An interesting example From Zertz of the kind of thing that makes developing AI's either hell or a hell of a lot
of fun, depending on your point of view and mood. On Boardspace.net, review Zertz games and see the
game "interesting/troiker-strange". The interesting/strange thing about this game is that the robot
seems to be giving away the everything in sight. Surely this must be a bug! But no, it's a really subtle
The fun begin about move 9, where the robot seems to be setting up a standard exchange
to gain two white balls, in exchange for 2 gray balls. Which would be a good deal. However,
the search finds that by sacrificing a black first, the exchange becomes 1 black+1gray for
2 White, which is better.
So, the robot sacrifices the black and looks at the position anew. It sees one ply
deeper, so it sees that the exchange is now 1B + 1G for 2W; but because it already
gave up the black (making a total of 2B+1G) it sees that it could transform the
exchange into 2G for 2W, which now looks better.
The same line of disastrous reasoning continues for a long time. It's a close relative
of the horizon effect, except instead of trying to avoid a disaster by postponing the
inevitable, it is creating a disaster by trying to make a small improvement.
Any topic that does not fit elsewhere
1 post • Page 1 of 1