Unexpected insights from an AI rock star

Richard Sutton literally wrote the book on a branch of artificial intelligence called reinforcement learning. He's recognized as one of the top AI researchers in the world.

The U of A computing science professor shares some of his thoughts about artificial intelligence.

It's Not Really Artificial

» It's really too bad we have this name "artificial intelligence." Let's just call it intelligence. It's not really so much making an alternative to humanity; it's understanding humanity. It's the desire to understand the world, the desire to make tools to make ourselves better. It's really totally human - the natural continuation of what humans are.

The Appeal of AI

» I think of it as a great scientific objective, a scientific prize, if you will, and maybe the greatest scientific prize of all time. We will understand how a mind works. We will understand ourselves, in outline, not in the specifics of ourselves; the important parts of ourselves; the ability to achieve goals and to understand the world. ... And that's a really big deal. You can compare it to Darwin's discovery of how evolution works.

Why We're Afraid

» The source of the fear, I think, is that we are creating a new kind of being, and people have a long history of being scared of newcomers. We've been scared of our technology since the '60s, and the Industrial Revolution long before that. In the end, they [newcomers and new technology] take the jobs we didn't want to do anyway, and we end up having plenty of things to do. It's true that some people have to change their jobs, but that's just change and it's improvement.

How AI Could Change Us

» It's very hard to say, even in outline, how things will change when we understand this. Even what we want to do will be different, our goals and objectives. What if we can digitize people? Then life and death are different. You don't have to die if you can have a backup and be restored from the backup. Then you have to decide, 'Well, if I die and I'm restored from my backup, am I the same person or am I a new person? And would that be satisfactory?' Suppose you are digitized and could be brought back from a backup. You could also be brought back twice; your digital mind would have to be instantiated in a hardware body but you could do it. And if you could do it once, why couldn't you do it twice, or 1,000 times?

The Deep Mind of Richard Sutton

When Richard Sutton stood in front of his first class at the U of A in 2003, he told the students he might not be around to finish the course. But he would try.

Sutton's cancer had returned. He had endured four major surgeries, chemotherapy and immunotherapy after aggressive melanoma spread to major organs, including his brain, years earlier. The cancer was in remission when he arrived in Canada to start his new position, but now the tumours were back.

"My odds were never very good, but we just kept fighting," Sutton says of his years of treatment.

Doctors at the Cross Cancer Institute in Edmonton treated Sutton with temozolomide, a powerful oral chemotherapy drug. His tumours were last seen in 2004, and he stopped all treatment in 2005. "Twelve years clear - it looks like I survived," he said in an interview last fall.

Sutton is considered a founding father of reinforcement learning - a key methodology in artificial intelligence research. Twenty years ago, he co-wrote Reinforcement Learning: An Introduction, which is still considered the definitive book on the subject.

Despite his already significant renown in the field of artificial intelligence, at the time Sutton's health issues made it a bit of a gamble to bring him to the university, says Jonathan Schaeffer, U of A dean of science.

"And it was a brilliant one."

Sutton is now one of the top dozen or two computing science superstars in the world, says Schaeffer. "If we had a Nobel Prize in computing science, his is the kind of name that people behind the scenes would be whispering about."

Sutton's field, reinforcement learning, is the computation version of how an animal or person learns, he explains. "You try something and if it works you get positive feedback. And if it doesn't, you get negative feedback. Eventually, you learn what is good behaviour and what is bad." The field has links to behavioural psychology, Sutton's first area of study as an undergraduate at Stanford in the 1970s.

Most recently, Sutton was asked to serve as distinguished research scientist for DeepMind, Alphabet's artificial intelligence research and development firm. (Alphabet is the parent company of Google, among other companies.) In July, DeepMind opened an office in downtown Edmonton, marking the first time DeepMind set up a research lab outside London. The move made serious waves through the world of AI.

"Would DeepMind be in Edmonton without Rich Sutton? Probably not," says Schaeffer. "They could have set up at Stanford or Carnegie Mellon - anywhere they wanted - but their first place was 'the Subarctic of Canada,' as one of my colleagues likes to call this place."

DeepMind's arrival here has opened the eyes of many other tech companies to Alberta's potential, adds Schaeffer. "The knock-on effect is enormous."

For all Sutton's global influence and renown, those who know him are struck by his essential modesty. "He is one of the truly great computing scientists working in the world today, but I don't think he's at all interested in fame or fortune. I admire that," says Martin Müller, a longtime colleague in the U of A's computing science department. Müller calls Sutton a "pure researcher and a deep thinker."

As to his own future, Sutton sounds like many people who have faced life-threatening illness.

"The first thing it does is make you realize that life is very valuable and not infinite, and so you should figure out what you want to do and work on what you want to work on."