Snowboarder Maddie Mastro and other athletes are using a new AI tool powered by Google DeepMind's computer vision models to prepare for the 2026 Winter Olympics
An AI model developed by Google DeepMind is giving athletes like Maddie Mastro new insights into how their bodies move
Context & Ripple Effects
Sports performance has been incorporating machine learning for several years: professional surfing coverage documented ML-based forecasting and biomechanics data, while teams were already using computer vision for injury prediction and tailored training.
This application also lands amid a broader push to build models that interpret the physical world from video and robotics data, reflected in the race to develop world models. It gives Google DeepMind a visible training use case for its computer-vision work beyond research demonstrations.
First-order effects
- Mastro and other participating athletes can use model-derived movement analysis to inform preparation for the 2026 Winter Olympics.
- Google DeepMind gains a real-world sports-training deployment for its computer-vision models, with athlete movement as the immediate input domain.
Second-order effects
- Coaches and performance staffs face pressure to incorporate comparable video-based analysis alongside established training judgment, extending the trajectory of computer-vision-led personalized drills.
- Sports-technology providers may need to differentiate through workflow integration and coaching usability, not just the ability to capture athlete video.
Third-order effects
- If such tools prove useful across sports, biomechanics analysis could shift from specialist, staff-led review toward more repeatable model-assisted feedback; access to high-quality video and training workflows would become a key differentiator.
- The story is an early example of advanced AI labs commercializing physical-world perception capabilities through vertical applications, though lasting adoption will depend on whether teams trust the outputs in competitive settings.
The trend: Computer-vision and world-model research is moving from general physical-world understanding into domain-specific decision tools for human performance.