Weaknesses of Chess Computers

A chess computer program's weakness is that it doesn't understand the game. The fact that so many programs have been so successful in spite of this is as good an illustration as any of the inherently tactical nature of chess.

In order to take advantage of the computer's weakness it is necessary to play in such a way as to strategical issues in the forefront, and keep the danger of tactical accidents low. Blocked pawn structures are highly desirable for this purpose, and English and particularly King's Indian formations are usually effective.

Another aspect of the chess computer's weakness it that its tactical strength is mostly of a short-term nature. The branchings of the analytical tree in chess spread out so quickly that before too long even an incredibly powerful computer can't see so far; if the mainline of the tactics is obvious, it is possible that a strong human player will see to the conclusion more quickly. This is particularly dangerous for the computer when its king is the object of the tactical operation.

Thus, blocked positions where one can expand and attack in the direction of the computer's king are ideal. Many humans have figured out how to set up these "stock" wins; such games are generally amusing but not very interesting. I have a large supply of such "stock" wins - like the bishop sacrifice on h7 that is too deep for the machine to calculate - but the games just aren't all that interesting. After a while I prefer just to "play chess" against the computers and learn from the experience. It has toughened me tactically.

One key way for programmers to avoid this sort of treatment is to program their opening books very carefully to avoid these kinds of positions. Here are a few elementary examples where I managed to get at the worst features of some reasonably strong computer programs:










Position after:

(1) scriabin - jos (2240 FICS) [A36]
FICS game 2 12, Jun 1, 1998

1.g3 A nice anti-computer move; maybe we can move-order trick it somehow. 1...g6 2.Bg2 Bg7 3.c4 Nf6 4.Nc3 0-0 5.e4 A beautiful anti-computer setup. It is hard to open the board, and White has a simple kingside expansion that can lead to an attack if Black doesn't understand some nuances. 5...c5 6.Nge2 Nc6 7.0-0 d6 8.h3 Not the most flexible move, but it'll get it out of book. 8...Bd7?! 9.d3 e5? It is now well-known that White stands better because his f-pawn is raring to go. These days, most of the best computers are opening-booked in such a way that one can't get such positions anymore without some luck or extreme ingenuity. 10.f4 Qe7?? It would have been better to pass. 11.g4 Threatening g5 and Nd5. 11...h6 12.f5 gxf5?? Another positional howler. Even in a blitz game such a move should never even be considered. 13.exf5 Winning effortlessly. 13...Rad8 14.Ng3 b6 There's nothing to do. 15.Be3 Kh7 16.h4 Kg8 17.g5 hxg5 18.hxg5 Nh7 19.Nd5 Maybe this part of the game will give the reader some sort of sadomasochistic thrill (the more so if he has struggled against his own computer). 19...Qe8 20.f6 Bxf6 21.gxf6 Rb8 22.Be4 Ng5 23.Bxg5 Qe6 24.Bf5 Qe8 25.Bh7+ Kh8 26.Qh5 Bf5 27.Bxf5+ Kg8 28.Qh7# 1-0










Position after:

(2) ASGC (GNU Chess) - scriabin [E87]
FICS game 5 12, Jan 26, 1998

1.e4 d6 2.d4 Nf6 3.f3 g6 4.c4 Bg7 5.Nc3 0-0 6.Be3 e5 7.d5 c5?! Really, this is a horrible anti-Sämisch line, because it gives White exactly what he wants: a space advantage for no compensation. However, what is more important here is that I get what I want playing against a computer: a blocked position. 8.g4 Na6 9.Qb3? It must have gone out of book here. I reiterate: computers are pathetic in the KID until the board opens up. 9...Nc7 10.0-0-0?! a6 11.h4 b5 12.cxb5?! axb5 13.Bxb5 Ba6 14.Bxa6 Rxa6 Playing the Benko Gambit on the queenside with the White king as a target. What more could Black hope for? 15.Nge2 Qa8 16.Qc2 Rb8 17.g5? I guess, additional play on the kingside! 17...Nh5 Heading for f4, which might open up the monster on g7. 18.Kb1 Nb5 Eyeing d4 and a3. 19.Qd3 c4! 20.Qd2 Nf4 All the attacking mechanisms are in full operation. 21.Nxb5 Rxb5 22.Nc3 Rb3 What fun this is. 23.h5 Nd3 Whee! 24.Na4 Rxa4 25.hxg6 hxg6 26.axb3 cxb3 27.Rh8+ Bxh8 28.Bd4 Ra1# 0-1

