May 12, 2007
Bot Scandal at Full Tilt Poker
Robopoker reports on a poker bot scandal at Full Tilt Poker that has spawned a discussion on TwoPlusTwo.com that has more than 1700 replies over four days.
2+2 user SukitTrebek's Poker Tracker hand history database revealed to him that a number of regular players in the Full Tilt $200 no-limit hold'em games are very likely bots. SukitTrebek carefully gathered evidence — including playing against the supposed bots and repeatedly taking advantage of giant holes in their game, until, first, the bot programmer apparently manually took over when SukitTrebek and the bot were head-up, and second, the bots would leave the table whenever SukitTrebek joined a game they were in.
SukitTrebek presented his evidence to Full Tilt's management, and kept goosing them from time to time. They asked him, in order to help their internal investigation, to not go public with his accusaution while their investigation went on.
The alleged bots disappeared from Full Tilt. Some weeks later, SukitTrebek received email from FullTilt saying that the investigation was complete. Not long after that, the bots reappeared! SukitTrebek went public on 2+2's Internet poker forum early Wednesday morning, and the thread exploded.
Late on Thursday, 2+2 user FTPSean, identifying himself as a representative of Full Tilt Poker, started a new thread:
After doing our due diligence in this case, we came to the following determinations:
- During the investigation we found the evidence to be inconclusive in supporting either determination (human or bot).
- After careful consideration, the evidence did not warrant the seizure of funds and permanent account closure.
- We stand by our decision. Having said that, re-opening an account after an investigation such as this one does not mean we have made an irreversible decision. We will continue to reevaluate this situation.
I particularly like the third bullet point in all its weaselly, waffly glory; We stand by our decision ... We will continue to reevaluate this situation. Robopoker says Full Tilt management is in "damage control mode."
Two things stand out to me from this story;
First, the bot operator is not terribly bright. If I were running a profitable bot on a poker site, and I had just had my account locked down and then reopened, I would not celebrate by firing up my bot at full capacity again. I think I would draw down my bankroll and lay low for a while.
Secondly, it is really surprising that this bot wins any money over time... except that the evidence is clear that it does. It was found out because the Poker Tracker data for a number of tight-aggressive players was uncannily identical, statistic for statistic. Robopoker claims that his own bots' statistics don't converge anywhere near that closely, because his pays attention to board texture and opponent playing style. SukitTrebek thinks this guy is not a very good poker player, but has stumbled into a formula for playing that is good enough to beat the games ... for now.
How can a bot that can't adjust to opponents still make money? That's easy: Mike Caro's Law of Least Tilt. A bot never tilts, It wins or loses according to the fall of the cards and the basic strength or weakness of its algorithm. Real human players almost invariably go on tilt — lose their psychological balance and play badly, perhaps without even realizing it. Tilt is responsible for a noticeable fraction of many players' losses. Bots don't tilt; and so a bot is going to have an edge against a human player who uses the same strategy as the bot.
Posted by abostick at May 12, 2007 09:49 PMWho's to say Full Tilt isn't running the bots.
High level tech skill and a few felony conviction can be found in Full Tilt top brass...
Posted by: 11001011 at May 13, 2007 12:10 PMWhy would they want to?
Those bots are taking 3BB/100 out of each table. At a $1-$2 NLHE table, that's $6 for every hundred hands played.
Meanwhile they are directly raking fifty times as much, $3 for every hand played.
The obvious motivation for an online poker site to run bots is to have them there as prop players for tables that are getting short or are just starting up. Running them in full tables just doesn't make any sense.
If I ran an online poker site, and I wanted to use bots as prop players, I would adjust their hand selection/agression to be slight losers. In a nine-handed $200 NLHE game in which $3/hand is raked, a site-owned bot that lost as much as 10BB/100 would be making the site a net profit. (Breakeven for a site-run bot would be -16.67BB/100.) Players would love to play against a bot that was known to be a money loser; but what would really be going on is that the bot would effectively providing rakeback, preferentially to the better players.
Posted by: Alan Bostick at May 13, 2007 03:25 PMJust wanted to add that I can find no way that these "bots" could produce these stats without ridiculously simple logic code.
Assuming they aren't reacting to players at the table (and that would be nearly impossible to do and still produce such congruent stats) they are probably making decisions based predominately on the raw EV of their hand with some modifications based upon the previous betting so far in this hand.
I'm shocked they've been profitable. They might have struggled to find the system, but to code something this simplistic wouldn't take more than an afternoon.
The debate seems to be drawing to a close at two plus two. I'll do a wrap up in the next couple of days. This whole event has been a primer for how bad bots get caught.
Posted by: Robopoker at May 14, 2007 11:28 AMA long time ago, when people played online poker on irc.poker.net for just for bragging rights, there was r00lbot, a perl script written by Greg Wohletz, that was a consistent winner despite being playing the lamest of a-b-c poker strategies.
The University of Edmonton's Pokibot was a more formidable opponent, but had the saving grace of having a repertoire of Steve Wright jokes.
Posted by: Alan Bostick at May 14, 2007 11:50 AM