With all the cheating poker scandals in the news, now is a great time to look at another potential threat to the industry - bots.
A bot is a computer program that plays a game with or against you. Bots are prevalent in lots of apps. If you’ve ever played a game ‘against the computer,’ you’ve been playing a bot.
Most bots only exist as entertainment or when a human opponent isn’t available. These basic bots are common and don’t require much processing power.
If you’re reading this on your phone, you’ll have several different bots in your hand as we speak.
These basic bots work fine in a computerized game of Monopoly or FIFA. But they cannot competitively play a more complex game, like poker.
Table of Contents
How Smart Are Bots?
Bots come in all shapes and sizes, but the rule is pretty straightforward: The more complex a bot becomes, the more computer power it needs to run.
The complexity of your bot is based on two factors:
1: How complicated is the game you are playing?
1. Computers are designed to process data systematically. This factor enables them to perform well in games of perfect information (like Connect 4 or chess). Things are more complex in games of imperfect information (games where you can’t see everything happening).
2. The difference between NLHE and Limit Hold’em is key vital too. The strict bet-sizing rules in Limit Hold’em make the game far more restrictive than its counterpart. So, a bot named Cepheus can play a near-perfect Nash Equilibrium Limit Hold’em strategy.
3. The betting flexibility that comes with NLHE makes it a far more complicated game. As a result, beating it requires a far more complex bot.
2: How competent are your opponents?
1. While computers don’t make clumsy mistakes, they are restricted by pre-programmed strategies. Simple approaches might be enough to beat weak opponents. But strong poker players will understand how to exploit these tactics.
2. A rigid set of pre-written perimeters governs straightforward bots. They are unable to adjust to an opponent that is exploiting them. So, stronger opposition can easily dominate them.
3. Bots can best more advanced competition through complex strategies. But these programs are challenging to write and require significantly more processing power.
Who Might Use a Bot?
As a winning player has no need to cheat, you might associate bot use with losing players. But the reality is that anyone could decide to use a bot. People forget when nursing aspirations of becoming a professional player that life as a poker pro is a real grind.
It requires discipline, dedication and fortitude. Some people don’t have the stamina. Moreover, volume is the only way to combat the variance. So, it’s easy to see the temptation of using a tireless bot to put in the hard hours.
Players that might use a bot include:
1. Player prone to tilt:
Computers have no emotional response. So, a winning player may use a bot if tilt is ruining their game. They may feel that tilt is depriving them of the winnings they (incorrectly) think they deserve.
2. Winning player unable/impatient to improve and to climb stakes:
Moving up in stakes requires a lot of dedication and hard work. The requirements are amplified as you climb in stakes. Many people don’t have the mental power or patience to compete.
3. Player who wants to clear a bonus:
Many sites offer volume-based rewards programs, and these bonuses can be very lucrative. Rake races or deposit bonuses are often significant enough to turn a break-even player into a reasonable winner. These bonuses are often tough to clear, so using a bot is an easy way to hit the required number of hands.
4. Selfish Players:
Using a poker bot is another unethical way to steal money. Stealing takes a poor moral compass, so those using bots are unequivocally selfish.
Poker sites are working hard to prevent their use, but there have been several bot scandals. A pretty interesting post on the Two Plus Two servers a while ago outlined a multi-computer bot scam targeting high stakes sit and go’s.
You can find the whole thread HERE.
Why Are Poker Bots Banned?
The money that comes with being at the top of poker is an excellent incentive to get there. But it’s impossible to make it without a powerful hunger and desire to be the best.
Poker is all about testing your skill and resolve against someone else. This fact appeals to those with a competitive streak.
Poker is as much about a battle with yourself as it is against someone else. To win, you must master your emotions, sense of discipline and other powerful human tendencies.
This self-conflict is one of the game's most beautiful and challenging aspects.
Emotionless, computerised bots ruin this magic.
For these reasons, bots have no place in poker… usually.
There have been several trials where poker bots have tested the competence of AI. But there is a massive difference between this regulated bot use and those heavily outlawed in the wider online poker world.
The players willingly participating in these controlled studies know they are playing a bot in advance.
Why Play a Bot?
These challenges typically offer the players an attractive financial incentive to participate. But the experiments also have the same intrigue that made movies like The Terminator, The Matrix and I, Robot so successful –
The theoretical war between human and artificial intelligence.
These simulated poker matches offer an arena for the two to battle.
Until recently, many believed poker was too complex for a bot to master. For example, there are 10160 different situations in a game of heads-up poker. As Tuomas Sandholm, one of the founders of the revolutionary poker bot Libratus, explains –
“There are more potential heads-up poker scenarios than there are atoms in the universe.”
