Why humans can still play chess

Ken Rogoff writes,

At one time, it did seem that computers would sound the death knell for chess, not to mention all human mind games. That was certainly my guess in the late 1970s, when the rise of computers was one of the main reasons I gave for retiring from competitive chess.

Pointer from Mark Thoma.

Tic-tac-toe is not much of a game. It has been solved, and the solution is easy to remember.

I believe that computers have solved the game of Othello, and with best play by both sides, White should win, 33-31. But humans still play in Othello tournaments.

The reason is that the “solution” to Othello is impossible to remember. There are many, many variations that lead to the optimal score. More important, there are many openings where if you have memorized the continuations and your opponent has not, you can get an outstanding result.

Another way to put this is that a computer has sufficient memory to put every playable variation into its “opening book.” The only positions that a computer will not recognize from memory will be ones where it is way ahead.

Humans do not have that much memory. You can take a human out of his opening book and force him to think, without putting yourself in a hopeless position.

I suspect that a similar solution to chess is attainable. Suppose that you turned a computer loose playing against itself millions of times, a la Alpha Go, and you discarded all of the variations where one side falls hopelessly behind, until you end up with the solution to chess. My guess is that this would consist of tens of thousands of variations that end in a draw. A computer can store all of these variations, along with refutations of plausible alternatives.

If you play against such a computer, you must either follow a line that the computer has memorized, leading to a draw, or pick a line that loses easily. Same as in tic-tac-toe.

If I am correct about Othello, a championship match involving computers would just result in a lot of 33-31 wins for white that the computers have already memorized. If I am correct about chess, then such a match would just result in a lot of draws that the computers have already memorized. No new lines would emerge.

But humans could still play the game, because they don’t have that much memory. As an aside, when he was pursuing the goal of achieving grandmaster status, Rogoff told me that he had 30,000 entire chess games in what he called “active memory,” with more in storage. Wow!

6 thoughts on “Why humans can still play chess

  1. The problem with chess is that memorizing openings is boring and doesn’t add to any kind of understanding of the world. What’s interesting is the heuristic reasoning, discovery of meta cognitive fault lines, dealing with logical uncertainty.

    Chess960 kind of solves this problem, but doesn’t have the critical mass of players.

  2. Rogoff’s 30,000 memorized games is exactly why I don’t like chess as a competitive game. I would just as soon compete in reciting epic poetry, being scored on how exactly I reproduce the content word for word. I would much rather spectate that, as well. To my mind, there are better things to spend time memorizing, but to each their own.

  3. I don’t think modern game-playing systems e.g. AlphaZero are memorizing to the extent suggested in this post. They’re learning from experience but that experience is stored in a highly abstract and largely inscrutable representation, not as some kind of lookup table that simply lists the optimal move for each positions.

  4. Rogoff’s memory is extraordinarily impressive. I rarely can beat the chess software on my computer. Unsurprisingly it is much easier to beat the computer playing scrabble, even at the expert level. The element of luck gives better odds. Plus the vocabulary lessons are interesting and sometimes useful. Also unsurprisingly, I prefer to play scrabble.

  5. I enjoyed playing the computes in chess, up to a point. Up util a few yeas ago, you could set the settings to ‘medium’, and have a good game. Now the computer comes back instantly with a move even when the setting is at ‘expert’. The fun went away. In the former case I was playing against the algorithm designer, a human.

Comments are closed.