HENRIETTA, N.Y. (WROC) — A team of researchers at RIT are helping to develop artificial intelligence capable of playing the popular strategy-based video game “Starcraft II.”
The team behind the project says that once it’s complete, this could serve stepping stone to advance technologies like self-driving cars and service robots, among other real world applications.
RIT researcher Christopher Kanan — who specializes in creating optimized algorithms to expedite the learning of these machines — says this technology is already in use today in search engines, and other products.
He says that “AI” is a very loosely defined term, but he says in this context it’s computing whose infrastructure is modeled on the neural patterns of our brain: one neuron connects to the other, and take information from each other.
Compare that to a more typical computing program that has two switches, on or off, to run algorithms to complete programs.
To “program” neural network AIs, Kanan says that you have to introduce it to a massive amount of data at once, and then the machine learns it all. This is different than how humans learn, which is incrementally.
This makes board games ideal to test a AI’s efficacy; there are only a limited number of moves, and layouts, even if they’re very complicate games. Previous games like chess and go have often been litmus tests for AI, but RIT explained to us why StarCraft was the next logical choice for them to try the artificial intelligence.
“In StarCraft, that’s not true. In StarCraft, there’s all this area that you cannot observe immediately, so you cannot know,” said Kanan. “So it’s a problem that you can attack in order to not only study what your opponent is doing, and to learn how to play this game, but you have to do with incomplete information. You don’t have complete information at all times.”
More on the funding of this project was provided by a RIT press release:
The project received nearly $210,000 from the Defense Advanced Research Projects Agency (DARPA) to work on phase two of the Lifelong Learning Machines (L2M) program this year. After having success with their research in phase one of the L2M, Kanan and his team were selected to work on this new project led by SRI International and includes collaborators from American University and Georgia Tech.