The Expert: What I Know About … Artificial Intelligence
Patrick Pilarski has a Ph.D in electrical and computer engineering from the U of A and he researches artificial intelligence, human-machine interfaces, machine learning, biomedical engineering and robotics.
By Caroline Barlott | September 1, 2011
Photography by Darren Wolf
Who: Patrick Pilarski
Experience: He has a robot in his office. But this isn’t unusual at the University of Alberta’s department of computing science, where he’s a postdoctoral fellow with the Reinforcement Learning Artificial Intelligence Group. Pilarski has a Ph.D in electrical and computer engineering from the U of A and he researches artificial intelligence, human-machine interfaces, machine learning, biomedical engineering and robotics. He helps develop prosthetics imbued with A.I. technology. Pilarski is also the co-inventor of technology used in a hand-held medical device that can diagnose diseases such as cancer. His right brain is also in top form; he’s the co-editor of DailyHaiku, an international journal of contemporary English-language haikus, and the author of several books, including Huge Blue, a collection of poetry that was published by Leaf Press in 2009.
“We develop artificial intelligence so that a machine can learn to the same degree or even better, than a human could. The smaller-level bits of [A.I.] involve solving problems like doing better Web searches or targeting ads to the right people. But everything from intelligent transportation, like driverless cars, to intelligent artificial limbs – one of my areas of focus – can use the principles of A.I.”
“Electrical prosthetics can operate using electrodes that adhere to the skin, on say a bicep or a tricep. You stick it on and it will record through the skin how the muscle works. When muscles flex and contract they give off very small changes in electrical potential. There are resulting electrical changes in the skin; this is what the sensors record, and we can then use these recording to control a prosthetic.”
“We’re working on prosthetics that use artificial intelligence, whereby the patient can actually give it feedback, called ‘reinforcement learning.’ The amputee can give a limb positive or negative reinforcement by pushing a button that uses a number system – positive numbers signify a high score and negative numbers signify a poor score. Now amputees are working like a team with a limb. The limbs aren’t on the market but will be in five to 10 years.”
“The learning part of the artificial limb can make predictions and improve on its predictions. It looks at situations, and determines what should be done based on past behaviours. So, maybe last time, the limb tried to pick up a cup, it spilled on the pants of the amputee and got a really bad negative reward. Through trial and error, they build up predictions about what will happen next.”
“A.I. prosthetics are especially useful for a new kind of amputee called a TMR, targeted muscle reinnervation, patient. If people lose their arms in accidents but their nerves are undamaged, they can be rewired into the pectorals or the biceps. So, now when the amputee thinks of closing his or her hand, the correct muscles flex and an electrode can monitor the movement and operate a robotic limb.”
“A lot of the mechanics of the algorithm that we use in A.I. reinforcement learning are often very similar to the processes that happen in the human brain. Our dopamine system helps the brain learn to predict future rewards, and the brain can then use these predictions to make decisions. This is very similar to the learning system used in intelligent artificial limbs.”
“Having a machine spontaneously decide that humans are the enemy is an extremely unlikely situation. If someday, we have machines with A.I. that can match or exceed human-level intelligence, the robot will be doing much better things with its time. It might be trying to calculate the trajectories of galaxies through space, or trying to figure out the relationship between butterfly wings and orange prices in China.”