By
Riley |
Published
Apr 20 2005, 09:28 AM
Bots exist. We'd like to ignore it, but we can't. They exist, and they're smarter than us. They can calculate pot odds and winning percentages faster and more accurately than we can count to ten, and they can tell us more about how we play than we can ourselves.
They're faster and they're smarter, but they have their weaknesses. I have studied the technology concerning the artificial intelligence of poker from screen scraping to bluffing strategies to opponent modeling, and I have even built a bot that was entered into the Poker Bot World Series held by Carnegie Mellon University. Although my particular program was dominated by others in the competition, I learned a lot about the possibilities and limitations of poker AI, and I would like to share some of those ideas for those of you who are overly anxious or worried. *
The threat of bots destroying the industry of online poker is minimal at best. Bots may be playing at all sorts of levels, high and low, but they are infiltrating less than .1% of the market (it is very unlikely a bot is sitting at your particular table), and the vast majority of those in commission can only outplay the most predictable and dumbest of opponents, employing techniques as primitive as selection of top tier starting hand requirements and calculation of post flop winning percentages.
The better ones are built with a much smarter intelligence: opponent modeling. The term may be more dangerous than it sounds. With a database of hand histories, sometimes with data collected from a network of sources, bots have the potential to be armed with an arsenal of information about how you play. There are even a few bots that have over 10,000 recorded hands for players who log enough daily hours! In particular, if a bot has the knowledge that you have a strong hand 80% of the time you lead out with a bet on the turn, it can easily fold when you make this particular move against it. Similarly, if a bot has information indicating that you check raise roughly 3/4x the pot 60% of the time you flop a nut flush draw, it will be more inclined to make you commit all your chips when it has a dominant hand.
With so much information about how people play, combined with lightning fast calculations regarding winning percentages, it may seem difficult (if not impossible) for these types of bots to lose. The (disappointing) answer is that they don’t. The fact is, there are so many predictable and dumb opponents thronging in online poker communities, any sensible algorithm, combined with even a small sample of opponent modeling, is sure to win. It’s the very same reason you, the average winning player, find yourself consistently in the black. People playing online poker are (in general) very, very bad players.
But despite the fact that bots exist, despite the fact that they are consistent winners, there's a good side to all this technology. By figuring out how to improve your game to outplay even the smartest of theoretical bots, you will be improving your game to outplay the toughest of opponents. So, without further ado, I give you:
3 Rules on How to Outplay a Bot (a.k.a. 3 rules for Good Poker)
Rule #1: Don’t Be Predictable.
Don’t always make the same play. When you have the nuts, try different techniques. Depending on your opponent and the hand you think he has, you may want to check raise, check call, bet big, or bet small. When you want to bluff, make similarly different moves. Any opponent will have a hard decision to make if 50% of the time you lead out with a bet on the turn you have the nuts, and the other 50% of the time you have complete trash.
Rule #2: Play your Opponents.
By playing your opponents more than your cards, inherently you will be making unpredictable moves. Knowing a particular player is a weak-loose calling station versus knowing a particular player is a weak-tight rock allows you to make very different plays given otherwise identical situations. Any opponent only looking at the statistics of how you play will be baffled by the fact that you folded middle pair in one situation and called someone down with the very same hand 10 minutes later.
Rule #3: Sculpt your Own Image.
Be aware of what others are thinking about how you play. If you are known to only show down monster hands, you should let people see your bluffs. Although many people think it’s an unwise decision to show your hand when you don’t have to, you should always do so when it counteracts your normal image. Offset the information that is known about how you play from your showdown hands by letting your opponents see the opposite when taking down a pot on the flop or turn. The best long-run image is to have no image. In doing so, you will force your opponents to play their cards when you, in turn, can play them.
As a general rule, if you make it difficult for strong opponents to make decisions against you, you will also be making it difficult for bots to make decisions against you. Play your best game and take money from the thousands of players that make this industry so profitable.
*If there proves to be enough interest in the topic, I will be happy to write a later article that describes (step-by-step) both cognitive modeling and the process bots use to calculate winning percentages and bluffing opportunities, and why certain situations (limit poker and heads up play as opposed to multi-table no limit) are more conducive to bot intelligence.