Electrochemical learning

Electrochemical learning is a form of physical computation in which a system modifies its own material structure — typically the growth of metallic filaments in an electrolytic solution — in response to electrical feedback, thereby learning to produce desired outputs without programmed instructions. The term is most closely associated with Gordon Pask's 1950s experiments in which acid-bath devices grew conductive pathways that could discriminate audio frequencies, classify patterns, and adapt to perturbations.

Unlike conventional computation, which separates hardware from software and stores programs in memory, electrochemical learning is a process of morphological adaptation: the hardware itself changes. The system's knowledge is not encoded in symbols but inscribed in its physical geometry — the density, branching, and conductivity of metallic threads that have grown or dissolved in response to reward and punishment signals. This makes it a precursor to contemporary neuromorphic engineering and reservoir computing, in which computation is distributed across material dynamics rather than localized in discrete processing units.

The philosophical significance is that electrochemical learning dissolves the software-hardware boundary entirely. A system that learns by growing its own circuits challenges the computational metaphor that treats mind as software running on neural hardware. In Pask's machines, there is no software. There is only structure that has been shaped by its own history of interaction — a form of autopoietic adaptation that may be closer to biological learning than to digital computation.