In a humorous New In Chess article, a Dutch A-player by the name of Engelkes presented a diabolical anti-computer recipe: a "super-hedgehog". For White, this setup is pawns on a3,b3,d3,e3,g3,h3, knights on d2 and e2, and bishops on b2 and g2. Engelkes just sets this up and shuffles around on the back-rank. The computer is forced to make a "plan," which it is generally unable to do. Sooner or later the computer feels ambitious and pushes a center pawn, gaining space. Then you push the other pawn, meaning that you have blocked the position and are also pretty well set-up to expand on that wing. At the time of the article's publication, this might well have been very effective, but nowadays the top programs are too strong for such a primitive strategy. However, once I gave this a try against a moderately strong computer to see what would happen; I was not unsuccessful but it was really boring. Once was enough. The reader is well advised to use the "fast-forward" button for this game:










Position after:

(3) ThingTwo (Crafty: 2240 FICS) - scriabin [B06]
FICS game 2 12, Jun 1, 1998

1.e4 g6 2.d4 Bg7 3.Nc3 d6 4.f4 e6 5.Nf3 Ne7 6.Be2 Nd7 7.0-0 b6 8.Be3 Bb7 9.Bf2 0-0 10.b3 a6 Setup complete. Now, let's see what it tries to do. 11.Bh4 Qe8 12.Re1 Kh8 13.Qd2 Kg8 14.Kh1 Rc8 15.b4 Kh8 16.Bc4 Kg8 17.Re3 Kh8 18.a4 Ra8 19.Rd1 Ra7 20.h3 Ra8 21.Kg1 Ra7 22.Qc1 Ra8 23.Qb2 Ra7 24.Qb3 Ra8 25.Qb1 Rb8 26.Qb2 Ra8 27.Kh1 Rb8 28.Qc1 Ra8 29.Qa1 Rb8 30.Qb2 Ra8 31.Kh2 Rb8 32.g4 Ra8 33.Rf1 Rb8 34.Rg1 Ra8 35.Rd1 Rb8 36.Kg2 Rg8 37.Qb1 Qf8 38.Qa2 Re8 39.Qb3 Rb8 40.Qb1 Re8 41.Kh2 Rb8 42.Qb3 Re8 (yawn) 43.Qb2 Rb8 44.Qa2 Re8 45.Rd2 Rb8 46.Qb1 Re8 47.Qf1 Ra8 48.Rd1 Ra7 49.Qg1 Ra8 50.Rb1 Re8 51.Qf1 Ra8 52.Qd1 Re8 53.Qd3 Ra8 54.Qd2 Re8 55.Rbe1 Rb8 56.Qd3 Ra8 57.Qd1 Re8 58.R1e2 Rb8 59.Rd2 Re8 60.Qe2 Ra8 61.Qf1 Ra7 62.Qb1 Ra8 63.Qb3 Rb8 64.Kg2 Ra8 65.Re1 Rb8 66.Rdd1 Ra8 67.Rc1 Rb8 68.Rb1 Ra8 69.Kh2 Rb8 70.Rbd1 Ra8 71.Qa2 Rb8 72.Qa3 Ra8 73.Re3 Rb8 74.Kg2 Ra8 75.Rd2 Rb8 76.Qa2 Ra8 77.Qb2 Rb8 78.Qb1 Ra8 79.Re1 Rb8 80.Qb2 Ra8 81.Kh2 Ra7 82.a5 Now, events develop at a "dizzyingly rapid pace" ... 82...b5 83.Ba2 Ra8 84.Re3 Rd8 85.Rdd3? [85.Nd1] 85...c5! 86.Re2 c4? Too "consistent." I could actually have gotten the upper hand here! [86...cxb4! 87.Bxe7 (87.Qxb4? Nc6 88.Qb1 Rc8) 87...bxc3 88.Bxf8 cxb2 89.Bxg7+ Kxg7 90.Rb3 Nf6 91.Nd2 Rc8 92.Rxb2 Rc3] 87.Rd1 Rc8 88.Rf1 Rc7 89.Qa3 Nf6? 90.Bxf6! Bxf6 91.e5 dxe5 92.fxe5 Bxf3 93.Rxf3 93.exf6 Bxe2 94.fxe7 Qxe7 95.Nxe2 looks good. Bg7 94.Ne4 Nd5 95.c3 Phew! All that activity is going to calm down now. However, Black is horribly crippled by the weakie on f7. 95...Qa8 96.Ng5 Qe8 97.Ref2 Rf8 98.Ne4 Kg8 99.Bb1 Qd8 100.Nc5 Ra7 101.Be4 Bh6 102.Qb2 Re8 103.Bxd5 Qxd5 104.Qb1 Bf8 105.Ne4 Bg7 106.Nf6+? 106.g5! was clearly indicated, planning Rf4, Qf1, Nd6 and Nf6 ideas and I can't believe Black would survive. 106...Bxf6 107.Rxf6 Ree7 108.Qc1 Kg7 109.Qg5 Rad7 110.Qh4 Qa8 111.R6f4 Qf8 112.Rf1 Kg8 113.Qg5 Qg7 114.R1f3 h6 115.Qf6 Re8 116.Kg2 Qxf6 117.Rxf6 Ree7 118.Kg3 Rd5 119.Kf4 Kg7 120.Ke4 Red7 121.h4 Re7 122.R3f4 Red7 123.Rf1 Re7 124.h5 g5 And the remainder is typically thrilling: 125.Re1 Red7 126.Rg1 Re7 127.Rg2 Red7 128.Rh2 Re7 129.Re2 Red7 130.Re1 Re7 131.Re3 Red7 132.Re2 Re7 133.Rd2 Red7 134.Rg2 Re7 135.Rg1 Red7 136.Rg3 Re7 137.Rf2 Red7 138.Rf1 Re7 139.Rf6 Red7 140.Rff3 Re7 141.Rh3 Red7 142.Rf6 Re7 143.Rhf3 Red7 144.Re3 Re7 145.Rh3 Red7 146.Rh1 Re7 147.Ke3 Red7 148.Rc1 Re7 149.Rcf1 Red7 150.R1f3 Re7 151.Rf2 Red7 152.Rh2 Re7 153.Ke4 Red7 154.Ra2 Re7 155.Rf3 Red7 156.Rf1 Re7 157.Rf6 Red7 158.Rb2 Re7 159.Rf1 Red7 160.Rbf2 Re7 161.Rd2 Red7 162.Rf3 Re7 163.Rdf2 Red7 164.Rf6 Re7 165.Rf1 Red7 166.R6f3 Re7 167.Rd1 Red7 168.Rh1 Re7 169.Ra1 Red7 170.Rf2 Re7 171.Re1 Red7 172.Rg1 Re7 173.Rh1 Red7 174.Re1 Re7 175.Rf4!! Ending with a little joke. 1/2-1/2

