On July 3rd through 6th, three teams of professional poker players will compete live against Polaris 2, the most sophisticated poker AI on the planet and the result of years of research and thousands of man hours invested by the University of Alberta’s Computer Poker Research Group, in the second Man vs. Machine Poker Championship, held at the Gaming Life Expo at the Rio All-Suite Hotel and Casino in Las Vegas. Each team will play two 500 hand duplicate matches, wherein the same series of cards is dealt each round with the human teammates playing opposite sides of the hole cards against the AI. At the end of each match, the total number of chips won or lost by the humans or the AI will determine the winner. One of this year’s participants is Matt Hawrilenko, often better recognized by his online handle of Hoss_TBF, who has been one of the mainstays in limit hold ’em’s nosebleed games for a couple of years now.
I had a chance to talk with Matt about the upcoming match.
Bryce: For those who are not familiar with you, let’s hear the 60-second-or-less retrospective on your impressive limit hold ’em career.
Matt: I started taking limit hold ‘em seriously exactly three years ago this WSOP. I moved up limits aggressively, moving from 15-30 to 300-600 in under a year. In 200-400 limits and higher, my win rate in heads up games, taking on all comers, is over 3.5bb/100 hands at a high level of statistical significance. That translated to a limit hold ‘em income well into 7 figures last year.
Bryce: It’s my understanding that you’ve recently been having some success in some other limit games, such as O8 and HORSE. Do you find it fairly easy play these games at a competitive level, given your limit hold ‘em experience and the fact that many other people playing these games may be coming from a no-limit or pot-limit background?
Matt: Obviously the adjustments between games aren’t all entirely natural and intuitive, but in general, poker is poker. The same concepts apply across all forms of the game, although different concepts are more salient in different games and thresholds may be radically different. I think anyone who tries to look at poker from a conceptual standpoint will be able to transfer their skills across games more easily. It’s pretty evident to me whether a new player sitting in my game, be he a PLO player or hold ‘em player moving up in limits, really gets poker or whether his strategy in one particular game has simply won the natural selection lottery. In other words, whether he has a clue or he has fallen ass-backwards into success by doing many of the right things for the wrong reasons. The interesting thing is that even if these two players are winning or losing the same amount in EV, I can often tell who’s going to get really tough and who is not.
Bryce: Down to business. Polaris 2 represents a tremendously difficult opponent. I’ve had the opportunity to play against it myself and I was very impressed. The AI is based on a Nash equilibrium (also sometimes referred to in poker as game theory optimal) solution for a mini-game with 10^12 game states, as opposed to real heads-up limit hold ‘em’s 10^18 game states. While that’s several magnitudes off of a full equilibrium solution, what’s your plan of attack here? Are you banking on the AI making enough mistakes that you’ll be able to beat it with a superior, balanced strategy, or are you hoping to be able to gather enough information over the few hundred hands that you’ll be able to out-adapt and out-exploit it?
Matt: I doubt I’ll play very exploitively. Although I do exploit human players, my base strategy is the strength of my game. One of the reasons stronger players can exploit weaker players quickly is because our thought processes are predictably similar, so strong players can apply lessons from one human to another. I don’t expect Polaris to fit those parameters.
Bryce: Do you expect to be playing much differently against Polaris 2 than you would against a human opponent?
Matt: I’ll get less fiendish glee from sucking out, but I expect that smashing the monitor will be more therapeutic than normal.
Bryce: How important is it for you that Polaris 2 is a learning AI, as opposed to a static bot that wouldn’t adapt at all?
Matt: I’ve more than held my own against some of the best exploitive human players around and I’ve found that they often make quixotic adaptations to my strategy based on observations over a small sample size. I think my strategy is balanced enough that Polaris will have a hard time finding much to exploit over a few hundred hands, so my hope is that, like the humans, Polaris impales itself chasing windmills. While it might dominate a weaker player by making adjustments, I hope that my strategy is strong enough that adjustments Polaris might make over 500 hands provide it with very little upside but a huge potential downside. With today’s technology, computers can compound programmer errors more than a million times per second, right?
Bryce: You recently had a contest on your blog asking readers for their suggestions for machine tilting trash-talk (http://hoss-tbf.livejournal.com/6543.html). Personally, I’m partial to “you’re a mechanical David Benyamine” myself. Any favourites?
Matt:
- With today’s technology, computers can compound programmer errors about a million times per second, huh?
