Members of The University of Texas at Austin's Learning Agents Research Group have been a resource for members of the media about how to teach artificial intelligence systems to learn.
Often, systems must learn their own way around the world, much like an infant. The algorithms responsible for these systems are left to roam around and make mistakes – slowly but surely gaining an understanding of how they are meant to function.
Some AI systems are given robot bodies – like the RC cars tested by Amazon written about in a recent Wall Street Journal article – and are sent out to run into things until they learn how to drive themselves properly.
The University of Texas at Austin's computer scientist Peter Stone, David Bruton, Jr. Centennial Professor, commented in the article on the car's algorithms that learn by such a reinforcement learning approach.
"The algorithms are complex because they gather data on their own, instead of being fed millions of images to learn from," Stone told the newspaper.
Stone also contributed to a recent research project that applied the same principles of reinforcement learning algorithms to listeners' music preferences.
Stone joined recent computer science Ph.D. alumnus Elad Liebman and Maytal Saar-Tsechansky, a professor of information, risk and operations in the Red McCombs School of Business, for the project, which created a 'DJ-MC' that queued up new songs based on user likes.
On how the technology works, Saar-Tsechanksy told McComb's Steve Brooks in a Medium article that the algorithm learns as it goes: "It's learning on the fly while you're collecting more information about the user."
The research is published in the MIS Quarterly journal.
Comments