It’s not often a poker story draws interest from the mainstream media. But a recently concluded match between a poker bot developed by researchers at Carnegie Mellon University and a quartet of top poker professionals has managed to capture the attention inside and outside the poker world. Each group appears enthralled and interested by the match for different reasons.
The general public and techies are intrigued by the AI aspects of the match and the dawn of the singularity. Poker players are more concerned about poker turning into the next “solved” game that only a small group of enthusiasts have any interest in playing.
Like the chess battles of old, and the current Go matches, these man vs. machine poker matches have been taking place for several years. The machines are steadily getting better and better, and tackling more complex games.
Still, I wouldn’t worry about a Skynet scenario, or the end of poker, just yet.
Carnegie Mellon’s poker AI improves by leaps and bounds
In May 2015, a team of heads-up no-limit specialists thoroughly beat Claudico, Carnegie Mellon University’s first attempt at an AI poker bot.
The team of pros that took on Claudico in 2015 were:
- Doug Polk
- Bjorn Li
- Dong Kim
- Jason Les
After being defeated by the poker pros, the team of programmers from Carnegie Mellon went back to the drawing board and came up with Libratus, which fared demonstrably better against a similar team of top-level poker professionals.
2017 pros battling Libratus:
- Daniel McAulay
- Jimmy Chou
Libratus did so well in the match (beating the pros for over 17,000 big blinds in 120,000 hands) that a Twitter parody account has sprung into being, often needling the human contestants.
Even though Libratus was over 12 years in the making, the jump in skill from Claudico to Libratus in a span of just 18 months is pretty remarkable.
Meanwhile, the University of Alberta’s Limit Hold’em poker bot, Cepheus has apparently “solved” Heads-Up Limit Hold’em.
Both of these developments are fairly concerning to the poker community, as it places the viability of beatable online poker games in jeopardy.
Perhaps even more concerning is the existence of lesser bots — some available for purchase — capable of beating non-professional poker players quite handily. So, even though I don’t think this is the beginning of the end for poker, it does present some challenges going forward. We will likely see the game evolve in new ways.
Is this how online poker withers and dies?
It’s possible to stave off the inevitable with bot detection software and by other means. But at some point in time, these bots are going to be light years beyond the skill of normal poker players, and the cost to acquire one will be minimal.
Unsavory types will find a way to stealthily employ them at the tables. But others will simply look at the bot software as a way to develop their own bulletproof strategies through game simulations.
This latter use of bots has consequences that reach beyond online poker and into the realm of live poker, as the skill level of the average player will improve. With fewer mistakes, there will be smaller skill gaps and less money changing hands.
However, this isn’t really a new development.
Even before AI poker bots, the internet drastically cut down poker’s learning curve, allowing players to put in tons of volume, parse tons of data, or run simulations and hand match-ups at will. Even with rudimentary programs, players were able to play and dissect more hands in a year than Doyle Brunson has likely played over the course of his entire poker career.
[i15-table tableid=20717][i15-table tableid=19346]
What it means for poker’s future
The good news is these bots are generally heads-up specialists. That reduces the number of possible decisions and requires the bot to learn the tendencies of a single player from just two table positions.
Heads-up, online poker is probably breathing its last gasps because of bots. Savvy poker players should start shifting toward six-max and full ring, which require more sophisticated bots, making it less likely people will be able to procure a winning bot that is also cost-effective.
Furthermore, game specific bots can be overcome by playing mixed games. Mixed games will likely grow in popularity to mitigate the impact of these game-specific bots. This is even more true of the never-ending string of variations people add to mixed game lineups. Every time a Big O, Razzdacey, or Drawmaha appears, it will confound a poker bot programmed for specific games.
Finally, Libratus relied on a specific strategy to beat the pros, massive river-over-bets that were balanced to include bluffs, the nuts, and everything in-between. Had the stacks been deeper (the players started each match with 200 big blinds) this strategy could have been nullified.
Final thoughts on poker bots
I don’t see Libratus as the end of poker.
I see Libratus and poker bots as the end of specialized poker, particularly heads-up games and shallow-stack games where the decision tree has fewer branches.
Bots are a very real threat to online poker, but there are ways to mitigate their impact. Poker players who have been choosing to specialize, and stick to highly exploitable games like heads-up, may want to start branching out into other formats. Their future may very well depend on it.
Image credit: OlegDoroshin / Shutterstock, Inc.