♠ AI and poker: how artificial intelligence changed the game forever

For decades, poker was the last refuge of human pride against the machines. Chess fell in 1997, when Deep Blue beat Kasparov. Go, long thought unassailable, fell in 2016 to AlphaGo. But poker seemed different: a game of hidden information, of bluffs, of reading the person across the table. Here, we told ourselves, the machine could never break in. We were wrong.

In just a few years, artificial intelligence didn't merely learn to play poker — it learned to beat the best players in the world, and along the way it completely rewrote how the game is studied, taught and played. This is the story of that transformation, and of why, against all expectations, it turns out to be good news for anyone who still wants to sit down at a table and get better.

The day a machine truly won

The first serious blow landed in 2015. A program called Claudico, built at Carnegie Mellon University, faced four elite professionals in heads-up no-limit Hold'em. It lost — but only barely, close enough to make clear that the gap was closing at an unsettling speed.

Two years later, in 2017, came the moment that changed everything. Libratus, the successor built by Tuomas Sandholm and Noam Brown, sat down again across from four of the best players on the planet at the Rivers Casino in Pittsburgh. Over more than 120,000 hands and twenty days of play, Libratus didn't just win — it crushed them. The pros, who took turns trying to find a crack in its game, described the experience as playing against something that adapted faster than they could invent strategies.

Pluribus and the six-player barrier

Heads-up (two players) was an enormous challenge, but a theoretically tractable one: with a single opponent, the problem has a clean mathematical structure. A full table was another matter entirely. With six players, the combinations of decisions explode and classical theory gets blurry — there is no single "perfect" strategy against five opponents at once.

That's why Pluribus, unveiled in 2019 by a team from Carnegie Mellon and Facebook AI, mattered so much. It was the first AI to consistently beat elite professionals at six-player poker, the format almost everyone plays online. And it did so with something that stunned the community: it trained itself by playing against copies of itself, without studying a single human hand, and it ran on a surprisingly modest amount of computing power. The myth that you needed supercomputers to conquer poker collapsed overnight.

The machine didn't win because it read bluffs better. It won because it bluffs at a mathematically perfect frequency, impossible to exploit — and it does so with no nerves, no ego, and no fatigue, ever.

From intuition to solvers: how study changed

The deepest impact of AI didn't happen in a casino full of cameras and reporters, but on the screens of millions of anonymous players. Because the same family of algorithms that created Libratus and Pluribus ended up packaged into accessible study tools: solvers.

A solver is a program that calculates the theoretically optimal play — what's known as GTO, or Game Theory Optimal — for almost any situation you feed it. You describe the hand, the stacks, the texture of the board, and it hands back exactly how often you should bet, check, raise or fold in order to become impossible to exploit. What used to be intuition accumulated over hundreds of thousands of hands is now a numerical answer you can look up in minutes.

That turned poker study upside down:

• The end of "magic rules." Sayings like "never call an all-in on the river with middle pair" fell apart. The solver shows that almost everything depends on context: sometimes that call is a mistake, sometimes it's mandatory.
• The language became technical. Concepts like range, equity, bet frequency, blockers and exact bet sizing went from the jargon of a select few to the everyday vocabulary of any serious player.
• The barrier to knowledge dropped. A beginner today can access, for a few dollars a month, strategic wisdom that fifteen years ago was the exclusive territory of a handful of professionals.
• The field leveled upward. The average player of 2026 is meaningfully stronger than the average star of 2010, simply because everyone studies with the same tools.

There's a healthy tension in all of this. GTO is a defensive strategy: it makes you unexploitable, but it isn't necessarily what wins the most money against an opponent who makes specific mistakes. Against an amateur who calls too much, the solver's "optimal" play leaves money on the table; there, exploitative play — deliberately deviating from theory to punish a specific leak — wins more. The best players today use the solver as a compass, not an autopilot: they study the perfect line to understand why it's perfect, and then decide when to step away from it.

The other side: the war on bots and collusion

If AI can beat the best humans, the uncomfortable question is obvious: what stops someone from hooking one up to an online table and stealing everyone's money? Nothing, in principle. And that's why poker rooms have been fighting a quiet, permanent war against bots for years.

The irony is that the best weapon against cheating AI is, precisely, more AI. Serious rooms deploy security systems that analyze how each account plays, looking for fingerprints a human is unlikely to leave:

• Superhuman patterns. A bot tends to play with a consistency that's too perfect, with mechanical decision times and frequencies that never waver with tiredness or emotion. Detection systems hunt for exactly that unnatural regularity.
• Device fingerprints. Room software tracks technical signals from the machine and from mouse behavior to detect automation or the use of real-time solvers during a hand.
• Collusion detection. The same technology tracks accounts that "conveniently" never clash hard, share information or pass chips back and forth — the classic signature of illegal team play.
• Human review and refunds. When the system raises a flag, a security team investigates, closes accounts and, at regulated rooms, refunds the stolen money to the victims.

It isn't a war that's been won — it's a constant arms race between those building subtler bots and those building sharper detectors. But for the honest player, the practical takeaway is reassuring: choosing a licensed room with a reputation for security matters more today than ever, and those rooms have AI working on your side.

What this means for the everyday player

This is where it pays to turn down the noise. It's easy to read headlines about unbeatable superintelligences and conclude that poker is "solved" or ruined. It isn't — not remotely. That a program exists capable of beating the best in the world doesn't change what happens at your micro or small-stakes table on an ordinary Tuesday night.

The sensible conclusion isn't paranoia, but studying smarter:

1. Use the tools, don't fear them. The very ideas the pros use are now within your reach. Studying a few concepts about ranges and frequencies puts you ahead of most opponents who still play on hunches.
2. The goal isn't to memorize the solver. No human can reproduce GTO from memory in real time, and you don't need to. What's valuable is understanding the principles: why you bet, how a range is built, when a bluff makes sense.
3. Against humans, the exploiter wins. Your low-stakes opponents make huge, repeated mistakes: they call too much, they bluff fearfully, they don't defend their blinds. Spotting and punishing those leaks pays more than any perfect textbook line.
4. The human side still matters. Managing tilt, bankroll discipline, reading a specific opponent's tendencies, patience — none of that lives in a solver, and it still decides who's ahead at the end of the month.

Put another way: AI raised the ceiling of knowledge, but it also left a ladder for anyone to climb. The player who sits down to study with curiosity has more resources today than at any point in the history of the game.

Where the game goes next

The future doesn't point toward tables full of robots, but toward something more interesting: a symbiosis between human instinct and machine analysis. The clear trend is assisted training — apps that review your played hands, flag your mistakes against the optimal line and drill you the way an infinitely patient coach would. Study becomes personal, measurable and far faster.

At the same time, the industry is pushing in the opposite direction to protect the live game: more closely monitored tables, formats that reward adaptation over memorization, and an increasingly mature conversation about which aids are legitimate and which cross the line. The debate over the ethics of tools — what you may use while playing versus only for study afterward — will be one of poker's big arguments in the years ahead.

What won't change is the essential part. No machine can feel the tension of an all-in for your whole stack, or the satisfaction of reading someone correctly and pulling the trigger at exactly the right moment. AI solved the theory; it can never take the game away from us. If anything, by clearing away so much fog, it made poker deeper: now that the "correct" answers are on the table, winning depends, more than ever, on how and when you choose to depart from them.