Bot complaint

[adamstewart]

[ QUOTE ]
So Sklansky writes that you can earn 1.5-2 BB/hr at 20/40 (whats that, 50 hands and hour at best?) Yet it is unbelievable that I can win 4bb/100 at 2/4.

I wonder exactly what DS and NPA meant when then said you should be crushing the low limit games and that solid player can make $50k a year at low limit. I didnt think they meant you needed to play 120 hours a week.

It's called MULTI-TABLING.


[/ QUOTE ]


By the way... your 12K "sample size" combined with your claim to be "beating $2/$4 for 4.7BB/100" discredits all of your other posts.

I, nor anyone else, will currently convince you of your rationale errors.

Threfore, save this thread Play 50,000 more hands at 2/4... then re-read your posts. As a bonus, you may even want to post your win-rate after 50,000 hands.

Later,

Adam

[bpb]

[ QUOTE ]
[ QUOTE ]
then the argument that a good-playing bot isn't possible no longer would hold water

[/ QUOTE ]

There is no argument. I am a software engineer and I have a 4.7BB/100 win rate at Party's 2/4 games. I do not doubt for a second that I could program a bot to beat the game. If it wasnt for my conscience, and lack of free time due to a full time job and taking masters courses in engineering, I would be writing one. Heck, you can download half the source code to begin with from SourceForge.

The argument that the best players could outplay a bot is irrelavant if they don't know its a bot to begin with. Knowing a player is a "bot" and is programmed to act "correctly" is far different then assuming the oppenent is human and prone to tilt, mistakes etc... Its further irrelavent because its not the bot's intention to challenge world class players to begin with. It may be up against one or two "better" players, but its still getting its share of the other eight. This, IMO is the biggest threat the bots propose. They are slowly and artificially bleeding the games and diluting your own bottom line.

Think of how "intelligent" a bot can be when it can has 10k hands youve played in its database. Simply logic structures can be coded to take action based on the bot's oppenents VPIP, agression factor, bluffs on the river % etc...

Regardless of all that, from what I've read, Winholdem was beating the game. And it wasn't even engineered worth a [censored]. Imagine Chris Ferguson using his PHD in Comp Sci and Artificial Intelligence and his world champ of poker expertise in programming some of the logic for Winholdem or his own "personal" application. Its just denial for anyone to think that there aren't simialarly talented people already doing this.

As far as Party being able to detect this software, impossible. Unless of course, you promote it, try to sell it and make the Poker room aware of its existence, they would never know youre running one in the same way the cable company can't detect youre getting HBO for free.

[/ QUOTE ]

Why, after reading this post, is everyone getting bogged down in the "you aren't a proven long term winner at 2/4 because you've only played 12K hands." He makes several valid arguments. Winning low-limits at party isn't brain surgery, and anyone who thinks that it takes more than ABC poker to pull off 1-2 BB/100 is seriously overestimating what it takes to beat that game.

To summarize:

ABC poker can be described by a rules based system.
ABC poker wins money at party low limits.
A bot can be constructed to follow a rules based system.
A bot can play ABC poker.
A bot can win money.

[Bytestream]

Thanks for your input, but I don't require you or anyone else to convince me of my rationale errors. I have a thorough education and understanding of stats and probability. I am well aware of the confidence interval of my "sample size" and the meaningfullness and interpretation of the data.

It is only PokerTracker that I am new to, I only began using it about 3 months ago, hence the relatively small sample size. However, I have been playing online poker recreationally for over 3 years and the amount of money I have won during the previous 3 months while collecting the data with PokerTracker has been fairly consistent with my previous earnings. I have plenty of experience with the game and recognizing when I have been running either good or bad or neither.

