How Chess Engines Should Try To Win

Anyone care to share an opinion on how a stronger chess engine should try to beat a weaker one?

My opinion is that they should try to make the position as complicated as possible:

* lots of pieces attacking each other

* lots of threats

* lots of choices of move

This will throttle off the weaker engine's ability to see ahead - the absolute life-blood of chess computers - and maximise the chances of getting them to make a mistake.

Other than that, maybe make a database of positions that chess computers play badly and train your net to like this kind of position!

Hi,
Assuming a stronger program, I am afraid steering it towards getting the game played differently (more complicated or any bearing) would weaker it. So you would need a large margin to afford such a strategy, and thus you wouldn't need such a strategy    ... except maybe for tuning the style of play. But you might get something like a procrastinator   ... at some point, the strategy for complication should move to simply achieve the win.

On another hand, I think such a strategy is great for anti-human play (I have in mind CT800 is somehow tuned in such a way).

If human A is a lot better than human B, then A should be able to beat B without having to find B's least favourite positions. However, in computer play, something different is happening: they're getting so good that they're able to get draws even against stronger players. It seems likely to me that computers will be able to become almost completely unbeatable. Until this happens, though, they are beatable, and in order to beat them, one must find types of positions where they might make a big enough mistake to be beaten.

Once the strong computer is 2-3 points ahead, it can revert to best play to find a reasonably quick win.

TheSelfImprover wrote:

Anyone care to share an opinion on how a stronger chess engine should try to beat a weaker one?

My opinion is that they should try to make the position as complicated as possible:

* lots of pieces attacking each other

* lots of threats

* lots of choices of move

This will throttle off the weaker engine's ability to see ahead - the absolute life-blood of chess computers - and maximise the chances of getting them to make a mistake.

Other than that, maybe make a database of positions that chess computers play badly and train your net to like this kind of position!

To that I would add a preference for avoiding trades and keeping many pieces on the board.
My assumption here is that weaker engine is lower rated by at least one class.

Creating smart winning chances should result in more wins and fewer draws, without incurring significantly more losses, and thus resulting in an overall better performance against such weaker engines.