When combining this complexity with computational limitations, many believed that writing a world-beating poker bot was too complicated.
Even with the steady increase in computer power, many rather romantically maintained that the poker bots’ inability to adapt was their downfall. Those human players would eventually identify a flaw that they could hyper-exploit.
They had good reason to do so too. The players selected to participate in these trials are the most elite worldwide. They were savvy and aware of the game's most innovative and cutting-edge strategies.
Even with the rapid advancement in AI technology, bots still need programming.
How could an inferior player write a bot good enough to compete with the sharpest minds in the game?
Well, until 2017, they couldn’t. A poker bot called Claudio had had some success against seasoned players like Laak and Esfandiari. But these players are not renowned for their one-on-one prowess.
The elite NL heads-up specialists had proved too much for poker bots time and time again.
Enter Libratus?
In 2017, four of the best players in the world took on the latest poker bot. Its name was Libratus, and it was a game-changer.
Clever projects usually have cunning names, and Libratus was no exception. The word stems from the Latin word ‘libra’. Libratus directly translates as ‘balanced’ a fitting name given the industries progression towards GTO poker. It’s the same root word as the star-sign Libra, which is why it’s represented by a pair of scales.
The players battling Libratus were Jason Les, Dong Kim, Daniel McAulay and Jimmy Chou. With two players involved in a 2015 victory against Libratus’ predecessor, Claudico, confidence was high, but it was also misplaced.
From the get-go, Libratus proved formidable, and the secret was in its programming.
With 600, 28 core nodes and 2.7 petabytes of data (yeah, we had to Google how much that was, too!), Libratus’ brain power was impressive. But it was not a bot that had studied how to play poker well. It had not learnt poker strategies beyond the basic rules of the game.
Instead, Libratus had the ability to solve games of imperfect information. It used its immense processing power to run continuous poker simulations to teach itself.
A big flaw with most poker bots is that when a player finds a hole in the bot’s strategy, they can exploit it relentlessly. The bot cannot reprogram itself in real-time. In an interview, Jason Les admitted that the players attempted to beat the bot in this way in the challenge.
Due to Libratus’ clever programming, it could evaluate the strategies that the human players were using as it was playing. While human players can do this too, Libratus’ ability to study and play simultaneously was at a level where human players can’t compete.
This pretty impressive robo-flex process removed a fatal vulnerability that hindered previous bots. It enabled Libratus to plug leaks that the human players may have been exploiting very quickly.
The Hunter Becomes the Hunted!
But Libratus wasn’t just defending itself either. Upon discovering what Sandholm described as potential ‘holes’ in its strategy, Libratus prioritised the computation of ‘better strategies for those parts of the game’.
So, Libratus would learn to plug its leaks and devise clever ways to exploit the human player’s new adjustments. The hunter became the hunted!
As well as its superior brain power, Libratus had other advantages over the human player. The 2017 challenge consisted of 120,000 hands over 20 days. Although it’s not uncommon for poker pros to play long sessions, it’s reasonable to expect fatigue to creep in during such a long gruelling challenge.
As a computer, Libratus had no such issues.
To make their strategies more water-tight, humans typically limit their different bet sizes. The more you vary your sizings, the more complicated your poker strategy becomes.
But it’s also more challenging to implement and balance them.
Given its super brain, Libratus could handle this complexity, enabling it to be more creative with its bet sizing than a human. Libratus could make unusual plays that made it more challenging to read and predict.
Towards the end of the challenge, the pros conceded that Libratus was just ‘too good’ and suffered significant losses.
It was a terrifying step forward for poker AI.
The Libratus Verdict?
While Libratus’ victory raised some eyebrows, few were panicking. As well as requiring computer power way out of the average consumer's reach, Libratus’ remained untested against multiple opponents.
Many questioned if the bot (and its hardware) coulc beat multiple opponents since this made the game more intricate.
This scepticism was short-lived, with Polaris (Libratus' successor) beating multiple opponents easily.
Thankfully, the bots’ designers refused to release details of Polaris’ programming ensure it did not fall into the wrong hands.
Should We Be Worried About Bots?
Poker sites are doing everything they can to protect players from bot use online. While they pose a threat to online poker in the long term, human beings are hard to replicate well.
Our emotional flaws are as random as they are powerful.
Scams like Fedor Kruse’s ‘dream machine’ ruse, for example (which you can read about in this magazine), were quickly identified. Fedor’s bot used bet sizing that was too unorthodox and complicated to be human.
Super bots do very unusual things. So, it’s hard to see how they could avoid detection for very long with an entire poker community on the lookout.
If you enjoyed reading about Libratus, you can find a short documentary below.
In the meantime, good luck at the tables.
