man vs machine

  • 28 replies
    • Hahaownedlolz
      Hahaownedlolz
      Bronze
      Joined: 24.04.2009 Posts: 1,755
      Limit players should probably fear for the future of their game to be honest. Since there is only a fixed amount you can bet or small which is usually quite small in relation to the pot. It wouldn't be too hard to make a pretty decent limit bot in comparison to how ridiculously hard it would be to make a no limit one.


      I think no limit players still have quite a few years before bots will become decent. But it'd take a ridiculous amount of work and effort to make a good no limit bot.. And that person would have to be very good at poker. Understanding the game very very well and be very good at programming. In which case he could probably make a lot of money with poker anyway..
    • GodlikeRoy
      GodlikeRoy
      Bronze
      Joined: 13.10.2010 Posts: 337
      This is why PLO is so great - we're still decades away from a bot that can win at PLO :-) Limit already has them, NLHE has some and will have more in the next few years.
    • fusionpk
      fusionpk
      Bronze
      Joined: 21.01.2010 Posts: 1,683
      There are already NLHE bots built that would crush nearly anyone, im confident some of these bots could beat midstakes+ infact I heard of a bot that took on some HSNL players before and it was a close match iirc.

      The point is, AI today is pretty advanced, bots are no longer just "if(board.isWet() && handStrength.isPair()) { betPot(); }"
      bots today are built using real AI, they use training sets and artificial neural networks to be able to piece together, history, board textures, hand strength etc. aswell as being able to calculate exact equities and frequencies. Infact if some of the bots got good samples of hand histories vs you, they could play near perfect games, which is why only the best plrs can beat them.

      Now ofc these bots are only wrote/built by big companies who are trying to achieve something, your average hobbyist doesn't have quite the time and/or resources to be able to do stuff like this, and the bots that they talk about in the article are just the standard 'follow rule set' bots.

      As for PLO bots, imo there are a couple of reasons why:
      1) The game is not as big as HE.
      2) The game is more complex than HE.

      Whilst I agree that it would be much harder to write a PLO bot thanks to the additional dynamics of the game, I think it would still be possible for good PLO bots to be built, infact I think some pretty mindfucking bots could be built. But again, it's not going to be hobbyists that write them. (Thank the good lord really else poker would be destroyed).
    • Classified181
      Classified181
      Bronze
      Joined: 22.03.2011 Posts: 111
      EVERY SITE- Ban the Bots!!! If you cant play poker and are a losing player reverting to the use of a bot (excuse the language), F OFF and play bingo with your grandmother where anyone can win, but do not bring bots in and ruin our great game simply because you are'nt good enough to compete or win at it!!!
    • Hahaownedlolz
      Hahaownedlolz
      Bronze
      Joined: 24.04.2009 Posts: 1,755
      Originally posted by fusionpk
      There are already NLHE bots built that would crush nearly anyone, im confident some of these bots could beat midstakes+ infact I heard of a bot that took on some HSNL players before and it was a close match iirc.

      proof? I've never heard about any good no limit bot. In theory it's ofcourse possible to make a bot that could beat high stakes online.. But if your that good at poker and programming then your wasting your talent :f_biggrin:
    • fusionpk
      fusionpk
      Bronze
      Joined: 21.01.2010 Posts: 1,683
      Originally posted by Hahaownedlolz
      Originally posted by fusionpk
      There are already NLHE bots built that would crush nearly anyone, im confident some of these bots could beat midstakes+ infact I heard of a bot that took on some HSNL players before and it was a close match iirc.

      proof? I've never heard about any good no limit bot. In theory it's ofcourse possible to make a bot that could beat high stakes online.. But if your that good at poker and programming then your wasting your talent :f_biggrin:
      lol *sigh* i dont want to get into some kind of war.. but I would have thought it was quite obvious these bots exist.

      From my time studying neural networks, ive realised that actually, for alot of people/groups/development teams, writing bots that play near perfect games vs opp's w/ big history isn't actually so hard to do. The point is however, generally things like this are illegal and require big teams and alot of time, hence why you don't hear about many, because they arn't everywhere. They do however exist and I believe there are many underground scenes where ppl have/do work on bots like this.

      I think you are quite naive to believe that writing a bot like this requires you to be good at poker, infact quite the opposite, most ppl that write bots like this probably arn't so great. I also don't see many great programmers making the money that a bot would make. The only reason more of these bots don't exist is for the reasons I explained previously, not because there arn't people capable.

