Hacker News new | past | comments | ask | show | jobs | submit login
Matt Maroon: Why I Quit Playing Poker For A Living (thepokerchronicles.com)
44 points by rms on Dec 3, 2007 | hide | past | web | favorite | 60 comments



Matt's implication that poker is more complex than chess is entirely unfounded. Computer scientists have been trying to create the best chess programs since the days of the first computer as a test of artificial intelligence. Poker has been ignored because a computer can easily figure out the odds (figuring out the strategy is the hard part), and it was never as popular as it is now. I would expect that a computer will be the best poker player in the world in a few years, one is already close now. http://www.news.com/8301-10784_3-9750210-7.html

Phil Laak (one of the best) didn't start playing poker seriously until he was 25. Contrast this with chess, where if you didn't start studying by the age of ten for hours every day, you don't have a chance at being the best.


You are very incorrect. Here's where you're wrong:

1. Phil Laak is not one of the best poker players. Not even top 100. The fact that you think he is shows how little you know about poker. Also, I don't even know who Ali Esmali is.

2. The game (which I think was No Limit Hold'em) played was the simplest form of poker for a machine. Even if one became better at this than humans (which probably is possible in the not too distant future) it will be far from beating a 2-7 Triple Draw player.

3. Figuring out odds is trivial, but virtually useless. Figuring out odds as well as a human, who has the ability to put his opponent on a range of holdings, is not.

4. No poker bot ever made could come anywhere close to me in an 8 man, limit hold'em match, and there are far better players than me.

5. I don't know if chess has had more man hours dedicated to it. It probably has. But Polaris has been worked on for 26 years. That's certainly not insignificant. Phil Laak has probably been playing for 10 or so.

6. I'm a mediocre programmer at best but could easily write a chess program that, with infinite processing power, would be perfect. (I'm not sure if it would be unbeatable, as it is possible that with perfect play one side will always win.) I could not do that with poker.

7. Chess may be harder to learn for a human (though even there I'm not sure) but that doesn't mean it is also harder for a machine. Calculating the 18th power of 2 is harder for a human than picking out an apple from a group of pictures. Not the same for my PC.


6. I'm a mediocre programmer at best but could easily write a chess program that, with infinite processing power, would be perfect.

Yeah, IF you had infinite processing power, it would be a nearly trivial exhaustive search program. IF I had a way to read my opponents emotions, etc like humans can, then I could write a fairly trivial program to do well at poker. But those are two very big "IF"s.

The problem is the two games are very different. Chess is a game of perfect information, where both players know the entire state of the game. Poker is not, plus it has an element of chance.

You can't really compare the two.


I'd say you can in terms of how knowledge plays a role in the game. In chess, the computer can calculate the best move given a situation. It doesn't need to know anything about the opponents strategy. In poker, it is pretty much the opposite. Your move depends partially on purely deterministic criteria, what people likely have, but more so on what you think everyone else thinks. Here you enter a realm I'd say computers are inherently bad at.

This, I think, is what mattmaroon is getting at.


What I meant was you can't really compare the "easiness" of the two for humans based how well a computer would do. Computers are theoretically good at chess, and theoretically bad at poker, but that says nothing about how easy or hard the two are for humans.

You can't say that A is easier than B for humans just because a computer can do A well but not B. For example, computers are excellent at doing large calculations very fast but terrible at recognizing objects in a scene visually, but humans are the exact opposite.


> You can't really compare the two

I'll still try. Both games have a philosophical and a computational side. The chess philosopher could ask questions such as

"everyone knows that dominating the centre is an advantage, but WHY EXACTLY is that the case"

But even if he understands this better than others, it won't help him that much because chess games are decided on the computational side. This is why you have to start at an early age in order to become a great chess player. And it suits computers.

Poker's computational side includes, for example, odds (trivial) and ICM. In most forms of poker, though, it is far outweighed by the philosophical side, at least as of today. The poker philosopher could ask questions such as

"everyone knows that being in position is an advantage, but WHY EXACTLY is that the case"

If he's willing to work hard on his understanding of the game, he might, in theory, learn the rules at any age and become the world's best poker player in just a few years. Getting direct monetary value from one's philosophical insights is an attractive proposition. Unfortunately there are big psychological drawbacks, which Matt Maroon describes so well.


