AI is getting better at measuring athletic performance, but turning those measurements into training that actually helps remains the harder test. In an opinion article for CGTN, Beijing Sport University professor and Noitom co-founder Liu Haoyang argues that coaches need reliable, sport-specific guidance, not another dashboard full of numbers.
Watch Desk analysis
What happened
Liu points to several uses already under way: AI-assisted tracking at the Milano Cortina Winter Olympics, motion analysis used by China’s diving team, and TacticAI, a Google DeepMind and Liverpool FC system that proposes alternatives for football corner kicks. In a blinded expert assessment, TacticAI’s suggestions were preferred to existing setups 90 per cent of the time. That shows experts found the options promising, not that they produced more goals; Liu notes that Liverpool players had not tried them on the pitch when the research was published.
The gap between measuring and improving performance is the article’s central point. Camera-based motion capture can estimate joint angles inaccurately by around four to five degrees, Liu says, while injury-prediction models may flag risks too broadly or too far in advance to guide a coach’s next session. Correlation with performance, he stresses, does not prove that changing a measurement will improve it.
Why it matters
A system can spot movement patterns that are difficult to see and still fail to tell a coach what to change. Athletes differ in anatomy, injury history and competitive demands, so a universal “ideal” technique is unlikely to be much use to everyone. The practical question is whether a recommendation is accurate enough, timely enough and specific enough to test with a particular athlete.
Liu’s proposed destination is a training loop: measure an athlete, suggest a specific adjustment, then check whether the next session produced the intended result. That would make the system accountable to outcomes, rather than simply rewarding it for producing precise-sounding analysis.
Our read
This is a valuable, practitioner-led reality check on sports AI, not a claim that the technology is useless. The examples show real progress in tracking and tactical analysis; they also show why a striking benchmark or slick assistant is not the same as better training.
Coaches should treat AI recommendations as testable suggestions, not instructions from an all-knowing touchline oracle. The useful measure is whether the advice improves performance for the athlete in front of them, and whether the system can show its working.
What to watch
- Whether TacticAI’s suggested corner routines are tested in real matches.
- Whether sports-AI systems report measurement error alongside their performance claims.
- Whether recommendations are validated over repeated sessions with individual athletes.
- Whether coaches can check how a system reached its advice, rather than being handed a polished answer.
Discussion spark: Should AI coaching tools have to prove that their recommendations improve an individual athlete’s results before coaches rely on them, or is useful measurement enough to justify trying them?
Sources and evidence
- Hawk-eye took 25 years. How long will the AI coach take? – CGTN (5 November 2026, 14:35 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.