Computer Go

British Go Journal No. 61. March 1984. Page 23.

Comments: Matthew Macfadyen

The world's first Computer Go tournament was held in January at the Covent Garden officesof Acornsoft Ltd. Games were played on 13*13 boards, and the programs all ran on BBc microcomputers (no peripheral memory was allowed). The winner of the £1000 first prize was Bronyslaw Przbyla, from Swindon. Most of the eight programs competing played something looking quite like 25 kyu human play, but his was the best at avoiding worthless moves. The final game is given below, with brief comments by matthew Macfadyen, who was one of the referees at the tournament.

Black: Bronyslaw Przybyla's program
White: Richard Granville's program
The game-file in SGF format.

Figure 1 ( 1-44)

• Black 17 creates a very tense situation - neither program explicitly recognises eyes, so one of the groups on the side is likely to die.
• White 18: White has no idea that his group is in trouble.
• Black 19: Przybyla's program looks for ways in which groups can extend towards the centre, and finds 19 because it was the only way for white to extend. Having played here, however, it loses interest until the group is down to 2 or 3 liberties.
• Black 23: He knows about ladders, and spots that white 8 can't be killed in one any more.
• Black 29: Lucky. As long as black 31 has not been played, the three stones 7, 11 and 15 have only two liberties and Black will not 'see' that the big white group has only three. After 31, though, he can 'see' white's danger, and adds another stone at 33, so that he can prove that white is dead.
• Black 33-43: Having played 33, Black can handle the fight up to 43. The game was by no means over after 43, but Black's advantage was clear.

Figure 2 (45-87)

• White 56 is more or less the losing move. Both of these programs explicitly recognise ladders, and neither would cut at 56, but Richard's program does not make the one extra step needed to avoid protecting against it.
• White 64 is an example of a blunder several of the programs made - trying to make good shape from dead stones. The shape of 58, 60 and 64 is good, but white is not 'strong enough' to use it properly. This is one of the trickier problems with using simple minded definitions of good shape.
• Black 71 is a wasted move, but many of the programs would have wasted two moves by capturing 64 and 66 completely.
• Black 87 has been the biggest point for some time - if white had taken this point instead of 86 he might still have had a chance. From here until 105, black's yose is almost perfect.

Figure 3a (88-110) BGJ had Fig 3a and 3b as one diagram, Fig 3.

Figure 3b (111-140) BGJ had Fig 3a and 3b as one diagram, Fig 3. Black 129 passes.

Black wins by 14 points.

Note that the tournament was played on the basis of Chinese counting, so that the many extra stones White played inside his own territory at the end cost no points, and the stones Black played to capture White's dead stones were necessary according to the rules.

The complete results of the tournament are as shown here - a fullreport including records of all the games is being produced by John Hobson. It is planned that the event should be repeated at the same time next year.

B. Przybyla   }
              } B. Przybyla   }
A. Lucas      }               }
                              } B. Przybyla   }
T. Hart       }               }               }
              } D. Skidmore   }               }
D. Skidmore   }                               }
                                              } B. Przybyla
R. Granville  }                               }
              } R. Granville  }               }
M. Scott      }               }               }
                              } R. Granville  }
J. Hobson     }               }
              } J. Hobson     }
M. Reiss      }