An overview of how machine learning is being used in sports to assess athletic talent, detect and predict injuries, improve the betting odds, and more
Craig S. Smith / New York Times :
Context & Ripple Effects
This 2020 New York Times survey by Craig S. Smith laid out the three lanes where machine learning entered sports — talent assessment, injury prediction, and betting odds — before most of the follow-on coverage existed. The arc since then has validated each lane: MLB's partnership with Uplift Labs turned two iPhone cameras into a scouting-and-injury-risk tool, and coaches' adoption of computer vision for injury prediction moved the idea from research paper to practice drills.
The betting lane has its own dynamic: Bloomberg's reporting on Sportradar and the firms that set odds for gambling apps shows data vendors building models not just to price bets but to keep people betting. Against that sits the caution from FT-covered research that ML systems are designed to always make a prediction, which matters when the prediction is a roster spot or a wager.
First-order effects
- Teams and leagues adopting these tools — MLB with Uplift Labs, coaches running computer-vision injury models — get earlier warning on player risk and a data layer that competes with traditional scouting judgment.
- Bettors face odds increasingly set by vendor models like Sportradar's, shifting the edge in sports gambling from the bookmaker's intuition to whoever owns the best data pipeline.
Second-order effects
- Sports data vendors gain leverage over both leagues and gambling apps: the same models that price bets can be sold upward to teams, making companies like Sportradar dual-use suppliers to the betting and performance sides of the sport.
- Talent scouting becomes contestable by software — Brazil's embrace of AI-powered video scouting apps pressures traditional scout networks and pushes leagues to standardize what their evaluators measure.
Third-order effects
- If prediction systems keep spreading across scouting, health, and betting, the structural question becomes accountability: research showing ML models will always output a prediction suggests sports decisions may inherit systematic false positives, from mis-flagged injury risks to mispriced odds.
- The likely endpoint is a split sports economy where model-rich organizations — leagues, data vendors, well-funded clubs — compound their informational advantage over smaller clubs and casual bettors, with the model itself becoming the scarce asset rather than the athlete data feeding it.
The trend: Machine learning is moving from sports' back-office analytics into its three money paths — talent, health, and gambling — with data vendors positioned to own the models that all three depend on.