Lifelong motor learning as collective motion in a space of gestures

Complex motor behaviors such as speech or gait consist of sequences of simpler gestures that recur with context-dependent variation. While previous work has examined how single gestures adapt and generalize, less is known about how a rich repertoire of gestures is maintained over a lifetime. Here, I propose a model of lifelong motor learning inspired by flocking dynamics. In the model, neural representations of vocalizations locally attract each other in motor space, analogous to how birds align in flight. This mechanism reproduces how Bengalese finches adapt a song syllable across contexts, balancing divergence with local generalization. Theoretically, it shows how the brain integrates two opposing strategies: decomposition and knowledge transfer.

A shared mathematical structure – an isomorphism – emerges between collective motion, motor learning, and evolutionary speciation. Each phenomenon involves game-theoretical decisions on which connections to maintain or break as objectives diverge. Breaking similarity-based links can also yield new, complementary ones, enabling entities to fit together within larger systems, such as song syllables in a melody or elements in a Turing pattern.

Finally, I suggest that disordered gesture connectivity may underlie abnormal movements after stroke, representing the brain’s attempt to regain stability through temporary motor module merging. Once representations are stable enough, training across gradually diverging contexts could be used to re-separate these modules. The proposed framework offers a quantitative basis for understanding and optimizing motor recovery.