Top scholarship for Waikato student to teach a computer to think

Teaching a computer to think may sound futuristic, but University of Waikato doctoral student Sam Sarjant has just been awarded a government-funded Top Achiever’s Doctoral Scholarship to do just that.

The scholarship, one of 14 awarded across NZ in 2009, is worth $93,000 over three years. The funding will allow Mr Sarjant, a former student at Hauraki Plains High School, to investigate ways to use reinforcement learning, which involves rewards, in a relational environment, where the computer builds up an internal map or model of the different objects in its environment.

Computer scientists currently use reinforcement learning to train computers to play games, such as chess or backgammon. Mr Sarjant says it’s a little bit like training rats in a maze.

“A rat in a maze learns how to get to the cheese; it’s sort of the same with a computer; it has to learn which actions give it the best result, and then infer new strategies from what it’s learnt.”

Last year Mr Sarjant came fifth in an international competition for creating the best computer learning strategies for the game of tetris. But in his PhD research, he is now hoping to devise a way to teach computers to think outside the confines of a game. The idea is ultimately to create agents for environments that may be difficult for humans to operate in, such as space exploration, mining or nuclear waste clean-up operations.

“I’m aiming to create a general learner that can learn in any environment,” says Mr Sarjant.

“It does this by receiving observations about the problem and acting on those observations with actions.

“For example, when playing poker, the agent needs to build up a picture of the significance of all the different hands. So if you have the observation of having a full house, a good action would be to bet high. The agent learns by receiving reward (in this case poker chips) and learning which actions and observations work best.”

Mr Sarjant is one of five University of Waikato post-graduate students to receive a 2009 Top Achiever’s Doctoral Scholarship. Three specialise in computer science, one in psychology and the fifth in engineering.

Mr Sarjant has also just garnered the best student paper award at an international conference on web intelligence. His paper All You Can Eat Ontology-Building: Feeding Wikipedia to Cyc’, co-authored with Catherine Legg, Michael Robinson and Olena Medelyan, won the award at the 2009 IEEE/WIC/ACM international conference held in Milan in September.