Harkness employs existing views in the philosophy of science to create a divide between functions and mechanisms: functions specify what some phenomenon of interest ought to be doing, they don’t specify how it actually does it. For that, a mechanism is needed which, in addition to specifying a functional role, also names the parts of the mechanism that perform this role (i.e., the realisers of the function), for example in the brain. This is thought to limit the explanatory power of FEP, which at its mathematical heart is just functionalist.
Whilst I accept the divide between functions and realisers, I don’t think there is much explanatory mileage in naming realisers. If I already know what functional role is being realized, I don’t come to understand a phenomenon better by being given the names of the realizing properties. This can be seen by imagining any mechanistic explanation (encompassing both functional role and realisers) where the names of the realizing properties are exchanged for other names. Such a move might deprive us of knowledge of which parts of the world realize this function, but this is not in itself explanatory knowledge. For example, I get to understand the heart by being told the functional role realized by atria and ventricles; I don’t lose understanding if we rename the atria “As” and ventricles “Bs”.
This is not to deny that we can gain understanding from learning about mechanisms. In particular, if I don’t know about a phenomenon of interest, then I might explore the realizer of a particular case, and thereby get clues about the functional role. For example, in the 17th century William Harvey was able to finally comprehensively explain the functional role of the heart by performing vivisection on animals. Indeed, the point of such an exercise is to arrive at a clear and detailed description of a functional role (recall the difference between behaviourism and functionalism is that for the latter, the functional role is not just an input–output profile but also a description of the internal states and transitions between states).
Importantly, exploration (e.g., via vivisection, or via functional magnetic resonance imaging) of a mechanism is not the only way to eventually arrive at explanations. There can be multiple contexts of discovery. In particular, there can be very broad empirical observations as well as conceptual arguments. In the case of FEP, a key observation is that living organisms exist in this changing world. That is, organisms like us are able to maintain themselves in a limited number of states. This immediately puts constraints on any mechanistic explanation, which must cohere with this basic observation. Further, since an organism cannot know a priori what its expected states are, there must be an element of uncertainty reduction going on within the organism in order to estimate its expected states, or model. In a world with state-dependent uncertainty, this must happen through hierarchical inference. With these simple notions, FEP itself is well on its way to being established.
So I don’t think it is explanatory power that is limited by being confined, as FEP fundamentally is, to functional roles. This mainly seems to impose a limit on our knowledge of which objects realize a given functional role, or it might curb our progress in finessing the functional role in question. Whereas it is right to say that FEP is limited because it is merely functional, this limit does not apply to its explanatory prowess.