On Monday, 15 July 1991, prior to the Ninth National Conference on Artificial Intelligence (AAAI-91) in Anaheim, California, over 50 scientists and engineers attended the AAAI-91 Workshop on Intelligent Multimedia Interfaces. The purpose of the workshop was threefold: (1) bring together researchers and practitioners to report on current advances in intelligent multimedia interface systems and their underlying theories, (2) foster scientific interchange among these individuals, and (3) evaluate current efforts and make recommendations for future investigations. The workshop was planned and coordinated by Mark Maybury (MITRE Corporation, Bedford, Massachusetts), Steve Feiner (Columbia University), Alfred Kobsa (University of Saarbruecken), and Bonnie Webber (University of Pennsylvania). The purpose of the workshop was threefold: (1) bring together researchers and practitioners to report on current advances in intelligent multimedia interface systems and their underlying theories, (2) foster scientific interchange among these individuals, and (3) evaluate current efforts and make recommendations for future investigations. Intelligent multimedia interfaces go beyond traditional hypermedia or hypertext environments in that they process input and generate output in an intelligent or knowledge-based manner.

Interactive simulation environments constitute one of today's promising emerging technologies, with applications in areas such as education, manufacturing, entertainment, and training. These environments are also rich domains for building and investigating intelligent automated agents, with requirements for the integration of a variety of agent capabilities but without the costs and demands of low-level perceptual processing or robotic control. Our project is aimed at developing humanlike, intelligent agents that can interact with each other, as well as with humans, in such virtual environments. Our current target is intelligent automated pilots for battlefield-simulation environments. These dynamic, interactive, multiagent environments pose interesting challenges for research on specialized agent capabilities as well as on the integration of these capabilities in the development of "complete" pilot agents. We are addressing these challenges through development of a pilot agent, ...

AI can't solve these problems, it isn't AI have been vocal about this approach. Case-based reasoning provides both a methodology for building systems and a cognitive model of people. It is consistent with much that psychologists have observed in the natural problem solving people do. Psychologists have also observed, however, that people have several problems in doing analogical or case-based reasoning. Although they are good at using analogs to solve new problems, they are not always good at remembering the right ones.

For many, scheduling time was probably the biggest challenge because the conference included numerous invited speakers, 189 technical paper presentations, 93 posters, a Mobile Robot Competition, 19 Innovative Applications of AI (IAAI) award-winning paper presentations, a Trading Agents Competition, a special track on AI and the web, and the vendor exhibit. Where Is AI Having an Impact? Frequently, many of us in the AI community get asked, "Where is AI being used these days?" There were several high-impact applications discussed at IJCAI that provided excellent new information to answer such queries, including Google News and a NASDAQ surveillance tool that detects insider trading and fraud. Google News, which is used several million times a day, employs several AI techniques to automatically generate and refresh the news from 4,500 news sources.

We describe the application of plan-recognition techniques to support human intelligence analysts in processing national security alerts. Our approach is designed to take the noisy results of traditional data-mining tools and exploit causal knowledge about attacks to relate activities and uncover the intent underlying them. Identifying intent enables us to both prioritize and explain alert sets to analysts in a readily digestible format. Our empirical evaluation demonstrates that the approach can handle alert sets of as many as 20 elements and can readily distinguish between false and true alarms. We discuss the important opportunities for future work that will increase the cardinality of the alert sets to the level demanded by a deployable application.

In the first phase of the analysis, I produced a conceptual reconstruc-Peter D. Karp In the next phase, I searched for patterns in the differences between successive My Ph.D. dissertation describes an A class knowledge base defines a tax-states of the biologists' knowledge. Patfocuses on a program of research in process knowledge base describes the terns in the differences indicate reamolecular biology that culminated in chemical reactions that can occur soning methods that were used to the discovery of a new mechanism of between the biological objects in this derive new theories from old ones. An experiment is described My analysis identified theory-modifiuation. In the first phase of my work, in a third knowledge base by creating cation operators that the biologists I performed a historical study of this the particular objects (instantiated used to modify their theories; these program of biological research in from the known classes of objects) operators form the core of the which I reconstructed the different that are present in the experiment. These patterns also supat different points in time and then called Gensim (genetics simulator) port the conjecture that scientists use analyzed the differences between predicts experimental outcomes by four different modes of scientific these successive theories.

