Here’s how computer scientists, who worry about the risks of Artificial Intelligence, hope the big wake-up call about the need to regulate it will happen. “There would be a kind of Goldilocks accident,” says Yoshua Bengio. “Large enough to wake people up. Not large enough to be harming our infrastructure or our institutions too seriously.”
The fear, continues Bengio, one of the world’s preeminent AI scientists, “is the boiling frog scenario where it comes gradually, and we get used to it… and we never react until it’s too late.”
The recent development of Anthropic’s new AI model, Mythos, which the company considered too dangerous to release publicly because of its ability to hack into software infrastructure systems, is a cautionary example. “Mythos is apparently having an effect,” Bengio says. “National security agencies of several countries, including of course the U.S. and China, are really concerned.”
For Bengio, worry about the risks of AI came to an inflection point much earlier — at the close of 2022.
The advent of ChatGPT was the reason.
Alan Turning, a 50s-era foundational figure in AI, had argued that a critical moment in the development of AI would be when machines master language. “He thought that once you master language, you basically master at least human-level intelligence,” explains Bengio, professor of computer science at the Université de Montreal, the founder and scientific advisor of Mila, and co-president and scientific director of LawZero, a non-profit startup dedicated to developing safe-by-design AI systems.
Not everyone shared his concern. “If you say that computers don’t actually understand language, it means you haven’t played with them, because it should be very clear — they do,” Bengio says.
Turing had anticipated that one day humans would build machines that would be smarter than us. “And they thought that if we didn’t figure out how to control them, it could be catastrophic,” Bengio continues.
Interestingly, in 2018, Bengio, along with Geoffrey Hinton and Yann LeCun, was a recipient of the A.M. Turing Award, considered the Nobel prize of computing for his foundational work in deep learning and artificial neural networks.
With ChatGPT, “I started thinking, wait, what’s going to happen if this trend continues? What’s going to happen to my children? What kind of world will they live in?”
In March 2023, he signed an open letter from the Future of Life Institute calling for AI labs to pause for at least six months. But it was ignored. In July of the same year, he testified in front of the Senate in the United States about the dangers of AI. “A minority has been taking it seriously, but not enough to move the needle,” he says.
“The forces of competition are at two levels: between companies and between countries, especially the U.S. and China,” Bengio says. The competition “is now forcing the hand of even the most ethical of the leaders towards taking risks that they wouldn’t otherwise take.”
Why do humans feel reluctant to stand in the way of progress? “That’s bullshit propaganda,” he retorts. “Progress is not a world where we play Russian roulette with our future. Progress is not a world where we play Russian roulette with our democracy. Progress is not a world where we play Russian roulette with creation of pandemics to kill half of the population. That is not progress. That is taking crazy and unacceptable risks for a few people’s benefit.
“Are seatbelts turning back progress on cars? Are security procedures in planes turning back progress on flying? It is a ridiculous claim that having safety protocols and careful design is turning back progress. No, it is progress. Progress is doing something that is good for us. It’s about going forward in a positive way for humans.”
Born in Paris into a Moroccan family who immigrated to Montreal when he was 12, Bengio started playing with computers in the late 70s with his younger brother, Samy, who is also a computer scientist. “We learned programming by ourselves, and it was so exciting…We got really hooked. It’s shaping something, almost with your hands, because you use the keyboard.”
Initially, he wanted to study physics in university. But he switched to computer engineering, receiving his undergraduate and master’s degrees from McGill University.
His work was “the scientific question of what is intelligence,” he explains. “Whether there are a few simple principles, like the laws of physics that can explain intelligence, that can give rise to intelligence both in us, and animals, and, of course, machines.”
Does he feel responsible for the Pandora’s Box his work has helped to open?
“Because I have a lot of expertise, and because of my reputation and international recognition, my voice counts. If I didn’t use these, advantages to move us towards a better world, I would feel bad. That’s the sense in which I feel responsible.”
His focus is now centred on Scientist AI, a system he’s developing at LawZero, the nonprofit organization launched last year that priorities AI safety over commercial imperatives. The Canadian government has announced funding. “And we’ve been negotiating with other governments, because the best way from an ethical point of view of developing AI is in a context where multiple countries work together and develop it as a global public good…Think of the international agreements around nuclear weapons. Think of the international agreement around the ozone layer, and think about the incomplete, but still incredibly activating issue of climate change … When we understand that there is a big risk, governments can move very quickly.”
LawZero has been successful in raising philanthropic funding, he says, and is actively recruiting researchers and engineers, many who feel a moral imperative.
“We think that AI, if it continues in the current direction, is going to be a crucial power tool for governments and economies, the military, and so on. If we don’t do anything, we’re going to be fully dependent on the goodwill of the two hegemons that Carney talked about . Which may work out — or not. So we should definitely have a Plan B, and the Plan B is to work with other Democratic countries who share the same concerns.”
The momentum of LawZero, which has developed a blueprint for “how we can build AI that will give us safety guarantees”, defies the moniker of “doomer” some in the AI community have given Bengio.
“Portraying me as a doomer is wrong. I’m a doer.”
“ Bigger tables, better narratives, broader impact”
Inez Jabalpurwala, President and CEO of the Public Policy Forum
By bringing together established leaders and emerging voices, our work produces resilient, practical policy ideas that serve all Canadians.