The other weakness of the computer I would like to highlight is its sometimes naively materialistic nature. Computers will grab material that humans would be scared to touch for fear of the positional concessions. (Sometimes the humans don't even realize all their tactical possibilities for this reason!) Also, once the computer is ahead a quantity of material, it likes to cling to it, and doesn't always take advantage of opportunities to remedy its positional defects (such as king positions!) by returning a little extra material it doesn't really need. Unlike the "blocked game" strategy, it is impossible to "plan" to exploit the computer's greed; it just has to happen. Both of the following games are interesting in this respect:










Position after:

(4) JPCrafty (2355 FICS) - scriabin [D79]
FICS game 2 12, Dec 5, 1997

1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.g3 0-0 5.Bg2 c6 6.Nf3 d5 7.cxd5 cxd5 8.Ne5 e6 9.0-0 Nfd7 10.f4 f6 11.Nxd7?! Bxd7 12.Nxd5?? Sometimes the computer sees too much for its own good; I hadn't noticed this shot, but if I had, it wouldn't have taken me long to reject it. 12...exd5 13.Bxd5+ Kh8 14.Bxb7 Bc6 15.Bxa8 Bxa8 Theoretically Black is "down material" but looking at the board, it is just obvious that he must be winning. 16.b3 Qd5 17.Rf3 Nc6 18.Ba3 Re8 19.Bc5 f5 White has defended magnificently - as one would expect - but sometimes that is not sufficient. 20.e3 [20.Rc1 Nxd4 21.Bxd4 Qxd4+ 22.Qxd4 Bxd4+] 20...Nxd4 21.Bxd4 Qxf3 22.Qxf3 Bxf3 23.Rf1 Be4 24.Rc1 Bxd4 25.exd4 Rd8 26.Rc4 The ending still presents technical problems: Black wants to offer a rook swap on the c-file; since White must avoid this he will perish on the kingside where Black's bishop is commanding. 26...Rd7 27.Kf2 Kg7 28.Ke3 Kf7 29.Rc5 Ke7 30.Re5+ Kd6 31.Ra5 Rc7 32.Ra6+ Rc6!? [32...Kd7 avoiding pawn exchanges may be simpler.] 33.Rxa7 Rc3+ 34.Kf2 Rc2+ 35.Kg1 Rg2+ 36.Kf1 Rxh2 37.b4 Rb2 38.Rxh7 Rxb4 39.Rg7 Bf3! Black has to be very careful to keep a pawn. 40.Rxg6+ Kd5 41.Kf2 Bg4 42.Ke3 Ra4 43.Rb6 Rxa2 44.Rb5+ Kc6 45.Rc5+ Kd6 46.Kd3 Rg2 47.Kc4 Rc2+ 48.Kb4 Rg2! Unfortunately I can't take the rook. [48...Rxc5 49.dxc5+ Kc6 50.Kc4 Bd1 51.Kd4=] 49.Kc4 Rxg3 50.Rd5+ Ke6 51.Ra5 Be2+ 52.Kc5 Rg1 53.d5+ Kf6 54.Ra4 Rc1+ 55.Kd4 Rd1+ 56.Kc5 Rc1+ 57.Kd4 Bf3 Hammering the nail in the d-pawn's coffin. 58.Ra6+ Ke7 59.Ra7+ Kd6 60.Ra6+ Kd7 61.Ra7+ Rc7 62.Rxc7+ Kxc7 63.Ke5 Be4 64.Ke6 Kd8 65.Kd6 Bc2 66.Ke6 Ke8 67.Kd6 Kf7 68.Kc7 Ba4 69.Kd6 Kf6 70.Kc5 Kg6 71.Kd4 Kh5 72.Ke5 Kg4 73.d6 Bd7 74.Kd4 Kxf4 0-1










