Session: The Embodiment of Medical Knowledge

Intelligent Documentation and Decision Support for the Evaluation of Adverse Events in Clinical Trials

Heiner Gertzen

Hoechst AG, Frankfurt

The efficacy and safety of a new drug must be investigated in a series of clinical trials before it can be released. In order to guarantee the safety of the drug and estimate its risk-benefit ratio, adverse events which occur in the clinical trials must be carefully evaluated using all the knowledge and experience available at that time. In particular, it must be decided whether a reported adverse event is related to the experimental drug or due to other causes (e.g., other concomitant drugs or diseases). An intelligent documentation and decision support system was designed to support the medical expert in this difficult evaluation and decision task. The system comprises three main components: 1.) an intelligent documentation component ; 2.) a decision support component; and 3.) a hypertext based interface.

A newly reported adverse event is processed by the intelligent documentation component. Subsets of biomedical knowledge and previously solved cases are retrieved, processed and applied to the case at hand, yielding an enriched case description. This is processed by the decision support component to reach a decision about whether the event is related to the drug under study, together with an explanation of the adverse event. At present, the algorithm actually used by medical drug safety experts is applied in this step.

A pilot system has been implemented in a Windows environment using MS Excel. Currently, knowledge acquisition techniques are used to enhance the biomedical knowledge base/intelligent documentation component, and case-based reasoning techniques are adapted to the needs of the evaluation task. We consider to incorporate additional decision rules into the system comprising phased decision strategies, heuristic decision models and connectionistic models.

Textbooks/Expert Critiquing Systems for Medical Education

Frank Puppe

University of Wurzburg, Germany

Humans learn best by doing. The computer can provide an attractive learning tool by simulating the problem environment, where the student can solve cases. Our concrete scenario is as follows:

  1. The student learns the basics of the domain in a conventional manner (e.g. with textbooks).
  2. The student can study formalized expert knowledge of the domain.
  3. The teacher presents (real) cases to the students and discusses them with respect to the material the student should know.
  4. The tutor system presents the student cases to be solved in a manner as realistic as possible, i.e. audio-visual and stepwise presentation of data.
  5. While solving the case, the student - if not in examination mode - has access to the textbook and formalized expert knowledge, preferably by an integrated hypertext system.
  6. The student is not interrupted when doing something wrong, e.g. not recognizing correctly an audio-visual presented symptom or ordering an unnecessary test, unless asking for comments.
  7. The tutor system can follow the student's actions and finally criticizes suboptimal performance.

The methodology necessary for this scenario is combining an expert system shell for building the knowledge base and for problem solving capability, a case presentation interface and a hypertext system. We present three example-applications in an advanced stage within our D3-framework in the domains of rheumatology, neurology and ECG-interpretation.

Parallel Computer Architectures for Artificial Intelligence and Knowledge Processing

Ivan Plander

Slovak Academy of Sciences, Slovakia

The paper provides a survey on the applications of massively parallel computer architectures in artificial intelligence (AI) and knowledge processing, discusses the most important areas of AI where the applications of massively parallel computers provide the greatest advantage and talks objectively about massively parallel SIMD architectures as the prerequisites to master truly massive parallelism represented by neural, optical, molecular and other massively parallel computers. The paper presents the main characteristics of massively parallel computers and defines the true parallelism where the number of processing elements is so large that it may conveniently be considered a continuous quantity. The most effective applications of massively parallel computers presented are:

The simulation of a parallel inference processor on a SIMD-type parallel computer is advantageous in the possibility to study the rule set processing process in a bit-serial word- parallel manner. AI is moving into a new phase characterized by a broadened understanding of the nature of knowledge, and by the use of new computational paradigms. A sign of this transition is the growing interest in massively and truly massively parallel computers, represented by neural, optical, molecular and other massively parallel analog computers.

Session: The Integration of Expert and Tutoring Systems

The Segmentation of Work with Expert and Intelligent Training Systems

Valerie L. Shalin

State University of New York at Buffalo

The functions of intelligent tutoring systems are designed to enhance learning. The function of expert systems, and workplace aids in general, are designed to improve user performance, often measured in the achievement of system goals with minimal resource expenditure. In complex work settings, the functions of intelligent tutoring systems may not serve as useful workplace aids for two reasons. First, the task decomposition represented in the expert model may isolate individual skills for pedagogical reasons, while in practice the skill interacts with a much richer context of goals and opportunities. In such contexts, the expert models within an intelligent tutor may in fact be quite impoverished, and provide little ultimate benefit to performance.

Second, the functions of performance evaluation within an intelligent tutor often address blatant procedural errors, which occur less often and with less ultimate cost to system success than daily ill-informed decision making and problem prioritization. A promising and feasible approach to expert aiding is to complement rather than duplicate the available human expertise with knowledge-based aids that represent useful information and knowledge from related parts of the work system, to support self- evaluation and improvement according to system-oriented goals and values. More ambitious approaches to aiding depend on broader models of expertise in complex systems, including processes of problem identification and prioritization, and expert processes for deciding resource tradeoffs, conducting self- evaluation and correction. Such models may ultimately enrich tutoring goals as well, making expert systems and tutoring systems more similar.

