Representation and rationality : foundations of cognitive science
In this essay I consider the foundations of a particular approach to cognitive science. There is a belief, among some, that cognitive research ought to proceed in two steps. In the first step, cognitive systems ought to be interpreted as rational beings, endowed with knowledge of their environments and motivated by goals and desires. Questions about the modularity of knowledge and about the sort of knowledge that is necessary or sufficient for a given competence are addressed at this stage. In the second step, inferences are to be made about the design of control systems that might instantiate these kinds of knowledge states, goals and rationality. Studies at the knowledge level are meant to serve as guidelines in the search for mechanisms. Problems arise as soon as one asks for the justification of this view. For instance, one often hears that knowledge is attributed by a process of interpretation that is subjective; it depends on the scientist having a 'manual' for interpreting behaviour that has no foundation in fact. Knowledge is essentially observer relative; it designates nothing intrinsic in a system. In Chapter Three I argue that this position is false. By introducing the notion of 'robustness 1 as the touchstone of realism I suggest that knowledge states are potentially as robust as any in science. Moreover realism about knowledge does not entail accepting what Fodor has called the Language of Thought hypothesis. We can reason about knowledge states in abstraction from the various ways knowledge can be implemented in a system. The language of thought is just one of many ways that knowledge can be used by a system. Hence there is no simple way to move from an account of what a system knows to how it uses or has access to that knowledge. In Chapter One I argue that the step from 'knowledge theories' to 'process theories' is more complicated than language of thought theorists suppose. In Chapter Two I discuss the basic methodology of research at the knowledge level. Any well-defined task imposes severe constraints on the way it can be accomplished. The discovery of these constraints and the consequences that flow from them is perhaps the central job of knowledge level research. I conclude the thesis with two chapters on the limitations of knowledge level research. Given that the more structured and rigid a task environment is, the more determinate the knowledge that is necessary or sufficient for task competence, we would expect that tasks and environments which are more open-ended, less closed to intervention from outside interference, would not submit to knowledge level research. Relying on a distinction between peripheral and central cognitive faculties, I question the prospects for knowledge level research of central faculties. Unlike the problem of vision or muscular co-ordination, the problem of deliberation is radically open-ended. Too many factors might become relevant to bound the class of task knowledge that might become vital.