Author: Vincent Diepeveen
Date: 11:30:59 04/14/02
Go up one level in this thread
On April 13, 2002 at 02:08:18, Slater Wold wrote: >[D]r2qk2r/pp3ppp/2p1pn2/4n3/1b6/3P2PP/PPPN1PB1/R1BQK2R b KQkq - 0 11 >In this position Deeper Blue found h5 in 121 seconds. After Murray moved, >Kasparov looked at him in disarray. After the match, Kasparov said "Sometimes >the computer plays very human moves." Can you explain an FM why h5 is so good? It looks awfully stupid move to me. Please remember that deep blue is a preprocessor. it doesn't see in the leaves that when it castles short that h5 is bad and suicidal. I agree here with Seirawan in the post mortem where he explains in june 1997 analysis that the move is bad. Kasparov no doubt has said it is a human move. He didn't mean that positive i bet. If you hear audio tapes from 1997 then you hear how he continuesly laughs about the machine. His only big stupid own mistake was game 6 of course. In short there is only 1 short statement from kasparov directly after the game saying it is a very human move. On the other hand Seirawan doesn't like it in analysis where he has very carefully analyzed the games. Note he was also giving online comments. Knowing computers i know that programs play h5 just like that. Junior, Fritz, and also DIEP are programs which play moves like h5 in bad times and in good times. DIEP doesn't consider to play h5 here with black. The real interesting thing is that Deep blue planned h5 already at 6 ply, then it planned castling short. In most mainlines where it shows o-o it also plans h5. Completely suicidal. During the game some weak old commentator mentionned that h5 was meant to play h4 after g4 control f4. However in the deep blue log files we do not see anything about controlling f4 by deep blue. In fact it is happy to allow f4 ;) During the match and short after of course there were no publicly available logfiles. They are now lucky. That shows how something stupid in a program (both o-o combined with h5-h4 is insane) sometimes can be shown as something good. The plan with which deep blue plans h5 is obviously bad for black. That says enough about the reasons to play h5. A completely misevaluated position. No wonder. There are hardly open files on the board... ...it's not a closed position but close to that. I feel h5 is a too unclear move to call it a good move. In fact i dislike h5. Note that because deep blue is a preprocessor, AFTER having played h5 it doesn't consider o-o anymore. Only o-o-o comes into account because of king safety :) Says something how silly a preprocessor is in theory. I do not know where the 121 seconds come from. It was completely doubting between h5 and o-o simply. In fact initially it wanted both moves. Scores are: -3 (h5) , -2 (o-o) , -15 (o-o ply deeper) , -5 (same ply) where a pawn is 128 units. So we talk about less than 0.09 difference for o-o and h5. If i use DIEP's mainlines here you see i also drop about a tenth for o-o shortly from 10 to 11 ply (note deep blue dropped from 9 to 10 ply a tenth). Lucky diep doesn't plan the stupid h5 move here, though i wouldn't blink if it did. 00:08 0 0 801513 (142367) 9 -0.001 Qd8-d7 a2-a3 Bb4-e7 Nd2-b3 O-O O-O Nf6-d5 Bc1 -d2 b7-b5 ++ e8-g8 00:16 0 0 1873488 (235405) 10 -0.140 Qd8-d7 O-O O-O Nd2-b3 b7-b6 Bc1-f4 Ne5-g6 a 2-a3 Bb4-d6 Bf4-d2 ++ e8-g8 00:20 0 0 2304533 (381133) 10 -0.020 O-O O-O a7-a5 Nd2-f3 Ne5xf3 Qd1xf3 Bb4-c5 B c1-e3 Qd8-d6 Qf3-f4 e6-e5 00:45 0 0 5572639 (889305) 11 -0.132 O-O O-O a7-a5 Nd2-f3 Qd8-c7 Bc1-f4 Bb4-d6 N f3-g5 h7-h6 Ng5-e4 ++ d8-e7 01:16 0 0 9742668 (1122545) 11 -0.125 Qd8-e7 O-O O-O-O Nd2-b3 Nf6-d5 Bc1-d2 Qe7- f6 Rf1-e1 Bb4xd2 Nb3xd2 ++ d8-c7 01:42 0 0 13128926 (1698339) 12 -0.163 Qd8-e7 O-O O-O Nd2-b3 a7-a5 a2-a3 Bb4-d6 Bc1-g5 a5-a4 Nb3-d2 b7-b6 Rf1-e1 ++ e8-g8 02:05 0 0 16353345 (2929632) 12 -0.070 O-O O-O a7-a5 Nd2-f3 Ne5xf3 Qd1xf3 Bb4-c5 Bc1-f4 Qd8-b6 Bf4-e5 Bc5-d4 Be5-d6 Rf8-d8 04:15 0 0 34379051 (8346896) 13 -0.165 O-O O-O Bb4-d6 Nd2-b3 Nf6-d5 Bc1-d2 Qd8-e 7 Nb3-a5 b7-b6 Na5-c4 11:04 0 0 87442465 (22628203) 14 -0.157 O-O O-O Bb4-d6 Nd2-b3 Nf6-d5 Bc1-d2 Qd8- e7 Nb3-a5 Rf8-e8 Rf1-e1 Ra8-d8 Qd1-e2 b7-b6 21:12 0 0 164121004 (25399278) 15 -0.178 O-O O-O Bb4-d6 Nd2-b3 Nf6-d5 Bc1-d2 Qd8 -e7 Rf1-e1 Ra8-e8 Ra1-b1 Nd5-b6 Nb3-c5 >The power supply to my new machine has not arrived yet, so I can only test this >move using Fritz 7. It found h5 in 11:47 and played it through the 15th ply and >over 20 minutes. (On an AMD 1200mhz.) This move (h5) has been confirmed by >several masters to be the "best" move. Can anyone get it faster? gotta be NMs rated 2000 uscf or so then. this 2300+ FIDE master, which is about 2600 USCF says it's a bad move. >This is move 11 from game 5 of the 1997 match.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.