Goal-directed cognition is often discussed in terms of specialized memory structures like the "goal stack." The goal-activation model presented here analyzes goal-directed cognition in terms of the general memory constructs of activation and associative priming. The model embodies three predictive constraints: (1) the interference level, which arises from residual memory for old goals; (1) the strengthening constraint, which makes predictions about time to encode a new goal; and (3) the priming constraint, which makes predictions about the role of cues in retrieving pending goals. These constraints are formulated algebraically and tested through simulation of latency and error data from the Tower of Hanoi, a means-ends puzzle that depends heavily on suspension and resumption of goals. Implications of the model for understanding intention superiority, postcompletion error, and effects of task interruption are discussed.

This paper presents a method by which a reinforcement learning agent can automatically discover certain types of subgoals online. By creating useful new subgoals while learning, the agent is able to accelerate learning on the current task and to transfer its expertise to other, related tasks through the reuse of its ability to attain subgoals. The agent discovers subgoals based on commonalities across multiple paths to a solution. We cast the task of finding these commonalities as a multiple-instance learning problem and use the concept of diverse density to find solutions. We illustrate this approach using several gridworld tasks. 1.

The congeries of theoretical views collectively referred to as "situated action" (SA) claim that humans and their interactions with the world cannot be understood using symbol-system models and methodology, but only by observing them within real-world contexts or building nonsymbolic models of them. SA claims also that rapid, real-time interaction with a dynamically changing environment is not amenable to symbolic interpretation of the sort espoused by the cognitive science of recent decades. Planning and representation, central to symbolic theories, are claimed to be irrelevant in everyday human activity. We will contest these claims, as well as their proponents ' characterizations of the symbol-system viewpoint. We will show that a number of existing symbolic systems perform well in temporally demanding tasks embedded in complex environments, whereas the systems usually regarded as exemplifying SA are thoroughly symbolic (and representational), and, to the extent that they are limited in these respects, have doubtful prospects for extension to complex tasks. As our title suggests, we propose that the goals set forth by the proponents of SA can be attained only within the framework of symbolic systems. The main body of empirical evidence supporting our view resides in the numerous symbol systems constructed in the past 35 years that have successfully simulated broad areas of human cognition. During the past few years a point of view has emerged in artificial intelligence, often under the label of "situated action " (henceforth, SA), that denies that intelligent systems are correctly characterized as physical symbol systems, and especially denies that symbolic processing lies at the heart of

It has long been recognized that hierarchical problem solving can be used to reduce search. Yet, there has been little analysis of the problemsolving method and few experimental results. This paper provides the first comprehensive analytical and empirical demonstrations of the effectiveness of hierarchical problem solving. First, the paper shows analytically that hierarchical problem solving can reduce the size of the searchspace from exponential to linear in the solution length and identifies a sufficient set of assumptions for such reductions in search. Second, it presents empirical results both in a domain that meets all of these assumptions as well as in domains in which these assumptions do not strictly hold. Third, the paper explores the conditions under which hierarchical problem solving will be effective in practice.

Learning and Problem Solving with Multilayer Connectionist Systems September 1986 Charles William Anderson B.S., University of Nebraska M.S., University of Massachusetts Ph.D., University of Massachusetts Directed by: Professor Andrew G. Barto The di#culties of learning in multilayered networks of computational units has limited the use of connectionist systems in complex domains. This dissertation elucidates the issues of learning in a network's hidden units, and reviews methods for addressing these issues that have been developed through the years. Issues of learning in hidden units are shown to be analogous to learning issues for multilayer systems employing symbolic representations.

This paper describes an approach to automatically learn planning operators by observing expert solution traces and to further refine the operators through practice in a learning-by-doing paradigm. This approach uses the knowledge naturally observable when experts solve problems, without need of explicit instruction or interrogation. The inputs to our learning system are: the description language for the domain, experts ' problem solving traces, and practice problems to allow learning-by-doing operator refinement. Given these inputs, our system automatically acquires the preconditions and effects (including conditional effects and preconditions) of the operators. We present empirical results to demonstrate the validity of our approach in the process planning domain. These results show that the system learns operators in this domain well enough to solve problems as effectively as human-expert coded operators. Our approach differs from knowledge acquisition tools in that it does not re...

This dissertation is dedicated to my parents, Bill and Gaye, who have always loved and believed in me and to my husband, Andy, whose love and support made it possible. ACKNOWLEDGMENTS Andrew Barto has been a great thesis advisor. He has helped me to become a better researcher by shaping my critical thinking as well as by improving my expressive skills. I also benefited greatly from having Rich Sutton as my second advisor during my first two years at the University of Massachusetts. I would like to thank the members of my thesis committee, Eliot Moss, Rod Grupen, and Neil Berthier for their feedback. Doina Precup and Kiri Wagstaff have been wonderful friends and supporters of my re-search. It is very helpful to have such smart women friends in CS. They provided support when I needed it and they pushed me when I needed that. I feel privileged to know Doina both as a mentor and as a friend. I thank Kiri for helpful feedback on drafts of my disser-tation as well as the motivation provided by exchanging and reviewing each other’s thesis

: To create more powerful knowledge acquisition systems, we not only need better acquisition tools, but we need to change the architecture of the knowledge based systems we create so that their structure will provide better support for acquisition. Current acquisition tools permit users to modify factual knowledge but they provide limited support for modifying problem solving knowledge. In this paper, we argue that this limitation (and others) stem from the use of incomplete models of problem solving knowledge and inflexible specification of the interdependencies between problem solving and factual knowledge. We describe the EXPECT architecture which addresses these problems by providing an explicit representation for problem solving knowledge and intent. Using this more explicit representation, EXPECT can automatically derive the interdependencies between problem solving and factual knowledge. By deriving these interdependencies from the structure of the knowledge-based system itself ...

We summarize selected recent developments and promising directions for improving the quality of models of human performance in synthetic environments. The potential uses and goals for behavioral models in synthetic environments are first summarized. Within that context, we examine relevant, current work related to modeling more complete performance, for example, on cognitive modeling of emotion, advanced techniques for testing and building models of behavior, new cognitive architectures, and agent and Belief, Desires and Intentions (BDI) technology. The report also considers the usability of these systems as an important but neglected aspect of their performance. A list of projects with high payoff for modeling human performance in synthetic environments is noted.

External representations (ERs) are effective in reasoning due to their cognitive and semantic properties. We investigated subjects' use of ERs in their solutions to analytical reasoning problems. Two sources of data were analysed. The first consisted of a large corpus of ERs (`workscratchings') used by students in their solutions to problems administered via paper and pencil tests. The second source of data was collected using switchER, a computer-based system that administered the problems, provided a range of ER construction environments for the subject to choose between and which dynamically logged user--system interactions. SwitchER was developed in order to study the process and time-course of ER use and to investigate the mechanisms (such as ER switching) by which subjects resolve impasses in reasoning. The results showed great diversity of ER use across subjects, allowing the utility of various ERs under differing task conditions to be studied. The range of ERs used by subjects ...