- 01111001 01101111 01110101 01110010 00100000 01101101 01101111 01110100 01101000 01100101 01110010 01100010 01101111 01100001 01110010 01100100 00100000 01101001 01110011 00100000 01110011 01101111 00100000 01100110 01100001 01110100 00100000 01110100 01101000 01100001 01110100 00100000 01110011 01101000 01100101 00100000 01101001 01110011 00100000 01101100 01100001 01110010 01100100 00100000 01100011 01101111 01101111 01101100 01100101 01100100 (translates to: your motherboard’s so fat, she’s cooled in lard)
- Hey Polaris, what’s 1 divided by zero?
Bryce: You’ve been a regular player in the big PokerStars games for some time, so I imagine you’ve had ample opportunity to play against your teammate for this event, IJay “Doughnutz” Palansky. Any interesting rival stories or comments on being paired up with IJay?
Matt: I know he’s done very well in 3-6-handed games, but I’ve never had the pleasure of playing him heads up so it’s hard to comment on his game.
Bryce: There have been a few bot scandals lately, most notably at Ultimate Bet. When you’re playing online are you ever concerned about the possibility that your opponent might be a bot?
Matt: I think it would be a bad move for most bots to tackle the biggest game. They seem so much better suited to toil in anonymity at mid-limits where they can keep smaller account balances and lose less when they’re eventually caught. I’m most afraid of bots scaring people away from online poker. While bots are getting tougher and tougher heads up, my understanding is that 3-handed games are hugely more difficult to tackle than 2 player games and the difficulty grows exponentially with each added player, so we have a good, long time before bots start to permeate 3-handed and higher games.
Bryce: There tends to be a bit of a mythos around equilibrium strategies these days in that many players believe that an equilibrium strategy is always the best possible strategy. In reality, of course, this isn’t the case, as exploitative strategies will always have higher expected returns against imperfect (re: human) opponents. What suggestions do you have for aspiring players who are interested in learning a bit more about Nash equilibrium (Game Theory Optimal) strategies?
Matt: Exploitive strategies will only have higher expected returns if you have an idea how they are deviating from optimal frequencies in the first place. This means that to develop a good exploitive strategy, you actually have to have some idea of what a strong, balanced strategy looks like. You can call me an idiot for check-raising hands on the flop that make no sense to you, but if I balance it well, you can try to exploit all you want but you’re not going to get anywhere. Simply focusing on working towards a balanced strategy, I’ve achieved win rates in ring games higher than most others I’ve heard reported.
For aspiring players, read the Chen and Ankenman book “Mathematics of Poker”. All the concepts I use are in the book and contrary to popular belief, you don’t have to be some kind of wunderkind to apply it to your game. Read the big ideas, usually highlighted at the end of the chapter, and question how they apply to every hand you play. Watch your opponents play and instead of dismissing them all as donkeys (as most people tend to do), think about how their plays, especially the outlandish ones, might fit into an effective strategy. People seem to think I’m some sort of super smart guy. While it’s an appealing thought and I do enjoy the rumor, the boring truth is that I am constantly trying to stay ahead of the curve by always questioning and never going on autopilot. I want to outwork everyone else.
Bryce: I know I bug you about this every second e-mail I send you, but is there any chance of you making instructional videos for one of the new poker training websites in the near future?
Matt: That’s up to you guys. There’s definitely a chance. It correlates highly with you agreeing to compensate me appropriately.
Bryce: Last question. If you could either bust Patrik Antonius or have somebody write you a check for 1.2 times his net worth, which would you choose?
Matt: You may as well interview Tiger Woods and ask him how sweet it would be to take down Michelle Wie, the main differences being that Patrik probably spends more time trying to look pretty and definitely shows more cleavage.
Kidding. Patrik is easily the toughest player I’ve faced online. We have very different styles, but it’s easily apparent that he has a better grasp of the game than anyone else I’ve played and pushes in all the right spots, every single one.
Results, hand histories, and video recordings of the preliminary matches between the human teams and Polaris, as well as news and results from the live finale at the Rio, can be viewed online at http://www.stoxpoker.com/man_vs_machine.html?refer=2+2 . The final matches will take place at the Stoxpoker booth at the Rio’s Gaming Life Expo, July 3rd through 6th. More information regarding the Computer Poker Research Group, the academic team behind the Polaris AIs, can be found at their website, http://poker.cs.ualberta.ca/ .
Additional coaching and tips from Bryce can be found at Stoxpoker.com.
