What happened
ESPN deployed an AI tool during the 2026 World Series of Poker Main Event broadcast in early July that analyzes player movements—eye motion, blink rate, posture, chip handling, and hand fidgeting—to predict hand strength (strong hand, drawing hand, bluff, etc.). The tool, designed by Luke Geel (an AI engineer for the US Air Force), was trained on footage from three camera-fed tables at the tournament, which drew over 9,000 entries.
Why it matters
Poker professionals argue the tool has fundamental limitations that undercut its value. Most players never appeared on the recorded tables, and even those who did sat there too briefly to give the AI enough data to reliably distinguish tells from situation-dependent behavior. Shaun Deeb, a two-time WSOP Player of the Year, notes that tells extend beyond what cameras capture—leg movements, verbal tics, breathing patterns—and that confidence or weakness alone does not reveal actual hand strength; context and psychology matter more.
What to watch
The tool will not be used for the final table, according to Omaha Productions (the company licensed by ESPN for WSOP coverage), though no reasoning was given. Geel acknowledged the tool benefited from a small sample size and reported mixed results in blind tests on other tournaments. High-roller tournaments with hundreds of hours of existing footage of top pros could theoretically fuel more advanced versions, though players like Deeb say in-person tells specialists still outperform filmed analysis.
Summaries like this, in your inbox every morning.
The deployment of the ESPN tells-detection tool marks a moment of tension between AI's potential and poker's deeply human nature. Unlike sports where quantifiable metrics (speed, distance, angle) map cleanly onto performance, poker tells are interpretive—a player's behavior reflects not just their hand strength but their personality, the tournament's stakes, their fatigue, and their opponent dynamics. Geel's system attempts to automate the pattern-recognition work that specialist coaches perform in person, but the fundamental constraint is data scarcity: a tournament broadcast captures only a fraction of participants, and even those top players appear on screen too infrequently to build the robust dataset that machine-learning models require.
Professionals like Michael Gagliano and Shaun Deeb articulate a second, deeper problem. Even if the AI perfectly identified when a player appeared confident or hesitant, it would still miss the context that converts observation into strategy—the psychological layers that separate a weak two-pair from a strong one. This gap between visual pattern and intention is not a limitation the tool's creators deny; Geel himself acknowledged that a larger sample would help and reported mixed results in blind tests. For now, the tool functions as broadcast entertainment rather than a competitive edge, a fact underscored by ESPN's decision to exclude it from the final table. Whether future iterations with more data could narrow the gap remains open, but for the moment the human tells specialist—watching in person, asking questions, building rapport—remains the gold standard.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
