[Computer-go] Beating old ManyFaces at 29 handicap stones

["Ingo Althöfer"]

Hello Don,

> I think the fear was that the player would beat the program the same exact
> way each time.    

Right.

> Is that likely to happen without the randomness?

Perhaps.

In 2002 I did an experiment with the commercial bot Handtalk on 13x13 board
and 4-dan amateur Guido Tautorat. Guido gave 3 handicap stones and tried to
win as clear ar possible (in Japanese rules). "Random" was switched off
in that games. Four games were intended, in one session without break.
Before the event Guido got Handtalk for six weeks of sparring.

One day before the event we phoned. He was optimistic "I get more than
+170 [on a board with 169 squares, IA] in almost every game." I asked him:
"How did you start the sparring games?" Guido: "I set up the handicap stones
manually." Ingo: "Oh no. When doing this, HandTalk plays less carefull. You
have to say the bot that you want to play at handicap 3. Then it plays more
defensive and is a harder opponent. And now you have another 20 hours to prepare
for that situation." Next evening Guido came ("I played sparring games almost
without any break - since our phone call"). Then he scored +171, +181,
+40, +181. The +181 had been his intended result. In game 1 he played one
"Fingerfehler" (oversight), in game 3 too.

Funnily, ManyFaces in those days with "Random" switched off was also
 non-deterministic. David had forgotten to make proper garbage collection,
and depending on the content in memory ManyFaces did not always react
identically.

Ingo.

 
-- 
Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de