H. G. Muller wrote on Wed, Nov 1, 2017 09:35 PM UTC:
A normal Chess engine should play this game disastrously poorly. It would be very easy to trick it. E.g. it will suppose its pieces are sufficiently protected when they are once protected, and attacked only by a single more valuable piece. E.g. wait until he moves up a Knight, protected by a Pawn. Than attack that Knight with your Queen. The engine will ignore it. It will also think QxN will not be your best move, as it sacs a Queen. So it will let you play that. Then it will try to recapture PxQ, and you refuse that. Bye bye, Knight. Repeat for the other Knight. And then Bishops...
It should indeed be very simple to modify an existing engine to play it. Just make it keep track of the score of the best two moves in each position, rather than only the best, and take the score of the second move as the score for the position. This will need a pair of moves to refute a line, rather than one, so it will increase the branching ratio by a factor sqrt(2), so you will lose some depth. To optimally use a transposition table it will have to hold the pair of refuting moves, rather than just the best. This shpuld be pretty trivial.
A normal Chess engine should play this game disastrously poorly. It would be very easy to trick it. E.g. it will suppose its pieces are sufficiently protected when they are once protected, and attacked only by a single more valuable piece. E.g. wait until he moves up a Knight, protected by a Pawn. Than attack that Knight with your Queen. The engine will ignore it. It will also think QxN will not be your best move, as it sacs a Queen. So it will let you play that. Then it will try to recapture PxQ, and you refuse that. Bye bye, Knight. Repeat for the other Knight. And then Bishops...
It should indeed be very simple to modify an existing engine to play it. Just make it keep track of the score of the best two moves in each position, rather than only the best, and take the score of the second move as the score for the position. This will need a pair of moves to refute a line, rather than one, so it will increase the branching ratio by a factor sqrt(2), so you will lose some depth. To optimally use a transposition table it will have to hold the pair of refuting moves, rather than just the best. This shpuld be pretty trivial.