The goal of an intelligent system is to translate high level commands or specifications automatically into lower level actions upon the environment or "plant," while fully utilizing any prior information as well as the information contained in the real time environmental responses. Such an intelligent system must interface between high level specifications and low level actions; between prior information and real-time information which may improve; modify or contradict it; between discrete state or command changes and continuous signals; between inductive and deductive learning; between real-time actions and off-line decisions; between various time scales of environmental response; action and computation; and between models, algorithms, and architectures. At the highest level, the intelligent system must be able to select classes of models within which to describe the environment, change between model classes or models in response to environmental changes such as failure modes or disruptions, and finally refine the parameters on-line to select a model within a class. At the operational level, consisting of the core of an intelligent system, it must deal with the qualitative aspects of discrete and continuous variables, data or actions, the temporal aspects of multiple-time scales of environmental response, measurement or action and the spatial aspects of decentralized or distributed operation, information or control. In the qualitative realm traversing the purely discrete domain, one has the issues arising from automata theory and supervisory control for reliable or safe operation of discrete event dynamics. With the introduction of time, one also has the ability to model performance measures such as cycle-time in manufacturing systems or throughput in transportation systems. This leads to several scheduling and optimization problems. Finally, one has the implementation layer of an intelligent system which involves a unification of software, algorithms whether parallel or sequential, computer architectures and ultimately their mapping into chips for the end user. Thus this layer gives rise to several issues such as data structures and computer-aided-design. While the above suggests a tripartite dissection of an intelligent system, our goal is to nevertheless design it to be transparent to the end user, in a fashion similar to the open systems interconnection architecture.
