It’s an arms race
It appears the time when computers will begin beating humans at Go is coming. The solution to this “problem” is to create better games. Chess is failing. Even without computers, chess at the highest level is being victimized by its own success. As more and more people play it at higher and higher levels of play, the opening play of the game is slowly being “solved”. There are too many openings that simply aren’t playable at the highest levels, because they inevitably lead to inferior positions, provided the opponent knows how, and these days, the grandmasters have studied thoroughly enough to know how. Computer now being mostly unbeatable only worsens the situation. How long can we keep playing such a game with some sense of satisfaction, knowing that nearly any box of metal with some very stupid set of rules can play it better?
What we need to do is improve our games to keep pace. Go is still beyond the ken of computer simply because of its size. But, as computers get faster, this will no longer be an obstacle. There is nothing special about Go that makes it a more difficult problem than Chess, except it’s sheer size. But, computers continue to get faster, and new techniques continue to be developed. Eventually, Go will fall.
So, what if we created new games that upped the ante? Create new games that increase the search space without increasing the complexity beyond what humans can handle? Humans seem to handle different search space sizes the same way. We simply apply more pattern-matching skill to the situation. For instance, in tic-tac-toe, we can brute force the solution ourselves, and so we do when playing it. In chess, we can brute force to some degree, but not enough, and so we use more pattern-matching ability to play it effectively. In Go, we can brute force very little, and so rely almost exclusively on pattern-matching. A bigger game doesn’t change this - we still won’t be brute forcing it, and so it makes no difference to us.
Further improvements can also come from the introduction of minor elements of chance. Just enough to destroy the deterministic quality of a game can severely hamper the computer’s ability to brute force a solution.
The advantages of embracing this games arms’s race are twofold: A) We get better games to play and B) AI developers get better problems to tackle. It’s win-win (but let’s not tell the game players that).
