Expert: Poker Bots 'Already Better' Than Many Players

In his new book, THE PERFECT BET: How Science and Math Are Taking the Luck Out of Gambling, author Adam Kucharski explores the long and complex relationship between science, math and gambling. The casino world has been a great “playground” for testing mathematical concepts, according to Kucharski.

Kucharski is a researcher at the London School of Hygiene & Tropical Medicine, an award‐winning science writer with a PhD in mathematics from the University of Cambridge and the winner of the 2012 Wellcome Trust Science Writing Prize.

According to him, the game of poker is not only an immensely complicated game rich in theoretical puzzles, but it is also one of the ultimate challenges for artificial intelligence. Computer scientists have long tried to build machines to conquer the card game, and in some sense they have or are very near doing so.

Card Player also had the chance to interview Kucharski about the luck vs. skill debate around daily fantasy sports and one that still continues around the game of poker.

Brian Pempus: In the United States there is a lot of discussion right now about whether DFS games are predominately skill or chance. What do you make of this ongoing debate?

Adam Kucharski: It is a really interesting debate around fantasy sports because sports betting has been very restricted in the U.S. outside of Nevada. In fantasy sports, you are in essence putting money behind your predictions being correct. Predominately that is skill. You can argue that you are putting your money into something that isn’t gambling, but how it is defined in the law can vary. There was an interesting case a few years ago involving Internet poker, operating in New York, although poker was specifically mentioned as a gambling game, as a game predominately of chance…but the outcome was that poker was a game predominately one of skill. I think it’s tempting to put something entirely to chance or skill, when it’s really much more of a spectrum. I think fantasy sports have made a lot of people realize that the line they drew between what is gambling and what isn’t gambling, what is skill and what is luck, isn’t necessarily where they expected it to be. It will be interesting going forward if this makes people think differently about sports betting as well.

BP: So, in your opinion this skill vs. luck debate sort of creates a false binary?

AK: In many ways there is a psychological bias as well. If something goes well you want to attribute it to skill, but if a decision goes badly you want to put that down to luck. I think that’s the same in many other games as well…In sports betting and fantasy sports you actually have a small group of people making all the money. If you look at their backgrounds they often have a lot of math and statistical or business knowledge.

BP: A very small number of customers on DFS sites also generate the majority of revenue. Can you talk about the importance for these companies to have these high-volume bettors and if they could be viewed as “automated gamblers” for these firms?

AK: Yeah, fantasy sports companies and bookmakers like gamblers who bet systematically, people who have any consistent betting patterns, even if it doesn’t work very well. I think a lot of the advertising is aimed toward that.

BP: With regards to poker, there are very similar debates about that game as well. A lot of lawmakers emphasize the fact that you can’t control which cards you are dealt in a given hand. Can you talk about how that’s a minor element to the game and how there is a lot of skill at subsequent points in the game?

AK: Definitely. With poker it depends on the time scale you are looking at. Of course with a single hand you don’t have any control over the two cards you are dealt. In many sports that is also the case. If you look at a single baseball pitch there’s an element of randomness. The question is whether, over a realistic duration of the game, skillful players usually come out on top. In poker, there is really a lot of evidence that this is the case. Over a plausible game or a tournament, the more skillful players will be making the money.

BP: Do you agree that it’s fair to say that over a large enough sample size of hands the cards break even? Everyone is receiving the same value of hands over the duration of a long enough session, but where the skill comes into play is how people play their range of hands.

AK: Yeah, deals are random in poker. How you handle those cards you are dealt is the difference. A study that tried to separate out the luck involved a man vs. machine match. In the game, each pro played the bot individually and swapped over the hands that were dealt. This was a way to try to control for the element of luck. In one game the humans got certain cards, and in the next game the bot got the same cards.

BP: In your recent article on the March Madness tournament, you mentioned a term called “path dependence.” Can you translate that concept over to poker?

AK: Path dependence is this concept that says the order in which you make decisions is important. For example, in March Madness if you get all your first round picks wrong then by definition you are going to get your subsequent picks wrong. The order in which you make your decisions is important, which is very similar in poker. One of the major examples is something like bankroll management. If you lose all your chips in the first round it doesn’t matter if you were going to play fantastically in later rounds because you have no money. I think really when these people use game theory to try to improve their strategy, one of the things they do is try to achieve this balance of making good decisions earlier on to get benefit but also in a way pacing their play so they can go through the whole tournament.

BP: With these poker-playing bots it has created this idea that you can “solve” poker. Can you talk about what it means to say you can solve a game and what implications that has for how humans play it in the future?

AK: A solution is much easier to think of if you have something like chess or checkers, a game where you have all the information in front of you. A solution is just a set of moves that if you played them you would limit, in the long-term, how much you would lose. In checkers, if both optimally play the game you will be expected to end in a draw. Poker is of course more difficult because you have the element of randomness involved. In heads-up limit poker it is possible to find a solution, in that there is a strategy that in the long-term, no matter what type of opponent you are playing, you won’t expect to lose any money compared to any other strategy. Of course, in competitive games this isn’t the full story. In poker, you might want a strategy that is very defensive, or if you are playing someone very weak you might want a more exploitative strategy that will take advantage of them. If you start to deviate from the defensive strategy it might leave you vulnerable to someone else. Poker is a very interesting game, because there is a big element in understanding what your opponent is doing…There is a lot of evidence that bots are getting much better year by year. In many cases, bots are already better than many of the people playing poker online. Poker sites have certainly clamped down on bots.

BP: In your book, you mention a lot gambling terms that the casual gambler might not know. Can you give a few really important gambling terms or concepts every gambler should read up on and be aware of?

AK: Sure. So, one thing that really stood out is this concept of bankroll management. One of the most well-known formulas for this is the Kelly criteria. This is a way of, if you have a certain edge in a bet and you have a certain amount of money, telling you how much money you should be putting on it. Obviously, you are putting money on a chance event so there is the possibility that even though you have the advantage in the bet you might not have any profit. This is a way of adjusting what you bet over time to maximize your profit and minimizing your chance of walking away with nothing. Another thing is testing. Say, you are picking a bracket. If you use a set of data to come up with a bracket that would have done best last year, you want to test that method on something that is outside the sample that you used to develop your method. You want to test it on unseen data sets. If you don’t do that, there is always the risk that you are explaining what happened last year very well, but it might not be a very predictive method.

Most Viewed Articles