Logics for Agents with Bounded Rationality
Zhisheng Huang

Abstract:
Bounded Rationality has two interpretations: a wide one and a narrow one. 
In the wide interpretation, {\em bounded rationality} refers to the phenomenon 
agents have limited cognitive resources and capabilities. In the narrow 
interpretation, bounded rationality refers to the notion raised by H. A. Simon.
He considers a general decision procedure for a rational agent who would not 
know all action alternatives, nor the exact outcome of each, and would lack a 
complete preference ordering for those outcomes. This thesis proposes several 
logics for agents with bounded rationality in both the wide interpretation and 
the narrow interpretation.

For the wide interpretation of bounded rationality, The thesis focuses on
the phenomenon of belief dependence in multiple agent environments, where 
{\em belief dependence} refers to the phenomenon that some agents rely on 
someone else about their beliefs, knowledge, or information because of their 
own limited information or beliefs. A general methodology for the study of 
belief dependence is presented. Several logics for belief dependence are 
proposed. The soundness, completeness and decidability of those logics are 
studied. Moreover, the logic for belief dependence is used to analyse the 
Schoenmakers problem. The possibilities and impossibilities of strategies for 
dealing with this problem were investigated.

For the narrow interpretation of bounded rationality, the thesis focuses on
the studies of action logics for agents with H. A. Simon's bounded rationality
in order to develop a formal language for social science theories, in
particular for theories of organization. we are knitting together ideas from 
various strands of thought, notably H.A. Simon's notion of {\em bounded
rationality}, G. H. Wright's approach to {\em preferences}, Kripke's
{\em possible world semantics} in combination with binary modal operators, 
Stalnaker's notion of {\em minimal change}, and more recent ideas from 
{\em belief revision} and {\em update}. The resultant logics are called ALX. 
The soundness and completeness of ALX logics are proved. Some plausible 
applications of ALX logics towards a formal theory of social agents are 
discussed.