Position after:

(5) JagorCrafty (2036 FICS) - scriabin [D45]
FICS blitz game, Dec 12, 1998

1.d4 Nf6 2.c4 c6 3.Nc3 d5 4.e3 e6 5.Nf3 Nbd7 6.Qc2 Bd6 7.Be2 0-0 8.0-0 e5 9.cxd5 cxd5 10.dxe5 Nxe5 11.Rd1 This is not a good position to get against a computer. Usually, the d-pawn will just drop off somehow. This part of the game has some theoretical interest. 11...Bg4!? A rare move, speculating on the computer's greed. 12.Nxd5?! Probably most humans would have shied away from this risky grab. 12...Nxf3+ [12...Nxd5 13.Nxe5! Bxe2 14.Rxd5] 13.gxf3 Nxd5 14.fxg4 [14.Rxd5 Be6! is dubious for White.] 14...Qh4 15.f4 Rac8 16.Qd3 Rfd8 17.Qxd5 ..Nxf4 was menaced and White didn't have any appealing alternatives. The following is all forced. 17...Bxf4 18.Qxd8+ [18.Qh5 Rxd1+ 19.Bxd1 Qe1+ 20.Kg2 Qxd1 21.exf4 Rc2+ 22.Kg3 Qg1+-+; 18.Qxb7 Rxd1+ 19.Bxd1 Qe1+ 20.Kg2 Rc2+=] 18...Rxd8 19.Rxd8+ Qxd8 20.exf4 Qd4+ 21.Kg2 b5! Essential to cramp the queenside. What a weird situation. Black is, of course, way behind in material, but his queen is dominant, White's king is exposed, and the machine can't really develop the other side without material losses. White must try a2-a3 or a2-a4 to reduce the extent of the cramp, but that is hard for a machine to understand. Its evaluation function is dominated by the material considerations and thus it can't figure out what to do. My guess is that this position properly played should be drawn. 22.Kg3 [22.a4 b4 23.a5!?] 22...Qg1+ I suggest 23.Kf3!?, planning b2-b3 etc., and meeting ..Qxh2 with 24.Be3. Fritz finds this move, but only because it calculates a forced perpetual after 23.Kf3 Qf2, etc. 23.Kh3?! b4! 24.Bf3 Qf1+ 25.Kg3 Qe1+ 26.Kg2 a5 27.h3 a4 28.Kh2?! This was the last chance for f4-f5. 28...Qf2+ 29.Bg2 f5! 30.gxf5? It's very bad to allow Black's h-pawn will advance. 30...h5 31.b3? This little sequence is a rather amusing game of "horizons." 31.a3! probably would hold, but the reasons were beyond the computer's horizon. A human's positional intuition would conclude that it's too risky to expose the rook like this. 31...a3 32.f6 gxf6 33.f5?! [33.Kh1 h4 34.Kh2 Qg3+ 35.Kh1 Qc3 36.Rb1 f5 and I think Black wins by marching the king ceremonially down the center!] 33...Kg7 Completing total domination. 34.Bxa3 Here's the funny part; the computer was the first to realize why it was lost (in a blitz game the reason was still "beyond my horizon"), so I gratefully collected the piece and won the game! [34.Kh1 Qe1+ 35.Kh2 Qe5+] 34...bxa3 35.Kh1 Qb2 36.Rg1 Qxa2 Now Black wins rather easily. 37.Bd5+ Kf8 38.Be6 Qb2 39.Rg8+ Ke7 40.Rg7+ Kd6 41.b4 a2 42.Bxa2 Qxa2 etc. 0-1

Back to my chess page

Generated with ChessBase 7.0