Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2000.Includes bibliographical references (p. 211-224).This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.We have long wanted to create artifacts that resemble us: artifacts that not only look like us, but act, think and feel like us, with which we can have sympathy and empathy. In particular, the purpose of this thesis work is twofold: One is to understand animal behavior, the role of the motivation system and of affect, in terms of the theory embodied in the software architecture for modeling interactive synthetic characters. The other is a practical one: to build believable synthetic characters that we can interact with, utilizing our best understanding of animal behavior in creating characters that are perceived as sympathetic and empathetic to humans. We have done this with the help of a specific understanding of the roles of the motivation and affect systems of animals. To accomplish these goals, I propose a creature kernel model which is largely based on the approach of Blumberg [7]. The creature kernel is modeled as a sum of four main parts, the perception, motivation, behavior and motor systems. Among these four components, despite the fact that the motivation system plays a crucial role in daily survival of creatures in nature, its functional importance has often been neglected in attempts to create intelligent artifacts because it has been thought of as the "opposite" of rationality. Thus, in the system proposed in this thesis, emphasis is placed on the roles of the motivation system and how it acts as the integrator of the four parts of the creature kernel, and enables a creature to exist as a functional whole. The plausibility of the proposed system is demonstrated through two projects, for which the characters were built using the creature kernel. The first project is called Sydney K9.0, in which the main(cont.) character is Sydney, a virtual dog. Human participants can train the dog to do certain tricks using various physical input devices: voice command, clicker sound, milkbone box and a training stick. Learning and training phenomena are observed as operant and classical conditioning, and it can be explained how each subsystem is functioning inside of the character's mind to implement that functionality. The other project is called (void*): A cast of characters, through which three distinctive characters--Earl, Elliot and Eddie--are introduced. In this project, a human participant can 'possess' one of these three characters using buns-and-forks interface and control the possessed character's dance movement by wiggling the interface in various ways. Learning and adaptive change of attitude through the interaction as well as expression of different personalities and its effect on interaction are emphasized. How well the personalities were represented is explored through the results of a survey of a number of novice users of the system.by Song-Yee Yoon.Ph.D.