One of this week’s major controversies in the online poker world regards a mysterious piece of software developed by a player going by the screen name “Skier_5,” to assist in hyper turbo heads-up sit-and-go play. Skier_5 has declined to explain publicly what the software does, arguing that sharing this information would undermine his edge. He has, however, allowed a select few other players to use his software, in return for a cut of their profits.
Details are sketchy, but two facts are certain: Firstly, that the software does in fact provide a significant advantage, as Skier_5 and the other players using the software have all moved up in stakes rapidly, profiting at levels far above where they played before. Secondly, that PokerStars has not merely tolerated the software, but actively declared it legal to use.
The most significant clue as to the nature of the software comes from a 2+2 thread on the subject, in which a poster going by “He I Se N Be Rg” states that he’s observed that three known users of the software all have identical behavioural statistics for preflop, flop and turn. Without access to their hand histories, it’s impossible to confirm this, but assuming this information is correct, there’s really only one possibility which can explain it: that the software is effectively a bot, suggesting moves based on its best approximation to a game theoretically optimal strategy. If the humans were doing much significant decision-making, then at least some small difference in their behavioural statistics would be expected; if they are in fact making each type of move with the same frequency, then it seems very likely that the software is feeding them moves and the human players are merely acting as a middleman. If the software were adaptive, you would likewise expect their statistics to vary due to playing different opponents.
The idea of an AI “ghost” player with a human middleman is exactly what poker AI expert Eric Jackson predicted in our interview. Almost everyone in the poker world feels this is over the line, except for those using the software, of course. The real question, though, is why PokerStars is adamant about allowing it.
What’s technically allowed?
Like all sites, PokerStars updates its terms and conditions occasionally, so the nitty gritty of what’s allowed has changed over the years. As it stands, there are five things which are forbidden:
- Anything which assists in collusion, for instance by sharing hole card data with a partner.
- Pooling data with other users, or making use of data not personally observed by the player.
- Gathering data from games that are observed by the player without actually playing.
- Anything which automates play without a human actually operating the poker client.
- Finally, “Any tool or service that offers dynamic, real-time commentary or advice on the current game state that goes beyond reporting data and statistics.”
This last item would appear to ban the use of an AI ghost with human middleman, so the fact that PokerStars has inspected the software and deemed it legal would seem to suggest that it is something a little more subtle than this. The fact still remains that the players using the software all seem to be playing identically. Without actually seeing the software in action, it’s hard to know what’s actually going on, but if it allows users to immediately begin playing near-identical, winning strategies, then it seems that the difference between it and a bot helper must be little more than a technicality. For example, starting hand charts are explicitly allowed, so if the software amounts to an extremely large and complicated set of predetermined charts for various situations, PokerStars may have decided that it doesn’t count as “dynamic” and “real-time.”
One likely reason for PokerStars’s willingness to allow the software is that there are serious practical limitations in their ability to ban it. The PokerStars software does give the company access to inspect other processes running on a player’s machine, but at most all they could do would be to force players to jump through a few hoops in order to use an AI “ghost.” There are a number of techniques a programmer could use to run such a program covertly, particularly if it is modified on a regular basis and not widely distributed.
Even if PokerStars took sufficient efforts to prevent this, however, the software could simply be run on a separate computer entirely. This would necessitate the extra step of entering card and bet data into the other machine by hand, or perhaps an additional software layer to interpret data from a camera pointed at the first computer’s screen. As annoying as this would be, if the potential profits are high enough, no degree of inconvenience can be counted on to be a sufficient disincentive.
Hypothetically, you could still attempt to catch AI ghost-users by statistical analysis of a player’s moves. What you’d actually be looking for would be the absence of certain patterns, since computers are better at randomness than humans. This would require a huge amount of effort (and therefore expense for PokerStars), however, and could result in false positives from humans using legal methods to assist in randomizing their play, such as dice or clocks. Furthermore, actually banning a user on this sort of basis could produce backlash; distinguishing a talented but unorthodox human from an AI is a subtle problem, and PokerStars would not be able to disclose their methodology lest they help cheaters circumvent it. There would be a risk, therefore, of appearing to be banning players arbitrarily.
Overall, you could therefore argue that PokerStars is refusing to ban the software simply because they know that such a ban would be unenforceable, and see explicit and transparent permission as preferable to driving it underground, where it might prove even more dangerous. I think there’s more to it than just that, however. I think that heads-up sit-and-gos are doomed as a format – particularly the hyper-turbos – and that PokerStars is aware of this and has elected not to postpone the inevitable.
The fundamental problem
At the heart of the problem is a simple fact of game theory, which is that two-player games are solvable and games with more players are not. So-called “equilibrium” strategies exist for games of all-sizes, but their meaning is different when you have more than two players.
Players’ strategies are said to be in equilibrium when no player can improve his results through a change in strategy. Poker is a symmetric game, in that all players are playing by the same rules; in a two-player symmetric game, an equilibrium necessarily means that both players will break even in the long term. A non-equilibrium strategy facing an equilibrium strategy will never do better than break-even, and will typically lose. In that sense, an equilibrium strategy for a two-player game represents “perfect” play.
No such guarantee exists in multi-way games, however. Even though the rules of the game are symmetric, that symmetry is broken by the presence of additional players. Most multi-player games present opportunities for two or more players to “gang up” on others. This can produce a situation which is technically an equilibrium in that no one can do better by changing to a different strategy, yet the cooperating players are coming out ahead while the targeted player or players are losing.
