Should chess grandmasters play with a computer as an aid during a championship game? (Or, should the current world chess championship become the Advanced Chess World Championship?) Hartosh Singh Bal (‘Chessmate’, International Herald Tribune, June 5 2012) offers some arguments for this claim, but fails to consider a possible unintended consequence and leaves an interesting angle unexplored.
First, I agree that allowing chess grandmasters to use the computer engines during a championship match could result in higher-quality games. (I say that as someone who, as a rather incompetent chess player, is not entirely clear about what ‘higher-quality’ chess means, but at the very least, the opportunity for cross-vetting of moves appears likely to result in interestingly new strategies.) However, I wonder if a tendency to defer to a computer’s moves would not manifest itself, slowly marginalizing a human player. (There is ample evidence from decision support system design literature that so-called ‘mixed’ systems–those that combine automation and human inputs–slowly become ‘pure’ as humans trust, and defer to, computer-generated outcomes.) As the power–and corresponding opacity–of the chess engines increases–as is extremely likely–it seems the human role would diminish as well. Extremely opinionated and confident grandmasters would perhaps be more inclined to resist such deference but such resistance, I’d suggest, would be fighting a losing battle in the long run, slowly turning the collaborative game into a game between computer chess engines. In this regard, I wonder if the best role for computer chess engines in the World Chess Championship would be to remain trainers, preparing humans for games against other humans. Of course, this does not preclude the possibility of a Computer Chess Championship.
Second, in the link that Bal provides for the claim that ‘computer engines play chess better than humans’ it is noted that the ‘open-source Houdini’ engine beat the ‘commercial Rybka’ chess engine (this is the wrong contrast by the way – it should be ‘open-source’ versus ‘proprietary'; an open-source engine can also be ‘commercial’). An open-source engine could benefit from one very important source of inputs: competent chess-playing programmers, who would be able to share their implementation strategies with each other to arrive at collaborative solutions for difficult chess situations.
Incidentally, as Bal wraps up his piece he claims:
[Allowing computers to play chess alongside humans] would also teach us important things about the world. Take, for example, a game that’s winding down with this particular configuration: rook and a bishop versus two knights. [link in original] This situation came up in a world championship qualifying game in 2007, and the match concluded in a draw. But computer analysis showed that the game was really a forced win for black in 208 moves. This revealed not just a strategic truth about chess, but also a phenomenological truth, a truth about reality, that would otherwise have remained inaccessible. [emphasis added]
The example Bal points to shows that a particular chess configuration has an outcome not found by human players for various reasons (in this case, the sheer amount of look-ahead required). Why is the existence of this winning strategy a ‘phenomenological truth’? Furthermore, computer chess engines can explore such outcomes, discover such ‘truths’ if you insist, on their own–generate game configurations, explore possible winning strategies–without being involved in human chess championships.