The internet solved one of humanity’s oldest problems: access to information. Knowledge that once required a university library, a specialist, or proximity to the right people can now be reached in seconds. Someone learning a new skill can watch demonstrations, read expert explanations, and compare approaches almost instantly. But access to information is not the same as access to expertise.
Knowing what good performance looks like is different from having an expert watch you perform, identify what matters, and adjust the recommendation accordingly. That distinction points to a potentially more consequential role for artificial intelligence. The next opportunity for AI may not simply be generating more information. It may be helping scale parts of expertise itself.
Expertise Is a Feedback Loop Much of today’s AI experience begins with a request and ends with an answer. Ask for a training program, explanation, or recommendation, and an AI system can generate one almost instantly.
Expertise usually works differently. A skilled professional operates through an iterative process: Observe → Assess → Recommend → Evaluate → Adapt Consider a coach working with an athlete. The coach observes performance, interprets it in context, recommends an adjustment, watches what happens next, and adapts again. The value is not simply the information being communicated. It is the continuous interpretation surrounding it.
AI Has Become Good at the Recommendation Layer Generative AI has made parts of this process dramatically easier to scale. Systems can synthesize information and generate explanations, programs, suggestions, and educational material.
Yet in physical and real-world activities, a fundamental limitation remains: a system cannot meaningfully evaluate an outcome if it cannot meaningfully observe what happened.
That creates an observation gap. A training application can recommend an exercise, for example. But that recommendation becomes potentially more useful if the system can also understand how the exercise was performed and whether the result supports the next decision. Computer Vision Could Help Close the Observation Gap Advances in computer vision and markerless motion capture are beginning to make physical activity more computationally observable.
A 2024 study published in the Journal of Biomechanics evaluated Stanford’s OpenCap, a smartphone-based markerless motion-capture system, against an established marker-based system during return-to-sport movements. Across 437 recorded trials, researchers found particularly strong agreement for hip and knee motion in the sagittal plane, while agreement was lower for frontal- and transverse-plane motion. A 2023 study published in Sensors similarly evaluated a MediaPipe-based approach against a Vicon motion-capture system for gait analysis, reporting promising agreement for several temporal gait measures while identifying limitations in others. The point is not that ordinary cameras have already replaced biomechanics laboratories. They have not.
The more interesting development is that aspects of human movement that once required specialized environments are becoming increasingly accessible to computational observation.
From One-Time Answers to Continuous Intelligence That creates the possibility of moving from one-time intelligence to continuous intelligence.
Instead of generating a recommendation and stopping there, an AI system could eventually participate in more of the feedback loop: observe a result, assess what changed, recommend an adjustment, evaluate the next attempt, and adapt again. Observe → Assess → Recommend → Evaluate → Adapt That is fundamentally different from simply generating an answer from a prompt. It creates the possibility of systems in which new observations continuously inform the next decision.
Scaling Expertise Without Replacing Experts None of this requires assuming that AI will replace coaches, clinicians, teachers, or other professionals.
Expertise includes judgment, experience, communication, accountability, and contextual understanding that cannot simply be reduced to an algorithm.
A more useful question is: Which parts of the expert process can technology responsibly make available to more people?
If AI can help extend observation, identify useful patterns, maintain context over time, or support feedback between human interactions, technology can expand the reach of expertise rather than eliminate the expert.
The internet made knowledge scalable. AI’s next opportunity may be making parts of expertise scalable too.
Sources
- Turner, Chaaban & Padua, "Validation of OpenCap: A low-cost markerless motion capture system for lower-extremity kinematics during return-to-sport tasks," Journal of Biomechanics, 2024. https://pubmed.ncbi.nlm.nih.gov/38905926/
- Hii et al., "Automated Gait Analysis Based on a Marker-Free Pose Estimation Model," Sensors, 2023. https://doi.org/10.3390/s23146489