If that sounds like collusion, well, it is. Or rather, it would be if it were agreed upon in advance by certain players. Ironically, however, one of poker’s biggest problems is also its saving grace, in that it renders multi-way poker considerably more resistant to AI than heads-up play. An AI which explicitly colludes with other AIs in order to produce a profitable equilibrium would be detectable by the same methods that are used to catch humans colluding; conversely, an AI which does not do so will inevitably be vulnerable to certain combinations of opponents’ strategies. This is not to say that equilibrium-like strategies are not useful in multi-way poker or that AIs will never be a problem there, only that these games will never be totally solved the way Heads-Up Limit Hold’em has been and are thus less likely to be killed entirely by advances in technology.
Once you understand this fact, and the difficulties involved in policing against AI ghosts with human operators, then it’s clear that heads-up online play has always been doomed, although it is still unclear exactly how long it will take for the end to come. Hyper-turbo heads-up sit-and-gos are by far the most solvable sub-format, because the shallow stack sizes simplify the problem, so they will be the first to go… but sooner or later all heads-up games will die, except in live environments where the players have no access to computer assistance.
Not necessarily a bad thing
It’s certain that the decision-makers at PokerStars understand all of this, but still, they could attempt to fight the problem and postpone the inevitable. Even if the most their security team could manage would be to throw some additional hoops in the path of those attempting to use this sort of software, those hoops would certainly slow the process down and keep the games viable for a while longer. If you think about it, though, it’s hard to see any compelling reason that they would want to do so. Indeed, many of PokerStars’s recent moves suggest that they would rather those games die sooner, rather than later.
The rise of Spin-and-Gos is the biggest indicator of this. Superficially similar to heads-up hyper-turbos – albeit with three players, rather than two – they add an additional element of randomized gambling which inevitably appeals to casual players and discourages sharks by adding to their variance. It’s therefore extremely predictable that they would poach casual traffic from the heads-up hyper-turbos of similar stakes, making the field for the latter both smaller and tougher, and gradually causing them to dry up.
Right afterwards, there were the hyper-turbo rake hikes. Although the backlash against these was sufficient that PokerStars backed down on that decision, the intent was likely the same. At the time, the sense among players was that these rake hikes were a short-sighted cash grab by the site, but it always seemed likely to me that they were a more calculated effort to drive players out of certain games and into others.
The reason that PokerStars might want to kill heads-up sit-and-gos is that they are a surefire way for losing players to go broke without ever acquiring sufficient bankroll to move up in stakes. The occasional big score is what causes most players to change stakes; this can be achieved naturally through larger fields, or artificially through Spin-and-Go style randomized payout structures or other jackpot mechanics. Lower-variance payout structures – heads-up and double-or-nothing sit-and-gos in particular – make it almost impossible that a player who is a statistical loser will ever come out ahead by enough in the short term to sit at a higher-stakes table. Some degree of upward mobility for losing players is essential to a healthy poker ecosystem, so if too many of them are being bled out at these games, the site as a whole will suffer.
Why not just kill them, then?
Having said all that, the final question is why PokerStars doesn’t just kill those games if they want them to die. It’s their site, after all, and they don’t have to spread any game that they don’t want to. I think the answer to that question lies in the infamous rake hike and its rapid reversal.
These games are the most popular among high-volume grinders, for the same reason that they are bad for the fish. If you can beat them, you will beat them consistently and you can play a lot of them. There are a lot of players specializing in hyper-turbo sit-and-gos who have played many tens of thousands of games, some of them even hundreds of thousands. These players produce a lot of rake for the site, even as they’re doing it harm by slaughtering the casual players and causing them to quit.
PokerStars doesn’t want these players to quit: it wants them to decide to play other games. Raising the rake was an attempt to nudge them to play elsewhere, but so many of these players operate at such small margins that the rake hikes didn’t just cut into the players’ profits – in fact, they turned a huge percentage of narrow winners into narrow losers. Grinders aren’t stupid, and won’t keep playing a game they can’t beat, but learning to beat a new game takes time. I think that raising the rake probably proved to be too fast a death, resulting in many grinders quitting entirely rather than migrating to new games; if that guess is correct, then the site’s profits probably took a nosedive, and they would have reversed the decision for that reason. Removing the games from the site outright would produce an even worse outcome.
In that light, Skier_5’s software might seem like a boon to PokerStars, especially given his tightly-controlled, gradual means of distributing it. If his stable of players continues to grow slowly, and if others come along with similar software and gradually fill the lobbies with computer-assisted grinders, the heads-up sit-and-gos might become unbeatable by increments. In terms of players’ bottom lines, it would end up being much the same as the rake being raised not all at once, but by a fraction of a percent each month.
The players frequenting these games will not find themselves in the red overnight, but rather see their profits dwindling slowly but inexorably as it becomes ever tougher to find an opponent who is not computer-assisted. Some will surely quit, but if profits diminish slowly enough, it’s likely that at least some of them will be incentivized to search for a new home in other games, rather than waiting for the bloody end. Eventually, only computer-assisted players will remain, at which point not even they will be able to profit. Once they begin to leave as well, then PokerStars can quietly remove the games, and few will be left to complain about it.
Alex Weldon (@benefactumgames) is a freelance writer, game designer and semipro poker player from Montreal, Quebec, Canada.