The ongoing war to rid poker of bots

Whenever peer-to-peer games and money are involved, you can bet your bottom dollar unscrupulous individuals will look to gain an unfair advantage. With online poker, that can involve the deployment of bots, which are autonomous software programs that surreptitiously face off against unsuspecting human opponents. Whether it be rudimentary programs built in someone’s bedroom or highly sophisticated AI orchestrated by organised bot rings, this scourge flouts the rules of pretty much all online poker sites.

Earlier this year, GVC-owned partypoker announced that it had closed 277 bot accounts and redistributed precisely $734,852.15 to affected players. This was followed by the shutdown of 94 bot accounts in April and over $150,000 seized. The crackdown continued into May with 42 bot accounts targeted and another $50,000 forfeited. And in June, 67 fraudulent accounts were shut down and more than $30,000 forfeited, bringing the total confiscated over the course of six months to around $1m.

Other online poker operators tend to keep their cards close to their chest regarding bot account closures and player reimbursements. However, in 2016, Sweden’s state-owned Svenska Spel revealed SEK3.8m had been returned to around 25,000 players who fell victim to a bot ring operating more than a dozen accounts. A year earlier, a suspected bot ring involving accounts mainly in Russia and Kazakhstan was accused of winning an eye-popping $1.5m on PokerStars in pot-limit Omaha cash games. Practically every poker site and network in the world has had bots at one time or another.

“As with security and fraud in general, with bot detection we are in an arms race against the bot creators,” says Alex Scott, managing director of poker at Microgaming. “We’ll invent a new means of detection, and for a couple of months we’ll have better than average detection and confiscation rates. Then the bot creators will figure out how they are being detected and find a way to evade detection. So, the challenge for the industry is to stay one step ahead, anticipate what the bot creators might try next, and be ready for it.”

Alex Scott, managing director of poker, Microgaming

Joining the dots to detect the bots

On a more basic level, potential indicators of a bot account could include always immediately joining a table when a certain number of players are seated, or betting and raising the same amount each time. A bot may also always take the same amount of time to act, never use the chat function or never move the mouse. However, just because a user behaves in a certain way or out of the ordinary doesn’t mean it’s a bot. Instead, you need to study the bigger picture, says Michael Josem, who has previously worked in various online game security roles at PokerStars.

“You can’t just look at one individual brushstroke in a painting; you’ve got to look at that whole painting to see how it works together,” he says. “For example, a lot of poker players simply don’t chat at the table, so you can’t use not chatting as an indication of misbehaviour. Similarly, some players who have physically disabilities use a keyboard to play, so you can’t use not moving the mouse as an indication of cheating. Instead, you have got to look at the whole totality of different information.”

While more industrious bot creators have wised up and reconfigured their software to behave in a more humanlike manner, the fact of the matter is many bots simply aren’t very good at poker. In fact, many of these automated programs end up losing money or are so crudely built with robotic-like behaviours that it doesn’t take a proactive site long to identify bots and close accounts. “I think that players tend to overestimate the ability of bots to win,” Scott explains.

“While winning bots do exist, they are a tiny subgroup. The majority of bots are profile-based and play by rote – it is difficult for such primitive bots to win except in the very weakest games.” The profitable bots that do eventually end up being banned will move to another site and try to win as much as possible before they are forced to switch to yet another operator.

Sharing notes on cheats

To try to prevent those behind bots hopping from site to site, Rob Yong, partner of GVC and owner of the Dusk Till Dawn poker room in Nottingham, England, is setting up an independent body to run the cheats out of town. The #Fairplay initiative involves online, land-based card rooms and live poker tours collaborating to share information on cheats, including those running online bot accounts. For instance, if a cheat with a bot is discovered and banned from partypoker, other sites signed up to the scheme would hear about it.

This seems a positive step to clean up online poker and restore trust, although you have to question whether sharing information on players falls foul of data privacy laws.

“As with security and fraud in general, with bot detection we are in an arms race against the bot creators” – Alex Scott, managing director of poker at Microgaming

“It’s a complete non-starter,” Josem states emphatically. “Firstly, I cannot imagine how in 2019 one company can say to another company, ‘hey, we have this customer who we think is a fraudster and a cheater’. That just throws up a whole range of data protection and defamation problems. I find it difficult to imagine that a publicly traded company would ever sign up for this because sites have no right to accuse someone of what is fundamentally a criminal act.”

