Dimensions of Intelligent Systems

Abstract : As intelligent systems have become more fully functional and commonly available, questions about their capabilities and relationship with humans have increased. This paper builds on the IS requirements ideas of Messina et al [2001] to explore middle ground between anthropomorphic approaches like the Turing test that rely on similarity to human behavior in an "imitation games" and the narrowness of tests of chess mastery. I contrast a system like Deep Blue which has a very fixed environment in which it performs to more complex types such as Associate technology. Deep Blue, I argue, is an example of system whose performance is expert, but whose competence is fragile and it may not satisfy extended definitions of competence and performance intelligence that we measure in dynamic environments. A clinical protocol system is used to explore the basic functional capabilities and knowledge. Beyond symbol processing and the knowledge level are grounded reactive intelligent within more of an environmental/systems perspective. I build on grounded systems to discuss the use of goals using, learning-based systems and & multi-modal logics that characterize "realistic" intelligent systems. It is argued that such characterizations will evolve as IS design matures into grounded intelligence and situated, rational agent systems. In the future belief models and measures of rational coherence might be used, as basic approaches to facilitate intelligent system performance in dynamic environments.