Neverlose

[turnipmonster]

[ QUOTE ]
And forgetting all else, a bot should not be that hard to beat in a short-handed game anyway, so I don't see what the big deal is.

[/ QUOTE ]

kevin I think you are underestimating things somewhat. HU bots are extremely strong and not at all easy to beat. although a bot might not exist to beat the big stars game, I can assure that someone can write a shorthanded bot that is anything but easy to beat.

--turnipmonster

[sammy_g]

[ QUOTE ]
I believe Mason has already written about this and if I remember right, his conclusion was that poker is a much more complicated game than chess. At least from a programming standpoint.

[/ QUOTE ]
Mason is a great poker player, but he is not a computer scientist. I enjoyed his Poker Essay books. The essays about computer poker were my least favorite, however. Just because poker playing software doesn't beat good players today does not mean computers will never beat experts. Remember the first chess computers were an absolute disaster. For a long time, the consensus among chess experts was that no computer program would ever beat a master, and even moderately skilled humans could take apart the early chess programs. Look at chess software today.

Computer poker is still in its early stages.

A tremendous amount of research and effort went into computer chess over many years. This is why chess programs are so good. This hasn't happened for computer poker yet, although with the recent poker boom I suspect it is on horizon.

[turnipmonster]

making suboptimal plays for purposes of play balancing is a slipppery slope and opens you up to all sorts of exploitation.

--turnipmonster

[bobbyi]

Your analysis of computer chess is very inaccurate. You refer to all of these advances in chess software, and then don't name a single one. And I can't think of any big ones. Yes, we've made some improvements, such as the massive library of openings and endgames that Deep Blue had, which certainly would not have been in a chess program in the 1970's (and couldn't, due to lack of disk space, etc.). And I believe it did have some clever tricks in its software. But ultimately, it played the whole middle game by using basic min-max with alpha-beta pruning. This is not new or innovative. This is the same technique we've always used. The ascent of chess AI was entirely a victory of hardware. Many years ago, computers were simply not fast enough to look far enough ahead to be decent chess players. You could take the best AI programmers today and give them an unlimited amount of time to write a program to play chess on a 1970's computer, and they wouldn't be able to make a player that could beat something an undergrad could write that has the advantage of running on modern hardware. Deep Blue won because it had a specialized chip designed to do alpha-beta in hardware. Thus, it could run the algorithm unreasonably fast. Remember, the whole reason Deep Blue was built was as a showcase for IBM's hardware-building talent, not their programming skills.

This isn't at all similar to the state of the art in poker software. What we need is good algorithms and, as far as I know, we don't really have them yet. The stuff coming out of Alberta and elsewhere is decent and it is possible that we can write software good enough to play profitably against the right opponents, but we are still software-bound, not hardware-bound, and those problems are much harder to solve.

[tipperdog]

I have no idea if Neverlose is a bot or not. But I do know that Stars personally contacts its 100-200 and other big game players to protect against fraud, money laundering and collusion. I'm sure a determined "bot master" could easily fool this screening by posing as the player, but wanted to note that *some* measure of personal contact is made.

[Kevin J]

[ QUOTE ]
making suboptimal plays for purposes of play balancing is a slipppery slope and opens you up to all sorts of exploitation.

[/ QUOTE ]

Against a live thinking player, I agree, but I don't see how a bot can exploit something it doesn't even know you're doing. When a bot has you pegged for playing too tight and you raise, it doesn't know that you're raising because you know it has you pegged for being too tight, etc. etc. It can only go off it's compiled data. When it evens out, it will stop folding, but by that time you'll have stolen lots of money and on be on to another way of exploiting what it "thinks" it knows about you. No?

The only experience I have with bots is playing Turbo hold'em and I'm sure it's nowhere near the same thing. Turbo doesn't make plays based on compiled data. Still, it's incredibly easy to beat!

I have a feeling that I'm talking out of my ass, because I know next to nothing about any of this. Part of me wants to issue a challenge that I'd play heads-up against any bot you could throw at me. But it would be a stupid bet on my part. For one thing, there are better players to handle that challenge. Another thing is that I'd be afraid you are right and it would prove incredibly tough and I'd get killed. But for now, I just have a hard time believing that.

[AviD]

Well, are we talking individual effort or team effort here?
As you involve more people, the chances of you keeping things "under wraps" becomes more problematic as well.

Regarding the codebase, again you are assuming it is a solid core codebase of which you can extend your own implementation. It would certainly cut down the implementation time for your own enhancement...but I'm a cynical kinda guy, so I hate relying on other people's code, and would have to really tear apart another developer's bot code to be sure my own efforts/changes/activity wasn't being somehow tracked, etc.

