Poker AI: what everyone should be talking about

The recent success of the Lengpudashi poker AI in China, on top of Libratus’s victory in January, adds fuel to the skill vs. luck debate and may have implications for the legal status of poker, says Paul Phua

Last week a team of poker players in China were resoundingly defeated by “Lengpudashi”. Meaning “cold poker master”, Lengpudashi is the new, even more improved version of the Libratus AI (Artificial Intelligence) programme that I wrote about back in January.
Not surprisingly, this latest AI victory has been big news: people have worried for years that robots equipped with AI will take over human jobs. Now not even poker is safe. Though computer programmes long ago proved their superiority in the classic skill game of chess, until now the bluffing and intuitive elements of poker – its very human elements – had made it hard for a machine to master.

Skill vs. luck in poker

But there is one interesting aspect to this story that people don’t seem to be talking about. Any poker player will tell you that poker is a game of skill; that a pro like Phil Ivey may lose several pots to an amateur through misfortune, but let him play a thousand hands, and Ivey will surely come out on top. The law, however, does not always see it that way.
In many countries, poker is legally classified as a game of chance, which places it under the more restrictive legislation reserved for gaming. It’s the reason why online poker is illegal in America (except within certain states). It’s also the reason why, in England, poker pros do not have to pay tax on their gains – they are classified as winnings, just as in a lottery, and not as earnings.
For many years, people have tried to demonstrate scientifically what we poker players know from experience to be true: that poker is a game of skill, albeit one in which luck plays a significant part. As the renowned poker expert David Sklansky puts it, if only luck was involved, everyone would get to the river and show their cards, letting random chance decide the winner. Whereas, in fact, an analysis of a million PokerStars hands by the software company Cigital in 2009 showed that 75% of them never made it to showdown.
There have been other studies. Levitt and Miles, for instance, analysed the results of more than 30,000 players in the 2010 World Series of Poker, and found that players classified as “skilled” earned an average return of 30%, whereas unskilled players on average lost 15%. Set against that, a 2012 study by the Journal of Gambling Studies found no difference between the performance of skilled and unskilled players out of 300 participants, and argued from this that poker is a game of luck – though others have countered that this study only took place over the course of 60 hands, which may not have been long enough to allow skill to assert itself over luck.
Suffice it to say that luck vs. skill has been a hotly contended issue, with very important legal consequences hanging in the balance. Does the success of Libratus and its successor, Lengpudashi, finally provide the missing argument to demonstrate conclusively that skill is the determining factor?
No doubt it will take hundreds of thousands more hands to prove. Perhaps a human champion will yet emerge, like John Connor in the film The Terminator, to take on the machines. Or perhaps the cold, hard, application of science and probability, allied to an ever-improving computer programme that learns from its opponents and gives off none of the human tells, will implacably defeat even the best and brightest of our current poker generation.
Either way, no doubt legislators will be following the development of this poker AI keenly. And as for us human players, the lesson we can draw from Lengpudashi is that it’s more important than ever to brush up on our poker skills.
So keep reading my strategy blogs at Paul Phua Poker, and subscribe now to the Paul Phua Poker School YouTube channel, so as not to miss my exciting new videos featuring tips from the top pros.