BIn 2012, Yoshua Bengio declared that mouse brains are infinitely more intelligent than any algorithm. “But ask me again in 10 years,” he added with remarkable foresight. Ten years later, the founder and scientific director of the Quebec Institute for Artificial Intelligence, Mila, and professor at the University of Montreal, is focusing on national security, deepfakes, disinformation, fraud, surveillance, social destabilization, systemic discrimination, and much more. We are sounding the alarm about potential risks. Loss of control of AI systems, etc.
Technological advances have accelerated so rapidly in recent years that he now believes superhuman AI could be developed within the next 20 years, “or even a few years.” The world's most widely cited computer scientist, one of the pioneers in pushing the boundaries of AI with more advanced learning algorithms, last year called for a complete halt to research, became an AI whistleblower, and accused people of similar What is the background that led to the prediction of a different revolution?Agricultural Revolution and Industrial Revolution? Here we explore his rapid rise in the scientific world, along with a number of potentially life-or-death puzzles.
“I've always been fascinated by the human mind – how the brain and intelligence work,” says Dr. Bengio. As a teenager, like any self-respecting computer nerd, he turned to programming first. When it came time to focus on his university studies, he explored connectionism and drew inspiration from the structure of the brain to design cutting-edge systems in AI. “What excited me then and still fascinates me now is that we can explain intelligence in the same way that we explain physics, using some scientific principles,” he says. . In 1991, Dr. Bengio received his PhD in Computer Science from McGill University. After gaining extensive experience abroad, first at the Massachusetts Institute of Technology in statistical learning and sequential data, then at AT&T Bell Laboratories in Holmdel, New Jersey, where he studied machine learning and computing in the field of vision, and on the faculty of the University of Montreal. I returned to Montreal as M found that his research has practical applications in speech and handwriting recognition.
Although AI research is currently booming, it was still in its infancy in the 1990s. “The system couldn't even handle basic tasks like recognizing handwritten digits or phonemes. Now these things are trivial, but back then we didn't have the algorithms or the processing power or data we needed. That's why only a few researchers have considered the potential for AI to be misused, he argues. “Only crazy people like me were saying you needed to be careful, but for a long time it never crossed people's minds that computers could pose a risk. .”
A life dedicated to academic research
How would you summarize a 35-page resume? Dr. Bengio's groundbreaking research in deep learning earned him the 2018 AM Turing Award, also known as the “Nobel Prize of Computing” along with Jeffrey Hinton and Yann LeCun. I won an award. Unlike his colleagues who worked at his Google and his Meta, respectively, Dr. Bengio spent most of his career doing academic research. In addition to his duties at U de M, he is also co-director of CIFAR's Machine and Brain Learning Program. He also serves as Scientific Director of IVADO, an interdisciplinary data assessment organization that brings together industry experts and university researchers.
When his colleagues joined Google and Meta, he sensed the potential pitfalls of AI. “These companies have jumped on the deep learning bandwagon,” says Dr. Bengio. “They were quietly considering using AI to power their online advertising systems.” Getting consumers to choose one brand over another is another. But “it was not hard to imagine that AI would influence other, more influential choices, not to mention political opinions, and that's exactly what we're seeing today.” he laments. But it took 10 years for the alarm bells to ring. Dr. Bengio, who also held the Canada Research Chair in Statistical Learning Algorithms from 2000 until 2019, said it was “a hard pill to swallow” to realize that he spent years researching and promoting potentially harmful technologies. states. , we are just playing with fire,” he says. “Before you take action, you need to understand the risks. To me this is obvious, but to many people it is not.”
“Science without conscience”
In 2017, U de M took the first steps towards creating the Montreal Declaration for the Responsible Development of Artificial Intelligence. His declaration, one of the first joint initiatives for the responsible development of AI, is based on ethical principles and fundamental values such as justice, welfare, privacy, democracy, and accountability. But it wasn't until 2023, when he joined his colleagues and other global technology leaders in calling for a moratorium on AI, that global awareness of the risks of AI came into focus. It was after that. “Obviously, companies have continued to do what they've always done since then,” he says. “But there is a fundamental shift taking place among public opinion, decision-makers and even researchers. Many were worried about the risks but did not speak up for fear of being criticized. did.”
This global effort is reminiscent of the nuclear pioneers who came together in their later years to protect humanity's future, which was endangered by their work. “AI Hiroshima is not here yet,” Dr. Bengio points out, but since the suspension he has been working to raise awareness in influential forums. Since 2023, he has been a member of the United Nations Scientific Advisory Board, which seeks independent advice on breakthroughs in science and technology, and was commissioned by the UK to chair the State of Science report on frontier AI capabilities and risks. I am. In early February, he testified before a federal committee examining Canada's legal framework for artificial intelligence. “The advances that are coming are likely to be highly disruptive, and it is impossible to predict when they will occur,” he said, calling for urgent regulation of the development and use of AI now. He emphasized that. “It is unwise to leave these decisions in the hands of individuals. The interests of some people do not align with the interests of the group. That's where things get serious.”
To put this problem into context, Yoshua Bengio puts advances in AI in the context of a tame grizzly bear that has become smart enough to escape from its cage to save itself, and a fish that is rewarded with I'm making an analogy. “Survival for AI means controlling or eliminating humans.” This is not a question of “existential consciousness.” It's simple mathematical reasoning. “If this being wants to maximize its rewards, the best course of action is to control the environment. And that includes us.”
