The chess game Hubert Dreyfus vs. Mac Hack Ⅵ () annotated


A few days ago American philosopher Hubert Dreyfus passed away. He was widely known as a critic of artificial intelligence (AI). In his famous book What Computers Can't Do1 Dreyfus provided a short account of early computer chess projects which all played very weak. He mentioned a loss of such a program against a 10-year-old novice in (p. ⅩⅩⅩⅠ), and still in there was no program available that could beat an amateur player (p. 223, note 45). Two years later he himself was willing to play a computer program by Richard Greenblatt and lost. It is easy to find the moves on the Internet, but I did not manage to find them annotated. So I did it myself. In his book Dreyfus called himself a rank amateur (p. ⅩⅩⅩⅠⅠ), but in my view as a chess coach he was simply a beginner.

Enable JavaScript to execute a script here!
[Event "The Dreyfus Match"] [Site "MIT"] [Date "1967.??.??"] [Round "?"] [White "Dreyfus, Hubert"] [Black "Mac Hack VI"] [Result "0-1"] [ECO "C55"] [Annotator "Ingram Braun"] [PlyCount "74"] [EventDate "1967.??.??"] 1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Nc3 Bc5 5. d3 O-O 6. Ng5 $2 {This attacking move is simply pointless. There is no threat against h7 or f7. And White is not well prepared for an attack with f4. White should proceed with developing his pieces instead. "Do not move a piece twice in the opening" is the general rule.} (6. O-O) (6. Bg5) 6... Na5 7. Bd5 $2 {Generally advancing the queen's pawn to d5 is a good plan for Black in so called open games (1.e4 e5). Here White helps Black to realize it by provoking c6. The exchange on b3 is going to happen anyway.} (7. Nxf7 Rxf7 8. Bxf7+ Kxf7 {is favourable for Black. The material is about equal, but White lacks attacking ideas against the weakened king's position since he gave up his best attackers Bc4 and Ng5. On the other hand, Black easily gets his additional minor pieces into play.}) 7... c6 8. Bb3 Nxb3 9. cxb3 $2 {This looks pretty much like a beginner's mistake. There is no need to open the c-file since nothing is going to happen there.} (9. axb3 { The general rule is that pawns should capture toward the center.}) 9... h6 10. Nh3 $2 {Was he still dreaming of an attack starting with f4?} (10. Nf3 { would limit the damage, although Black is clearly better here.}) 10... d5 { [%cal Gc8h3]} 11. exd5 Bg4 12. f3 Bxh3 13. gxh3 {White's pawn structure is totally ruined.} Nxd5 14. Nxd5 Qxd5 {Material is still equal but White is completely lost here. The black king is safe while there is no shelter for the white one. Black's pieces are actively placed (B, Q) and he can comfortably finish his development by moving the rooks to the central files. White has not developed a single piece yet (after move fourteen!) and it is hard to imagine where the bishop should go? White has four pawn islands, Black only two. And since most white pawns are on light squares, the dark squares a pretty weak. The dark-squared black bishop is a monster.} 15. Bd2 (15. Qe2 Rfd8 16. Be3 Bxe3 17. Qxe3 Qxd3 18. Qxd3 (18. Qxe5 $4 Re8 {[%cal Ge8e1]}) 18... Rxd3 19. O-O Rd2 20. Rf2 Rad8 21. Raf1 Kf8 {This is the relatively best White can achieve but it is still a very bad endgame.}) 15... Qxd3 16. b4 Be7 {[%cal Ge7h4,Gh4e1]} 17. Rg1 (17. Rc1 Bh4#) 17... e4 $2 {From a today's point of view it is embarassing that a computer does not win the exchange in a position where it is clearly the best option.