Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Big data and poker-playing bots are blurring the line between man and machine

kernelmag.dailydot.com

Post reply on HN
Print view View original

21–30 of 51 posts

Re: Big data and poker-playing bots are blurring the line between man and machine

#21

Post by vintermann »

doug1001 wrote: post #5

i worked in anti-fraud for a large online gaming house for about six years from 2003 to 2009 so my information is not current. At that time, there were no bots (whether in the wild or in academic, eg, University of Alberta) that could competitively play no-limit, likewise for tournaments (which account for a large portion of the total hands played in online poker Sites). Anyone know if this has changed?

Yes, game playing AI has progressed immensely since 2009, as evidenced by AlphaGo. I haven't looked at poker playing specifically, but I'm convinced that if there's a high-stakes game played online that doesn't have known superhuman bots, that's merely because the people who have made those bots make a fortune on keeping it secret.

Re: Big data and poker-playing bots are blurring the line between man and machine

#22

Post by frgewut »

lettergram wrote: post #14
skizm wrote: post #9

Earlier quoted context omitted.

As someone who works in anti-fraud, would you be able to detect someone who was playing manually, but had a bot set up to tell them what to do (and then they would manually input the move)? Is that even illegal or against the rules?

You should be able to tell. It's fairly straight forward outlier detection; if someone is doing 20% better than anyone you've seen before you monitor them. If they continue to do better constantly you manually look, then you can either make a choice manually or have a system auto remove them. They may even remove you just because you are costing them too much money (depending how the game is setup).

It should be easy to artificially lower bot's advantage to 1-5%. It still would be profitable, but wouldn't stand out among other players.

Re: Big data and poker-playing bots are blurring the line between man and machine

#23

Post by stevoski »

arisAlexis wrote: post #7

There was a big scandal 3-4 years ago with heads up nl chinese bot ring that got 1.5m approximately before getting shutdown.

That was on PokerStars' single table "double or nothing" sit&goes if I recall. As a consequence , PokerStars has changed the pay-out on such tournaments. Instead of all five winners getting the same amount, the prize pool is split between the last five remaining players based on their stack size (number of chips) when the sixth placed person is knocked out.

Re: Big data and poker-playing bots are blurring the line between man and machine

#24

Post by stevoski »

doug1001 wrote: post #5

i worked in anti-fraud for a large online gaming house for about six years from 2003 to 2009 so my information is not current. At that time, there were no bots (whether in the wild or in academic, eg, University of Alberta) that could competitively play no-limit, likewise for tournaments (which account for a large portion of the total hands played in online poker Sites). Anyone know if this has changed?

Limit Hold'em Heads-up is 'solved'. Over the long run, the AI will break even no matter who the opponent is (rake excluded, of course).

IIRC this involved making precalculations for the entire sample space, which is massive, but tiny compared to no-limit full-ring.

Re: Big data and poker-playing bots are blurring the line between man and machine

#25

Post by lordnacho »

I know a guy who was talking about the roulette prediction thing. He was an MIT guy, but a few years ago smartphones weren't that sophisticated and besides you'd have issues using them at a casino.

Rough plan:

- You can identify both the ring of the roulette table (transforms into an ellipse), the ball (shiny), and the green zero on the spinner (it's different from red and black).

- You can use a Kalman filter to guess the current location of the ball while updating your guesses, especially when the ball goes out of view. Same for the spinner phase.

- You can calibrate how fast the ball's orbit descends from a given speed.

- You know where the diamond shaped thangs (the things that perturb the path of ball right before it falls onto the spinner) are. There's a fixed number of them, evenly spaced. From the orbit decay you know which one it will strike.

- From the speed of the spinner, you know roughly where the ball will be when it arrives on the numbers.

- From experiments, you have a rough probability distribution of where the thang will distribute the ball.

- You don't need to be right every time. There's 33/34 numbers, and the house's margin is only one or two of those numbers (there's variants of roulette tables). Being able to rule out an equivalent amount of numbers tips you over the margin.

- Use Kelly criterion to maximise your equity. With finance you'd be crazy to use full Kelly, but I guess a roulette table doesn't have regime changes.

Re: Big data and poker-playing bots are blurring the line between man and machine

#26

Post by fabriceleal »

jaaames wrote: post #16

Former professional poker player here, no limit holdem cash games, mid stakes. NLHE cash games are considered the hardest to solve given the depth of the decision trees. Other formats like sit and gos, spin and gos, and tournaments can be modeled and solved in a more simplistic way. A group of us (cash game pros) discovered a group of players on the Ongame network around 2010. We combined our databases and found a fe…

> They made very annoying, aggressive plays that were difficult to counter, not taught anywhere in professional videos, forums or training sites

Mind sharing more on these strategies? I'm curious about what sort of strategy a poker pro would find so unique as to not recognize it from anywhere

Re: Big data and poker-playing bots are blurring the line between man and machine

#28

Post by doug1001 »

skizm wrote: post #9
doug1001 wrote: post #5

i worked in anti-fraud for a large online gaming house for about six years from 2003 to 2009 so my information is not current. At that time, there were no bots (whether in the wild or in academic, eg, University of Alberta) that could competitively play no-limit, likewise for tournaments (which account for a large portion of the total hands played in online poker Sites). Anyone know if this has changed?

