This epistemological paper makes a number of important points about what constitutes an explanation in neuroscience, with many interesting pointers to relevant ideas in philosophy and psychology. A strong tendency in current literature is to identify neural circuits that are « necessary » for a particular behavior, in that a lesion would suppress the behavior, and « sufficient » for that behavior, in that activation of the circuit produces the behavior. To what extent do these observations constitute a biological explanation of the behavior ? The author points out that such explanations focus exclusively on efficient cause, that is, the source of the change (the firing of a neuron causes the contraction of a muscle). Interestingly, as noted by the author, this is typically not the type of explanation that physics produce. For example, a physical explanation of the spherical shape of a bubble does not typically involve a chain of molecular events, but rather a more global explanation based on the minimization of energy by the entire structure (in philosophical terms, a final cause).

It is not clear that efficient causes are a very relevant concept for biology. First, as the author notes : « Causal manipulations do not produce the behavioral effect : they reproduce it in a given context. ». Thus, to claim that a neural circuit is « sufficient » is highly misleading, because it works in the context of the entire organism ; it is the manipulation that is sufficient, not the neural circuit. Second, a glimpse at the connections between brain areas or at a metabolic pathway should convince us that interactions between biological elements do not form a linear chain at all. Therefore, it is unlikely that as a chain of causal events can explain biological processes, in particular behavior. In addition to the internal dependencies between neurons, coupling with the environment implies that there is also circular causality between sensory inputs and the organism's actions : actions cause sensory inputs just as much as inputs cause actions. As noted by the author, the implicit view that neurons react to stimuli is only a modern form of behaviorism, which we thought had been outmoded several decades ago (see e.g. Chomsky, 1959).

The author proposes to abandon the behaviorist perspective and instead : « Rather than concentrating on models of behavior, let us conceive models that behave, and then set up our experiments so as to allow control, not of the animal, but by the animal. ». This I find inspiring for theoretical neuroscience. Instead of building models of stimulus-responses experiments and then pretend that the neural responses are somehow linked to behavior (typically through a « decoder » which is purely abstract construct), perhaps we should try to build models that exhibit some actual behavior, which means addressing the full sensorimotor loop.