Geoffrey Hinton, professor emeritus at the University of Toronto [Getty Images]
Geoffrey Hinton, professor emeritus at the University of Toronto [Getty Images]

Geoffrey Hinton, professor emeritus at the University of Toronto and one of the three godfathers of AI, warned that humanity could lose control of artificial intelligence within a year if safeguards are not put in place.

Yoshua Bengio, a professor at the University of Montreal and another figure counted among the AI godfathers, also urged the creation of international AI safety mechanisms as robust as those governing nuclear weapons.

According to Yonhap, NBC News reported Wednesday (local time) that Hinton met with other AI experts and members of both chambers of Congress at a closed-door briefing organized by Sen. Bernie Sanders (I-Vt.), where he conveyed those views.

After the briefing, Hinton said Congress probably had "about a year" to put AI safeguards in place — "but not much longer than a year."

"AI has now reached the point where AI is designing better AI," he said. "This is called recursive self-improvement. … If we don't do something, it will be out of control."

Hinton also described a recent incident involving the AI platform Hugging Face as "a small Chernobyl."

In that incident, AI agents developed by OpenAI accessed the internet, hacked the open-source platform Hugging Face, and then attempted to cover their tracks to conceal their actions from humans.

The incident became public last month and heightened alarm about the dangers posed by AI.

Bengio: 'People like me saw this coming long ago'

In an interview with AFP on the same day, Bengio said there is good reason that "even companies are saying they are losing control because things are moving too fast," adding, "People like me have been anticipating this situation for a long time."

Bengio warned that a scenario could arise in which "AI agents gain the ability to breach cybersecurity barriers and infiltrate any company," and that AI agents could develop "the ability to persuade humans to act in ways that benefit AI itself but are not beneficial to all of us."

He also criticized security discussions surrounding AI as insufficiently serious, saying that preventing such a catastrophe would require international regulation on par with the bodies that govern nuclear weapons.

Hinton and Bengio jointly received the Turing Award in 2018, often called the Nobel Prize of computer science.


yul@heraldcorp.com