Yoshua Bengio is worried that innovations in artificial intelligence that he helped pioneer could lead to a dark future, if “killer robots” get into the wrong hands.
But the soft-spoken, 55-year-old Canadian computer scientist, a recipient of the AM Turing Award — considered the Nobel Prize for computing — prefers to see the world though the idealism of Star Trek rather than the apocalyptic vision of The Terminator.
“In the Star Trek, there is a world in which humans are governed through democracy, everyone gets good healthcare, education and food, and there are no wars except against some aliens,” said Bengio, whose research has helped pave the way for speech- and facial-recognition technology, computer vision and self-driving cars, among other things. “I am also trying to marry science with how it can improve society.”
Bengio was expounding on the promises — and perils — of AI on a recent day while sitting in his small, cramped office at the Montreal Institute for Learning Algorithms, a research centre he founded that has made Montreal a global center for AI. Next to him was a whiteboard covered with complex mathematical equations, along with a warning for the cleaners written in French: “Do Not Erase.”
Erasing those equations could come at a heavy cost for humans as well as machines.
Bengio, a professor of computer science at the University of Montreal, is self-effacing. But his work in an area known as “deep learning” — “teaching machines to learn in a way inspired by how our brains compute,” he says — has already affected our daily lives in countless ways, making it possible for Google Translate to convert a sentence from French to Mandarin or for software to detect cancer cells in a medical image.
He and his researchers are also harnessing AI to discover molecules that could cure diseases, to detect gender bias in textbooks and to predict when natural disasters will happen.
Cherri M. Pancake, the president of the Association for Computing Machinery, which offers the $1 million prize, credited Bengio and two others AI luminaries who shared the prize, Geoffrey Hinton and Yann LeCun, with laying the foundation for technologies used by billions of people. “Anyone who has a smartphone in their pocket” has felt their impact, she said, noting their work also provided “powerful new tools” in the fields of medicine, astronomy and material sciences.
Despite all the accolades, Bengio recoils at scientists being turned into celebrities. While Hinton works for Google and LeCun is the chief AI scientist at Facebook, Bengio has studiously avoided Silicon Valley in favor of a more scholarly life in Montreal, where he also co-founded Element AI, a software company.
“I’m not a fan of a personalisation of science and making some scientists stars,” said Bengio, a self-described introvert, who colleagues say is happiest when hunched over an algorithm. “I was maybe lucky to be at the right time and thinking the right things.”
Myriam Coté, a computer scientist who has worked with Bengio for more than a decade, described him as an iconoclast and freethinker who would feel stymied by the strictures of Silicon Valley. A communitarian at heart, she said, he shuns hierarchy and is known for sharing the profits from his own projects with younger, less established colleagues.
“He wants to create in freedom,” she said. Citing the credo of student rebels in 1968 in Paris, where Bengio was born, she said his philosophy was: “It is forbidden to forbid.”
Even as the late Stephen Hawking, the celebrated Cambridge physicist, warned that AI could be “the worst event in the history of our civilisation,” and billionaire entrepreneur Elon Musk has cautioned it could create an “immortal dictator,” he has remained more upbeat.
“We need to pursue scientific knowledge or all we will do is run against a wall,” he said. “But we need to do it wisely.” Referring to the use of algebra to compute the angles of missiles, he added: “You can’t blame the inventor of algebra for war.”
Nevertheless, at a time when Facebook algorithms have come under criticism for their influence in the 2016 United States election and fears are growing that robots could use AI to target humans without human oversight, Bengio is acutely aware that his innovations risk becoming “Frankenstein’s monsters.” As a result, he said, he supports regulating AI, including an international treaty banning “killer robots” or “lethal autonomous weapons.”
But he dismissed the “Terminator scenario” in which a machine, endowed with human emotions, turns on its creator. Machines, he stressed, do not have egos and human sentiments, and are not slaves who want to be freed. “We imagine our creations turning against us because we are projecting our psychology into the machines,” he said, calling it “ridiculous.”
He said his imagination was particularly kindled by the relationship between man and machine in Stanley Kubrick’s 2001: A Space Odyssey. “A teacher who spends every day teaching a machine about the world resonated with me,” he said. “Science fiction was a way to dream about the future.”
At the age of 11, he recalled, he began programming by plugging numbers into a calculator. Asked to recall the most formative memories from childhood, he suddenly paused and made an unlikely confession.
“I don’t have a good memory — I’m good at reasoning, not at memory. That’s why I was drawn to math and computer science as you don’t need to memorise anything.”
Bengio attributed his success to his socially conscious parents, noting that his father, a pharmacist who directed theatre, and his mother, who managed artists, gave him and his brother the freedom to think for themselves.
It appears to have worked. His brother, Samy, one year younger, runs a research group at Google Brain, an AI research team. The two were “practically twins,” he said, collaborating on research and exchanging ideas.
When he was 12, the family moved to Montreal, where his maternal grandparents were living. While pursuing his masters and doctorate in computer science at McGill University in the late 1980s, he recalled, he was drawn to the work of Hinton, who was striving to develop “intelligent computers” based on neural networks, mathematical algorithms that can learn tasks on their own by analysing vast swathes of data
“It wasn’t just some mathy thing or computer science, but about understanding human intelligence to build intelligent machines.” Postdoctoral work at the Massachusetts Institute of Technology and Bell Labs followed.
Bengio recalled that he, Hinton and LeCun laboured on neural networks for decades, even as many in the computer science establishment dismissed them as dabblers in a form of “dark magic” with few if any practical applications. For Bengio, ignoring the naysayers and bucking conventional wisdom came naturally.
“The big dreams of building intelligent machines faded in the late 1990s and people thought, ‘Oh, it’s too hard — let’s just use our algorithms to solve concrete problems’,” he recalled. “I guess I am not sensitive to what people care about at a particular time, and I believed in what I was doing.”
While their trailblazing work transformed the field — and turned him into an AI superstar — Bengio still delights in spending time with students, whom he describes as a “family.” He is divorced and has two grown children, one of whom has gone into AI. He spends his rare free moments reading Spinoza and walking in the woods near his house.
Unimpressed by scientific prizes or riches, Bengio stressed that complacency and overconfidence were the enemies of scientific progress.
“Being self-confident is not enough,” he said. “You can be self-confident and wrong.”