Behavior of an Adaptive Self-organizing Autonomous Agent Working with Cues and Competing Concepts Csaba Szepesvári * Jànos Bolyai Institute of Mathematics Andràs Lórincz † Hungarian Academy of Sciences A brain model-based alternative to reinforcement learning is presented that integrates artificial neural networks and knowledge-based systems into one unit or agent for goal-oriented problem solving. The agent may possess inherited and learned artificial neural networks and knowledge-based subsystems. The agent has and develops ANN cues to the environment for dimensionality reduction (data compression) to ease the problem of combinatorial explosion. Here, a dynamical concept model is put forward that builds cue models of the phenomena in the world, designs dynamical action sets (concepts), and makes them compete in a spreading-activation neural stage to reach decision. The agent works under closed-loop control. Here we examine a simple robotlike object in a two-dimensional conditionally probabilistic space. Key Words: adaptivity, artificial neural networks; knowledge-based system; self-organization; activation spreading; autonomous system 1 Introduction The history of developments in knowledge-based systems (KBSS) is lengthy (see, for example, Bundy, 1990), as is the history of research on artificial neural network (ANN) models, the latter dating back at least to the original work of Hebb (1949) (see, for example, Hertz, Krogh & Palmer, 1991). Ever since their universal approx- imator nature was proved (Hornik, Stinchcombe & White, 1989), there has been renewed interest in ANN systems. Just a few years ago, KBS and ANN were con- ceived of as different. Recent articles, however, report on ANN models that solve KBS problems (Peng and Reggia, 1989; Thagard, 1989). It is, in fact, difficult to make a clear distinction between KBS and ANN models. One might try to define KBSs as systems having a collection of if-then rules. However, the simplest neural system is a receptor neuron connected to a motor neuron and that similarly works as jinos Bolyai Institute of Mathematics, Attila József University of Szeged, Szeged, Hungary H-6720; szepes@obehx.lkLk£k1.hu u t Department of Photophysics, Institute of Isotopes of the Hungarian Academy of Sciences, PO. Box 77, Budapest, Hungary H-1525; lormcz@obellx.ikLk£kLhu @ 1994 The Massachusetts Institute of Technology at UNIVERSITY OF ALBERTA LIBRARY on September 1, 2010 adb.sagepub.com Downloaded from