Luck, Bayesian Inference, and Poker Skill

For those who aren't familiar, Bayesian inference or Bayesian updating refers to the process of adjusting probabilities based on new observations. I think most people are probably Bayesian updaters, and I think most of us are familiar examples of Bayesian updating gone wrong. One that immediately comes to mind has to do with Sept. 11's terrorist attack (or staged terrorist attack - whatever you prefer - let's avoid debating details )

The chance of a terrorist attack anywhere near you is probably like one in a million, but if you lived in the USA after Sept. 11 you witnessed people who thought the probability of a terrorist attack was pretty much a coin flip on any given day. Most of these people had never given terrorism any thought at all prior to Sept. 11, but the magnitude of that day changed their minds completely. I lived in Alabama at the time, so what I saw was probably worse than most... For instance, people were trying to seal their houses with duct tape to prevent airborne biological attacks - like they didn't realize they would suffocate if they were successful...

So to get to the point: I was thinking about a particular possibility concerning luck in poker: the idea that a player with a lot of potential might never become interested in the game simply because he ran bad when he was starting. I've also read some arguments analyzing high-stakes players. One idea that stuck out was that "feel-based" or "non-mathematical" players had to run good through the micros and low stakes where a more patient, mathematical, nut-peddling style of poker was favored. I'm sure there's a grain of truth in this, and it was while I was thinking about these two instances of luck in poker that I thought about an application of Bayesian updating.

I realized that there might be a separate and somewhat subtle way in which luck determines the development of poker skill. I'm referring to the idea of Bayesian updating gone wrong as in the Sept. 11 case. I was thinking that a player who sees a large, skewed sample at the beginning of his poker career may have his instincts completely fucked up. Likewise, even a player who sees a small but intensely skewed sample at the beginning of his career may develop really misguided instincts concerning the game.

I think most people develop a strong instincts pretty fast - definitely in well less than 10k hands. So it's easy to imagine that many if not most players develop skewed instincts of varying degrees. Of course once these instincts are created it's much more difficult to adapt them, and even a strong, fundamental knowledge of statistics is probably not enough for most people to avoid falling victim to Bayesian updating. Often decisions are influenced or even controlled by the subconscious so it is reasonable to imagine even a wary, informed poker player could be led astray such that leaks generated by a skewed sample of observations could take a huge hand sample to correct.

Was wondering if some of the players here had an opinion on this topic. Obv LP is a pretty skewed sample in its own right Prob composed of people who strictly run good despite what the blogs say ^.^

Being good at poker is having enough experience to realize when bayesian interference is for the weak minded. Having enough experience to realize what's correct and the ability to overcome every emotion blocking you from achieving the correct play.

It's very clear, however, that the early days of a poker player's career have much basis on their successes long term. Someone who runs bad and is playing well may just give up the game because he thinks he has no edge. Whereas someone like myself with a huge ego and good success early on made me believe i was a better player than I ever was.

People learning to play poorly due to running unusually is a case of updating badly. Bayesian updating is the solution, not the problem. The difficulty is that humans are really poor at explicit Bayesian reasoning, so universally we rely on the approximations and heuristics that are the way our brains actually work. You could argue that the people whose thought processes best approximate true Bayesian updating are the people who succeed (besides the life-heater-blessed), but I don't see much practical application down that line of thought.

I expect a lot of people who would have succeeded at poker ran bad in the beginning and never came back. Lots of people ran good and thought they were god's gift (Marshall28 for a local example). Bayesian reasoning doesn't seem especially more relevant to that than to anything else in life.

I think you are right, there are many good players out there who quit because they started on a bad run, also many who went broke later when the good run stopped. The survivors (making money) are those who adapted correctly to the bayesian updating and didnt start out on a bad run.

of course this has probably happened a lot in history. luckily for the majority of this forum, we started from 2004-2007 and knew what bankroll management was so starting at the lowest limits with a decent fundamental understanding of the game meant you probably have a 10ptbb+ winrate even just starting out at the lowest possible limit, so losing over any worthwhile sample size was super unlikely.

but for those with potential starting in tougher times, tougher games, etc, it may have been much more difficult to learn a style they could be confident was a winning one. there was a good post on 2p2 by strassa about something along these lines a few years ago