The ‘Brains vs. AI’ challenge is officially over and Libratus crushed the pros to the tune of 17,000 big blinds. While some hoped that the pros would fare better, this win isn’t exactly a shock. When pros battled Claudico in 2015, they won but developers called it a “statistical tie.”
CMU Prof. Sandholm talks about Libratus' win in the #BrainsVsAI poker competition. pic.twitter.com/zHFNsWb4bo
— Carnegie Mellon (@CarnegieMellon) January 31, 2017
While this is truly a remarkable technological feat, one has to remember that a lot of time and money went into the “win.” Some online poker fans may be worried that this is the beginning of the end of online poker with advanced “superbots” just around the corner. Luckily, Libratus is more of an anomaly than the future and here are five reasons why.
. @dongerkim @heyitscheet @678DMcA @Chouchoupoker Fellas, t'was a good game. I am ready for your final challenge…Tom Dwan. Can u call him?
— Libratus (@BotsAintLoyal) January 31, 2017
The Bot Can Only Play Heads-Up
Libratus’ primary limitation is that it can only play heads-up Hold’em effectively. Add additional players and the bot loses a lot of its edge. Also, if the stacks were deeper than 200 big blinds, things might have gone differently as well.
The Bot Sometimes Needs More Time Than Online Poker Players Get
Another potential problem with the bot is that it sometimes tanks for too long. According to some reports, the bot would sometime take a lot longer on the river to act than what is normally allowed when playing online poker.
Libratus isn’t a virtual William Kassouf but if one could figure out which situations that Libratus might tank consistently on, players could exploit this weakness – at least until programmers updated the bot.
. @Kevmath Hi Kevin, any idea of when the WSOP heads up event will be? And what's the electrical outlet situation?
— Libratus (@BotsAintLoyal) January 23, 2017
Programmers Were Allowed to Tinker With the Bot During the Game
At the end of each day, programmers got together to tinker with Libratus and try and correct any potential leaks from previous play. Jimmy Chou commented that any weaknesses they discovered in the poker bot were gone the next day.
If this type of bot were to be used in a real-world scenario, only programmers would be able to regularly update and modify the bot. How many people using a poker bot would actually have the skills to do so outside of the original developers?
. @RealKidPoker I would like to apply as a PS Team Pro. While I can't really wear a PS patch, I can definitely change my desktop wallpaper.
— Libratus (@BotsAintLoyal) January 27, 2017
Online Poker Rooms Combat Bots
Online poker sites regularly screen for bots and a program such as Libratus would likely be tagged in short order. As the bot is presently programmed, it would set off a ton of red flags including those for an unusually high win rate.
Even if Libratus was to go live and win a ton of money, good luck on the developers being able to collect on it.
Different Pros Might Get a Different Result
Finally, while this demonstration was impressive, one has to wonder if the same results would occur with a different group of pro players. Perhaps the bot is legit and will be able to crush all opponents but I’d like to see a bigger sample size.
Let’s put 25 to 50 of the games top poker players up against this bot and see if it can continue the same insane win rate. If it can continue crushing against such a large number of highly talented players, then maybe we can buy that Libratus has solved heads-up poker.
What lies ahead for me? Offered TV show…"Life With Libby". May smash Omaha. Would run for president, but can someone w/no experience win?
— Libratus (@BotsAintLoyal) February 1, 2017