I am not the parent, but:

1. Subjective, with a nice insult thrown in there.

2. Do you really think the simplest form of poker for a bot would be a No Limit game?

3. Are you implying poker bots don't take opponents possible hands into account? wrong

5. No reason to cast doubt, chess has had much more effort put into it. Polaris has been in the works for 17 years, not 26. Compared to over 50 years for chess engines.

6. I don't think writing any chess engine is easy, but your main point, yes chess is a finite game. A computer isn't expected to win every time at rock, paper, scissors, so why poker? Like blackjack, there is a certain amount of available information, and poker bots can make the best play off that.

7. Amateurs with 1 year of poker under their belt win a WSOP bracelet.


2. Heads up no limit, yes. A player who simply pushed all-in every hand would only lose something like 45% against another player using optimal strategy, and would actually be a long term winner against more players than you'd think. I could write that bot.

3. They may, but they do a very poor job of it.

6. A computer can easily tie at roshambo, and the roshambot beats most humans in the long run. And poker is much different and far more complex than blackjack. The two arent comparable at all.

7. Irrelevant. I never contended that a bot couldn't win a tournament. They probably have. That's not really a good measure of ability.


I have actually programmed game AI and I can say none of your arguments is convincing. It is a very bold claim you are making, with very little to support it.


indeed.


Chess and poker are fundamentally different problems. Unfortunately this argument is going to come down to arguing about the definition of "complex" and things like that.

But I will still reiterate these points:

1. A great deal more effort has gone into computer chess than computer poker. When that amount of effort is put into computer poker, computers will be better than humans at poker.

2. The argument can be made that poker is harder or more complex than chess for computers, but the opposite is definitely true of humans today. If you don't study chess hard from the age of ten and then dedicate your life to it starting in your teens, you can't be the best. The amount of study (not playing) is probably an order of magnitude difference. In practice this means that any great chess player could switch to poker, study hard, play hard, and have a chance at making it, but no great poker players could switch to chess.

Fundamentally, sports in which real money is gambled will probably never reach the competitive level of play of non-gambling sports. Chess has a rating system and is setup so that the players of similar calibar play each other. Contrast this with poker where professionals are always actively seeking to play against worse players (with higher bank rolls).


1. I disagree, as do the few people I've met who work on it. 2. The dedication required is, in large part, due to the number of people who study the game, social attitudes towards it, and the wealth of information available. It has nothing whatsoever to do with complexity.

Poker is far more competitive than chess. Show me a chess game that runs every night with buy-ins in the hundreds of thousands.


The poker bots out there mostly suck because they don't need to be great to win lots of money from the weak players. The real problem the bots face is getting around anti-bot security measures.

But, although almost no effort has been put into poker bots, they're already pretty damn close:

http://cosmiclog.msnbc.msn.com/archive/2007/07/25/289607.asp...

If poker bots analyzed a massive database of how pros play like humans or the chess bots do then I'm certain that the poker bots would blow you away.

> No poker bot ever made could come anywhere close to me in...

You'll be seeing this quote again:)


>The poker bots out there mostly suck because they don't need to be great to win lots of money from the weak players.

That's dumb. They'd win 5x the money if they were great.

>You'll be seeing this quote again:)

I meant made yet, maybe I should have specified. I've no doubt that computers will one day overtake humans. It just is nowhere near having happened yet.


The poker bots right now are also only good at heads up.


2. The game was limit holdem, which is easier for a computer to solve than no-limit.


I'm pretty sure it's much harder when heads up.


Limit is definitely easier to solve in a standard ring game. Why would limit be more difficult to solve than no-limit heads up?


Chess doesn't even need processing power in theory, only a very large database with state lookup in the big book of chess.

There is a nice proof in game theory that any turn based game with 2 players, limited number of board positions, and finite max length is subject to the creation of the perfect game book - basically a simple table of board positions and the ultimate correct decision for every board position. using it you are guaranteed to win/draw optimally (depends on first mover advantage) against any player.

You just can't create a "big book of poker" it doesn't meet the requirements.

PS: Go is also subject to this treatment despite being several orders of magnitude more complex than chess, it is still a finite 2 player turn based game.


