Alphago, a computer program with a neural network engine, recently beat Lee Sedol, a top level Go professional, four to one. This fact has implications for the field of artificial intelligence and for high level Go players. It has also changed my personal relationship with Go. Let me say it: I no longer love the game of Go.
One of my earliest posts alluded to my love and respect for Go. I still play – a quiet, serious game with a clock timer – once a week. I really, really enjoy the game. But the mystery is gone. The black and white patterns no longer fascinate. They are just interesting puzzles to solve. The ideal, the search for the perfect move, has been made mechanical – a search through historic patterns and tree-based practice games played in a flash of time.
Learning to play Go at all competently for someone of my limited capacities was a matter of learning stories, analogies, proverbs and common opening sequences. Alphago has access to every recorded Go game. It wins by being a superior pattern recognizer. Alphago also plays games with itself (one million in one day), something humanly impossible obviously in terms of time but also because people can not bifurcate their minds to play unbiased games.
When I win a game of Go against a human player, I like to think that I had a superior strategy – using outside influence, a leaning attack, etc. If I played against Alphago, it would be my story versus its totality of historical game knowledge. Any strategy I might know would have come from teachers/book authors who have condensed their study of past Go games into a simple mnemonic. Even if I got to the endgame, Alphago would play perfectly choosing the highest point move every time. I would have no chance.
Alphago has revealed the game of Go to be an exercise in pattern sorting and calculation. It was always so. I just couldn’t see it. My approach to the game given my limited error prone human brain was to create a mythology. I now know I have been worshiping a false god.