Finally got around to doing something I’d wanted to do for a while, setting up the chessboard in front of the computer and playing the classic and historic MacHack chess program against an unexceptional modern program . MacHack is the famous chess program developed in the 60s at the MIT AI Lab by Richard Greenblatt, famous among other things for being the first computer chess program capable of tournament-level play, and also famous for beating the AI skeptic and philosopher Hubert Dreyfus who had claimed that computers would never be able to play a good game of chess.
And in the other corner, we have a program called Shredder, which is very good in some of its PC implementations, but I wanted to run the lesser version that I have on my humble Android tablet. It actually has some serious issues at lower settings like sometimes letting an important piece be captured for no apparent gain (the author claims it mimicks a human player by “deliberately” making mistakes), but these failings seem to go away at higher settings. Still, I was skeptical about its proficiency. To at least get rid of obvious weaknesses I cranked it up to what it alleges to be ELO 2350 (supposedly “master” – it goes all the way to 2600 - “grandmaster”, although I don’t believe this is at all accurate) and away we go …
MacHack was written for the PDP-10 timesharing system (actually, originally for its predecessor, the PDP-6) and for anyone curious, I was running it on the SIMH PDP-10 simulator. Also for the curious, an MIT research memo about MacHack [PDF].
This was the game. Shredder was white, MacHack played black.
Not only did the humble tablet beat MacHack at the latter’s default settings, but it did it quite readily and decisively. I have never been able to beat MacHack myself (so it was sweet to see it humiliated!), though I’m a pretty poor chess player. I can beat Shredder, but only when set to somewhere around “casual player” level and it makes a stupid mistake. I guess cranking it up to the “master” range was probably unfair, but even the humble tablet was coming up with its moves almost instantaneously at that setting. Also surprising, after about the first 6 moves or so, when entering MacHack’s moves, on at least three occasions Shredder warned “that last move wasn’t so good, do you want to take it back?”. I have new respect for that program, at least when it’s allowed a far amount of scope in the search tree.
The default MacHack settings I was using were the “standard” rather than tournament level settings. These were: default search depth of 4, number of plausible moves to consider on each ply 7, 7, 6, 6. The tournament setting is a depth of 6, plausible-move settings 15, 15, 9, 9, 7, 7. I think I’ll try that next.
It may seem “obvious” that MacHack would be beaten by a program with presumably deeper searches, but MacHack has long had a reputation for excellence due to its amazing heuristics, and it also tends to create interesting games because of its aggressive tactics. Shredder seems to me to be more plodding, and, as said, even makes dumb mistakes at certain settings.
Tablet MacHack
---------------------------------
1 P/K2-K4 P/K2-K4
2 N/KN1-KB3 N/KN1-KB3
3 B/KB1-QB4 N/KB3*P/K5
4 N/QN1-QB3 N/K5*N/QB6
5 P/Q2*N/QB3 P/QB2-QB3
6 O-O P/Q2-Q4
7 N/KB3*P/K5 P/KB2-KB3
8 Q/Q1-KR5 P/KN2-KN3
9 N/K5*P/KN6 P/KR2*N/KN3
10 Q/KR5*R/KR8 P/Q4*B/QB5
11 R/KB1-K1 K/K1-KB2
12 B/QB1-KR6 B/QB1-K3
13 Q/KR8-KR7 K/KB2-K1
14 R/K1*B/K6 B/KB1-K2
15 R/QR1-K1 K/K1-Q2
16 R/K6*B/K7 Q/Q1*R/K2
17 Q/KR7*Q/K7 K/Q2-QB1
18 B/KR6-KB4 N/QN1-Q2
19 Q/K7-K8 Mate