Why is it unfounded that "poker is more complex than chess"? Chess lacks chance and bluffing. There is perfect knowledge of the state.

The long apprenticeship in chess doesn't mean chess is inherently tougher for computers than poker. It means for humans, chess is a difficult game that requires a lot of training. Humans would take a long time to memorize and incorporate book knowledge. Computers could suck in that knowledge in no time.

Babies acquire visual recognition and speech in a smaller time period than adults go from novice to grand masters in chess. So it depends on what you are learning, not just the time it takes.

It's also interesting that the computer vs human poker contest tried to remove "luck":

"The man vs. machine poker contest was designed to eliminate the luck factor by dealing the same cards in each hand, but in reverse. So in one hand, Laak might get lucky with two aces, but in the other room Esmali would be unlucky as the computer was dealt two aces. To finalize a winner, officials added the two human scores and two computer scores separately, and the highest number won."


Also, interestingly enough, since poker is all about optimization, it would still be hard to play perfectly even if you had perfect knowledge of the state. Even sometimes knowing what someone had, it's hard to decide the perfect play.


Chess lacks chance compared to blackjack too. Does it means chess is simpler than blackjack?


No it doesn't mean that, blackjack too has an even smaller number of knowledge state (based on cards already dealt/hand/dealers card). There is a finite and small number of cards dealt and in a shoe and simple probabilistic calculation can be made to determine the absolutely correct action.

Poker is not finite, at least the wagering part of it, the interaction between the players. Even in the exact same situation as defined by the mechanics of the round, the correct answer might be different, depending on a host of psychological factors.


No. Try an analogy with chance + bluffing + non-fixed, unlimited betting + imperfect information. Better yet, try to analyze poker.

My argument is not that EVERY game with chance + bluffing is more complex for a computer than chess. It's that poker is a very different beast than chess, and it makes no sense to argue complexity based on amount of required human training. Otherwise, we'd have computers with visual perception before chess skills.


Cool idea for removing the luck, but I think it has no bearing on the computer's strategy, only on the analysis of the results by the human observers.


That is a novel way to attempt to do so (though I still think is far from absolutely successful). Most previous attempts dealt only with sample size.


you are incorrect imo

chess computers are better than people because chess computers brute force the problem, they can think millions of moves in advance and plan out every possible game path, a human can only think several moves in advance, so huge advantage for the computer.

in poker, there are only a handful of cards in play, only a few decision spots per hand, a pro poker player could probably program a computer to be as good as he is by giving it the rules which he plays by, but it is unlikely a computer will ever surpass the talent of the best players imo


A few minor corrections. First, most chess programs don't literally "brute force" the problem. A lot of work goes into pruning the search tree and only looking along the most promising paths. Second, humans can actually think many moves ahead along certain paths.

As other's have stated in this thread chess is particularly well suited to be played well by a computer program, but that does not make it trivial.


They don't brute force only because processor power isn't sufficient to do so. Eventually they will cease to prune and will be unbeatable.

I'd never say chess is trivial, it's just far less complex than poker.


> Eventually they will cease to prune and will be unbeatable.

No, they won't. The universe isn't big enough for a computer that would play brute force in a reasonable amount of time.


Computers can brute force poker too. The way computers and humans play chess is also very different, although there is superficial similarity in that they both enumerate future moves.


Here's an honest question. Why isn't more research being done into solving Poker compared to the amount put into Go or Chess?

> It seems to me that it is an interesting question that combines elements of computation, AI, and psychology.

> One can see instant practical results by playing your bot for real money.

> Poker is a fun game. :)

Any answers would be appreciated, especially as I'm thinking about doing a PhD and am currently exploring open problem domains.


There is quite a bit of research being done about poker in the last few years. In fact poker related research has got best paper awards at a couple of the major AI conferences in recent years.

Check out http://poker.cs.ualberta.ca/. They're probably the leaders in this field.


Chess has been researched much, much more than Go.

In research, you are looking for problems that can be solved in isolation, not ones that combine many diverse topics.

Perhaps poker is not researched more because gambling is tainted negatively by the society.


Figuring out pot odds is an elementary part of poker. Most who play poker well can do that correctly, that is just as well as a computer. On the other hand combining those odds with an estimate of what other players and how they may react is a damn interesting problem for a human player or a computer programmer.

