Google's DeepMind Defeats the Toronto Maple Leafs

Three members of the DeepMind team during the opening skate.

TORONTO, ON - DeepMind defeated the Toronto Maple Leafs by a score of 5-0 in exhibition play last night. It was the first time an AI has won a hockey game against a professional team. Demis Hassabis describes this as "Phenomenal."

"We've had a lot of difficulties trying to teach DeepMind to play team sports. Up to this point it was mainly basic video games and board games. But this goes to show just how versatile DeepMind is," Hassabis continued. "Especially when DeepMind lacks important details, such as opposable thumbs or knees that bend, or even having legs at all."

DeepMind is designed to learn in a way similar to humans. By practicing and observing games, the AI learns the rules, and soon develops combinations of how to respond to situations as they arise within the rules. By observing it's environment, it learns and adapts. So how did DeepMind learn Hockey?

"We started with some video games," Hassabis went on. "We had DeepMind play thousands of games on NHL '95 to get a feel for the basics. Then we had him run hundreds of thousands of reviews of recent games to understand most of the standard plays and changes to the rules since the Super Nintendo era. After a slight neural pruning process, we set him up against Toronto, just to test it out, and the results were better than we had hoped."

Their next goal is to test DeepMind against better teams, including the Montreal Canadiens, who hold the record for most Stanley Cup Wins, as well as against more recent and relevant Stanley Cup Champions, the Chicago Blackhawks, Los Angeles Kings, and Pittsburgh Penguins.

DeepMind staff hope to have enough data from the World Cup of Hockey to prepare for a best of 8 series against Team Canada late next year, with a possible upcoming series against Russia.