Abstract

Traditional epistemology has it that the pursuit of knowledge is predicated on two inter-connected goals: the generation of meaningful truths and the avoidance of error. This is neatly summarized in the conventional definition of knowledge as justified true belief In the following thesis I trace the evolution of an alternative account of knowledge which is predicated not on the avoidance of error but on the capacity to learn from error. I contend that the connectionist model of artificial intelligence provides the necessary framework for an understanding of cognition in which knowledge emerges as a dynamic product of learning. Epistemic content in this alternative is not comprised of fixed representations. Instead, content is encoded within shifting patterns of activation among large numbers of processing units. However, a connectionist approach does not give rise to a new epistemology. Rather, when integrated with Quinean naturalism, it fulfills the project of naturalized epistemology in ways that psychology could not. The convergence of connectionism and naturalized epistemology then embodies the normative principle that we should be fallibilistic about beliefs and realistic about believers.