HOW CANADA GOT AN EARLY START ON AI
Many players anxious to get foothold here with our homegrown talent, strong startup support
This is the first instalment of a regular column that will explore innovations being made in the vast area of artificial intelligence.
It was a big news item for the startup community when Microsoft announced last week the acquisition of Maluuba — a Montreal-based deep learning research lab for natural language understanding, and the doubling of its AI campus there. But the reality is, an acquisition by a major player is quickly becoming old news in the world of artificial intelligence (a term that is also applied to the disciplines of deep learning and machine learning).
Maluuba was founded in 2011 by Sam Pasupalak and Kaheer Suleman after they graduated from University of Waterloo. “We went through a number of accelerator programs including Velocity and Communitech when we started,” Pasupalak says.
In 2014 they set their sights on Montreal to join the AI ecosystem of more than 150 deep learning researchers with University of Montreal and McGill University. “It’s the highest concentration in the world,” Pasupalak says.
Like many projects that fall under the AI spectrum, deep learning and natural language is an esoteric discipline few laypeople relate to. But what isn’t hard to grasp is Maluuba’s success, since many players are anxious to get a foothold in Canada’s AI community.
One of the reasons is the early investment efforts of the Canadian Institute for Advanced Research (CIFAR). Another is the pioneering work of such visionary thinkers as Geoffrey Hinton at University of Toronto and Yoshua Bengio at University of Montreal, whose efforts to build machine learning research and development programs have placed Canada second only to the U.S., according to a peer review panel for CIFAR.
CIFAR’s interest in artificial intelligence dates back to 1982, says John Hepburn, vice-president of research. “There were all sorts of exciting things going on because computers were getting powerful and everyone was buying into the way of the future.”
The program was stopped in 1995, though, because technology required for AI applications wasn’t quite up to speed, but a new program came into being in 2002 as a result of Hinton’s renewed participation.
Hepburn says Hinton’s appeal was that his first degree was in psychology and not computer science. “Until then everyone had given it up as a sort of sideshow that was interesting but purely academic. Hinton brought together psychologists, neuroscientists, ophthalmologists and electrical and computer engineers to make computers intelligent. He saw the potential of this field of neural networks and deep learning.”
“As early as 1987, CIFAR made a very nice job on research opportunities at U of T,” says Hinton, who is now working with Google as vice-president, engineering, Southern Group. “I was doing machine learning with neural nets when very few people were doing it, and managed to attract all the best Canadian graduate students from Canada and elsewhere. I was able to get a lot of interaction with a number of people in Canada, one of the most important being Yoshua Bengio (at University of Montreal).”
The startup community followed suit around 2013 as people began realizing that deep learning could transform everything from the way companies do searches and translations to medical research and drug discovery.
“AI came from being an academic interest to a central interest for major companies,” Hepburn says. “The industry became a feeding frenzy.”
Those industries hungry for talent and/or acquisitions included Amazon, Google, Microsoft, Uber, Facebook, Twitter and Apple, among others.
Canada’s position continues to be strong thanks to homegrown talent and strong support for the startup community, he adds. “We’re in competition with the big kids south of the border. But Canada has both the research leadership as well as extensive startup activity. A number of groups are working actively to support the startup community in Canada, such as Creative Destruction Lab, Element AI and NextAI, to name a few.”
Daniel Mulet, associate director of machine learning with Creative Destruction Lab at U of T’s Rotman School of Management, says increasing numbers of apps are being created and used in real time to solve real business cases and real world problems. Some research teams in fact are being snapped up even before they can form a company. “We’ve seen that happen more and more.”
But there are also a growing number of researchers who are creating their own businesses. “They see much more value in being acquired. There’s a lot of appetite on the part of large corporations and VC firms for machine learning companies, which has led to a flurry of activity on the investment side.”
Success stories on that front are legion, including Deep Genomics, Thalmic Labs, Atomwise, Blue J Legal, Eigen Innovations and Winterlight Labs. “At the core of all of these is some form of machine learning to solve a real world problem,” Mulet says.
Last year his program had 25 companies on board, half of which received financing of $1 million apiece. This year the number has doubled to 50, including some based outside Canada, he notes. Investors fly in from Europe and the U.S. several times a year to seek out investment opportunities. “They like Canada because there is less competition for deals, and investors are able to find interesting entrepreneurs developing interesting technologies they can’t find elsewhere.”
The demand for expertise around the world, however, is putting pressure on Canada to keep pace. Brendan Frey is a former U of T deep learning research team leader who is now president and CEO of Deep Genomics, where AI is being used to detect and treat disease genetically. “At one point the best technology in AI came from Toronto. It was such a huge success. So much so, it caused many good people to jump on other opportunities.”
Frey is now working with the community to rebuild those capabilities. The goal is to combine the efforts of academia, industry and the accelerator community to build a central entity for developing and training engineers and research scientists.
Hinton concedes that Toronto is now in a curious state, since many of those former students are now running AI groups at big companies. What is needed to replenish the supply is a free-standing AI institute devoted to research and startup activity, he says. “We want to help startups and get those big companies to create labs here and encourage researchers from other universities and companies. The governments seem sympathetic, and companies are very interested, so we’re very optimistic.”