Wednesday 31 May 2017

AlphaGo

The AlphaGo program not only continues to beat the best Go players in the world, but it is also influencing how the game is being played. Talking with some Go playing friends, they were amazed at how AlphaGo was demonstrating ideas and concepts that had been considered bad, were in fact playable. As a result, top level professional players are reassessing how the game is being played.
It seems that this effect may be even more profound than the effect computers had on chess. While computers probably taught the modern generation the increased importance of tactical calculation, and probably helped resurrect some openings that had been considered less than optimal.  the underlying strategic concepts did not really change. Computers did the same things that humans did, just faster and better. With AlphaGo, it seems that its learning method of recognising good and bad moves based on patterns and previous games has not only come up with better moves, but also enabled it to recognise better structures.
AlphaGo has just completed a series of matches in China, against some of the worlds leading players. At the end of the match the AlphaGo developers have released a set of 50 games which AlphaGo played against itself, to show some of the new ideas it has learned. You can play through all 50 games here.

No comments: