computer-go: k-best modes in (commercial) go programs

[althofer@xxxxxxxxxxx (althofer)]

Hi, I am a newbie to this group.

In the last two decades I made several experiments with man-machine
combinations in the game of chess (3-Hirn, Double-Fritz with Boss, List-3-Hirn, 
...). It turned out that a team with two different (commercial) chess programs 
and a human amateur player was able to play very strong chess: The computers 
were run and made their move proposals. Then the human had the final choice 
amongst these proposals. In other experiments I used one chess program with a 
k-best mode (in k-best not only the best move is computed but instead the k best 
moves, in the opinion of the program), and the human boss selected one of the k 
proposals for execution.

Recently I have started to try similar approaches in other games. For instance,
in backgammon (JellyFish) and Othello (WZebra) there exist commercial
programs which are very suited for such an approach (with k-best proposals).
Last weekend I made a first step in man-machine play in go. Xiaoding told
me (in rec.games.go) that EZ-GO (Shareware version of Bruce Willcox' EGO) has a 
hint function which gives proposals for the human player (often exactly one move 
is proposed, but sometimes two or more proposals (with "subjective" evaluations) 
are given).

I found a 2-dan player (Achim Flammenkamp, mathematician at the University of
Bielefeld) who was willing to participate in the following experiment:
For Black the moves of EZ-GO was played. When it was White's turn to move I 
pressed the hint buttom of EZ-GO and told Achim about the proposals of
EZ-GO. Then Achim had to select one of these moves.

We played a single game, taking about 65 minutes altogether: The "EZ-GO&Achim" - 
team lost by 60 : 115.
Most of the time Achim was not happy with EZ-GO's proposals. Very often (in
about 90 percent of the time) exactly one move was proposed. In many of the
other cases the moves proposed were rather similar to each other (and weak).
During the game Achim became more and more unhappy. Around move 80 or so
he told me: "Do not say 'It is your turn to move'. Instead say 'It is White's
turn to move'. I am not responsible for this position." At the end of the game
Achim gave me the following hint: "Don't let us play more games now. Better,
go home and think about what has happened." I agreed.
By the way, as a warm-up before the experiment Achim had played a sparring
game with EZ-GO which he won by 229 : 17.


Now, my questions to this mailing list:

(1)  Do there exist other (strong and commercial) go programs which have a
     "multi-hint" feature like EZ-GO   ("multi-hint" means that at least
     sometimes more than one move is proposed)?

(1') Do there exist go programs which evaluate all legal moves in a position.

(2)  Which of the strong commercial programs have comfortable "take back"-
     features? (I would need this for 3-Hirn experiments: two go programs
     making their move proposals - and then a human who selects one of
     these proposals. Of course, the program with the unchosen move has
     to be set back and the other move entered.)

(3)  Which of the commercial programs typically propose different moves?
     (For 3-Hirn it would not make much sense to have computers which almost
      always propose the same move.)

Thanks in advance for your replies!     Ingo Althofer.