Skill Is Pruning Degrees of Freedom

01The evidence

• Minimum-jerk formulations applied to redundant arms show that a unique solution requires an additional constraint, which is the mathematical signature of reducing dimensionality in a redundant system.IET Control Theory
• Controllers that produce smooth movement with feedback do so by projecting a redundant system onto a low-dimensional policy, which is the mechanism a dimensionality account requires.Neuromorphic control

02The tension

Skill is measured by outcome: faster, more accurate, more consistent. The Bernstein insight that experts coordinate rather than freeze is usually treated as descriptive. Dimensionality is rarely measured, and when it is, the results are interpreted as a correlate rather than as the learning itself.

03The theory

A limb has more degrees of freedom than a task requires. Learning a task means finding a low-dimensional subspace in which the movement is reliable, which is equivalent to freezing some combinations while allowing others to vary.

If that is what learning is, then dimensionality should fall before performance improves, because the reduction is the precondition for consistency. It also predicts that the specific subspace matters: two learners who reach the same score with different subspaces should transfer differently.

The prediction is measurable with existing motion capture. Estimate the dimensionality of the variability across trials, and track it through learning. The account says it should drop first and performance should follow, which is a temporal ordering that a performance-only account cannot produce.

This also explains why experts can be perturbed without losing performance: they are controlling fewer dimensions, so the perturbation has fewer places to enter.

04What this is not

05Predictions

• P1Movement dimensionality should fall before performance improves.
• P2Learners who converge on different subspaces should transfer differently despite equal scores.
• P3Experts should be more robust to perturbations that act outside their controlled subspace.

06What would refute it

Dimensionality unchanged while performance improves would leave coordination as a correlate.

07What it would change

It would give skill a measurable internal variable, and change how training and rehabilitation are assessed.