There are a host of strategy games that the world has come to love. The most well known is chess, but that’s just one of a countless number of games which pit mind against mind. The most complex game, according to strategy theorists, is an ancient Chinese strategy game called Go. Recently, developers at Google’s Deep Mind designed an AI Go player named AlphaGo.
The Google developers set up a challenging competition to test their AI player. They pitted AlphaGo against some of the best players in the world including the world famous Lee Sedol. AlphaGo won the tournament, 4 games to 1. However, the game itself produced some really interesting new understanding of how AI works.
For starters, Go has some pretty traditional strategy that even amateur players know well. There are certain moves that are outside the scope of normal strategy. Lee was playing Go with this built in perspective on strategy. However, AlphaGo was not. This meant that AlphaGo was able to make moves that were unthinkable to a human player, but finally resulted in victory. In their first game together, AlphaGo played a move that was completely unexpected from a strategy viewpoint, and Lee was shocked. However, as the game moved on, it was clear that AlphaGo’s move had been strategic at an incredibly deep level. As Lee watched, AlphaGo controlled the section of the board and ultimately was victorious. The strength of the AI was that it was able to analyze moves from a non-traditional perspective and literally think outside the box of Go strategy. In fact, the AI continued dominating Lee with moves that were surprising. When traditional players would have attacked, AlphaGo defended, and ultimately won. The AI also sacrificed some stones in order to control large sections of the board that were open. Through the first four games, AlphaGo kept Lee on his toes and managed to control the game in a way that was remarkable.
However, in the final game, Lee made some moves that were outside the scope of normal moves and this made AlphaGo begin to make mistakes that were even below amateur level. The AI was unable to cope with Lee’s changed strategy, even as Lee took a page from the AI playbook. While Lee played according to convention, the AI was able to think ahead of Lee’s strategy, but as Lee shifted his thinking, the AI was perplexed. The results indicate that the learning for the AI is still a work in progress. Lee’s ability to evaluate at a higher level and make moves that would confuse the AI show that the human thought process is still far more complex than the AI system.
The results are being analyzed, but representatives from DeepMind have already said that they will not be using the AI to play Go anymore, and AlphaGo will hang up its stone bag. Instead, they’ll be focusing on using the technology to help with medical analysis and other real world problems where high level thinking is required. Keep up to date with the AI market here.