Behavioral Decomposition Must Precede Neural Investigation

This methodological argument asserts that behavioral decomposition is epistemically prior to neural investigation — a claim with deep consequences for how neuroscience should be practiced. The core problem is that the field habitually operates in reverse, recording from neurons and attempting to reconstruct behavioral explanations post hoc. David Marr's three-level framework — computational, algorithmic, implementational — serves as a corrective reminder: one must first specify what problem the organism is solving and what algorithmic strategies could solve it before asking which neural circuits implement the solution. Without this prior decomposition, neural data lacks an interpretive frame.

The examples are pointed. The spinalized cat walking on a treadmill reveals genuine properties of spinal central pattern generators, but it tells us nothing about balance, gait selection, or terrain navigation — the reduced preparation has excised the behavioral phenomenon of interest while preserving neural activation. The complete C. elegans connectome, despite cataloguing every neuron and synapse, does not yield behavioral prediction. Jonas and Kording's microprocessor study drives the point home with engineered precision: standard neuroscientific analysis methods applied to a known computational system fail to recover its actual architecture.

The proposed sequence is disciplined: identify the natural behavior, decompose it algorithmically, generate competing either/or hypotheses about mechanism, and then — and only then — use neural data to arbitrate. The sound localization case is paradigmatic. Knowing that an animal must lateralize a sound source, and that two distinct algorithmic strategies (spatial coding versus temporal coding) could accomplish this, permits the design of experiments where neural recordings become genuinely decisive. A connectome, in this view, functions as a library — enormously valuable once you arrive with the right question, but inert as a source of explanation on its own.