Ultimately, if it were me (and no I'm not wasting my time to write a bot, I have this strange fixation will actually learning how to beat the game myself, even stranger...I value beating the game myself over making money with a bot)...I would right a custom bot by myself from scratch with some serious security spoofing going on. Call me paranoid, but computer fraud has become a serious crime in this day and age...and in the US it carries a pretty heavy sentence, if you ever got extradited to another country for it...who knows what would happen. Bottom line, I don't want to be in those shoes.

Onto opponent modeling, good topic but extremely complex. Realize (as I'm sure you do) that players (majority) are humans, and the practicality of modeling human behavior is obviously prone to error. And the biggest downside to modeling human behavior of your opponents isn't actually beating them, its alternating winning and losing strategy to offset your wins and losses such that you maintain a +EV increase without making your play or your bot identifiable. So you have tiers of complexity within the opponent modeling.

To be honest, the problem with humans is greed. Everyone has this image of bots ruling the poker world and making billions of dollars (OK maybe millions), but it is certainly more practical and more profitable (in the sense of not being detected and risk losing everything you've earned) to run a bot a couple hours a day and make just a little money at a time at middle limits. In fact, the best approach would probably be to just run two alternating bot implementations, TAGBot v1.0 and LAGBot v1.0, where TAGBot maintains a slightly +EV playing style over LAGBot. Much easier to design, implement, and control. Forget all the AI, opponent modeling, matrix of playing styles and adjusting real-time to address them.

And all of this in the end means sh|t, if and only if the poker sites step up and implement some security protocol to prevent such trash from happening. If they ignore it, people will do it and continue to do it, and continue to get better at doing it...which is why PokerStars ignoring or slighting the discovery of a bot (if this is indeed true) can be an extremely bad thing over the course of the future of poker play online (if other sites follow suit).

I've played online games (non-money related) for years, and cheating/bots has always been a problem. Poker sites are so far behind on controlling it, its not even funny...and the sad thing is, with how sh|tty their software is not (at least IMO) I can't imagine their security implementations being even remotely effective.

But in the end, there will be more fish than bots and thats still +EV for all the legitimate poker players. Fish won't "catch on" to bots, they are there to gamble and have fun...which often results in losing...so they won't notice the difference. Winning poker players will eventually learn to identify bots and avoid them (just as we have in other games). Granted this will cost money and is total BS, but the +EV of beating fish over the -EV of running into occassional bots still ends up +EV for the winning poker player. Then there are always casinos and lives games, where if you bot (i.e. cheat), you are sure to get a finger cut off! [img]/images/graemlins/wink.gif[/img]

If bots get so overwhelming that poker sites start losing business (yes this is certainly possible, are you going to continue playing on a poker site notorious for bots??!! I'm not), then they will be force to act or they will lose an insane amount of money. So whatever happens with the bots...I don't think it will hurt the overall scheme of poker more than cheaters do in live games. There are always better games around, it will just be an addendum to better game selection.

That's my .04! [img]/images/graemlins/grin.gif[/img]

[Kevin J]

I agree. I use to fancy myself a decent chess player and was quickly humbled by playing against the computer. There's no doubt that programming computers to play poker is still in it's infancy and will become MUCH better. But I always thought this was interesting...

I was discussing this subject with a friend of mine who is quite an excellent backgammon player. I was amazed to hear him say that a computer cannot even come close to beating a world class backgammon player. More so than chess, backgammon is a game of math, right? I would think a computer could easily come up with the exact mathematically best play in backgammon and stomp even the best human player. But this isn't the case. I don't understand it all, but it has something to do with the doubling cube and other intangibles that a computer cannot take into account.

Anwyay, I was floored by this because I always thought chess was a much more creative game (than backgammon), with infinite ideas, positions, etc. And I think poker is more akin to backgammon than it is to chess. There's just too many intangibles. I think it will be a while before a computer can beat the best poker player. Just my opinion as I know nothing about programming, etc.

[turnipmonster]

kevin I'm not saying anything about your play at all, I don't even know you. I'm just talking theoretically here. that said, the latest and greatest HU publicly available bots are poki, at http://www.poki-poker.com/ . as someone else mentioned they get a lot better the longer they play you, and they seem to play pretty well. so if you are interested that might be something to check out.

--turnipmonster

[Kevin J]

Very informative. Correct me if I'm wrong, but is it even a fact yet that a computer has been built that the best human player cannot beat? I know a human lost to one recently, but was that conclusive? Being human, maybe he just had an off match. Has this computer been proven to be able to consistently beat grand masters? Thanks.