A Blog by Jonathan Low

 

Sep 13, 2019

Will Artificial Intelligence Ruin Poker?

Amateurs provide the bulk of the profits professionals glean from poker, both on and off line. But if  amateurs are beginning to sense they are being fleeced by online versions of the game, they may stop playing and deprive 'the industry' of much appreciated revenues. JL


Thomas Hale reports in FT Alphaville:

New research covers “the development of superhuman artificial intelligence beating professionals at poker”, which it says poses a risk to the online gaming ecosystem.Computers have already reached “superhuman” levels at limited version of the game, which limits the range and size of bets. Bots are already common on poker sites. If amateur players stop seeing poker as a game of skill, that poses a risk to the revenues they provide. Any business based around moving human behavior on to the internet stands to be impacted (for better or worse) by software that replicates that behavior.
Artificial intelligence has by now been linked with pretty much every imaginable line of business. As a concept, it risks being diluted into oblivion.
So it’s a breath of fresh air when someone provides concrete details of the changes AI technology is driving – usually at the fringes of business models.
Morgan Stanley has a new piece of research out this week, which covers “the development of superhuman artificial intelligence beating professionals at poker”, which it says poses a risk to the online gaming ecosystem.
Computers have already reached “superhuman” levels at limited version of the game, such as two-player pot limit poker, which limits the range and size of bets massively. They have of course already reached this level for pattern-intensive games like chess, checkers, Go, and rock paper scissors (see how you fare here).
Scientists at Carnegie Mellon have been working on a poker artificial intelligence program that they say is better than human professionals at six player games, with no pot limits. It’s called Pluribus, and its strategy was developed over eight days at a “cloud computing cost” of $144. From the report:
Poker has in general been especially difficult for AI given the computer doesn't know the cards their opponents have, and it involves a level of deceit or bluffing. Pluribus' approach, outlined below, is a milestone in AI development and for poker, which could, in Brown & Sandholm's view, be adapted "to basically any other form of poker" including, say, nine player variants or tournaments.
Pluribus’ strategy is based on self play, which was also used for backgammon, Go, and Starcraft 2. It also uses “abstraction”, where slightly different hands or bets are treated as the same thing, to reduce the complexity of the game.
The broader implication here is one that will be familiar to anyone who has played online chess: the risk of players using software on their computers to assist their decision-making. (It’s not clear from the report that the new technology is definitively better than human experts, by the way).
One full-time player we spoke to says bots are already very common on poker sites. The Morgan Stanley report mentions that professionals see them as “relatively easy to spot” (and often a source of easy money), but the same is probably not true for the generic “net depositor” player, from whom the bulk of revenues are ultimately sourced.
If these amateur players stop seeing poker as a game of skill, Morgan Stanley argues, that poses a risk to the revenues they provide. A bigger threat is to the professional players, as their expected returns would fall if bad players use better software.
One striking aspect of the report is the response of the platforms. Major players like PokerStars have “dedicated integrity units” devoted to identifying bots, and partypoker has closed over 500 bot accounts since December. This is highly reminiscent of other parts of the online economy, particularly Amazon reviews. Amazon says it uses “machine learning” to analyse incoming and existing reviews, and estimates that 90 per cent of inauthentic reviews are computer generated. Glassdoor, the employee review site, uses similar language.
Basically, any business based around moving human behaviour on to the internet stands to be impacted (for better or worse) by software that replicates that human behaviour. Games of software cat and mouse are likely to have political and economic ramifications far beyond the relatively small niche of online poker.
That world is edging towards a dystopian scenario, where people sit inactive in front of screens, playing Pluribus against Pluribus rather than poker, like some kind of self-driving car race. But don’t worry, a tech company will soon come along with a radical solution: the pictures on cards will be printed out on organic materials, so that you can hold them in your hand, maybe in a saloon, and play without the risk of algorithmic interference.

1 comments:

Online Shopping Lucky Draw Winner said...

Nice Post thanks for the information, good information & very helpful for others. flipkart lucky draw For more information about Online Shopping Lucky Winner, Flipkart, HomeShop18, Shopping Lucky, Draw, Contest, Winner, Prize, Result, 2018 - 2019 Flipkart Lucky Draw Winner

Post a Comment