Good Bot, Bad Bot
How bots have destroyed poker platforms and built AI innovation
The Tail of Two Bots
In the most recent years, poker bots have become the benchmark for AI development and innovation. For poker sites like the industry leader PokerStars, bots have become a nightmare with tons of bots flooding their sites, beating human players, and destroying the online poker ecosystem. In other cases, companies like Facebook and Google are using poker bots as a way to innovate their own AI programs. Why is there such a different approach to poker bots between these two entities?
Bot Collusion and Forming Bot Rinks
Online poker has become a gathering place for bots, pre-programmed software that has the ability to outplay and outcompete human players. In the past few years, bots have slowly grown and taken over the largest online poker sites in world. Most high-level bots can easily beat low to mid-tier players in poker. In some cases, they are even able to compete at the highest levels of poker. To counteract the rise in bots, may poker sites are implementing new measures to strike down all bot activity.
Although bots can be destructive on their own, the biggest challenge about counteracting bots is preventing them from working together. Bots at the same table can be programmed to identify one another and set each other up for optimal success. By colluding, groups of bots can create rinks that dominate any table. Overtime, bot rinks can accumulate thousands and in some cases millions of dollars. Until recently, detecting collusion and bot rinks can be extremely difficult and in many cases very devastating to the online poker industry.
Bots Rinks in Online Poker
In the past few years alone, there have been many cases of bot rinks being discovered on some of the biggest poker websites in the world. One case that took the industry by storm was the 2015 bot collusion case with Poker Stars, the industry leader owning over 70% of the market share. The case involved a huge bot rink originating in Eastern Europe which accumulated over $1.5 million by playing on low-stake PLO (Pot Limit Omaha) tables. After discovering the bot rink, Poker Stars had to pay out and refund every single victim that was impacted by the bot rink. As the biggest player in the industry, if Poker Stars wasn’t safe, no one is.
Since that incident in 2015, there have been new cases of bot rinks on other websites and it hasn’t stopped. In this year alone, three large poker sites have been impacted by poker bots. In March 2019, reports came out about Zynga Poker being stormed by bots, affecting millions of their players. Although Zynga had known it had been compromised for over 12 months, the company failed to address any issues regarding the bots. In late 2018, the site experienced a surge in activity but have not been able to distinguish which players are human or bots. These bots are extremely sophisticated as well, making it even harder to identify bots on the site. More recently, Zynga Poker has seen a decline in activity, which many people blame on the increased activity and unfairness of bots on the platform, making it harder for human players to enjoy online poker.
Partypoker, another online poker site, recently began implementing anti-bot initiatives that have allowed them to identify over 270 bots in March alone. Following their first bot raid, they made another huge discovery in July 2019 when they discovered an additional 120 bots. All of these bots were scattered among their platforms including both their NA and EU servers. Since then, they have identified and banned over 600 bot accounts and seized over $1 million in winnings.
After hearing about their competitor Partypoker, Americas Cardroom also began their anti-bot efforts. In April 2019, Americas Cardroom was able to implement new technologies that was able to detect over 270 bots on their platform. By accumulating the winnings of these bots, Americas Cardroom took these winnings and reimbursed over 4,000 players their money, estimated to be around $176,000.
New Anti-Bot Initiatives and Bot Detection Technologies
Many platforms have been able to increase their efforts to prevent bots on their platform and generally have been widely successful. In some cases, platforms like Poker Stars have created detection software that has a near 90% success rate at identifying bot collusion and 89% at detecting all bots. Even the smaller poker sites like Partypoker and Americas Cardroom have implemented systems that have been successful on some degree. If we see this trend continue and bots are being banned from every poker platform, what is next for bot players?
Facebook and their AI program
Unlike poker sites, companies like Facebook have approached poker bots with the opposite perspective: opportunity. While many poker sites are trying to get rid of bots, other companies are using poker bots as a way to create more complex and self-learning AI programs. In 2019, Facebook partnered with a team from Carnegie Mellon University to create a poker bot called Pluribus. Although it may seem strange to enter the poker world, poker provides an essential factor for AI development: imperfect information. Unlike playing games like chess or checkers, poker provides a lot of unknown information. In games like chess, bots play within a closed system in which they have access to all the game’s information. Within a closed system, bots can see both their pieces and their opponents, allowing calculate different strategies and predict what their opponent will do. As a result, AI bots can become gods just after a couple hours of playing. Compared to closed systems, poker bots only have vision for their own hand and the cards on the table, but are unaware of their opponents, making it difficult to create sound strategies for winning.
With hidden information, bots and AI have to work so much harder, increasing time and resources and putting it all towards computing power. For companies like Facebook, their goal in creating Pluribus is to development a bot that can create complex strategies with incomplete data. Rather than testing the bot against humans, Pluribus was force to play itself, making the bot hone its skills and perfect different strategies. Instead of using different inputs, Pluribus was able to learn from itself, evolving its core strategy and getting better and better. When it faced off against humans, it easily swept.
Pluribus was more advanced than ever, playing poker games with 5 other players. This feat was something that was never done before, and luckily for Pluribus, it was an easy obstacle to overcome. Over the course of 10,000 hands and 12 days, Pluribus did a clean sweep over 15 different players. It won by an average of $5 per hand, making over $1,000 an hour. Not only was it smarter, but used less computing power than any of its predecessors as well. Pluribus was undoubtedly the smarted AI bot in the world.
What does this mean for the future of AI development?
AI development and poker has just hit the surface of its true potential. In early 2019, Libratus, an older AI poker bot was signed to the U.S. Army for a $10 million contract. The AI’s ability to create strategies with incomplete information can be applied to any industry from military to marketing. With the strides made in AI development through poker, you can expect to see poker becoming the standard for future forms of AI.