} (17... Bh4+ 18. Rg3 {Only move!}) 18. fxe4 $2 (18. Qe2 Bh4+ ({or} 18... Rad8) 19. Kf1 {Black still has a winning position but White is not as bad as in the 17...Bh4+ line.}) 18... Bh4+ 19. Rg3 Bxg3+ 20. hxg3 Qxg3+ 21. Ke2 Qxh3 ({Objectively better is} 21... Qg2+ 22. Kd3 (22. Ke3 $143 {simply looses the Bishop.} Rad8 {[%cal Gd8d2,Gg2d2]} 23. Qe2 (23. Qf1 Qxd2+) (23. Rc1 f5) 23... Qg5+ 24. Kf2 Rxd2) 22... Rad8+ 23. Kc3 Qxe4) 22. Qg1 h5 $2 {This move is hard to understand. Maybe there is a bonus for advancing passed pawns. But here it weakens the king's position. The obvious plan is to get the rooks into play.} (22... Rfe8) (22... Rae8 {[%cal Gf7f5]}) 23. Bc3 { [%csl Gg7]} g6 (23... Qg4+ {Most humans would happily get the queens off the board here. It is not the objectively best in this position but Black "sacrifices" a small amount of his edge in order to eliminate White's counterplay completely. The resulting endgame is easily winning.}) 24. Qf2 (24. Qd4 {[%csl Gg7] threatening mate immidiately.} f6 25. Rg1 Kh7 {[%cal Gh3d7]} ({ Even} 25... Rad8 {is possible:} 26. Rxg6+ $2 Kh7 27. Qg1 Rd3 $1 28. Rg7+ Kh8 29. Qg6 {[%csl Gh7]} Qf3+ 30. Ke1 Rd1#)) 24... h4 $4 {Unbelievable! Black ignores the threat of mate. Of couse, it is a kind of horizon effect. As far as I understand Greenblatts account of his program it had a quiescence search implemented (called 'feedover') but did not do it on simple checks. Then the checks could have simply pushed the mate move out of the search tree.} 25. Qf6 {In contrast to 24. Qh4 Black's f-pawn is blocked and can not inhibit the mate on g7. Objectively the game is drawn now because Black has nothing better than a perpetual check.} Qg4+ 26. Kd2 $2 {He hopes to hide his king on the queen's side and then mating if there is no check left. But this should not have worked.} (26. Kf1 {He has to stay on the king's side.} Qh3+ 27. Ke2 Qg2+ 28. Ke3 $11) 26... Rad8+ $2 {The wrong rook!} (26... Rfd8+ {[%csl Gf8] This vacates the f8 square for the king.} 27. Ke3 (27. Kc2 Qxe4+ 28. Kb3 (28. Kc1 Qe3+ 29. Kc2 Qe2+ 30. Kb3 (30. Kc1 Rd1#) 30... Qe6+) 28... Qe6+) 27... Qg3+ 28. Ke2 Qd3+ 29. Kf2 Kf8 $1 30. Qh8+ Ke7 31. Qxh4+ (31. Qf6+) ({Winning back the exchange does not work out:} 31. Bf6+ Ke6 32. Bxd8 Qxd8 {[%csl Gh4] Protects h4 and white can not avoid exchange of queens.} 33. Qc3 (33. Qh6 Qf6+ 34. Kg1 Rd8 {White should trade queens in order to protect his king but this ends up in a completely lost rook ending.}) 33... Qf6+ 34. Kg2 Qxc3 35. bxc3 {The rook endgame with two extra pawns is easily winning for Black.}) 31... Kd7 {Black is still winning but the game goes on.} {Fi.} 32. Qf4 Re8 33. Qxf7+ Re7 34. Qf3 Qxe4) 27. Kc2 $2 {This is the final blunder since White can not avoid exchange of queens now.} (27. Ke3 {and draw by perpetual check as seen above. Black has nothing better than infinitely checking with the queen.} Qg3+ 28. Ke2 Qd3+ 29. Kf2 Qc2+ 30. Kg1 Rd1+ 31. Rxd1 Qxd1+ 32. Kf2 Qc2+ 33. Kf3 Qd3+ 34. Kf2 Qc2+ 35. Kf3 Qd3+) 27... Qxe4+ 28. Kb3 (28. Kc1 Qe3+ 29. Kc2 Qe2+ 30. Kb3 (30. Kc1 Rd1#) 30... Qe6+) 28... Qe6+ 29. Qxe6 fxe6 30. Rh1 Rf4 31. Be1 Rf3+ 32. Ka4 h3 33. b5 Rd4+ 34. b4 $2 cxb5+ 35. Kxb5 Ra3 $2 (35... Rd5+ 36. Ka4 b5+ 37. Ka5 Ra3# { is the correct move order.}) 36. Kc5 $2 (36. Bf2 Rd5+ 37. Bc5 b6 38. Kc4 bxc5) 36... Rd5+ 37. Kc4 b5# 0-1


1: What Computers Can't Do: a Critique of Artificial Reason. New York: Harper & Row . ISBN 06-011082-1, 259 pp.

words, ≈ characters