Game Development Reference
Delayed gratification by the fox shatters the hounds' plan.
There is indeed. The fox can pin three hounds if it stays close to them, as shown
in move 34 (see Figure 6.9). The hounds will not open a hole with the fox right
there to exploit it. The fox AI will not move this way, but the game allows the user
to make moves. Instead of moving down toward freedom, the player moves the
fox back into the pocket to hold the three hounds fixed. The left-most hound is
too far to the left to help, proving that move 32 was an error. The hounds can
stall, as shown in move 36, but as long as the fox keeps the pressure on the pocket,
that only leads to move 38. With no free moves left, the hounds have to break
open the pocket holding the fox. Our fox does not see this because our fox does
not look ahead when it has a path to freedom.
The hounds' delayed gratification strategy is flashy but flawed. When writing
game AI, avoiding AS—that is, artificial stupidity—is a higher priority than
increased brilliance of play. There is also a lesson here about making plans that
survive contact with the enemy. The hounds' AI needs to evolve.
The hounds look ahead to reform their line. Rather than change the evaluation
function, the hounds' method for comparing different boards needs to change.
The best move for the hounds, when the line is broken, is the earliest move that
reforms the line. This equates to ''slamming the door now to make the win
inevitable.'' The hounds stop ''over thinking'' the situation and quit toying with
the fox. If the line is intact, they take the board with the highest rank, which means
the board with the fewest black squares. Things look promising for the hounds, but
they still could lose. The subtle difference between the simple-to-compute ''fewer