      I'll leave you with a thought, say we have 1m hands on villain in our HEM db. As a reg vs reg at high games we would definitely be analysing these hands, and we would pickup an edge vs opp. Now consider that a neural network takes these 1m hands as a training set, the neural network now knows EXACTLY how opp plays. It doesn't just know his WTS is 25%, it knows what board textures opp will call light, it knows what frequencies opp is c/r'ing as a bluff, which boards he is willing to stack off his overpair, which boards he might trap. Infact, the neural network learns so much that villain has to seriously adjust his game to exploit the bots tendency to exploit his game. Even then im sure some people are capable of writing bots that run 2 neural networks that compare entire history with session dynamic & history. Infact some of the stuff thats possible is unbelievable.
    • Hahaownedlolz
      Hahaownedlolz
      Bronze
      Joined: 24.04.2009 Posts: 1,755
      Originally posted by fusionpk
      Originally posted by Hahaownedlolz
      Originally posted by fusionpk
      There are already NLHE bots built that would crush nearly anyone, im confident some of these bots could beat midstakes+ infact I heard of a bot that took on some HSNL players before and it was a close match iirc.

      proof? I've never heard about any good no limit bot. In theory it's ofcourse possible to make a bot that could beat high stakes online.. But if your that good at poker and programming then your wasting your talent :f_biggrin:
      lol *sigh* i dont want to get into some kind of war.. but I would have thought it was quite obvious these bots exist.

      From my time studying neural networks, ive realised that actually, for alot of people/groups/development teams, writing bots that play near perfect games vs opp's w/ big history isn't actually so hard to do. The point is however, generally things like this are illegal and require big teams and alot of time, hence why you don't hear about many, because they arn't everywhere. They do however exist and I believe there are many underground scenes where ppl have/do work on bots like this.

      I think you are quite naive to believe that writing a bot like this requires you to be good at poker, infact quite the opposite, most ppl that write bots like this probably arn't so great. I also don't see many great programmers making the money that a bot would make. The only reason more of these bots don't exist is for the reasons I explained previously, not because there arn't people capable.

      I'll leave you with a thought, say we have 1m hands on villain in our HEM db. As a reg vs reg at high games we would definitely be analysing these hands, and we would pickup an edge vs opp. Now consider that a neural network takes these 1m hands as a training set, the neural network now knows EXACTLY how opp plays. It doesn't just know his WTS is 25%, it knows what board textures opp will call light, it knows what frequencies opp is c/r'ing as a bluff, which boards he is willing to stack off his overpair, which boards he might trap. Infact, the neural network learns so much that villain has to seriously adjust his game to exploit the bots tendency to exploit his game. Even then im sure some people are capable of writing bots that run 2 neural networks that compare entire history with session dynamic & history. Infact some of the stuff thats possible is unbelievable.

      Dont want a war and then call me naive :f_mad: I'll just leave it at this.
    • fusionpk
      fusionpk
      Bronze
      Joined: 21.01.2010 Posts: 1,683
      Actually thats very wrong. The whole point is that the neural network adapts to the opponent itself. You don't teach the bot, the bot learns from the training set and figures out the optimal adaptation.

      Your also pretty wrong in thinking that the bot just takes advantage of a hud, and that it would take a good player to program the bot to be able to adapt. The fact of the matter is, the bot doesn't use the hud thats the whole point. A bot doesn't play like a human player in that respect. The bot learns it's entire adaptation from the history.

      Think like this, everyone knows if you are playing vs aggro 3b'er we can adapt in a few ways: tighten up range, 4b more etc etc. everyone knows the theory behind it. A GOOD player is the player that picks the right moves in the right spots and exploits this tendency from villain. A bot knows exactly what the optimal move is thanks to it's 'brain' being adapted round villains strategy. Therefore it's very possible that for example, I couldn't beat you in a HUSNG but I could instruct a team on how to build a bot that would be able to adapt and beat you in a HUSNG.
    • jasmineB
      jasmineB
      Basic
      Joined: 08.04.2011 Posts: 2
      I've never heard about any good no limit bot. Man created machines so, they're superior than robots.
    • supeyrio
      supeyrio
      Bronze
      Joined: 11.11.2009 Posts: 3,106
      @fusionpk: i think you forgot about adaptions.. a simpler game like rock scissors paper, sure a bot can estimate my frequencies etc e.g. i choose rock on higher frequency. the bot will then choose scissors on a higher frequency to win. however as human, i would have spotted that and adapt. while the bot still relies on the old data it has.

