The World Series of Poker returns to ESPN for the first time since 2021, and the broadcast arrives with an AI system trained to look for the physical tells players spend careers erasing. The tool was built by independent AI engineer Luke Geel and enlisted by Peyton Manning’s Omaha Productions. It analyzes past results and player behavior — posture, blink rate — to flag moments when someone may be bluffing or holding the “nuts,” poker slang for the best possible hand.
Omaha is being careful with it. The AI-assisted coverage focuses on players only after they have been eliminated, which keeps the analysis away from anything that could affect a live hand. Geel’s system scans video for repeated physical patterns — a blink, a shift in posture, a change in stillness — and compares them against what happened in earlier hands. Getting there was harder than he expected. “I can’t just, like, upload a YouTube URL and say, ‘find their tells,’” he said. He is also skeptical it would find anything on the best players. Asked about running it on Daniel Negreanu, he told Sportico: “He’s not going to have any tells. He’s probably worked hard to remove them.”
Poker AI is not new. Libratus and DeepStack beat professionals at heads-up no-limit Texas hold’em years ago. Those systems played the game; this one watches the people playing it. The same idea is spreading elsewhere — in soccer, researchers have built models that read a penalty taker’s movement before the kick. David Freire-Obregón, an associate professor in Spain who has studied the problem, said future systems may help goalkeepers learn a player’s individual habits, then added: “I don’t think AI will ever make penalty shootouts predictable.” Geel said better cameras and more data could let future systems track heart rate or facial flushness. Similar tools could interest car dealers studying what excites shoppers, or coaches identifying an athlete’s tendencies.
For now this entire system is a broadcast feature rather than a factor in play. Once players know a machine is cataloging their blinks, they can train the blink away, or fake it. A tool that reads deception in a game built on deception is one more thing to deceive.
Read more at ZME Science.