Is online poker dead? Has the virtual felt transformed into a perfect GTO (game theory optimal) environment? A place where robots battle it out to rob unsuspecting punters of their stacks?
This type of concern is becoming more widespread in the modern game.
And it all revolves around a simple question -
Is Poker Solved?
If poker is solved, a perfect solution to the game is already known. If enough players (or poker bots) are using the solution, winning at online poker over the long term will become impossible for regular players.
Does the advent of poker solvers signal that it’s time to find a new hobby?
Perhaps we don’t need to be so hasty; let’s consider the topic of solving poker in a little more depth.
Cepheus Solves Poker
In 2015, the poker bot Cepheus solved heads-up Limit Hold’em.
It's entirely impossible for a human player to beat Cepheus over a large enough sample of hands.
This feat initially may have appeared as the harbinger of doom for online poker. But it’s crucial to keep things in perspective.
- This result only applies to heads-up Limit Hold’em - a game rarely played online. More complex games, such as No-Limit Hold’em or poker games with more than two players still have no complete solution.
- Heads-up Limit Hold’em is still not fully solved. A full solution is a strong solve. Cepheus’ solve is a weak solve. By contrast, a solution from a poker solver falls into the category of ultra-weak solve.
Let’s get a basic understanding of what the different levels of solve represent to understand the issue further.
Strong vs Weak vs Ultra Weak Solve
Let’s break down the different types of solutions:
Strong Solve – This response is the best from any given game state. It specifically includes non-optimal game states where both players have made mistakes to arrive at the current situation.
Weak Solve – This perfect strategy is from start to finish against any possible move from Villain. But Hero must always take the perfect action. If Hero deviates from an ideal response, the solution to the subsequent scenario may not be known.
Ultra-Weak Solve – The technical requirement for an ultra-weak solve is to prove which player will win given the initial game state. In poker terms, think of an ultra-weak solve as any attempt at a solution that falls short of the weak solve.
The State of Poker Solvers
The good news for us is that a solution from a no-limit poker solver falls a long way short of a weak solve. In other words, solver output is not the same as the feat performed by Cepheus back in 2015.
Cepheus was able to generate a solution to heads-up limit poker by considering every single possible game state. Limit Hold’em has a finite number of betting options on each street. So, the “small” game tree made this possible.
And by “small”, we mean “eye-wateringly massive”. So, a powerful computer can provide a solution that factors in every game state within a reasonable time frame.
For example, it took Cepheus 68 days to solve heads-up limit in 2015 on a supercomputer equipped with 48 CPUs.
Theoretically, the same type of algorithm can exist for solving no-limit poker. But the extra complexity of bet sizing means that even the most powerful computers wouldn’t reach a solution in our lifetime. And this is just for a single-stack depth, a variable that changes from hand to hand.
Poker solvers work because the game tree is heavily simplified. Players may only choose one small number of bet sizings at each decision point. So, poker solver output is an ultra-weak solve at best.
That said, poker solvers are valuable tools. Let’s consider some of the pros and cons of using an ultra-weak solution.
Problems with an Ultra Weak Solve
A fundamental problem with an ultra-weak solve is that it cannot effectively deal with every action Villain might take.
Many no-limit solutions revolve around a single set of preflop sizings. For example, the answer might have a single preflop range for defending the BB when facing a BTN open.
The problem is that this range will be based on a specific sizing, such as a 2.5bb BTN open.
How should a GTO player or bot respond if BTN makes a 3bb open instead?
Technically, the defending range should be tighter. But the exact strategy is not covered by the solution. A GTO bot will typically use the closest solution (the 2.5bb sizing, which is inaccurate in this case).
So, it’s technically possible to ‘exploit’ a GTO bot by choosing sizings that don’t exist as part of its game tree. This strategy would subsequently force inaccuracies.
Advantages of an Ultra Weak Solve
Although ultra-weak doesn’t fill us with confidence, the truth is that this poker solver output is extremely powerful. The average human opponent doesn’t stand a chance against a robot-following solver output.
In the modern era, there is an abundance of ‘pre-solved’ (ultra-weak) solutions to poker available commercially for low prices.
This fact heavily lowers the barriers to entry for players looking to gain an unfair advantage by mimicking solver output. This type of cheating is referred to as RTA or ‘real-time assistance’.
Many players are concerned about what this means for the future of online poker. For example, can we protect against RTA cheaters?
Protection vs RTA
Firstly, recognise that poker sites have security teams actively monitoring for RTA. RTA play exhibits a specific signature that’s very different from human play. Any large-scale attempt to automate an RTA strategy is unlikely to survive security surveillance.
Manual (non-automated) attempts using RTA might evade capture for longer. But there are serious issues with such an approach.
Rate of play will be excruciatingly slow due to manually referencing GTO tables at every decision node. Even if a human had the patience, the difficulty of accurately executing a perfect GTO strategy is complex.
Even in the worst-case scenario, RTA bots in the player pool don’t mean that winning is impossible. Bots aside, there are usually players in the pool who are stronger than us.
Making money is feasible if we can effectively limit the number of hands we play vs stronger opponents.
The State of Other Variants
The problems with RTA apply almost exclusively to No-Limit Hold’em. The threat of RTA is virtually non-existent with other poker variants.
PLO (Pot-Limit Omaha)is the second most popular online poker game. Commercial solvers are available, but PLO solutions are nowhere near as widespread as NLHE.
PLO game trees are also more complex than an average NLHE game tree due to the sheer number of starting hand combinations.
Most other games do not even have commercially available solvers. Although not impossible, it’s unlikely that there is RTA activity for these variants.
The Future of Online Poker
It doesn’t appear that NLHE will be strongly solved or even weakly solved any time soon. Currently, no online poker strategy exists which is not open to counter-exploitation.
Of course, playing against ultra-weakly solved solutions using RTA cheaters is not ideal. The good news is that such activity is relatively minimal and doesn’t preclude us from winning at poker.
Even in the worst-case scenario where the NLHE games become unbeatable 20 years from now, we’ll have the option of playing fresher variants relatively untouched by efforts to find a solution.