      poker is a game of adaptation, whoever can adapt faster and better wins.

      lets not forget, handhistories do tell a part of the story on how one would play his cards, but lets not forget, a player will play differently vs a different player. if u have btm set on a A29r board, you definitely would stack off vs fish, but would u stack off against a 5/5 nit with 1.0 aggro factor and 20% WTS who you suspect has a better set?

      that is where the bots are failing.
    • zloiadun
      zloiadun
      Bronze
      Joined: 12.02.2009 Posts: 31
      I think it is obvious that the good NLHE bot will be created by many persons, and most likely some very big IT corporation(s) is(are) working on it right now. IBM already beat humans at chess, why not beat us at all types of poker? :)

      Also we all should keep in mind that such bot is build for academic/research/advertising purposes, not to win big on-line :)
    • erptech
      erptech
      Platinum
      Joined: 15.03.2008 Posts: 659
      @supeyrio

      I think you are correct at this moment, but future brings AI that will be more intelligent and adapt even faster than humans. Only difference will be that they will be able to process hundreds of thousands hand histories in a second, and human will have his lame HUD.
    • supeyrio
      supeyrio
      Bronze
      Joined: 11.11.2009 Posts: 3,106
      @erptech: future beholds possibilities. few if any thought humans could fly. so i wouldnt bet against that. but till then..
    • skiffcz
      skiffcz
      Bronze
      Joined: 27.03.2011 Posts: 229
      Well, you dont need NN to build a pokerbot. I dont think the magic potion is technology used, nah. Its rather finding an +EV strategy as simple as possible.

      What we need:

      1, +EV strategy playing bot
      2, Is able to find and exploit fish
      3, Is able to recognize sharks and protect itself from exploitation by those (i.e. - gtfo and find another table)

      What we do not need:

      1, That it crushes everyone
      2, That it has massive bb/100
      3, That it adapts heavily (except that it plays only against players that it can beat and ignore others)

      To me, that seems like a task suited for a master thesis for group of like 4 IT students. You dont need nuclear reactor and Ethan 'Bubblegum' Tate to achieve this.

      But the problem with this approach is - you really need to be a good poker player (remember, the trick is to develop a poker strategy). But in fact, there is a much better and completely legal way for a programmer with poor poker skills to make money through pokerbots. Sell them. Ppl will buy them no matter how crappy they are :D I mean - cmon and sue me idiots lol. GL with that :D
    • konkey
      konkey
      Bronze
      Joined: 29.12.2009 Posts: 385
      Originally posted by skiffcz
      you really need to be a good poker player (remember, the trick is to develop a poker strategy).
      Sorry but really not. You probably don't know much about computer science (not programming). With our knowledge it really is impossible to create strategy and bot playing according to this, plaing at least not terrible ( for game as complex as NL Holdem). As fusionpk said, for these purposes serves something that is called artificial neural networks. And they really learn themselves as they play. Actually it's easy to programm bot playing tictactoe, at first not knowing anything that learns when he plays. And is getting only better. Altough I doubt there are such bots already, at least somewhat decent. (for NL). For HU FL already exist bot that competed best FL pros like Ali Nejad, and did well. He competed 6 pros and to ensure variance is on noones' side, two people played the same set of hands, but with the bot on on one side in first match, and with him on the other side in second match. Not sure where I fonud this but you can definitely google it. But to develop neural network for NL is way more difficult. It's not about developing strategy. It's about developing neural network and then getting shitloads of hands to learn him something.

      And another thing is, unlike for us, humans, it's way easier to programm bot playnig HU than bot playnig SH or FR. Because there's only one opponent to focus on.
    • Wriggers
      Wriggers
      Bronze
      Joined: 21.07.2009 Posts: 3,250
      Surely it's easy to create a winning FL bot due to the fact a lot of FL is purely mathematical, whereas NL is a lot more complex due to the fact that betting has no limit. For example, would a bot be able to calculate implied odds in NL? Or would it simply fold it's OESD because Villain didn't give him pot odds. Whereas a human may be able to see that the effective stack sizes are deep and put Villain on a hand that he is always going to stack off with.

