This is a reply to a critique of Bickhard’s interactivist model (as well as of other action-oriented cognitive theories). The authors address various misunderstandings, and the most important one is the preconception that a representation is some kind of data structure that we use. This resonates with the last part of my essay on neural coding (Brette, 2018), where I argue against idea that there are neural “codes” that the brain uses, as if the coding variables were some kind of processor registers that the brain can manipulate, whereas in reality they are measurements of events (spikes).

The authors note that we seem to have a tendency to structuralize processes, but that contrary to our intuitions, data structures (eg “codes”) are not necessary. There is an interesting formal argument from abstract machine theory, which are about automata with transitions between states. Each time a binary variable needs to stored for later recall and manipulation, this can be equivalently obtained by inserting a transition to two submachines, the one in which the variable is 0 and the other in which it is 1. The authors conclude that “Items, memory, data, structure, etc. can do nothing relevant except influence process flow, and those influences can, in principle, always be built directly into the process organization”. Hence the claim that representing is not something we use, but something we do.