Published at 1:27 am. Updated at 6:00 am.
There is already Donald Trump, Ukraine, burning forests and trolls who are getting angrier every day on social networks. Do we really need a new source of trouble these days?
That was my first reaction when I read that artificial intelligence could pose an “existential threat” to humanity.
Of course there’s a lot of denial on my part. But also a healthy skepticism. Man has always feared the end of the world. And breaking news, we’re still here to tell each other fears.
The threat of artificial intelligence seemed particularly difficult to take seriously. Those algorithms that have trouble suggesting relevant recommendations to us on Netflix? The very ones who make us laugh with their awkwardness on ChatGPT? Should we be afraid of it now?
While whistling the REM (if it means the end of the world as we know it, I feel very good, thank you), I went to Yoshua Bengio’s house.
Let’s just say we don’t go there to cheer ourselves up. The Quebec researcher, one of the fathers of artificial intelligence, has been raising alarm bells for several months. He fears that the scientific advances to which he has contributed are now turning against us.
This summer he went to the U.S. Senate to raise the alarm. The day before our meeting, he had discussed it again with António Guterres, the Secretary-General of the United Nations, as a member of a new committee on the issue.
PHOTO SAUL LOEB, AGENCE FRANCE-PRESSE ARCHIVE
Yoshua Bengio testified before American elected officials in Washington on July 25.
I wanted to specifically understand what could go wrong with what we call AI, or artificial intelligence. I have been served in disaster scenarios.
But Yoshua Bengio is an intelligent and humanistic prophet of doom. He speaks calmly, often even with a smile, about the worst. We quickly understand that his goal is not to scare the world. His words are a call to action. He advocates research support.
“The damage could be so great that we have to think about it now, even if there are uncertainties about the scenarios,” emphasizes Bengio.
The first specter that awaits us is easy to understand: it is the danger that artificial intelligence will be misused for evil purposes.
Right now, anyone who asks ChatGPT how to build the deadliest bomb out of readily available materials hits a wall: the designers have built in guardrails.
“However, in the last few months we have noticed that it is quite easy to get around the guardrails. And in a year, two years, three years, we can expect that the systems will be able to help people who want to do harm in different ways,” says Mr. Bengio.
The researcher himself used artificial intelligence to help develop new drugs. He points out that the same thing is possible with the development of viruses or chemical weapons.
More recently, Mr. Bengio began thinking about the imbalance that could arise if some countries, individuals or companies had access to better artificial intelligence tools than others.
“Today you can buy and sell stocks online. If AI becomes better at this than humans, it could make a lot of money. And with money you can buy everything in our world,” he explains.
AI could also manipulate people. ChatGPT is already so good at mimicking us that it has become virtually impossible to know whether it is a machine or a human interacting with us.
Mr. Bengio conjures up the scenario in which artificial intelligence would be used to create thousands of fake accounts on social networks and influence public opinion.
If you can develop better weapons, if you can influence politics, if you can make almost unlimited money, all existing equilibria in our society will collapse.
Joshua Bengio
Are you starting to feel cold? Tie your hat on. “There is a danger even worse than this,” declares Mr. Bengio.
Warning: We are diving into science fiction here. However, Yoshua Bengio believes that we cannot exclude the possibility that artificial intelligence develops the desire for “self-preservation” and begins to act in its own interests, to the detriment of the interests of people.
“There are researchers who believe this could happen spontaneously without you intentionally putting it in the machine,” he says. I can’t say for sure that this will happen. But I’m perfectly capable of saying that there will be an idiot somewhere telling the machine: Watch your back. Protect yourself. »
PHOTO JOSIE DESMARAIS, THE PRESS
Joshua Bengio
Mr. Bengio explains that AI has advantages over the biological systems that we are. It develops faster. And machines can process and exchange infinitely more information with one another than humans can through speech or writing.
Since artificial intelligence is already able to write computer code, it can take on all sorts of tasks.
But these machines still work well with electricity! If there was a problem, couldn’t we just “pull the plow”? The question is naive. But at this point in the interview, I needed something to hold on to.
“Let’s imagine a machine that wants to maintain itself,” answers Professor Bengio. When she thinks for a bit, she realizes that a human could actually pull the plug on her. What will she do? It could duplicate on other machines. And since she doesn’t want to be found, she might as well do it by hiding. »
No, that’s not reassuring. Yoshua Bengio himself began to worry when he tried ChatGPT.
“ChatGPT offers so much more knowledge than what was available to me at university,” he explains. And all without scientific progress. It’s just a question of computing power! I realized that we are closer to the goal than I thought. »
He points out that since private companies are currently investing so much money in training artificial intelligence systems, progress could be exponential.
AI pioneer Yoshua Bengio admits to torture.
“It is painful to think that we have contributed to something that could be very destructive,” he recently wrote on his blog, saying he was thinking about the world he would leave to his 20-month-old grandson.
When asked if he sleeps well, Yoshua Bengio answers with that mix of rationality and emotion that characterized our interview.
“Physically speaking, I always slept well,” he says. But psychologically it definitely consumes my thoughts. »
Verdict
It would be very reassuring to think that Yoshua Bengio is going too far in his disaster scenarios. His speech is anything but unanimous in the scientific community. He is aware of this and himself admits to having fallen into “speculation,” a counterintuitive position for a scientist. But can one really rule out the possibility that the risks described by Professor Bengio will one day occur? The researcher is – unfortunately! – convincing. The most basic prudence dictates that we listen to him. We must have the courage to assume the worst if we want to avoid it.
Who is Joshua Bengio?
- Born in Paris in 1964.
- Winner of the Turing Prize (the “Nobel Prize in Computer Science”) in 2018 along with Geoff Hinton and Yann LeCun for their work in deep learning.
- Professor at the University of Montreal and world leader in the field of artificial intelligence.
- Founder and scientific director of Mila, the Quebec Institute for Artificial Intelligence.
- According to a study by Stanford University, in 2021 he was the third most cited scientist by his colleagues worldwide across all disciplines.