Artificial intelligence can be used as a force for good — but there are also big risks involved with the generative technology as it gets even smarter and more widespread, “godfather of AI” Geoffrey Hinton told the Collision tech conference in Toronto on Wednesday.
In a Q&A with Nick Thompson, CEO of The Atlantic magazine, Hinton — a cognitive psychologist and computer scientist who is a University Professor Emeritus at the University of Toronto — expanded on concerns he has recently expressed about the technology he played a key role in developing.
“We have to take seriously the possibility that [AI models] get to be smarter than us — which seems quite likely — and they have goals of their own,” Hinton said during a standing-room-only event at the conference, which was expected to draw nearly 40,000 attendees over three days.
“They may well develop the goal of taking control — and if they do that, we’re in trouble.”
Hinton, who recently left Google so he could speak more freely about AI risks, was one of several U of T community members scheduled to speak at Collision, which is billed as North America’s “fastest-growing tech conference” and counts the university as an event partner.
The government of Ontario used the occasion of the conference to announce that the Vector Institute — a partnership between government, universities and industry where Hinton is chief scientific adviser — will receive up to $27 million in new funding to “accelerate the safe and responsible adoption of ethical AI” and help businesses boost their competitiveness through the technology.
During his talk, Hinton outlined six potential risks posed by the rapid development of current AI models: bias and discrimination; unemployment; online echo chambers; fake news; “battle robots”; and existential risks to humanity.
When Thompson suggested that some economists argue that technological change over time simply transforms the function of jobs rather than eliminating them entirely, Hinton noted that “super intelligence will be a new situation that never happened before” — and that even if chatbots like ChatGPT only replace white-collar jobs that involve producing text, that would still be an unprecedented development.
“I'm not sure how they can confidently predict that more jobs will be created for the number of jobs lost,” he said.
Hinton added much of his concern stems from his view that AI may soon demonstrate the capacity to reason.
“The big language models are getting close — and I don’t really understand why they can do it, but they can do little bits of reasoning,” he said, predicting that AI will evolve over the next five years to include multimodal large models that are trained on more than just text, including videos and other visual media.
“It's amazing what you can learn from language,” he said. “But you're much better off learning for many modalities — small children don't just learn from language alone.”
Maximizing the creative potential of AI and minimizing its harms requires distinguishing between its potential risks, Hinton added, noting many in the tech sector have downplayed his warnings over the existential risk since he began speaking out.
“There was an editorial in Nature yesterday where they basically said fear-mongering about the existential risk is distracting attention [away] from the actual risks,” Hinton said. “I think it's important that people understand it's not just science fiction; it’s not just fear-mongering — it is a real risk that we need to think about, and we need to figure out in advance how to deal with it.”
Thompson pointed out that fellow AI luminary Yann LeCun — who jointly won the 2018 A.M. Turing Award (often referred to as the “Nobel Prize of computing”) with Hinton and Yoshua Bengio for their work on deep learning — has suggested that the positive aspects of AI will overcome any negative ones.
“I’m not convinced that a good AI that is trying to stop bad AI can get control,” Hinton said. “Before it's smarter than us, I think the people developing it should be encouraged to put a lot of work into understanding how it might go wrong — understanding how it might try and take control away. And I think the government could maybe encourage the big companies developing it to put comparable resources [into that].
“But right now, there’s 99 very smart people trying to make [AI] better and one very smart person trying to figure out how to stop it from taking over. And maybe you want to be more balanced.”