Re: computer-go: Insight of a human

[Joan Pons Semelis <joan@xxxxxxxxxxxxxxxx>]

Andre Engels, engels@xxxxxxxxxxxxxxxxx wrote:

>Joan Pons Semelis wrote:

>> I think best way to use pro games is to select games that end with
>> an even result (winner by less than 5 counting komi for instance)
>> assume that that eveness has been there all the game long, and tune
>> evaluation of positions. Of course sente/gote has to be taken into
>> acount, or if you prefer, the heat of the game.

>I don't think that that would work very well - according to this
>method, a program can improve its 'apparent quality' simply by
>reducing the value of everything.

No!, at the end, chinese counting, each player must have about
half the board!
Even if some moves serves to decrease the other side (potential)
territory, in the long run each player's controled area must
increase.

Joan