A hypothetical question about draws

[Dann Corbit]

Suppose that I made an engine which had as its goal achieving a draw instead of achieving a win. Would this engine play better or worse against a stronger engine (one that analyzes more deeply, has a better eval, etc.)?

[muxecoid]

Large positive contempt value is proven to be disadvantageous when playing against equally strong or stronger opponent. By extrapolation I guess negative contempt can help when all you need is draw.

[Dann Corbit]

Large positive contempt value is proven to be disadvantageous when playing against equally strong or stronger opponent. By extrapolation I guess negative contempt can help when all you need is draw.

I imagine too, a change in evaluation. The engine will search for repeated positions and things of that nature more aggressively if that is programmed as a goal.

It is an interesting question to me, for some peculiar reason.

[JVMerlino]

Large positive contempt value is proven to be disadvantageous when playing against equally strong or stronger opponent. By extrapolation I guess negative contempt can help when all you need is draw.

I imagine too, a change in evaluation. The engine will search for repeated positions and things of that nature more aggressively if that is programmed as a goal.

It is an interesting question to me, for some peculiar reason.

I would assume that the engine would still consider a win as a better outcome than a loss (draw = 1, win = 1/2, loss = 0)? Or maybe you had envisioned (draw = 1, win/loss = 0)? I think the results would be very different depending on this decision.

The former would certainly be easier to crudely implement in any engine, I would think.

jm

[bob]

Suppose that I made an engine which had as its goal achieving a draw instead of achieving a win. Would this engine play better or worse against a stronger engine (one that analyzes more deeply, has a better eval, etc.)?

I think it would play better. how much better is a question, but you could do things like encouraging trading pawns, locking pawns, etc...

In a chess engine, you want to avoid draws against weaker players and encourage them against stronger players. Some (Crafty, for example) do this automatically based on the ratings. But there is more to it. Locked pawns are good if your position is weaker than your opponent's. By the same token, I would not go for a draw against a stronger player if my position justifies confidence. Those things are _difficult_ to do. To know when to go for a draw vs avoid a draw is not exactly black-and-white science.

[bob]

Large positive contempt value is proven to be disadvantageous when playing against equally strong or stronger opponent. By extrapolation I guess negative contempt can help when all you need is draw.

That's not the same thing. One would change the eval to do things that are more drawish. For example, trade pawns whenever possible, lock pawns up so that pawn levers and potential passed pawn problems don't happen, dig in and defend your kingside. Etc...

[wgarvin]

Some humans are adept at anti-computer play. Don't they do these things? Lock up the position, etc.

I guess the major difference is that anti-computer humans are playing for a win, but they maneuver the engine into a position that it can't make any sense out of, and they wait for the engine to wreck its position, and then they counter-attack. And they are relying on their experience and intuition to see deeply ahead to make sure this counter-attack will be possible (I guess deeper than the computer can search).

So the question I started wondering about after reading Dann's question, is "Could we make an engine that uses anti-computer techniques to defeat other engines?"

I don't see how it could work, because the engine doesn't have human intuition to guide it, so it would probably get just as confused as its computer opponent.

[Ferdy]

Suppose that I made an engine which had as its goal achieving a draw instead of achieving a win. Would this engine play better or worse against a stronger engine (one that analyzes more deeply, has a better eval, etc.)?

I think it really depends on your skill to achieve the goal , assuming also that the development of stronger engine will be stopped. I believe there are still a lot of techniques that are yet to be discovered to improve engine strength. There are also techniques that we have not known yet, but were already known by others.

[Mincho Georgiev]

Let's not forget that is about what engine thinks it's better, not what actually is better. In that context, draw could mean a pure loss in a particular game.

[Kempelen]

At first think, an engine that play for a draw must have a special kind of evaluation that gives priority drawish positional features. But I think this is incorrect, I think it is easier to play for a win, and when you have it at your hand, then force a draw (gives pieces, make no progress, or things like that). So the strongest engines would be the betters to play for a draw.