Toronto’s New AI Hub Will Let Scientists Do Whatever the Hell They Want
Photo courtesy of Johnny Guatto/ University of Toronto


This story is over 5 years old.

Toronto’s New AI Hub Will Let Scientists Do Whatever the Hell They Want

Can the Vector Institute compete with Silicon Valley?

Artificial intelligence as we know it was largely developed in Canada, but Silicon Valley is reaping the rewards. Now, the Canadian government is placing its faith in a new Toronto-based AI research initiative, the Vector Institute, to bring some of that expertise—and the attendant profits—home.

The full-court press of media events that took place during Vector's launch on Thursday made it clear that hopes are high. The morning saw Geoffrey Hinton, a Canadian computer scientist known as the "godfather of deep learning," give a presentation on the technology to journalists at Google's Toronto headquarters. In the afternoon, Prime Minister Justin Trudeau highlighted the government's AI funding in a press conference. That evening, Ontario Premier Kathleen Wynne, Finance Minister Charles Sousa, and Hinton gathered in downtown Toronto's MaRS building to announce the Vector launch.


It really seems like everyone wants Vector to be a huge deal. But despite $150 million in funding from the provincial and federal governments, as well as a smattering of companies including Google, the Vector Institute is up against stiff competition in Silicon Valley, the UK, and around the world. What's Vector's angle?

"If someone's got a strong intuition that something's going to pay off in the future—we want research like that"

"We can offer people the chance to do any mix of basic research and applications that they want, and they're going to have lots of data, particularly from hospitals," said Hinton in an interview. Hinton will be living in Toronto permanently to act as a research adviser for Vector, essentially giving suggestions on where to spend its money.

It's unclear what sorts of data Vector will be taking from hospitals, but Toronto is also a major hub for biomedical and stem cell research. Hinton noted that AI could soon be a popular tool for medical imaging: for example, reading an X-ray for anomalies.

According to Hinton, Vector's philosophy will be to let researchers do whatever they want with as few strings attached as possible, whether that's basic research that may take years to see the light of day, or commercial services. And if anyone knows the value of basic research without an immediate payoff, it's Hinton.

"I had an idea about five years ago that I've been working on ever since, and it still doesn't work very well," Hinton said. "Google has been very good about giving me lots of time to work on this even though it hasn't worked out yet."


Read More: A New AI Hub In Canada Is Welcoming the Scientists Trump Is Pushing Away

"If someone's got a strong intuition that something's going to pay off in the future—we want research like that," he continued.

The vision for Vector seems like a long-term one. During his speech at the launch, Meric Gertler, president of the University of Toronto (which is affiliated with Vector), compared the Institute to how the US government funded Arpanet in the 1970s, which eventually became the internet.

This is just as well, because the history so far of AI is also one of delayed satisfaction. Just one example is the neural network fine-tuning technique of backpropagation, which was invented in the 1980s but only became really useful recently. Computers just weren't powerful enough back then.

Despite all this fanfare, there are potential pitfalls, like focusing less on research and more on pumping out degrees. One of Vector's selling points, highlighted by chair of the Vector board and former bank CEO Ed Clark during his speech at the opening, will be "[training] more Masters and PhDs in the field of deep learning than any other institution in the world."

Still, pushing the envelope of basic research seems to be exactly what Vector can do best. Hinton said that we shouldn't be satisfied to tinker endlessly with the tools he invented—we should be looking for new and better ones.

"I happen to know," he said, "that the neural networks we have today are just something people made up when they came to work."

Subscribe to pluspluspodcast , Motherboard's new show about the people and machines that are building our future.