A Knowledge-Level Analysis of Informing

Publisher Summary This chapter reviews how informing modifies the knowledge of an agent and the manifestation in terms of its problem solving performance. Robust intelligent systems must work in many differing situations. When the environment or computational resources change, a system must modify its behavior accordingly. To make matters even worse, information available to an agent is often incomplete, especially in complex domains. Designing intelligent agents that are able to adapt to their environment, to handle partial information, and to learn from experience has been a challenging task for AI researchers. Then it becomes important to build informable agents that are able to accept declarative information at runtime and put that information to use without the intervention of a human programmer. Intuitively, informing increases the amount of knowledge held by an agent. Informing adds knowledge to an agent. For an agent to be truly informable, it has to share the same conceptual primitives as the informing agent. Control knowledge may appear as various forms of procedural hints to the agents, such as prescription of specific actions, elimination of specific actions, constraints on actions, and preference among actions etc. Even though procedural hints are generally not in the deductive closure of the agent's body of domain knowledge, they do not necessarily enable the agents to solve more problems. With more knowledge about the problem-solving process, which may help cut down combinatoric explosions, an agent is able to solve problems more efficiently. Thus informing is a desirable feature for intelligent adaptive agents. Furthermore, Declarative representations such as predicate calculus are expressive enough to cope with incomplete and incremental information.