저 포커 선수, 블러핑일까? AI가 판독한다

Comment Loader Save Story Save this story Comment Loader Save Story Save this story For serious poker players, the ability to sniff out the “tells” that expose an opponent’s intentions is nearly as important to winning as the cards themselves. Many gamblers have made careers out of their ability to decipher the meaning of everything other players do at the table—their conscious movements, their body language, and their subconscious tics, all of which might reveal their strategy—as a method of gaining an edge in this game of incomplete information. It’s understandable, then, that ESPN’s use of a new “ AI tells detection ” tool during the 2026 World Series of Poker Main Event broadcast stoked some serious debate within the poker community. The tool began appearing periodically during the first few days of the tournament’s live broadcast in early July. A text overlay displayed various live metrics on a player’s movements, plus a “hand strength model” chart breaking down different possibilities of the type of hand a player might be holding. The tool looks slick, but a viewer might naturally wonder how accurate its data is, or how the AI came to know the players’ tics and gestures well enough to venture such a guess. Is the tool just a neat party trick—or a silly one, depending on your sensibilities? Or is it an attempt to haphazardly stuff AI into the inherently human pursuit of poker, threatening the game’s soul and future? Do Tell Hundreds of pros on the poker circuit specialize in spotting tells. This new tool, designed by an AI engineer for the US Air Force named Luke Geel, purports to digitize that process. It’s watched every hand captured on camera in the 2026 WSOP Main Event to build a tells database on various players. The system gathers inputs on the players ranging from eye movements and the rate at which they blink, to the players’ posture, chip handling movements, “hand fidget” metrics, and more. It analyzes that data and the outcomes of each hand to predict the likelihood of which general hand type a player might have: A strong made hand, a drawing hand, a bluff, and so on. The poker experts I spoke to are skeptical about the tool’s effectiveness—especially since it was trained on such a small amount of data. The 2026 edition of the WSOP Main Event tournament drew over 9,000 entries, but the vast majority of those players never spent time at one of three tables that were being recorded by cameras. (The same camera feeds used for the broadcast were also used to train the AI tool). Even those who did sit at those tables weren’t there long enough for the system to build a robust dataset that covers the vast range of situations possible in poker. “The streams are varied enough that you don't get the same players too frequently,” says Michael Gagliano, a 17-year poker professional who made the Main Event final table this year and is playing for the $10 million top prize this week. Gagliano, who started the final in eighth chip position, says he went back through every second of ESPN’s live streams during the two-and-a-half-week break after the final table was reached in mid-July, combing for any tells or info he could pick up on his remaining opponents. But that lack of screen time for any one player limited his ability to spot the tells, even for players who made it all the way to the final table and spent lots of time on camera. “I don't know how much actual information I'm going to be able to act on from what I saw,” he says. Any AI looking at the footage would face the same issue, even for a tournament as long as the Main Event. More Than Just Cookies Tell detection is nuanced work, and pros are dubious that a camera-based AI tool can effectively do it better than a human. Most nonplayers’ exposure to the importance of poker tells comes from the penultimate scene in the 1998 film Rounders . Matt Damon’s character, Mike McDermott, folds a monster hand to John Malkovich’s Teddy KGB after recognizing that the gangster has him beat—and McDermott discovers this after spotting a tell based on KGB’s habit of eating Oreos at the table. It’s arguably the most memorable poker scene in movie history because it perfectly expresses the battle of wits that underlies every poker game, even though in reality it’s quite reductive. “To reference the Rounders Oreo cookie tell, it’s a little more abstract than that,” says Shaun Deeb, a two-time winner of the WSOP Player of the Year award and one of the most recognizable players in the game. (Deeb also made a deep run in the 2026 Main Event, finishing 15th.) “Physical tells are so much more expansive than I think the public realizes,” Deeb says. “There are leg tells, checking tells, verbal tells, breathing tells, pulse tells. There's an insane amount of tells available, and most of those can't be picked up by a camera.” An AI can track visual and audio patterns, but it can’t deduce intention; the former is only so valuable without the latter. Even if the tool was hypothetically perfect at determining when a player was projecting confidence or weakness, that alone isn’t a road map to deciphering their actual hand. “How strong is two pair to one player versus another player?” says Gagliano. “Maybe someone is extra confident with a hand that’s actually weak for the situation, but for some reason they think they have the best hand, so they’re really confident. “Maybe if I was playing a casual tournament, I would think my two pair is extremely strong. But in the Main Event I'm still a little nervous, because it's a high-stakes situation. So maybe my body language is referencing the situation rather than the hand strength.” For a broadcast entertainment tool, those flaws aren’t necessarily a deal-breaker. No one is expecting some all-knowing oracle—Geel, the tool’s creator, least of all. He’s transparent about the fact that a larger sample of hands would be better for his tool, telling WIRED via email that he’s run some blind tests on other poker competitions with mixed results. Maybe the feature adds value for some ESPN viewers, though players like Deeb are skeptical even of that. “I think they randomly found something to try to make it like another sport, and I just think it was swing-and-a-miss,” Deeb says. While viewers saw the tool in action during portions of the tournament broadcast in July, a representative from Omaha Productions, a company licensed by ESPN for WSOP and other sports coverage, said in a text message that the tool would not be used for the final table. The representative declined to provide any reasoning for that decision. Watching the Detectives As AI continues to improve, even skeptics concede it’s possible tools like these evolve rapidly and will likely be applied for financial gain. Within poker’s “high-roller” tournament scene, where the buy-ins frequently reach six figures, a small pool of mostly recognizable professionals play each other in events that are often broadcast. It’s possible that hundreds or even thousands of hours of footage exist of these top players, many of whom play dozens of such events every year. It’s already common for poker players to study streamed and broadcast footage to gather info on their regular opponents. Could improved AI optimize that very human process? Deeb, for one, isn’t worried. As a top pro, he’s frequently been hired to coach players as they make deep runs in the Main Event; he says that process has often included bringing in a hand-picked live tells specialist to observe both opponents and the client themselves (to see if they have any glaring tendencies that should be corrected). A close friend of Deeb’s was watching the streams during his run this year as well, doing the same thing on his behalf. While he’ll use recorded footage if it’s the only option available, Deeb says the filmed route isn’t optimal. “The teams I hired, we always had a spot for the person spotting the tells to be watching the player in perso