Edmonton leading way in artificial intelligence research
“How the hell did you guys build a world-class A.I. research group in the sub-Arctic?”
That question directed at University of Alberta computer scientist Jonathan Schaeffer came from an American professor in a warmer city. It was prompted by the CSrankings, a new way to rank computer science research teams around the world based on the number of publications they have in the top computing science journals.
The University of Alberta now ranks third in the world for artificial intelligence research, behind only Carnegie Mellon University and Tsinghua University in China.
The U of A has led the way in such A.I. fields as computers beating the best humans in checkers, poker and Go.
Schaeffer has been at the forefront of the Edmonton effort as a researcher and as an administrator, leading the computer science program for a time and only recently resigning as the dean of science to return to full-time research and teaching.
“Like most people in our department we were surprised in a way,” Schaeffer said of the high ranking. “We knew we were producing high quality research and we’d done so for a long time, but we had no idea we were No. 3 in the world. It was immensely gratifying.”
Schaeffer says the U of A program has a few things going for it, starting with Alberta government funding from agencies like Alberta Innovates. There’s also a friendly working environment, with colleagues eating lunch together, sharing ideas and working on projects together. Finally, there’s Edmonton itself, Schaeffer said, with its affordable housing and great public schools.
“This is safe, it’s clean, it’s a nice place to live. People like myself and many of my colleagues have had opportunities to go elsewhere. And quality of life is important. Which means we stay here.”
Schaeffer, who is co-author of a new book Computer Chess Man vs. Machine: Challenging Human Supremacy at Chess, has witnessed a transformation in approach to artificial intelligence during his career. At first, there was a kind of snobbery against those who studied issues like beating the best humans at checkers or chess, he said, largely because some researchers didn’t approve of the “brute force” computer processing that Schaeffer and other A.I researchers employed.
Some researchers felt the whole idea of A.I. was to try and mimic the processes of the human mind, only with fast processing and extra memory. It sounds like a simple matter to do this, save for one thing — we don’t understand how the human mind works, Schaeffer said.
“Our brain is like a black box … The brain is an amazingly complex, sophisticated machine.”
Grouping super computers
Schaeffer and others realized that beating the best humans at complex games might well be accomplished by grouping ever greater number of super computers and building ever better chips.
“It was taking a sledgehammer to a nail. It was throwing massive amounts of computing at a problem to come up with a simple chess move … People were deriding it, saying, ‘That’s the easy way out. You’re not advancing the state of the art. That’s just going faster and faster.’
“But what people learned and what I learned very quickly is that if the human brain is a black box, why shouldn’t our artificial intelligence systems be a black box? The bottom line should be performance. Does it play like a grandmaster? … What goes on underneath the hood doesn’t matter.”
A human can process two chess moves every second. When Schaeffer started out building a chess program in 1978, his computer could do 300 moves per second. Deep Blue, the chess program that beat world chess champion Garry Kasparov in 1997, looked at 200 million positions per second.
The idea of computers using a massive amount of data and computing power has now been thrown at other problems. For example, computers struggled with facial recognition as recently as a decade ago but are now able to analyze hundreds of partial images of shadowed faces entering a mall and pick out individuals, Schaeffer says.
“They’re now super human at recognizing faces … Almost anything to do with visual recognition, we’ve been leapfrogged and computers can do a better job than us and this was something we didn’t expect.”
I’ll end with Schaeffer’s thought that he’s not fearful about a malicious A.I. one day harming humanity.
“You’ve got to believe in humankind. If you don’t believe in that, this is a pretty dangerous world. We’ve managed to survive to this point with potentially deadly technology, and A.I. has the potential, but you’ve got to believe we’re going to do the right thing.”