Abstract

On many of the idealized models of human cognition and behavior in use by philosophers, agents are represented as having a single corpus of beliefs which (a) is consistent and deductively closed, and (b) guides all of their (rational, deliberate, intentional) actions all the time. In graded-belief frameworks, agents are represented as having a single, coherent distribution of credences, which guides all of their (rational, deliberate, intentional) actions all of the time. It's clear that actual human beings don't live up to this idealization. The systems of belief that we in fact have are fragmented. Rather than having a single system of beliefs that guides all of our behavior all of the time, we have a number of distinct, compartmentalized systems of belief, different ones of which drive different aspects of our behavior in different contexts. It's tempting to think that, while of course people are fragmented, it would be better (from the perspective of rationality) if they weren't, and the only reason why our fragmentation is excusable is that we have limited cognitive resources, which prevents us from holding too much information before our minds at a time. Give us enough additional processing capacity, and there'd be no justification for any continued fragmentation. I argue that this is not so. There are good reasons to be fragmented rather than unified, independent of the limitations on our available processing power. In particular, there are ways our belief-forming mechanisms—including our perceptual systems—could be constructed that would make it better to be fragmented than to be unified. And there are reasons to think that some of our belief-forming mechanisms really are constructed that way