Smyslov_Fan wrote on 02/08/12 at 15:48:39:
Vladimir, the goalposts are, can a human work out that only Bh3 wins while a computer can't?
You say that the computer sees Bh3 as winning and no other move is quite sufficient to win. I'd accept that if it reproducible. The computer can't have had Bh3 already played though.
As I said, the goalposts seem to have moved because everyone was previously wondering if computers could even find the move, like the very quote I prefaced my first post with. And amidst all of this incredulity, did either of you look at the two links provided? Several engines find 47...Bh3 in reasonable time.

Notably, though, Stockfish seems to fail at finding the move.
Here is Houdini 1.5a, with the score still going up and up on the 29th ply before I decided to finally terminate the search. Note that Houdini doesn't really show its periodic score increments on separate lines like Critter does, so the final line's +5.60 is only after several smaller increases were displayed during the search:
Houdini_15a_x64:
26/54 00:49 553,861,944 11,081,000 +1.48 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-d4 Kd6-c6 Kf2-e2 Kc6-b5 Ke2-d2 Kb5-c4 Bd4-b2 d5-d4 Kd2-e2 d4-d3+ Ke2-d2 Kc4-b3 Bb2-a1 a4-a3 Ba1-c3 a3-a2 Bc3-h8 Kb3-c4 Kd2-e3 Be4-f3 Ke3-d2 Bf3-d5 Bh8-c3 Bd5-b7 Bc3-a1 Bb7-e4 Kd2-c1 Be4-a8 Kc1-d2 Ba8-d5 Kd2-e3 Bd5-e4 Ke3-d2 Be4-f3 Kd2-e3 Bf3-h5 Ba1-e5 Bh5-g4 Ke3-d2 Bg4-e2 Be5-c3 Be2-h5 Bc3-b2 Bh5-g4 Bb2-a1 Bg4-e2 Ba1-c3 Be2-h5
27/54 01:02 694,243,666 11,165,000 +1.48 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-d4 Kd6-c6 Kf2-e2 Kc6-b5 Ke2-d2 Kb5-c4 Bd4-b2 d5-d4 Kd2-e2 d4-d3+ Ke2-d2 Kc4-b3 Bb2-a1 a4-a3 Ba1-c3 a3-a2 Bc3-h8 Kb3-c4 Kd2-e3 Be4-f3 Ke3-d2 Bf3-d5 Bh8-c3 Bd5-e4 Bc3-f6 Kc4-b3 Bf6-e5 Kb3-c4 Kd2-c1 Kc4-d5 Be5-a1 Be4-f3 Kc1-d2 Kd5-c4 Kd2-e3 Bf3-h5 Ba1-e5 Bh5-g4 Ke3-d2 Bg4-e2 Be5-c3 Be2-h5 Bc3-b2 Bh5-g4 Bb2-a1 Bg4-e2 Ba1-c3 Be2-h5
28/54 01:22 929,349,860 11,264,000 +1.48 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-d4 Kd6-c6 Kf2-e2 Kc6-b5 Ke2-d2 Kb5-c4 Bd4-b2 d5-d4 Kd2-e2 d4-d3+ Ke2-d2 Kc4-b3 Bb2-a1 a4-a3 Ba1-c3 a3-a2 Bc3-h8 Kb3-c4 Kd2-e3 Be4-f3 Ke3-d2 Bf3-d5 Bh8-c3 Bd5-e4 Bc3-f6 Be4-c6 Bf6-a1 Bc6-b7 Kd2-e3 Bb7-d5 Ba1-h8 Bd5-e4 Ke3-d2 Be4-b7 Bh8-a1 Bb7-e4 Kd2-c1 Be4-a8 Kc1-d2 Ba8-d5 Kd2-e3 Bd5-e4 Ke3-d2 Be4-f3 Kd2-e3 Bf3-h5 Ba1-e5 Bh5-g4 Ke3-d2 Bg4-e2 Be5-c3 Be2-h5 Bc3-b2 Bh5-g4 Bb2-a1 Bg4-e2 Ba1-c3 Be2-h5
29/63 01:57 1,338,366,562 11,349,000 +1.48 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-d4 Kd6-c6 Kf2-e2 Kc6-b5 Ke2-d2 Kb5-c4 Bd4-b2 d5-d4 Kd2-e2 d4-d3+ Ke2-d2 Kc4-b3 Bb2-a1 a4-a3 Ba1-c3 a3-a2 Bc3-h8 Kb3-c4 Kd2-e3 Be4-f3 Ke3-d2 Bf3-d5 Bh8-c3 Bd5-e4 Bc3-f6 Kc4-b4 Kd2-c1 Kb4-b3 Bf6-a1 Kb3-b4 Kc1-d2 Kb4-c4 Kd2-c1 Be4-a8 Kc1-d2 Ba8-d5 Kd2-e3 Bd5-e4 Ke3-d2 Be4-f3 Kd2-e3 Bf3-h5 Ba1-e5 Bh5-g4 Ke3-d2 Bg4-e2 Be5-c3 Be2-h5 Bc3-b2 Bh5-e2 Bb2-a1 Be2-f3 Kd2-e3
29/64+ 03:07 2,134,991,667 11,382,000 +1.56 Bf5-h3
29/78+ 13:15 9,214,503,132 11,583,000 +5.60 Bf5-h3
Here is Critter 1.4 with the session file disabled and deleted just in case. Similar to Houdini, the score kept rising on the 28th ply before I terminated the search.
