Watching complex, strategy-based games is difficult unless you understand the play mechanics. And it’s even less entertaining when the action isn’t at 60 frames per second but instead on the seconds hand of a chess clock.
Now a team from MIT is trying to make 1,500 year old game into an e-sport, starting with the Millionaire Chess Open championship in Las Vegas. Throughout this week’s tournament, they’ve been testing a system called DeepView (a play on Deep Blue, the IBM super computer that beat chess champion Garry Kasparov in the 1990s). It combines algorithms, leaderboards, and real-time game visualizations to turn chess into a smart spectator sport.
“In the run-up to the tournament I gathered data on the top 25 players,” says lead MIT researcher and data scientist Greg Borenstein. “All these statistics are being used as part of the on-screen graphics.”
Announcers called the game while it was being played, and instead of the traditional method of tracking positions on the board–‘rook to queen bishop 4’–they gave odds on who was most likely to win based on the current board positions.
Borenstein basically created a scoring system for chess. It took more than stats know-how: He had to learn the game. “I’m not a chess expert. Maurice Ashley, a grandmaster, is my collaborator on this project,” he says. “Before this, I knew no more about chess than the average person.”
One thing he discovered is that there aren’t many tools that describe different types of players, or how likely they are to win a chess match based on the moves they’re making. So Borenstein did what any data scientist would, he analyzed 750,000 chess games.
“I used a chess engine to evaluate how strong a position that is,” he says. “And then I built up a statistical model of how likely each position is to result in victory for that side.”
Different chess players value different aspects of the game. Some aim for mobility, others focus on space, threats, pass/pawns, or king safety. There are aggressive early game players who attack immediately to shake the other player up. And there are more structural players who build toward a later game. Borenstein’s team analyzed them all.
“I can characterize them as players,” says Borenstein. “I can also look at how they’ll match up head-to-head and detect weaknesses and eccentricities in their game, when it will pay off and they’ll win and when it won’t.”
While live-casting chess matches won’t replace baseball anytime soon, DeepView does create a more accessible experience for anyone who wants to watch masters play (and learn from them).
It also represents a human-focused application of data science. Instead of building an artificial intelligence engine–using computers to win games and replace people–it’s about using the data to tell the story of each game.
“There is a change in how computer science uses data, a move from trying to replace people to understand them,” Borenstein says. “It’s the Google and Facebook way. Except that this uses those techniques to not sell things, but to make games more exciting to more people.”
Even if you’re more into Defense of the Ancients than watching play-by-play chess, this is worth checking out. The reason: DeepView’s open source software was built with a larger long-term goal of modeling how other games can become spectator sports. Including video games.
Multiplayer championship sports like League of Legends or Defense of the Ancients are extremely complicated. “They make chess look much simpler,” says Borenstein. “Tens of millions of people watch The World Championships of DotA, but if you’re a novice viewer trying to watch you have a very hard time making sense of it.”