Isn't strategy the hard part of chess after all?


It's widely accepted in AI circles that Poker is a much more difficult to solve than chess. The main reason is that chess is a game of perfect information whereas Poker is a game of imperfect or partial information. It's much more difficult to find an optimal (or even a consistently successful) strategy in Poker.


No matter how well you do something, sooner or later you reach the point where you question the value you are producing for others.

This sorta reminds me of the guy who got paid a fortune to dig holes in the beach, only to see them filled in by the ocean each night. No matter; he just got paid to dig them again the next day. Wasn't long before he thought, "This doesn't make any sense - I wanna build something that lasts and that others can use."

I wonder how big a role this kind of thinking played. The fact that you're part of ycombinator suggests that it's > 0.


Little to none, though I'll cover that in a later part.


Read the Myth of Sisyphus by Camus...


I was hoping for a few juicy poker details, but, that aside, thanks for the very honest and engaging post.


There are years of archives if you want to read about poker. Matt's blog was the only poker blog I ever got into. I think I found it from Fark.


What sort?


I've always been curious as to whether the whole reading people thing was exaggerated for the tourists. I'm sure poker strategy is a lot about trying to gauge what an opponent is doing, but does it really come down to people scratching their noses funny?

I used to play soccer quite a lot, and playing well is a lot about knowing whether both your players and their players are going to go. The thing is though that it has nothing to do with explicit tells - its all about the geometry, the state of the game and how they are playing.

I've always though past a certain level there would just be no way a poker player was going to reliably give themselves away.


Reading people is what poker is all about, but tells (mannerisms) are overly emphasized to a large degree in the media. They really aren't much a of a factor once you move beyond total amateurs, though at the same time, a relatively large % of the players you run into at the WSOP main event are playing their first live tournament anymore.

Really, though, it's more about observing how your opponent plays (both long term and short) and putting him on a range of holdings.


You know, like, "I was playing 30-60 at the Bellagio, and I got cleaned out..." Something along those lines.


sounds a lot like trading.


the cool thing about poker v trading though is you need solid 6 figure bankroll to make a stable living trading, you can start poker with a few K or less


There are still some sites that will give you a free $50 to start playing. I made more money exploiting affiliate schemes like that than I did playing poker. For a while, Party Poker required a $50 minimum to open an account and paid out $65 for referred account + around $25 in bonuses + free polo shirt, hat, and jacket. So I bought a Fark ad and gave out a bunch of Party Poker accounts.

Party Poker wouldn't pay me my money for 6 months, but they got a new affiliate manager who eventually cut me a check. I wonder if they cut their losses or if my players actually generated enough rake to pay me back.


He he. Some time I'll tell you the story of why I would be a millionaire many times over by now if Neteller had not stopped accepting Americans. And why, after they did, I ended up driving home from Canada through a blizzard.

And why I own a whole bunch of cheap Dell PCs and registered copies of VMWare.


I had a similar setup back in the day, and I think you did just tell the story.


Online you could turn a $2k bankroll into six figures in a year, and it's actually fairly easy.


It's only easy if you have an extraordinary amount of self-discipline. I couldn't do it, though I am sure many here could if they really worked at it.


True.


Please show me how..I'll be eternally grateful


fairly easy? How do I go about doing this?

Edit: Never mind - realized this does not scale.


The advice comes down to "Start at a low limit. Win consistently. Move up to a higher limit. Win consistently. Move up to a higher limit. Win consistently."

It's the winning consistently that you have to worry about.

Also, if you want to start playing poker using someone else's money, I would recommend these guys -- http://www.pokersourceonline.com/free-poker/money. They give you free money at one of the poker rooms.


It's pretty easy really, and nothing secret. Read good books. Practice. Discuss with friends. Start off low, build your way up using proper bankroll management (which is found in said books).


The hardest thing I found was building up my way with proper bankroll management.

I played too high of limits for my bankroll, so I would do well but my negative swings would deplete my whole bankroll.

That, and I found myself cashing out because I was in school at the time.


Would you care to elaborate? Which books are good? What kind of practice? I'm sure many here are interested.

Thanks.


You can see my list of recommended books on my blog. They really are the books I endorse.




Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: