Check mates: AI goes beyond every chess composition you’ve ever dreamed
On International Chess Day, we marvel at how AI has taken this timeless game to new heights – and how robotics can make it more accessible.
This year’s International Chess Day on July 20 is a celebration of marvellous human minds and, these days, artificial intelligence and machine learning.
Chess and the 3000-year-old Chinese board game Go have been something of a proving ground for elegant examples of AI forging its own deep neural networks.
In 2016, Google DeepMind’s game-playing AI AlphaGo beat the world’s best player of Go, which is said to be even more difficult than chess. During a challenge match against South Korean world-ranked Go professional Lee Sedol, experts and 60 million fans in China alone watched in awe as AlphaGo’s moves were placed on the board, with one move described as “really creative and beautiful”. Sedol resigned in short order.
As headline grabbing as the events were, DeepMind’s ambitions for its AI are way beyond defeating genius-level boardgame players. “There are so many application domains where creativity in a different dimension to what humans can do could be immensely valuable,” explained DeepMind lead researcher David Silver.
Next came AlphaZero. While its predecessor was trained on algorithms based on data from past games played by masters, AlphaZero had no human knowledge in its algorithms and taught itself to play Go, the Japanese game and chess in a matter of hours.
After beginning playing chess randomly against itself in the morning, by dinner time AlphaZero had ascended to what AI researcher and DeepMind founder and CEO Demis Hassabis described as a “superhuman level of play.” (Hassabis was also a child chess prodigy.)
Chess champions pre-empt and thwart opponents’ moves with a shrewdness that would make Machiavelli blush. Yet as AlphaZero defeated chess champions, they noticed how it played within the regulations but beyond human cognition.
Traditional chess engines – including the previous incumbent super program Stockfish – have a huge database of rules to direct plays. AlphaZero knew the parameters of the game, but learned how to play it through testing itself.
“Because it had taught itself, it might play the game in a completely different way from the way we play it,” explains chess champion Natasha Regan, who DeepMind’s Hassabis brought in to challenge AlphaZero and who has since written a book about her experience and the promise of the AI, Game Changer. “It’s a check on everything we’ve taught ourselves as humans since chess was devised … and it feels like it’s got a lot of potential to do a lot of other things.”
As the possibilities of AI applications continue to expand, on International Chess Day we also salute how technology is helping to make the beloved, brain-testing board game more accessible.
In 2017, the same year that AlphaZero was defeating champions, ABB and Spanish startup Irisbond joined forces to create a chess-playing robot that allowed MS sufferer Marga de la Torre to direct her moves on the board using facial and eye movements. By monitoring the physically challenged player’s face with an eye-tracking camera, the robot identified her desired moves and made them for her on a physical board. Chess-loving De la Torre described the experience as “like a dream”.
Just as chess is timeless, the AI dream is just beginning. Wishing Happy International Chess Day to human and AI champions alike.