The first reason is a need to help the computer user solve problems that require specialized knowledge or expertise. In many situations, users need guidance and counseling in order to solve the problem at hand. The solutions to many problems in business, science, and engineering depend on the application of sophisticated numeric algorithms or techniques. In such situations, users often need help in determining which specific algorithm or technique should be employed and in interpreting any computed results. In other situations, the need is more basic--for guidance in determining whether the problem at hand can be solved and, if so, whether the resources that can be brought to bear are sufficient.

Loan underwriting is the process of evaluating a loan application to determine whether the loan should be funded. The process often starts with a potential borrower walking into a branch office and requesting a loan to purchase or refinance a home. A processor asks the borrower to fill out an application, setting in motion a lengthy information-gathering process in which as many as 1500 data-element pieces will eventually be collected. This loan information includes items about the borrower's employment, income, assets, liabilities, and monthly expenses. During the process, a credit report and appraisal will be ordered from a third-party vendor.

When building a controller for a physical process, traditional control theory requires a mathematical model to predict the behavior of the process so that appropriate control decisions can be made. Unfortunately, either many real-world processes are too complicated to accurately model, or insufficient information is available about the process environment. In addition, optimal control strategies can themselves exhibit undesirable complexity. However, a human controller can often acquire relatively simple strategies to effect near-optimal control from operational experience. This article reports on experiments performed using a black-box simulation of a spacecraft.

Over the past decade, it has become clear that one should go beyond the level of formalisms and programming constructs to understand and analyze expert systems. I discuss the idea of inference structures such as heuristic classification (Clancey 1985), the distinction between deep and surface knowledge (Steels 1984), the notion of problem-solving methods and domain knowledge filling roles required by the methods (McDermott 1988), and the idea of generic tasks and task-specific architectures (Chandrasekaran 1983). Such a synthesis is presented here in the form of a componential framework. The framework stresses modularity and consideration of the pragmatic constraints of the domain. A major question with knowledge engineering is (or should be) that given a particular task, how do we go about solving it using expert system techniques.

The annual conference of the American Association for Artificial Intelligence (AAAI) is the largest and most important meeting of AI theoreticians and practitioners in the United States This year, the conference was held in Seattle, Wash, and paid attendance was just under 5100 Last year's Philadelphia conference drew 5400 The drop in attendance was primarily the result of competition with the International /oint Conference on Artificial Intelligence, which took place in Milan a few weeks after AAAI I took advantage of the AAAI-87 exhibits to visit with a number of industry leaders and discuss the state of the U.S. AI marketplace. Vendors everywhere tend to be optimistic about their products and industries. Certainly, the AAAI vendors displayed optimism, but there was also an underlying current of doubt-or at least question. Questions raised by the vendors include the following: Does the recent lull in the market mean that the luster of AI and expert systems in particular is wearing thin? Are there more than a handful of significant expert systems in operation today?

Research at the Artificial Intelligence Laboratory of the University of Connecticut is currently focused on a number of projects addressing both fundamental and applied aspects of next-generation expert systems and machine learning. We believe that these next-generation expert systems will have to be based on cognitive models of expert human reasoning and learning in order to perform with the ability of a human expert. Cognitive expert systems should display three characteristics. First, because expert human reasoning and learning rely in part on qualitative causal models and large-scale event-based memory structures, cognitive expert systems should rely on similar knowledge. Second, because human experts are skilled at acquiring knowledge, often through natural language interaction, cognitive expert systems should learn through real-world natural language interaction.

He points out that one of the key features these systems lack is "a suitable verification methodology or a technique for testing the consistency and completeness of a rule set." It is precisely this feature that we address here. LES is a generic rule-based expert system building tool (Laffey, Perkins, and Nguyen 1986) similar to EMYCIN (Van Melle 1981) that has been used as a framework to construct expert systems in many areas, such as electronic equipment diagnosis, design verification, photointerpretation, and hazard analysis. LES represents factual data in its frame database and heuristic and control knowledge in its production rules. LES allows the knowledge engineer to use both data-driven and goaldriven rules.

Induction and deduction are two opposite operations in data-mining applications. Induction extracts knowledge in the form of, say, rules or decision trees from existing data, and deduction applies induction results to interpret new data. It starts with existing database technology and performs both induction and deduction. The integration of database technology, induction (from machine learning), and deduction (from knowledge-based systems) plays a key role in the construction of ILDB systems, as does the design of efficient induction and deduction algorithms. This article presents a system structure for ILDB systems and discusses practical issues for ILDB applications, such as instance selection and structured induction.

An artificial laboratory is a hypothetical computing environment of the future that would integrate mathematical and statistical tools with AI methods to assist in computer modeling and simulation. An integrated approach of this kind has great potential for accelerating the rate of scientific discovery. Theory guides and directs the course of experimentation, and experimental results subsequently suggest ways in which theory must be modified. Some theories can, in fact, be discarded altogether. Over the past 30 years, computer modeling and simulation, analogous to theory and experimentation, has frequently guided scientific investigation.

What Works and What Doesn't? AI has been well supported by government research and development dollars for decades now, and people are beginning to ask hard questions: What really works? What doesn't work as advertised? What isn't likely to work? This article holds a mirror up to the community, both to provide feedback and stimulate more selfassessment. The significant accomplishments and strengths of the field are highlighted.