Discussion Group: Corporate Memory

Introduction to Corporate Memory

Gerhard Strube

University of Freiburg, Germany

Corporate Memory (CM) is an initiative to help companies cope with the problems of managing information that is useful and important for the company's operations. Technically, CM strives to integrate existing technologies of databases, information systems, knowledge-based systems, and also technologies currently under rapid development, like multimedia, hypertext, and electronic documentation. The importance of organizational aspects, and hence, of methods of workflow, information flow, and job analysis was highlighted. CM was discussed in parallel to human memory, which is also not a passive receptacle, but an active system for task-oriented and context-driven retrieval. Finally, costs and benefits were discussed, as well as feasibility.

Example for an Application in Aerospace

Markus Durstewitz

EURISCO, Toulouse Cedex, France

We understand corporate memory (CM) as an assistance for knowledge sharing within a corporation. Especially, in highly interconnected domains such as aerospace industry it becomes necessary to provide people undertaking a task with complementary knowledge about other sectors. Knowledge acquisition and representation is based on process (task), product, and operator models. The CM encounters three parts of knowledge:

  1. normative knowledge (that already exists in form of standards and norms),
  2. procedural knowledge (procedures as the set of activities of the process),
  3. episodic knowledge (experience, facts).

The use of the knowledge determines the functionalities and tools to be realized:

  1. intelligent reference management;
  2. adaptive interfaces;
  3. case bases.

Session: Networks and Multimedia

Hypercomposition & Instructional Explanations

Baruch B. Schwartz

Hebrew University, Jerusalem, Israel

l Research in teaching has shown that instructional explanations - the contributions of teachers and texts to learning - are complex goal states that demand high-level skills to be achieved. Important differences have been detected between experts and novices. For example, the identification of the problem (which elucidates the purpose of the explanation); or knowing the representational system (i.e., the set of devices that are communicative for the specific audience and on which explanations will be grounded; or finally completing verbal explanations to bridge between the representational system and the principles to be explained). The study that is reported here analyzes how the use of a hypercomposition tool could help the teacher articulate instructional explanations. It is shown that multiple-layered explanations turn to be feasible, a fact which suggests that instructional explanations with hypercomposition tools may lead to metacognitive learning. In addition, I show that most of the generally difficult kinds of instructional explanations were articulated during lessons in history and in mathematical problem- solving.

Problem Solving and Hypothesis-Testing with Intelligent Tutoring Systems

Claus Moebus

University of Oldenburg, Germany

The talk has five main topics. It will be shown that (1) a psychological theory of knowledge acquisition is necessary to derive some (2) design principles for intelligent tutoring systems. One of the design principles is that the system is sufficiently knowledgeable to (3) test hypotheses: students' solution proposals. It is discussed (4) in case studies how the hypothesis testing approach can be realized in three domains within three intelligent problem solving environments:

It is shown how the same functionality especially concerning hypothesis testing can be achieved despite differences in the domains. These differences make it necessary to use very different artificial intelligence techniques:

In the last part of the talk (5) an epistemological motivated overview is given how students problem solving and hypothesis testing fit into a more general frame of theory revision. It is argued that learning in the sense of theory revision occurs in impasse situations where the student gets system-generated feedback to his hypothesis. Learning occurs by self-explaining the feedback contents on the basis of student heuristics and repairs. Thus it is not expected that expert knowledge is directly implanted. This is in accordance to cognitive science and cognitive psychology based research.

The Intelligent Learning Environment with Multimedia under Networking

Toshio Okamoto

University of Electro-Communications, Tokyo, Japan

The paradigm of educational computing has changed from architectures with mixed initiatives like conventional ITSs (Intelligent Tutoring Systems) to educational software which prompts and supports a student's active learning, i.e. interactive learning environments and open-ended learning goals. Such learning environments form Micro Worlds, where the environment reacts susceptibly to a student's activity and reminds the students of particular issues. It means that the environment itself reflects its process to student-thinking as a mirror. One educational point of view states that the students would thereby acquire the ability of metacognition with which the students cannot only learn the knowledge about the specific domain but also monitor the process of self-cognition. The process needs the learning cycle of hypothesis-testing and emphasizes its thinking process. The importance of situated learning is pointed out on the basis of this learning paradigm. That is, we should incorporate the circumstances of problem solving according to the situation, the scenario and further the function of something like role- playing into the system. The intelligent multimedia system of "macro economics" domain described in this paper has been studied and developed from these points of view.

In the macroeconomics system, the students can play the role of an agent in "macro economics world" on the basis of socially situated learning by using the function of something like a game/simulation under a network environment. The system incorporates an expert system representing the virtually smart student and the chairman-expert system which controls the whole system. The idea is derived from the concept under which the computer companion should be embedded in the system in order to support learning by observation (modeling learning).

The students have an imitative life-experience by the intelligent simulation system with multimedia on the recovering and growing process of Japanese economics until the present age after World War II. There, they try to play the game of social political situations for each age from an economic viewpoint. It seems that the students can learn the principle of macro economics. The domain contents of the system are on the Japanese economics and its contemporary history. The learning environment is suited are for industry people, university students and high school students.

The system is built with multimedia technology (pictures, sound, animations, graphics, and text) including an electronic dictionary. The simulator of the model of macroeconomics and the expert systems with the rule-base of the characteristic information are employed to realize the virtual environment of macro economics.

Return to Home Page