The main result of this thesis is the clarification of some central concepts of agent theories, namely, the concepts describing the informational aspect of intelligent, resource-bounded agents. The main technical result is a framework for establishing direct connections between an agent's knowledge and his available resources. In the thesis two epistemic concepts -- the concepts of explicit knowledge and algorithmic knowledge -- will be introduced and characterized axiomatically. It will be shown that these concepts are important for resource-bounded reasoning about knowledge and useful for describing rational, but realistic and implementable agents.
Although the thesis is primarily concerned with (Distributed) Artificial Intelligence, I am convinced that it will have a considerable impact on other fields of research, especially on philosophy and game theory.
In the philosophical literature, epistemic logic has been frequently criticized for not being able to model agents realistically. Several researchers have therefore drawn the conclusion that epistemic logic is either not possible, or it is not useful for epistemologists interested in actual knowers in the actual world ([Hoc72], [Hal95]). My attempt to model realistic, resource-bounded reasoners can be seen as a defense against those attacks. By actually specifying a theory of knowledge that can be verified empirically I will provide the evidence that epistemic logic is indeed possible.
In the field of game theory and mathematical economics, resource boundedness has been a primary concern for a long time ([Sim57]). Since Aumann's seminal work ([Aum76]), game theorists have become interested in the role of knowledge (and especially common knowledge) in games. Recent works on the epistemic foundations of games (e.g., [Bin90], [Wal92], [Bac94]) have made clear what implicit assumptions concerning the players' knowledge are made when modeling a game. Because these assumptions are recognized as too strong for realistic agents, several attempts have been made to weaken the underlying epistemic logic in order to describe players more realistically ([Bac94], [LM94], [Hei95]). My investigation can contribute to the search for a more suitable epistemic foundation of game theory.
Some results of this thesis have been published previously. Chapter 4 is based on [Ho95] and [Ho97]. Parts of chapter 5 are based on [Ho98].