Neon Qwerty writes, by email:
Hey Dennis!
[snip]
I also have a trivial question re: computer chess and faulty engine evaluations. Everyone agrees that computers will often evaluate a side as winning when a human can intuitively tell that it's a dead draw. The question is: what is the most [egregious] evaluation that a computer can give a drawing position? Two pawns up? Three? A rook?
Just wondering and thought that you might know,
Alan.
Alan,
That's a fun question, though one difficulty in answering is that new engines are coming out all the time and this is a problem programmers regularly work on. So the answer is that I don't know, but as an opening bid I offer the following position:
White is up two rooks and a bishop, but although the position is hopelessly drawn, most of my software programs don't get it.
Shredder 9 wins the palm on this one, awarding White only a very modest .35 edge (at depth 37), which perhaps reflects the abstract possibility that White could put a rook en prise and Black mistakenly capture it.
Fritz 9: White has a 4.5 pawn advantage.
Rybka 1.0, 32-bit: White +12.95.
Fruit 2.2.1: White +12.96.
Junior 9: White +15.61
Hiarcs 9: White +16.56!
Not a proud display for computerdom, but Shredder 9 shows that it's possible to write code for this. In any case, readers are invited to top this example, and to find situations Shredder doesn't handle as well.
White: Kh2, Qg7, Rd7, h3, g3, f4
Black: Ke3, Qc5, Rd6, g6
The position isn't exactly drawn, but the computer played 59. Qg7-e7??, probably with a +7 or greater score for winning my rook. The computer was under no time pressure whatsoever, but I was! I played the first move I could find: 59... Kf3, which forced immediate resignation.
This probably doesn't rank as high as your position, but hopefully it is equally inspiring.