Abstract

Register to attend in person

What does it mean when we say that the brain implements a computation? In this paper, I build on recent work on idealisation to suggest that we should re-think this question about computational implementation. First, it is a mistake to approach the problem in the abstract, by reflecting on physical computation in a topic-neutral way. It is essential to have an idea of why theorists apply the notion in certain domains, why they feel motivated to provide a specific computational model of a physical system, and what benefits they regard flow from doing so. Second, an underappreciated feature of computational descriptions is that they involve a major degree of abstraction and idealisation. Normally, only a handful of physical properties of a target physical system feature in a computational model and these are themselves idealised in ways that depart from reality. The dynamics of a select, idealised group of properties are the fare of a computational model. I suggest that one should expect this rationale to be reflected in conditions of computational implementation. I argue that this explains the appeal of rival, incompatible theories of implementation among philosophers: in the real world – and in particular, in cognitive neuroscience – implementation is often constrained in different ways for different ends.