He continues: “Second of all, sites make mistakes. That is a huge defamation risk. There is a credible argument for regulators to share the details of convicted offenders with operators, but the idea of operators sharing information on players, especially when they have such little recourse, could be a huge problem.” Furthermore, there is also the concern that sites could sign up to #Fairplay and cease to invest in the resources needed to catch cheats and bots, choosing instead to rely on fellow members to warn them of fraudulent accounts.

Deep thinking  

As bots and AI become more and more advanced – and harder to spot – it begs the question: is online poker facing an existential threat? Think back to 1997 when IBM supercomputer Deep Blue stunned the chess world by beating Russian grandmaster and then-reigning world champion Garry Kasparov in a six-game match. Deep Blue’s shock victory marked a seminal moment in the evolution of AI and demonstrated that computers could make optimal plays in order to beat the best, and wouldn’t succumb to pressure or make rash, ill-judged moves.

Almost 20 years after Kasparov’s humbling defeat, a computer developed by Alphabet’s Google DeepMind triumphed over a human at the board game Go. With more board configurations than there are atoms in the observable universe, sceptics doubted whether a computer could outfox a proficient human opponent at this 3,000-year-old Chinese game. But the machine did indeed prevail, beating South Korean professional Go player Lee Sedol four games to one in a five-game match. This enthralling encounter in 2016 between man and machine was turned into a Netflix documentary, AlphaGo.

“AI is a threat for sure,” says Scott. “Backgammon was pretty much wiped out online because of the threat of AI. And a game like rummy, which is hugely vulnerable to AI, will never grow outside of one or two regions because of this. But rummy and backgammon are fairly static games. Poker will survive, as it has many times in the past, because it’s constantly changing.”

Evolving tastes

Scott points to the fact playing habits change and other poker variants become the game du jour. For example, five-card stud made way for five-card draw and seven-card stud. No-limit hold’em supplanted limit hold’em, then Omaha became popular with cash players and now 6+ Hold’em is a new action-packed, short-decked variant. Despite the appeal of cash games declining, lottery-style jackpot games and tournaments are as popular as ever, the latter representing more than two-thirds of online activity, according to Scott.

Multi-table tournaments with hundreds or thousands of players are very hard – but not entirely impossible – for bots to master. These games come with high variance, meaning a huge slice of luck is needed to survive and overcome so many opponents. “While AI is a threat of course, the constantly evolving nature of poker will ensure its longevity,” Scott says. “The current resurgence of online poker is in large part due to this and, in my opinion, poker today is very different from the game we knew 10 years ago.”

Microgaming poker

That said, the threat AI poses to online poker was brought into sharp focus this month when a bot built by researchers at Facebook and Carnegie Mellon University crushed human pro players in six-handed games of no-limit hold’em. This landmark achievement came more than two years after AI program Libratus, developed by the same team, defeated pro players at heads-up no-limit hold’em, winning $1,776,250 over the course of 120,000 hands. However, beating multiple opponents at one table is seen as a huge milestone in the evolution of bots and AI research.

Pluribus, as it’s dubbed, took on a bunch of successful pros over 10,000 hands and won an average of $5 per hand, equating to winnings of around $1,000 an hour. According to the team behind Pluribus, this was a “decisive margin of victory”. What’s more, this program was built in just eight days using a 64-core server with just 512GB of RAM. It also runs on two CPUs compared with the 1,920 used by Google’s machine in AlphaGo and, as the researchers noted in a blog post, it would cost just $150 to train a similar bot using cloud computing.

A busted flush

So, does Pluribus spell the end of online poker? Probably not, although news of its victory generated headlines around the world, with some outlets coming out with hyperbole proclaiming online poker to be dead. Due to poker being a game of incomplete information, it was doubted a decade or so ago whether machines would ever be able to beat the best human players. For a start, bluffing was cited as one aspect of the game AI would fail to accomplish with consistent success. But here we are, 21 years since the very first poker site launched, and an inexpensive computer program has whipped a coterie of top players.

In the meantime, the constant arms race between the sites and the bot creators is sure to rumble on. “In the old days,” Scott says, “finding unsophisticated bots used to be incredibly simple – you’d just get the process list, and see the bot running right there. These days, the level of sophistication on both sides is incredible.” And it’s not just poker players turning to bots; sometimes it’s individuals from outside the world of poker who spot an opportunity to make money by underhand means. For example, the biggest ever cheat partypoker caught using a bot had never played one genuine hand of poker on the site. “It is an ongoing cat-and-mouse battle,” Josem stresses. “An ongoing war, and I don’t think either side will win.”