Mechanisms in psychology: ripping nature at its seams

Recent accounts of mechanistic explanation suggest that models are only explanatory insofar as they map neatly onto lower level components (Craver, 2006; Piccinini and Craver, 2011; Kaplan and Craver, 2011). In the case of extensions into the cognitive sciences, this has meant interpreting the ‘integration’ of cognitive psychology and neuroscience to mean treating cognitive models as scaffolding for more detailed neural mechanisms. Bechtel (2008) and Piccinini and Craver (2011) describe this sort of integration as a seamless process where the functionally-defined components of cognitive models are localized in parts of the brain. One critic of this view (Weiskopf, 2011) has responded by denying that cognitive models provide mechanistic explanations, and arguing for the independent value of functional explanations of cognitive phenomena. I think the problem with these accounts of integration lies elsewhere. I argue that models may be mechanistic even if they don’t align tidily with lowerlevel ones. The problem with seamless accounts of cognitive-neural integration is their seamlessness, not that they take cognitive models to be mechanistic. The non-componential view of mechanisms described in Machamer et al (2000) allows for cognitive and neural models that cross cut one another, and for cognitive models that don’t necessarily decompose into neural parts. I critique three aspects of seamless accounts of integration: the claim that cognitive models are elliptical mechanism sketches; the assumption that cognitive mechanisms must be decomposable and localizable in neural parts; and this assumption being promoted to the status of a normative constraint on mechanisms. I argue that the integrations we can realistically expect are more partial and patchy than the mosaic unity Craver (2007) describes.