[Computer-go] Narrow wins

Stefan Kaitschick stefan.kaitschick at hamburg.de
Sun Jun 9 12:11:15 PDT 2013


@Don:

> To me the biggest issue isn't winning by 0.5 but not fighting when it's
> losing.

It can sometimes feel that way, but when a bot is timid in a losing
position that's because it can't see the loss.

> I have a feeling that most of this is not about improving the
> play of the bot although it's always cast that way,  but in reality it's
> about not offending our own sensibilities.    We just hate to see a bot win
> by 0.5 when it probably could have won the entire board or most of it.

True. The way a bot plays in a winning position is mainly a luxury problem.

> The standard proposed solution is komi-manipulation in one way or another.

A bot might use an artificial komi, but by endgame time it must use
correct komi.

> But what we are probably really after is something to do with "opponent
> modeling",  giving the computer more of a fighting spirit instead of playing
> overly logical.    I sense there must be a better solution as
> komi-manipulation just seems to me to be really illogical.    It's like
> fixing the books because we don't like the numbers.

Here's a suggestion: when playing against a human opponent, give a
bonus for higher point wins, but for the humans moves only.
This would model the human as greedy, and not so sure of his points.
That could be superior to taking an outlandish risk in a position
where the bot is slightly trailing a human.

> Some positions will return bigger "wins" than others (or
> smaller losses) but we know scoring this way is terrible.    But maybe that
> information can be effectively used to shape the tree towards positions that
> win bigger?

I think bots are using bonuses for higher wins. But the bonuses must be tiny.

Stefan



More information about the Computer-go mailing list