bots analyzed by gnu

[yappay]

when i analyze the luck factor of the games i am playing with fibs bots with gnu.
gnu calculates an incredible luck factor for the bots.
can i trust gnu or the bots?

[socksey]

Luck comes in streaks.  No one can measure this factor over a period of time until after the fact. 

Gnu IS a bot.   

socksey



Think occasionally of the suffering of which you spare yourself the sight.  Albert Schweitzer

[NIHILIST]

I believe most of the bots at FIBS are gnu-based. So, if you use gnubg to evaluate play, gnubg will almost always tell you that the gnu-based bots played flawlessly.

There are certain types of games where the gnubots are completely clueless, ie, hitting its opponents last 2 checkers while having no board development of its own. The bots seem never to slot points or do anything to create a position from which to win. I watched such a game that lasted almost 200 moves with the bot losing, having never bettered its starting position. A post game analysis by gnubg rated the bots play SUPERNATURAL. The truth is Ray Charles would have played the position better than the bot.

Bob

[PersianLord]

In addition to what NIHI correctly pointed out, the bots also face massive challenges in defending against a developing mega-backgame, specially during the middle and final stages. They can't analyze the positions correctly and usually fall in traps.

PL

[dorbel]

The "bots can't play backgames" theory is in my opinion a myth, dating from the early versions of Jellyfish and Snowie which didn't always do well in backgame situations. Modern bots play them quite well and certainly better than 99% of humans. You can construct positions in which they make some very strange plays, but positions that occur in real life backgammon are normally handled with robotic efficiency.
Are there humans with deep insight into playing and defending against backgames, who can teach Gnu or XG a thing or two? It's possible, but I doubt if they contribute to Fibsboard.