Researchers and technology developers from the National Aeronautics and Space Administration (NASA), other government agencies, academia, and industry recently met in Pasadena, California, to take stock of past and current work and future challenges in the application of AI to highly autonomous systems. In our lifetime, through the eyes of simple robots, grand vistas on other worlds have been unveiled for the first time. Enigmatic questions compel us to go further, to touch these distant landscapes and learn the secrets of the solar system. However, in trying, we find our reach wanting, limited by the link to Earth on which our probes depend. We are learning that to explore further, these probes must go alone, and to go alone, they must become much more intelligent.

IAAI seeks out applications of artificial intelligence that either demonstrate new technology or use previously known technology in innovative ways. IAAI particularly seeks out examples of deployments of AI technology that tackle the problems of demonstrating value and planning for long-term deployment. The five articles we have selected for this special issue are extended versions of papers that appeared in the conference. Two of the articles are deployed applications that have already demonstrated practical value. The remaining three articles are particularly innovative emerging applications.

Detection of fraud, waste, and abuse (FWA) is an important yet challenging problem. In this article, we describe a system to detect suspicious activities in large healthcare data sets. Each healthcare data set is viewed as a heterogeneous network consisting of millions of patients, hundreds of thousands of doctors, tens of thousands of pharmacies, and other entities. Graph-analysis techniques are developed to find suspicious individuals, suspicious relationships between individuals, unusual changes over time, unusual geospatial dispersion, and anomalous network structure. The visualization interface, known as the network explorer, provides a good overview of data and enables users to filter, select, and zoom into network details on demand.

GLISI' is a high-level language that. is compiled into LISP It provides a versatile abst art,-dnt.at.ypc facility with hierarchical inheritance of pl oprl ties 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 cfflcient as handwritten LISP An integrated programming environment is provided, including automatic incremental compilation, interpretive programming features, and an intelligent display-hased inspector/editor for data and data-type descriptions GLISP code is relatively portahlr; the compiler and the data inspcrtor are implemcntcd for most major dialects of LISI' and arc availablr flee or at nominal cost This research was supported in part by NSF grant SED-7912803 in the Joint National Science Foundation - National Institute of Education Program of Research on Cognitive Processes and the Struct,urr of Knowledge in Science and Mathematics, and in part by the Defense Advanced Resealrh Projects Agency rmdel contract MDA-903-80-c-007 Author's present address: Computer Science Department, University of Texas at Austin, Austin, TX: 78712 GLISP contains ordinary LISP as a sublanguage; LISP code can be mixed with GLISP code, so t,hat no capabilities of the underlying LISP syst,cm arc lost. GLISP has also been ex-!,cndcd as a hardware description language for describing VLSI designs. GLISP Statements GLISP provides several kinds of statements that arc t,ranslated into equivalent, code in 1,ISP; each is identified by a key word at, the front of a list containing the code for the stwtemcnt. Many of these statements are similar to t,hose provided by I'ASC!AL: If..then.else While Repeat Case ...Do ..Until These control st,atements provide A compact Given a. set of name/value pa,irs, the A function creates a new tlat,a st,ruct,urc having t,ht: specified values: (A CIRCLE WITH RADIUS Given the earlier ob.jcct dcscript,ion for CIRCLE, this will compile as: (LIST (APPEND '(0 0)) R) The A function works interpretively as well as wit,hin caonlpiled code Context and Type Inference One of t,hc design goals of CLISP is that program code should be independent of the irnpl T le t,at,ic,rls of the structures manipulated by the code to the grcat,cst, dcgrcc possiblc Inclusion of redundanl t,ype declarations in program code would make the code dependent on the actual inplementwtion of structures; instead, GLISI' relies on type inference and its compile-time context, mechanism to tletcrmine the types of object,s.