Agent Models of Intelligence -limitations and Prospects

AI applications are increasingly moving to artiicial agents, i.e. systems that independently handle parts of the problem based on small independent knowledge bases Gros 1 10, 27]. This study focuses on a speciic class of agents called \reactive agents", which use minimal communication between agents, and rely mostly on changes in the environment as their cue for action. Some early successes of this model, especially in robotics, has led to an intense debate over this class of models as a whole. One of the issues on which attention has been drawn is that of connicts between such agents. In this work we investigate a cyclic connict that results in innnite looping between agents and has a severe debilitating eeect on performance. We present some new results in the debate, and compare this problem with similar cyclicity observed in planning systems, meta-level planners, distributed agent models and hybrid reactive models. The main results of this work are: (a). The probability of such cycles developing increase as the behavior systems become more useful. (b). Control methods for avoiding cycles such as prioritization are unreliable, and (c). Behavior reenement methods that reliably avoid these connicts (either by reening the stimulus, or by weakening the action) lead to weaker functionality. Finally, we show how recent attempts to introduce learning into the behavior modules will also increase the likelihood of cycles. 1 \The study of logic and computers has revealed to us that intelligence resides in physical-symbol systems."