Models of Knowledge for Resource Bounded Agents

Abstract

We know things about the world in spite of our cognitive limitations and imperfections. Occasions of stress impact memory retrieval, resources for attention can be depleted by non-epistemic factors, and our visual system has limited resolution and discriminatory ability. Yet we know many propositions, ranging from the mundane to the arcane, and we often are able to know that we know these things. In this dissertation I explore the relationship between our cognitive limitations and the limits to what we know, and what we know that we know. I begin by considering a simple model of knowledge. Because it is difficult (perhaps impossible) to have intuitions about many higher-order or iterative knowledge claims ("I know that you know that she knows that I know that ..."), a modeling approach can help clarify and explain how various cognitive limitations impact knowledge and higher-order knowledge. In Chapter 2 I discuss the epistemic requirements for the rational coordination of our actions. While it may seem that coordination is rational only if each coordinating member has what may be called "common knowledge" of some relevant proposition, the model of knowledge I employ helps show the informational complexity of common knowledge. I argue that common knowledge is unattainable. In Chapter 3 I discuss epistemic closure. Perfectly ideal agents may know every deductive consequence of what they know, but if the aim is to understand how deduction extends human knowledge then it is necessary to model our cognitive access to information. In Chapter 4 I turn to the issue of higher-order or iterative knowledge. I argue that memory limitations and various information processing errors all result in failures of higher-order knowledge. The argument I give does not require epistemic closure or a principle of self-knowledge. I conclude, in Chapter 5, by discussing interpretive issues for models of knowledge and I discuss our awareness of what we know and what we do not know.