Smash Vegas

I think pokerbots are a natural source of curiosity for anyone interested in the mathematics of gambling.  However they are also a source of much contention in the online poker world.  One side argues that bots will destroy the game - even if they never reach expert play they will populate the low stakes world and kill off new players.  The other side argues that online poker is already in a process of change and is inherently unfair.  They argue that pokerbots are not a black and white issue - how do you classify the advanced player who uses an odds calculator and pokertracker with a collaborative database to aid his identification of player weaknesses?  They see pokerbots as a simple continuation along this path.

Whatever your viewpoint, I think it’s very interesting to see what pokerbots can / can’t do.  I have some experience with them so I feel able to give you a pretty fair breakdown of what’s going on in the world of bots.  (I should note here that I am refering to pokerbots which play ring games and are not designed solely to play heads up.)

When I say “pokerbot” your first thought is probably a program which checks through a list of actions under specified circumstances eg:

I have pocket aces and no one has bet yet and I’m in late position so …

I’ve flopped top pair but there’s a flush draw on the board with a raise in front of me so…

etc…

While pokerbots like this do exist and can be successful it is incredibly difficult to make sure you cover every eventuality.  They are known as “static” bots and are usually people’s first attempts.

The other end of the spectrum is a “dynamic” bot which has no hard coding at all and relies on computing probabilities forward through the hand.  In essence they look at their pot odds and make an estimation of their probability of winning and see if it’s worth playing.  The challenge here is adjusting the estimation of the probability of winning based on opponent actions.  We don’t know what cards they hold and can only judge their strength based on betting behaviour (which may be deceptive).  Another challenge is that the game of Hold’em is so vast that it is impossible to sum probabilities to estimate our chance of winning.  No computer in the world could do it fast enough.  The only choice is to simulate many hands to approximate a probability of winning (this is known as a “monte carlo” simulation).

In practice most bots are a hybrid of these two approaches, as each has benefits and disadvantages.  If you really want to learn about how these bots work I would recommend reading some of the papers produced by the University of Alberta’s Poker Research Group.  They have been at the forefront of poker A.I. for many years and invented many techniques which have now become standard.  A good start is this paper from 1999.  Much of their recent research has been into heads up play, where it is possible to use interesting bits of game theory (like approximating Nash Equilibria) to create very good (but not yet world class) bots.  Although this stuff is in many ways more advanced it is not commonly used in ring game play.

The above paper from 1999 is pretty much where the pokerbots of today are at.  It is easier to program a bot for limit hold’em because then you don’t need to worry about bet sizes (the exception to this is a no limit bot playing a short stack strategy…), so there are more limit bots around.  The thing to emphasize is that bots are able to play basic poker well but that’s about it.  Lots of research has gone into opponent modelling, but bots are slow to adapt to players.  It is inherently difficult due to the lack of information - you only see the opponents’ cards when they go to showdown.  In summary:

• A bot can probably estimate odds of success better than you in average circumstances.
• You can estimate the strength of opponents much better than a bot, so you will deal better with unusual circumstances.
• The real secret to a bot’s success is long, grinding persistance.  It plays good basic poker all the time.  It doesn’t get bored or drunk or pissed off.  Depending on the level of play this can be enough to be profitable.

There are many sites offering poker bots for sale with “no programming required”.  I have to say that in my experience and opinion these are mostly scams.  The old adage of “if it seems too good to be true then it probably is” applies.  If their poker bot works why are they selling it and not using it?  If you’re really interested in botting then head to the Winholdem or Openholdem (an open source spinoff) communities.  Read the forums and you’ll find a passionate community who make bots for a hobby and will testify how hard it is to make a winning bot.

What is really noticable about both these communities is how close they are in content to the famous 2+2 forums.  Many botters come to a much deeper understanding of the game as they try to “teach” a computer to play it, “botting taught me to play poker” is a common sentiment.