[disjunction]

[ QUOTE ]
disj, not sure if you are agreeing with me or not but what you say is my point. bots can evaluate how dry a flop is relative to a hand range far better than i can. imagine if i had an hour to make every decision. i would run pokerstove and play perfectly against my opponents estimated range. big advantage imo. a bot could go even further bc it wouldnt have to estimate the range. bigger advantage.

[/ QUOTE ]

I am definitely not agreeing with you, although I admit that I am not that confident in my point.

My point is this: programs aren't magic, if you can say what you want the program to do, then you can program it. But the problem is, you can not say what you want it to do. When you start to get specific, you will find that you have problems.

For instance, you steal-raise from the button, and the BB calls. The flop comes, he checkraises. What does he have? Well maybe that depends on whether the flop is K62 rainbow or 987 single-suited. If the former, then maybe he is using a different rule for checkraising. Maybe his range depends on the fact that he is estimating what your range is, and he thinks he can blow you off of a hand on the dry flop. So it's not enough to merely enumerate hands, say that he'll checkraise on any pair or draw, and estimate your equity. You need to do something more. It's easy to screw up, and I don't know if it's possible to do a great job. I've told the story before of playing a bot at the AAAI competition, and I could bluff-raise it with air on any dry board. I changed my range so that its computations were useless. If the programmer fixed this, he might introduce a new problem.

As for the state of the art, I've only gone to one talk and skimmed a paper or two, and not from the best bots, but what I was gathering was that publishable papers are weak from a poker standpoint. They abstract away information that you need. Maybe they can prove 5% of optimal or whatever but a poker player can drive a truck through that 5%. So you're left trying to encode things rule-by-rule, but when you actually try to do it, it's not that easy. Flop texture, opponent profiling, thinking one step ahead to the turn and river, even the ramifications of backdoor straight draws would be a pain in the neck. I have learned to be very wary of programs that sound easy when you wave your hands, but not when you sit down to write them. I'm sure you can cook up something to beat fish, but I'm not sure that you can rigorously describe, let alone program, a bot that can beat yourself.