In Edmonton for a reason
MAR 28, 2018
In the cutting-edge field of artificial intelligence and machine learning, Alphabet-owned DeepMind is a very big deal.
And in July 2017 when the company announced its plan for its first lab outside London, U.K., in Edmonton, suddenly the city's position as a leading centre for AI research vaulted into the global consciousness.
Artificial intelligence deals with tasks that one normally would think require human intelligence. Machine learning takes it a step further with algorithms that allow a computer to learn without extensive knowledge input by humans. DeepMind's work focuses on building general systems inspired by how our brains work that can perform well across a wide range of tasks.
Three professors were the draw to Edmonton for DeepMind. Richard Sutton, Patrick Pilarski and Michael Bowling, all leaders in the field who have been teaching at the University of Alberta, were recruitment targets for DeepMind. But they didn't want to leave the city or the teams they had built at U of A. DeepMind's founder, Demis Hassabis has long had close ties with Sutton and knew it could make even greater progress in terms of research breakthroughs by collaborating even more closely with the University and the academic community. So DeepMind established a beachhead in Edmonton, with 15 researchers by the end of 2017, and plans to grow in the future.
"We thought there was a great opportunity to try in an industrial setting to accelerate the research going forward and start building the next step in artificial intelligence," Mike Bowling says. "That aligned with DeepMind. But at the same time we loved the training component we do with the university. And we think Edmonton is a great place to see these things combined together."
The U of A has been turning out world-class graduates in this field for some time.
"I think one of the reasons DeepMind was excited about the three of us is many of our graduates ended up at DeepMind," says Bowling.
Other AI labs, including the university's Alberta Machine Intelligence Institute (AMII) and RBC's Borealis AI lab are also considering moves to the downtown core.
"It creates an opportunity to build up a bit of an ecosystem where the lack of office space near the university makes that a bit more difficult," Bowling says.
Bowling says there is a growing AI community in Edmonton, in research and commercialization. After DeepMind announced its expansion to Edmonton, there was an uptick in graduate students interested in the U of A program, which further widens the talent pool.
Plus there is entrepreneurial activity growing.
"If you’re getting trained inside the U of A and you have a bent to those big long-term problems, that’s great. But there’s a lot of people who are excited — let me take the technology we have now — I don’t need to make it any better. Let me apply it in places I can really help people. it’s great to see that happening too," Bowling says.
"The talent pool here is almost totally untapped because there isn’t the equivalent tech giant you have to compete with. DeepMind is going to be the one place nearly every graduate who is interested in machine learning is going to want to come to."
DeepMind, established in London in 2010 by Hassabis, was acquired in 2014 by Google and is now part of the Alphabet Group. It recently added a lab in Montreal and has a DeepMind applied team in California.
Bowling says the team in Edmonton is involved in projects that include researchers in the other centres. For instance there are some people in the city working on a project out of London that is focused on reducing energy consumption; DeepMind artificial intelligence has already cut the energy required for cooling Google's data centres by up to 40 per cent.
Much of the work focuses on fundamental research.
"It's largely around reinforcement learning. We have systems learn on their own to accomplish goals," Bowling says. They build on that research to create general algorithms that can be applied to a large array of problems.
"We don’t typically take one problem and see how far we can push it, but we push the generality. That’s a lot of DeepMind’s philosophy too. The mission is solving intelligence and making the world a better place. The better we do that first part, the better we build intelligent algorithms, the better we do the second."
Many of the news headlines around DeepMind have centred on games. DeepMind's AlphaGo project developed the ability to play the complex game Go better than any human champion. A further iteration taught itself chess from scratch in four hours and beat the reigning computer chess program in an early December 2017 tournament.
Why games? Bowling explains games are "a great way to test algorithms. If we want to apply this to the real world they need to be generalizable." Bowling adds that researchers can talk with each other about games without getting into the detail of commercial applications.
DeepMind in Edmonton is currently housed in the Manulife building in the heart of downtown. Bowling says the intention is to grow and find more permanent headquarters, likely still downtown, within a year.