We propose a flexible frame-structured representation and agenda-based control mechanism for the construction of production-type systems. Advantages of this architecture include uniformity, control freedom, and extensibility. We also describe an experimental system, named WHJXZE, that uses this formalism. The success of MYCIN-like production systems 141   has demonstrated that a variety of types of expertise can be successfully captured in rules. In some cases, however, rules alone are inadequate necessitating the USC of auxiliary representations (e.g.
Problem Solving in Frame-Structured Systems Using Interactive Dialog Harry C. Reinstein IBM Palo Alto Scientific Center 1530 Page Mill Road Palo Alto, Ca. 94304 ABSTRACT This paper provides an overview of the process by which problem solving in a particular frame-like knowledge-based system is accomplished. The interrelationship between specialization traversal and entity processing is addressed and the specific role of the user interaction is described. I INTRODUCTION Semantic networks Cl1 and frame-like systems have emerged as powerful tools in a variety of problem domains [Z&l. In many of these systems an initial knowledge base is used to drive an interactive dialog session, the goal of which is the instantiation of the particular knowledge base elements which represent a solution to the problem being addressed. In a system developed at the IBM Scientific Center in Palo Alto [3,41, a dialog is generated from a KRL-based c51 semantic network for the purpose of generating a well-formed definition of a medical sensor-based application program.
MULTIPLE-AGENT PLANNING SYSTEMS Kurt Konolige Nils J. Nilsson SRI International, Menlo Park, California ABSTRACT We analyze problems confronted by computer agents that synthesize plans that take into account (and employ) the plans of other, similar, cooperative agents. From the point of view of each of these agents, the others are dynamic entities that possess information about the world, have goals, make plans to achieve these goals, and execute these plans. Thus, each agent must represent not only the usual information about objects in the world and the preconditions and effects of its own actions, but it must also represent and reason about what other agents believe and what they may do. We describe a planning system t??at address es these is sues and show how it solves a sample problem. INTRODUCTION Certain tasks can be more advantageously performed by a system composed of several "loosely coupled," cooperating artificial intelligence (AI) agents than by a single, tightly integrated system.
Reasoning-based problem solving deals with discrete entities and manipulates these to derive new entities or produce branching behavior in order to discover a solution. This paradigm has some basic difficulties when applied to certain types of problems. Properly constructed arithmetic functions, such as those using our SNAC principles, can do such problems very well. SNAC constructions have considerable generality and robustness, and thus tend to outperform hand coded case statements as domains get larger. We show how a SNAC fimction can avoid getting stuck on a sub-optimal hill while hill-climbing. A clever move made by our backgammon program in defeating the World Champion is analyzed to show some aspects of the method.
E. & pelt Slar@rd University, Stanford, California SRI International, Menlo Park, California ABSTRACT This paper reports recent results of research on planning systems that have the ability to deal with multiple agents and to reason about their knowledge and the actions they perform. The planner uses a knowledge representation based on the possible worlds semantics axiomati7ation of knowledge, belief and action advocated by Moore . This work has been motivated by the need for such capabilities in natural language processing systems that will plan speech acts and natural language utterances [1, 21. The sophisticated use of natural language requires reasoning about other agents, what they might do and what they believe, and therefore provides a suitable domain for planning to achieve goals involving belief. I. WI-IAT,A KNOWLEDGE PLANNER MUST DO Consider the following problem: A robot named Rob and a man named John arc in a room that is adjacent Both Rob and John are capable of moving, reading clocks, and talking to each other, and they each know that the other is capable of performing these actions.
This paper focuses on the general A-MIN search-control method and characterizes the problem spaces to which it may apply. Essentially, A-MIN provides a best-first search mechanism over the space of alternate interpretations of an input sequence, where the interpreter is assumed to be organized as a set of cooperating expert modules.'