      Although a bot for the micros could be very easily made I believe, as you just have to set up a basic strategy and assume everyone there is a fish :D

      Take everything I say with a pinch of salt btw, since I know next to nothing about programming or computer science :D Although I probably shouldn't have said that, that's just made it possible for someone to bullshit me and me not have a clue about it :D
    • skiffcz
      skiffcz
      Bronze
      Joined: 27.03.2011 Posts: 229
      Originally posted by konkey
      Sorry but really not. You probably don't know much about computer science (not programming).
      I have a bachelor degree in Computer Science and will be finishing my masters this semester. My bachelor thesis is touching directly the soft computing area (cellular automatons), so I assure you - your read is false. Feel free to attack my poker skills and knowledge (those are lame), but Im a bit touchy about situations where my intelligence is being questioned.

      Originally posted by konkey
      With our knowledge it really is impossible to create strategy and bot playing according to this, plaing at least not terrible ( for game as complex as NL Holdem).
      I will reiterate the goals here, in case you have missed them for the first time:

      1, The bot must recognize fish and the rest of the players.
      2, The bot must have simple, solid strategy, which beats fish often enough to be profitable.
      3, The bot must have mechanism to prevent better players to exploit it.

      I will now elaborate on each point and list my reasons to believe this is doable without NN:

      1, You can simply integrate the bot with statistical software such as HEM and use stats that it provides to classify the players.
      2, All over this site, there are simple, strict, algorithmic strategies trying to accomplish exactly this (BRM rules, SHC, SSS ... anyone?). Also, in probably more than half of the beginners videos is clearly stated, that newbie should not deviate from these strategies, but rather stick to them closely, unless he wants to get bust. I feel a bit of a contradiction between this fact and your statement, that "is impossible to create strategy and bot playing according to this, plaing at least not terrible". Please note, that I do not attempt to resolve this conflict. You may as well be right, but than there is something wrong with ps.com.
      3, The most straightforward strategy to avoid shark is to leave the table. I think we can agree on this.

      More so, in my first sentence I clearly stated you dont need (imo) NN based approach to create an +EV NLHE bot fulfilling above requirements. Thats why I find further debate on neural networks or beating Ali Nejads unfruitful. You dont need to be faster that a lion to survive. What you need to is to be faster than at least one other person from your safari group. It is obvious, that NN based solutions have the potential to beat anyone, as human brain is of course itself a NN solution.

      P.S. I didnt mean to be in any case rude in my reply. The strict tone it may carry is simply my attempt to be as unambiguous as possible. There is very high probability, that there is still a flaw in my reasoning (I have never attempted to build a poker bot myself, nor am I any good at poker). However, I doubt it is caused by my lack of knowledge in the computer science area - it is most probably somewhere else. Feel free to point it out, Im here to learn, after all.
    • fusionpk
      fusionpk
      Bronze
      Joined: 21.01.2010 Posts: 1,683
      Originally posted by supeyrio
      @fusionpk: i think you forgot about adaptions.. a simpler game like rock scissors paper, sure a bot can estimate my frequencies etc e.g. i choose rock on higher frequency. the bot will then choose scissors on a higher frequency to win. however as human, i would have spotted that and adapt. while the bot still relies on the old data it has.

      poker is a game of adaptation, whoever can adapt faster and better wins.

      lets not forget, handhistories do tell a part of the story on how one would play his cards, but lets not forget, a player will play differently vs a different player. if u have btm set on a A29r board, you definitely would stack off vs fish, but would u stack off against a 5/5 nit with 1.0 aggro factor and 20% WTS who you suspect has a better set?

      that is where the bots are failing.
      Don't forget the bot will have seen you adapt before to people exploiting you. For example the bot will know how you reacted vs aggro opponent that c/r you on certain flops etc.

      Also im pretty sure any dev team that created a bot like this would run 2 neural networks, one adapted for optimal play vs opp and one adapted for optimal play vs session dynamics and between them they would figure out the likely 'best move'

      It's quite sick what you can do with AI nowadays.
    • fusionpk
      fusionpk
      Bronze
      Joined: 21.01.2010 Posts: 1,683
      @Skiff - Yeah ofc you are right, and there are many public (and private) bots that just follow a generic set of rules. I've seen scriptable NLHE bots before etc. And yeah there are some that grind at pretty reasonable winrates, but these bots likely wouldn't do well HU and likely wouldn't do well at higher stakes where they are put in much tougher spots and where people will recognise their tendencies.
    • 1
    • 2