I am certaintly aware of possible errors in this rationale as well. However, I am making the conjecture that my winrate will continue to exceed 3bb/hr over the next 38K hands as well. We are not talking about Vegas 20/40, this is Party 2/4, the majority of players are absolutely horrendous** and such winrates shouldn't be so unbelievable.
I used to play in a home game on campus a few years back when I started to take the game seriously and picked up my first poker book (Thursday Night Poker, Steiner?). After reading that and some others, I NEVER lost a single night for the next year and half. The players were simply "that bad" because it was all just beer, luck, fun and gambling. While the players at Party 2/4, are ironically, "better" they are not very much so. Yes, I do have my fair share of losing sessions at this limit but it doesnt take much to beat the game. You can certaintly expect to achieve a much higher winrate then you would expect at 20/40 or above.


** Only 40% are winners in my db, again I am aware of the small sample size, however I am interested to see how this number changes with a much larger sample size. Im sure it is still below the majority if anyone cares to post thier own data)

[null]

[ QUOTE ]
Bots do play on Party, and I'm not talking about WinHoldEm either, that is a steaming turd of a bot. The problem is that Party cannot admit to the problem as it would be a PR nightmare. Think how much money Party would lose by admitting that Poker AI exists on their site. It's like when a bank gets hacked into, it's much easier to just keep it quiet than to prosecute those who have gotten away with it. The most hillarious part of bot discussions however, is posts like these which assume a bot would be detectable due to obvious errors in game play. This is not true.

Edited: Just wanted to throw one other thing out there. The bots being used are not some kid in his basement either. We are talking large dollar amounts in profts and extortion for silence.

[/ QUOTE ]

Do you have any proof? Just asking because your post sounds kind of tinfoil hat-ish.

[Rudbaeck]

[ QUOTE ]
The chances of there currently being winning bots in play are probably 90%.

The chance that there is any bot at all that could master NL is 0%.

The AI required for a winning bot at limit is inevitable. There is currently a Heads-Up bot that is said to be nearly unbeatable.

Programming a bot for NL is exponetially more difficult.

[/ QUOTE ]

Uhm, almost everything that makes live nl more complex than live limit is gone online. The reason so many feel it's more complex is because many of the decisions are based on unquantifiable data. Tells are way more important for example. There are basically no online tells.

The math of nl is even simpler than limit, and frankly so is much of the decision making. Bluffing and snapping off bluffs can be perfectly worked out using game theory, and in fact must be done for both limit and no limit bots.

The decision tree for what a limit bot should do in a large pot with many contestants is more complex than that for a no limit bot actually. Bot is SB, has tptk in a 11sb pot on a non coordinated flop, there are five people left to act. A bet doesn't protect his hand, it's possible even a check raise can't protect his hand, but there is most likely a way to protect his hand, if nothing else on the turn.

You make more decisions that concern a big part of your stack, but they aren't necessarily harder to make. And the advantage you gain from an understanding of the 'art' of no limit might well be made up by the bot never ever making an error in the 'science' part. Even a minor steam every 10-12 hours on your part might be enought to make up the difference, or if you multitable and ever mistake two opponents and make the correct play against a calling station against a tight-aggressive player.

Tracking opponents style and varying your play based on that is more important in nl, but it's still very important in limit, and can probably make up for a leap from 1BB/100 to 3BB/100 once you got the ABC part down. Has to be done for both.

[adamstewart]

Bytestream,

You come off as an intelligent-sounding poster. However, your story begins to degenerate....

First, you claim a 4.7 BB/100 winrate at Party $2/$4. Then you offer a 12,000 sample size as support (and claim to have the education to understand the implications of such a small sample size, yet continue to believe that this is actually representative of your longterm winrate).

I don't disagree, though, that you are a statistical wizard. Not everyone can magically turn a 2.7 BB/100 winrate over a 12,000 sample size, into a 4.7 BB/100 winrate over 12,000 - OVER THE COURSE OF 10 DAYS.

For those who are wondering what it is I am talking about, I refer you here.

Only time (and 50,000 to 100,000 more hands) will tell. I look forward to hearing from you in a month or two.

Adam

[crockett]

I have a few things I'd like to add:

1. If your posting replies along the lines of "Programming a bot to beat Poker would be nearly impossible"

Just Stop. Your making a fool of yourself. Some people just have no concept of the power of computers to make quick repitive decisions based on INCREDIBLE amounts of information.

