there is only a 1% probability,” he said. “No one knows how to estimate it.”
What is certain, the Nobel laureate said, is that the window in which such a disaster could occur has shortened. "I used to think it was 30 to 50 years, and then it was shortened to 10 to 20 years," he said. "Now, I think it might be 10 years, or even shorter. Others think it's much faster than that."
As for how this might happen, Hinton cited risks ranging from cyberattacks to AI developing deadly viruses. Although Hinton did not say why this timeline was significantly shortened, leading AI companies including Anthropic and OpenAI have invested billions of dollars in scaling up their models, and their recently released products have shown significant improvements over models from just months ago.
Hinton’s comments come after Anthropic researcher Coxon announced his resignation on the eve of the company’s highly anticipated IPO. Coxon, who previously worked at OpenAI, said the AI giants are "gamble with our lives."
“Those developing AI sincerely believe that AI could kill us all before the end of this decade,” Coxon wrote on
Hinton said that humans may embark on a path of self-destruction, or they may find ways to "deal with the dangers posed by AI" while realizing the potential of AI. He reiterated his call for stronger global regulation, noting that while countries may not agree on all issues, they all share a common desire to safeguard human survival.
“Countries have the same interest in preventing AI from replacing humans. Therefore, they will eventually cooperate. The question is, can they take action in time?” he said.
Comments