AI can describe what players do on the field but stumbles badly beyond that. A new study tested top AI models on sports footage and found them floundering.

Researchers at two universities gathered 35,000 hours of basketball, soccer, and hockey videos, alongside millions of annotated plays and expert commentary. They wanted to see if AI could master perception, reasoning, simulation, and agency — all crucial for understanding group dynamics in sports.

The models nailed perception about 74 percent of the time. That means they could identify who did what and when, but that success rate would get a beginner announcer fired. For context, humans easily outperform this in real-time commentary.

More complex tasks exposed AI’s weaknesses. Causal reasoning — explaining why a play happened — hovered around 40 percent accuracy. When asked to predict what happens next, AI’s guesses were no better than coin flips. Simulation of player trajectories and longer sequences fell apart.

Most damning was AI’s attempt at agency: analyzing stats and trends like a human broadcaster. Accuracy dropped to a miserable 5 percent. AI could not decide which moments mattered or anticipate outcomes before the game’s final seconds.

This failure isn’t just bad news for automated sports commentary. It reveals a fundamental gap in AI’s ability to understand cause and effect and plan in dynamic, multi-agent environments. Sports provide clear outcomes, making them ideal for testing these abilities.

Humans have spent a lifetime developing causal reasoning and social intelligence. AI, despite recent advances, hasn’t caught up. Its training focuses mostly on pattern recognition and description, not on interpreting intentions or predicting complex group behavior.

Attempts to use AI in sports go beyond analysis. Some leagues eye AI for refereeing objective calls like out-of-bounds decisions to reduce human errors. But even here, the human element matters. Subjective calls, context, and game flow remain beyond AI’s reach.

Robot athletes and AI players showcase physical prowess but miss the emotional core of sports. The thrill comes from human struggle, drama, and unpredictability — things AI cannot replicate or appreciate.

Sports highlight a broader truth: AI thrives on clear rules and visible data. It falters when context, intent, and nuance dominate. Jobs requiring judgment, anticipation, and decision-making will remain human domains for now.

For knowledge workers fearing AI automation, this study offers relief. AI can describe facts but cannot yet understand or anticipate complex events. It struggles with what really matters in teamwork and strategy.

The future of AI in sports and beyond depends on bridging this gap. Until then, human analysts, coaches, and broadcasters remain indispensable. AI’s promise is real but incomplete — it can assist, never replace, the human mind on the field or off.