2. If your posting replies along the lines that "Chess is easier for a computer to beat because it's a game of complete information" or "Poker is not a simple game (in comparison to Chess)"

Just Stop! Your making a fool of yourself.

Look, I don't care what you call Chess; either a game of complete information or not a game of complete information, but to say Poker is one and Chess is the other is foolish.

In Poker I DO KNOW YOUR TWO HOLE CARDS. It is 1 of 168 possiblities (I hold one of the 169 possiblities). After the flop I narrow it further, the turn even further, etc.

In Chess I DO KNOW YOUR NEXT MOVE. It 1 of X moves based on the current layout of the board.

To a computer both these scenarios are a game of COMPLETE information. It has everything it needs to know to make the BEST decision.

A chess bot predicts what you will most likely do next based on many different variables; your previous decisions, layout of the board, history of how you play (in some cases), ranks your best decisons to worst decisons, etc.

A Poker bot will just do the same but it knows you only have a lousy THREE decisions to make. Yes there are other variables such as your position vs. the bots position, previous decisions, actions of others, etc. but these variables are minimal in comparision to the # of scenarios a chess bot must calculate. And why is everyone assumming that a Poker bot won't take this in to account? It will, it will take EVERYTHING in to account, it's decision will be based on it's position, it's opponents position, it's opponents tendencies, it's opponents likely holdings, pot size, the current board, current limit it is playing, players left to act, texture of the table as a whole (passive, loose, aggressive, tight, etc), and even more I'm sure. Of course, this bot is not one written by a high school programmer but to think that this isn't a simple thing for experienced computer programmers in combination with experienced hold'em players is ridiculous. In fact, it is very likely that it is being done right now as we type about it.

Two clowns devised a laser device to calculate the speed of a roulette ball. This laser device sent information back to a computer through a cell-phone (modified to be a transmitting device). The computer interpted the data and sent back what it thought were the most likely numbers for the ball to land on. It worked. In fact, it worked too well.

Oh yeah, us dumb humans could never program a computer to beat limit Poker. Gosh a simple game like chess yes but Poker no. Give me a break.


Finally, as time goes on, programs to beat Poker (limit) will be more and more sophisticated. They will start incorporating databases such as Poker Tracker, they will know your tendecies. For example, they will know without a doubt after playing a large number of hands against you that there are only certain hands you play or raise from UTG. This will of course give their decisions more weight becasue they have considerably narrowed your possible holdings.

Finally, I'm not saying a bot would ever have a bug that failed to recognize the nuts on the river. But one thing I do know, is it won't do it again. Any half-brained programmer would see it whie playing back the hands and correct the bug. A human will not do this. He will eventually misclick again and fail to bet the nuts on the river.

[Sephus]

oops.

[Freakin]

I honestly can't believe that the biggest hang up people are having in this thread is whether or not bytestream is a winning player at 2/4.
It would not be too difficult to program a bot to interact with Party poker. It's been done.
It would not be too difficult to program a bot to play ABC poker with no read-dependent decisions.
It would not be too difficult to believe that ABC poker with no reads can win on party poker.
It would not be too difficult to take your feet out of your asses and get off your high horse about sample size & win rate. Didn't Homer's post just show that at 480,000 hands you still could only be 95% certain that your winrate was w/n .5BB of your true number? And even at 96,000 hands, it was still +/- 2BB. So those of you saying "Play another 38k hands and then we'll see who's right" clearly are wrong about something too. Does that mean everything you say from this point on is meaningless? Good grief. Accept the fact that bots can win money, and that people can clearly make poker bots. Do you think that even a player who uses Winholdem and has programmed their bot to win is gonna be posting all over the place talking about it? They certainly won't need to ask the stupid questions like "How do I keep my bot from folding Aces preflop?"

[itsmarty]

[ QUOTE ]
Not everyone can magically turn a 2.7 BB/100 winrate over a 12,000 sample size, into a 4.7 BB/100 winrate over 12,000 - OVER THE COURSE OF 10 DAYS.

[/ QUOTE ]

Heh.

Martin