Áòøøðððððò Òò Úúóööð Óùòòòööö×

S ott A. Walla e and John E. Laird Arti ial Intelligen e Laboratory University of Mi higan Ann Arbor, MI 48109, USA fswalla e,lairdg umi h.edu ABSTRACT In this paper, we examine Newell's de nition of an agent's intelligen e. The de nition implies a very stri t orrelation between the agent's knowledge, its mind, and its behavior. Based on this orrelation we argue that a on ise and well{formed representation for the agent's behavior is often essential for measuring its intelligen e. To meet this need, we present Behavioral Bounding, a method that an be used to al ulate a on ise, high{level representation of behavior. We examine both the theoreti al ommitments of this approa h, as well as pra ti al limitations.