Mechanistic explanation claims that in order “[t]o explain a phenomenon, […] one has to know what its components are, what they do and how they are organized […]” (Craver & Kaplan 2011, p. 269). It does not suffice to merely be able, e.g. to accurately predict a phenomenon. Craver & Kaplan (2011, p. 271) show this by referring to the example of a heat gauge on a car. Despite the fact that the gauge represents engine heat and that one can also predict when the engine will overheat by looking at the gauge, it doesn’t explain why the engine is overheating. It only states that it is—not how it came about. Thus, mechanists introduced the “model-to-mechanism-mapping” (3M) requirement for explanatory models:
(3M) A model of a target phenomenon explains that phenomenon when (a) the variables in the model correspond to identifiable components and organizational features of the target mechanism that produces, maintains, or underlies the phenomenon, and (b) the causal relations posited among these variables in the model correspond to the activities or operations among the components of the target mechanism. (Kaplan 2011, p. 272)
This requirement can serve as a demarcation criterion as to when a model can actually be seen as explanatory. But how does mechanistic explanation progress? Two principal approaches are described by Craver & Kaplan (2011): reductionism and integrationism. The former tries to reduce mental phenomena into ever-smaller entities. Its most radical form, “ruthless reductionism”, is advocated by John Bickle (2003), who states that neuroscience should reduce “[…] psychological concepts and kinds to molecular-biological mechanisms and pathways” (Bickle 2006, p. 412). In other words, mental phenomena should be explained with low-level concepts. The integrationist approach, on the other hand, claims that explanations can be found across a hierarchy of mechanisms (Craver 2007), since every mechanism is itself embedded into a higher-level mechanism. Consequently, reductionism isn’t the only option, since “[…] mechanistic explanation requires consideration not just of the parts and operations in the mechanism but also of the organization within the mechanism and the environment in which the mechanism is situated” (Bechtel 2009, p. 544). In particular, multilevel mechanistic explanations consider three viewpoints on any given mechanism: the etiological, constitutive, and contextual aspects (Craver 2013). At the etiological level, the causal history of a given mechanism is investigated at the same level of the hierarchy. Yet mechanisms can also be broken down into smaller, more specialised mechanisms. When investigating the internal mechanisms that give rise to a mechanism at a higher level, one can speak of the constitutive aspect of mechanistic explanation. This strategy resembles reductionism most. But, as mentioned before, every mechanism is also embedded in a higher-level mechanism. Thus, one must also investigate how a given mechanism contributes to the next higher-level mechanism. This has been termed the contextual aspect, because it situates a mechanism into a higher-order context. After this short introduction into mechanistic explanation, the next section will show how this relates to the problem above, i.e., that applications of the free energy principle operate with functional concepts and thus can’t serve as full explanations.