"Learning", the incorporation of additional knowledge into expert systems, ranges from human data entry (learning by being told), to data gathering (learning by observing), to full-fledged theory formation (learning by discovery). One important kind of learning is the compiling of descriptive meta-knowledge into strategic form, recasting it into a form in which it can be evaluated efficiently. Much of what we earlier called strategic meta-knowledge may be seen to be operationalized "caches" of descriptive or systemic meta-knowledge. For instance, R9 and R10 can be converted from systemic to strategic form by slight rewordings of their actions.

This wolk was supported in parts by the Defense Advanced Rz-search Projects under ARPA contract no. MDA 903-77-C-0322 and National Institutes of He...Ith grant no.

The UNIT Package Is a frame-structured, hierarchically-organized knowledge representation and acquisition system. It was originally developed for the MOLGEN project at Stanford University [Stefik, 19701 [Friedland, 19791 [Stet ik, 19801 Tho package contains a sot of data structures and access functions for program manipulation of those structures. In addition, it contains a sophisticated Interactive editor, called UE. This editor enables a domain export (not necessarily a computer specialist) to construct a knowledge baso through direct interaction with tho computer; that is, tho transfer of expertise from domain export to machine flood not be mediated by a computer specialist. This document is intended to servo several purposes and parts of It can be Ignored by some readers.

Abstractions include not only small, simple concepts like hierarchies but also more complex notions like concavity and convexity or particles and waves. A model for an abstraction is essentially an interpretation for the symbols that satisfies the associated axioms. Different task domains can be models of the same abstraction (as biological taxonomy, geological time, and organization charts arc instances of hierarchies): or, said the other way around.

The selection of what to do next is often the hardest part of problem solving. This selection can be structured by di,'Mguishing decisions about the problem from decisions about the problem solving process. When planning decisions are structured in this way, we find that many of the most important decisionc, are about the planning process itself. This exercise tends to expose a variety of decisons, which are usually made implicitly and sub-optimally in planning programs with rigid control structures. This paper develops a layered approach for meta-planning, that is, for planning about planning. It is part of a course of research which seeks to enhance the power of a problem solver by enabling it to reason about its own reasoning processes.

Problem solvers need more than just the facts and logic of a problem domain. To work effectively, they also aced meta-knowledge, that is, knowledge about how to use the facts. For example, hierarchical planning uses meta-knowledge to distinguish between the important considerations and the details of a problem. This knowledge is added to the facts and logic of a domain and used to focus the generation of inferences. It relieves a hierarchical planner from trying to deal with everything at once.

The language designer typically designs that language with one particular application domain in mind: as subsequent types of applications are tned, what had originally been useful features are found to be undesirable limitations, and the language is overhauled or scrapped. One remedy to this bleak cycle might be to construct a representation language whose domain is the field of representational languages itself. One remedy to this bleak cycle might be to construct a representation language whose domain is th field of representation languages itself, a system which could then be tailored to suit many specific applications. Toward this end, we (Professor Douglas Lena and 1) have designed and implemented RLL-1, an object-centered2 Representation Languange Language.3 A representation language language (r11) must explicitly represent the components of representation languages in general and of itself in particular.

Section 1 1 Introduction The MOLGEN project has focused on research Into the applications of symbolic computation and Inference -to the field of molecular biology. This has taken the specific form of systems which provide assistance to the experimental scientist in various tasks, the most important of which have been the design of complex experiment plans and the analysis of nucleic acid sequences. During the period of further research proposed in this document, we plan to expand and improve these systems and build new ones to meet the rapidly growing needs of the domain of recombinant DNA technology. We do this with the view of including.

There are several different types of goals, and each has a corresponding set of parameters. For example, the goal (obtain (coefficient 6 x 2)) means to obtain an expression for the coefficient of x2 in g6, either to print it out or pass as argument to some MACSYMA command. Note that this can be done either by finding an already computed expression (stored, say, as the value of some variable) or by computing it anew. Either implementation is satisfactory so long as it computes the desired expression.

This talk is about "knowledge representation," particularly some of the fundamental assumptions in' olved in the way we handle knowledge in current Al and cognitive science research. The whole enterprise seems to have at-.'andoned Namely, we have a:7ccd to assume that knowledge is something that can be represelted--that knowing something means having a data structure stored away that stands for what is known. In other words, we treat knowledge as an object, a representable object. The advantage of this way of looking at things is that there is a very simple relationship between the things we say a person or system knows, and the memories, knowledge, or data structures we say he or it has stored.