Something MUST be done to slow computers exchanging Q's.

[S.Taylor]

Humans tend to favor keeping queens on, as much as possible, as it keeps the game alive and who knows, maybe it will get a winning possibility, and the queen will be very helpful for the one who is winning.

But computers see absolutely no reason to keep queens on. On the contrary, if a computer sees a possibility of gaining a hundredth of a pawn in a close by evaluation, after exchanging, it will not hesititate to exchange. It will do this even if this will throw away all the potential in the position.

Programmers! don't you realize this is happening?

[cdani]

Humans tend to favor keeping queens on, as much as possible, as it keeps the game alive and who knows, maybe it will get a winning possibility, and the queen will be very helpful for the one who is winning.

But computers see absolutely no reason to keep queens on. On the contrary, if a computer sees a possibility of gaining a hundredth of a pawn in a close by evaluation, after exchanging, it will not hesititate to exchange. It will do this even if this will throw away all the potential in the position.

Programmers! don't you realize this is happening?

We realize the elo win or lose

Of course something can be done to avoid this, for example adding some uci parameter, but it will not play optimally any more with this enabled.

Anyway is interesting to see that as the best engines at long time controls tend to play so strong, is possible that the elo difference with such parameter active will tend to be zero, but only at those long time controls, but nobody will be able to test this for the moment.

[Dirt]

Of course something can be done to avoid this, for example adding some uci parameter, but it will not play optimally any more with this enabled.

It should be tied to contempt. In fact, I think Komodo might do this.

[AlvaroBegue]

There is a comment in the evaluation function of our program Ruy-López that says "as long as there is a queen, there is a fight". What we actually do is multiply the score by something slightly less than 1 (which changes with the phase) if there are queens on the board.

So you want to exchange queens if you are ahead (unless it destroys your attacking score, for instance), and you want to keep them on the board if you are behind. At least, that's what we told our program.

[hgm]

Actually this is a weakness of the FIDE game rules, and a consequence of the fact that the game is rather young and not yet fully evolved. The more mature forms of Chess typically develop rules against the early loss of the key piece(s) that make the game interesting and fast.

In Chu Shogi, for instance, the Lion (worth about 1.5 Queen, and very dominant) is kept in the game by the rule that it is not allowed to capture Lions in two successive plies. After Lion x Lion recapturing the Lion immediately wins; after other x Lion counter-striking against the opponent Lion immediately loses. That makes trading Lions pretty hard.

In Maka Dai Dai Shogi the Teaching King and Buddhist Spirit (each moving as Queen + something much stronger than Queen) are protected from disappearance from the game by the rule that every piece that captured one of them immediately promotes to what it captured. So they stay in the game, although it could be for the other side...

That being said: Of course it is not good, Elo wise, to trade a Queen for mere 10 milliPawn advantage. Going on that track would leave you at +0.05 by the time all material has been traded, a sure way to achieve a dead draw. One should always keep in mind that one has to end at about +1.5 by the time everything has gone. As a Queen represents about 20% of everything you have, trading it for less than an advantage of 0.3 makes you hurry towards a draw, rather than a win. This is a real problem in Xiangqi, where the draw margin is about +5. You really have to be careful there that trades gain you enough.