Dynamical-systems view of motor cortex

The dynamical-systems view holds that motor cortex is better understood as a pattern-generating dynamical machine than as a look-up table that represents movement parameters. Its activity is modelled as a state that evolves by its own rules, dz/dt = f(z) + inputs, where preparatory activity sets an initial condition and the ensuing movement-period activity is largely the autonomous, roughly lawful unrolling of the dynamics that ultimately drive the muscles. Under this account, the single-neuron responses that look like tuning for direction or speed are read out as byproducts of the population's trajectory rather than as an explicit representational code.

This reframing organizes several otherwise puzzling findings — rotational structure, the large condition-independent signal at movement onset, preferred directions that change over time — as expected features of a low-tangling dynamical system, and it suggests that BCIs should decode the latent dynamics that generate movement rather than instantaneous kinematic labels. It is not a claim that representation is meaningless or that inputs do not matter: the honest position is that representational and dynamical descriptions are complementary levels, and that dynamics constrains, but does not by itself identify, the underlying circuit.