As someone who works in anti-fraud, would you be able to detect someone who was playing manually, but had a bot set up to tell them what to do (and then they would manually input the move)? Is that even illegal or against the rules?

great questions; here's where we drew the line (our Terms of Service (TOS) were fairly standard): it was within the TOS to use a "poker assistant"--an app that worked with the Site client (eg, PokerStars, or in my case, bwin.party)--provided that this app was populated with hand histories that you have personally generated against the opponents you are playing against. In other words, you cannot mine hand histories in bulk, feed them to your "assistant" and then let it tell you how to play.

keep in mind also that the whales in online poker are not high stakes players but "grinders"--folks who quit their day job to play a very disciplined style on low-stakes limit tables comprised mostly of beginners, and which requires multi-tabling (maybe 6 - 10 tables at once) 8 hours/day. So against that background, a bot playing 10 tables simultaneously for 10 hours each day, from a distance doesn't look so odd.

Re: Big data and poker-playing bots are blurring the line between man and machine

#29

Post by jaaames »

jsprogrammer wrote: post #19
ec109685 wrote: post #18

Earlier quoted context omitted.

He said it: their strategies were very similar and never seen before, nor taught anywhere.

So, some people got together, or one person was playing many accounts, and cleaned the amateurs out. Similar, novel strategies does not imply a bot net.

I'm glad you're playing devil's advocate, and you're not wrong in that a single player could have multiple accounts, this was especially true on the Ongame network, as it provided the platform which sites could whitelabel to run their own poker room, with individual promotions/branding etc, while leveraging the benefits of a large shared player pool.

The thing that stood out was the length of the sessions played by these players. The regular professionals (regs) would play multiple tables at once, usually 4-8 tables (Ongame software was shitty and prone to crashing) for a few hours at a time (poker is mentally taxing). These guys were playing 8-12 sessions on single tables.

In terms of amateurs, I'm not taking offence, but many of us were playing full time for a living at the time, and their winrates were beating many of us, without any game selection (explicitly choosing to play with weak players).

The strategies were indeed novel - the one I can describe in layman's terms - they never just called a bet before the flop, they always folded or reraised. This is a huge paradigm shift from all conventional wisdom at the time, and a strategy that is very difficult to execute well (hard to go into details without getting into poker terms/theory).

While poker strategy was (and still is) constantly evolving the overall stats/strategies deviated so far from the norm that we all agreed we were up against a bot. Again, without going into details the strategy had a much more mathematical/game theory approach than the far more commonplace exploitative strategies that everyone was using at the time.

Re: Big data and poker-playing bots are blurring the line between man and machine

#30

Post by DanBC »

lordnacho wrote: post #25

I know a guy who was talking about the roulette prediction thing. He was an MIT guy, but a few years ago smartphones weren't that sophisticated and besides you'd have issues using them at a casino. Rough plan: - You can identify both the ring of the roulette table (transforms into an ellipse), the ball (shiny), and the green zero on the spinner (it's different from red and black). - You can use a Kalman filter to gue…

The Newtonian Casino talks about some students who designed a predicting computer, and smuggled it into a casino in their shoe. It was published in 1985, so anything they built would have been considerably less powerful than a smart phone, even one that several years old.

(It's a great book, and probably available).

https://www.amazon.co.uk/Newtonian-Casino-Penguin-Press-Scie...

One interesting part of the book was that croupiers knew the wheels so well that they were able to spin them, and throw the ball to land predictably.

EDIT: In the US it's called "The Eudaemonic Pie" https://en.wikipedia.org/wiki/The_Eudaemonic_Pie

> The book focuses on a group of University of California, Santa Cruz, physics graduate students (known as the Eudaemons) who in the late 1970s and early 1980s designed and employed miniaturized computers, hidden in specially modified platform soled shoes, to help predict the outcome of casino roulette games. The players knew, presumably from the earlier work of Shannon and Thorp,[1][2] that roulette wheels obey Newtonian physics, and that by capturing the state of the ball and wheel and taking into account peculiarities of the particular wheels being played they could increase their odds of selecting a winning number to gain a 44 percent advantage over the casinos.[3]

Post reply on HN