Critter_1.4_64bit_sse4:
26/49 00:43 446,703,523 10,260,790 +1.95 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-b4+ Kd6-c6 Kf2-e2 Kc6-b5 Bb4-e7 Kb5-c4 Ke2-d2 d5-d4 Kd2-c1 Kc4-b3 Be7-f6 d4-d3 Kc1-d2 a4-a3 Bf6-c3 Kb3-c4 Bc3-e5 a3-a2 Be5-a1 Be4-f3 Ba1-f6 Bf3-a8 Bf6-c3 Ba8-d5 Bc3-a1 Bd5-e4 Ba1-c3 Be4-f3 Bc3-a1 Bf3-a8 Ba1-c3
27/49 01:05 665,685,130 10,161,890 +1.95 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-b4+ Kd6-c6 Kf2-e2 Kc6-b5 Bb4-e7 Kb5-c4 Ke2-d2 d5-d4 Kd2-c1 Kc4-b3 Be7-f6 d4-d3 Kc1-d2 a4-a3 Bf6-c3 Kb3-c4 Bc3-e5 a3-a2 Be5-a1 Be4-f3 Ba1-f6 Bf3-a8 Bf6-c3 Ba8-d5 Bc3-a1 Bd5-e4 Ba1-c3 Be4-f3 Bc3-a1 Bf3-a8 Ba1-c3
28/49 01:41 1,045,130,803 10,351,724 +1.95 Bf5-e4 g2-g3 f6-f5 Kg1-f2 Ke6-d6 Bc3-b4+ Kd6-c6 Kf2-e2 Kc6-b5 Bb4-e7 Kb5-c4 Ke2-d2 d5-d4 Kd2-c1 Kc4-b3 Be7-f6 d4-d3 Kc1-d2 a4-a3 Bf6-c3 Kb3-c4 Bc3-e5 a3-a2 Be5-a1 Be4-f3 Ba1-f6 Bf3-a8 Bf6-c3 Ba8-d5 Bc3-a1 Bd5-e4 Kd2-c1 Be4-f3 Kc1-d2 Bf3-a8 Ba1-c3
28/66+ 03:17 2,124,799,077 10,784,796 +2.07 Bf5-h3 g2xh3 Ke6-f5 Kg1-f2 Kf5-e4 Bc3-b2 f6-f5 Kf2-e2 f5-f4 Bb2-c1 f4-f3+ Ke2-d2 d5-d4 Bc1-a3 d4-d3 Kd2-e1 Ke4-d5 Ba3-b2 Kd5-c4 Bb2-c1 Kc4-c3 h4-h5 g6xh5 h3-h4 Kc3-c2 Bc1-d2 a4-a3
28/71+ 04:34 3,012,321,706 11,002,990 +2.37 Bf5-h3 g2xh3 Ke6-f5 Kg1-f2 Kf5-e4 Bc3-b2 f6-f5 Kf2-e2 f5-f4 Bb2-c1 f4-f3+ Ke2-d2 d5-d4 Bc1-a3 d4-d3 Kd2-e1 Ke4-d5 Ba3-b2 Kd5-c4 Bb2-c1 Kc4-c3 h4-h5 g6xh5 h3-h4 Kc3-c2 Bc1-d2 a4-a3
28/74+ 06:01 4,023,197,764 11,136,910 +2.64 Bf5-h3 g2xh3 Ke6-f5 Kg1-f2 Kf5-e4 Bc3-b2 f6-f5 Kf2-e2 f5-f4 Bb2-c1 f4-f3+ Ke2-d2 d5-d4 Bc1-a3 d4-d3 Kd2-e1 Ke4-d5 h4-h5 g6xh5 h3-h4 Kd5-d4 Ba3-e7 Kd4-c4 Be7-d6 Kc4-b3 Bd6-c5 Kb3-c3 Ke1-d1 Kc3-c4 Bc5-f2 a4-a3 Kd1-d2 a3-a2
28/74+ 08:50 5,995,119,336 11,307,534 +3.05 Bf5-h3 g2xh3 Ke6-f5 Kg1-f2 Kf5-e4 Bc3-b2 f6-f5 Kf2-e2 f5-f4 Bb2-c1 f4-f3+ Ke2-d2 d5-d4 Bc1-a3 d4-d3 Kd2-e1 Ke4-d5 h4-h5 g6xh5 h3-h4 Kd5-d4 Ba3-e7 Kd4-c4 Be7-a3 Kc4-b3 Ba3-c5 Kb3-c3 Ke1-d1 Kc3-c4 Bc5-b6 a4-a3 Kd1-d2 a3-a2
28/74+ 14:32 10,021,279,498 11,493,228 +3.65 Bf5-h3 g2xh3 Ke6-f5 Kg1-f2 Kf5-e4 Bc3-b2 f6-f5 Kf2-e2 f5-f4 Bb2-c1 f4-f3+ Ke2-d2 d5-d4 Bc1-a3 d4-d3 Kd2-e1 Ke4-d5 h4-h5 g6xh5 Ba3-b2 Kd5-e4 Bb2-g7 Ke4-e3 Bg7-h6+ Ke3-d4 Ke1-d2 a4-a3 Bh6-g7+ Kd4-e4 Kd2-e1 a3-a2 Bg7-f6 Ke4-e3 Bf6-g5+ Ke3-d4 Bg5-f6+ Kd4-c4 Bf6-a1 h5-h4 Ba1-f6 Kc4-d5 Bf6-g7 Kd5-e4 Bg7-f6 Ke4-e3 Bf6-g5+ Ke3-d4 Bg5-f6+ Kd4-e4 Bf6-g7 Ke4-e3 Bg7-h6+ Ke3-d4 Bh6-g7+ Kd4-e4
Trw, SMP searches are non-deterministic. That means that exact times are not reproducible, and that separate runs could find the move faster or slower than before, and even on different depths. For example, if I run through them again, they could find the move after 5 minutes, or get lucky and find it in under 3. Or maybe they find the solution faster without the slowdown from tablebases, who knows.