Machine learning has always been an integral part of artificial intelligence, and its methodology has evolved in concert with the major concerns of the field. In response to the difficulties of encoding ever-increasing volumes of knowledge in modern AI systems, many researchers have recently turned their attention to machine learning as a means to overcome the knowledge acquisition bottleneck. This article presents a taxonomic analysis of machine learning organized primarily by learning strategies and secondarily by knowledge representation and application areas. A historical survey outlining the development of various approaches to machine learning is presented from early neural networks to present knowledge-intensive techniques.

GLISP is a high-level language that is complied into LISP. It provides a versatile abstract-data-type facility with hierarchical inheritance of properties and object-centered programming. GLISP programs are shorter and more readable than equivalent LISP programs. The object code produced by GLISP is optimized, making it about as efficient as handwritten Lisp. An integrated programming environment is provided, including automatic incremental compilation, interpretive programming features, and an intelligent display-based inspector/editor for data and data-type descriptions. GLISP code is relatively portable; the compiler and data inspector are implemented for most major dialects of LISP and are available free or at nominal cost.

Early this year fifty people took an experimental course at Xerox PARC on knowledge programming in Loops. During the course, they extended and debugged small knowledge systems in a simulated economics domain called Truckin. Everyone learned how to use the environment Loops, formulated the knowledge for their own program, and represented it in Loops. At the end of the course a knowledge competition was run so that the strategies used in the different systems could be compared. The punchline to this story is that almost everyone learned enough about Loops to complete a small knowledge system in only three days. Although one must exercise caution in extrapolating from small experiments, the results suggest that there is substantial power in integrating multiple programming paradigms.

This report covers two main AI areas: natural language processing and expert systems. The discussion of each area includes an assessment of the state-of-the-art, an enumeration of problems areas and opportunities, recommendations for the next 5-10 years, and an assessment of the resources required to carry them out. A discussion of possible university-industry-government cooperative efforts is also included.

Among the difficulties in evaluating AI-type medical diagnosis systems are: the intermediate conclusions of the AI system need to be looked at in addition to the "final " answer; the "superhuman human" fallacy must be guarded against; and methods for estimating how the approach will scale upwards to larger domains are needed. We propose to measure both the accuracy of diagnosis and the structure of reasoning, the latter with a view to gauging how well the system will scale up.

Probabilistic rules and their variants have recently supported several successful applications of expert systems, in spite of the difficulty of committing informants to particular conditional probabilities or ";certainty factors"; and in spite of the experimentally observed insensitivity of system performance to perturbations of the chosen values. Here we survey recent developments concerning reasoned assumptions which offer hope for avoiding the practical elusiveness of probabilistic rules while retaining theoretical power, for basing systems on the information unhesitatingly gained from expert informants, and reconstructing the entailed degrees of belief later.

Editors' Note: Many expert systems require some means criticisms of this approach from those steeped in the practical of handling heuristic rules whose conclusions are less than certain issues of constructing large rule-based expert systems. Abstract the expert system draws inferences in solving different problems. Doyle's paper argues that it is difficult for a human expert "certainty factors," and in spite of the experimentally observed insensitivity of system performance to perturbations of the chosen values Recent successes of "expert systems" stem from much Research Projects Agency (DOD), ARPA Order No. 3597, monitored In the following, we explain the modified approach together with its practical and theoretical attractions. The client's income bracket is 50%, can be found (Minsky, 1975; Shortliffe & Buchanan, 1975; and 2. The client carefully studies market trends, Duda, Hart, & Nilsson, 1976; Szolovits, 1978; Szolovits & THEN: 3. There is evidence (0.8) that the investment Pauker, 1978). Reasoned Assumptions (from Davis, 1979) and would use the rule to draw conclusions whose "certainty factors" depend on the observed certainty Although our approach usually approximates that of Bayesian probabilities, accommodates representational systems based on "frames" namely as subjective degrees of belief.

The primary goal of the Artificial Intelligence Laboratory is to understand how computers can be made to exhibit intelligence. Two corollary goals are to make computers more useful and to understand certain aspects of human intelligence. Current research includes work on computer robotics and vision, expert systems, learning and commonsense reasoning, natural language understanding, and computer architecture.

Among the difficulties in evaluating AI-type medical diagnosis systems are: the intermediate conclusions of the AI system need to be looked at in addition to the "final " answer ; the "superhuman human" fallacy must be guarded against ; and methods for estimating how the approach will scale upwards to larger domains are needed. We propose to measure both the accuracy of diagnosis and the structure of reasoning, the latter with a view to gauging how well the system will scale up.

This article describes a theory of reasoning about uncertainly, based on a representation of states of certainly called endorsements. The theory of endorsements is an alternative to numerical methods for reasoning about uncertainly, such as subjective Bayesian methods (Shortliffe and Buchanan, 1975; Duda hart, and Nilsson, 1976) and Shafer-dempster theory (Shafer, 1976). The fundamental concern with numerical representations of certainty is that they hide the reasoning about uncertainty. While numbers are easy to propagate over inferences, what the numbers mean is unclear. The theory of endorsements provide a richer representation of the factors that affect certainty and supports multiple strategies for dealing with uncertainty.