Hazard: An Expert System
transferability to other diagnostic-testing settings.
exchangeability, i.e., ‘plug-in compatibility’ for other knowledge bases.
Second generation, ‘deep level’ ES.
HAZARD is an expert system (ES) under development at our Institute, assisting environmental regulators and public health personnel in detecting potential hazards at waste-sites of environmental chemicals. Expert systems are programs (i.e. software products) that model expertise in a specific task domain to achieve a high level of performance for problems that are considered difficult and require a great deal of expertise. ES normally uses one type of knowledge representation, e.g., production rules or semantic network frames. One perspective of our research is to test the outcome of the interaction of different types of representation methods.
a knowledge base (KB) that consists of scientific facts on a limited number of high priority toxicants, exposure information, and expertise on plausible reasoning, e.g., a set of •IF-THEN’ decision rules that associate with certain premises and facts certain conclusions.
an inference engine (IE), consisting of a set of recursive chain rules that
produce an exhaustive depth-first search based on an opti¬mal search algorithm (Gottinger, 1977),
are judgemental to permit inexact inferences and Bayesian updating (Gottinger, 1980).
a communication system which is responsible for interaction with the users. It is able to explain HAZARD’S decisions upon request, and to understand and use nearly natural language, which is a sine-qua-non-condition for high acceptance by users.
a modified system which enables a user to change and enlarge the knowledge base and/or the decision rules.
The knowledge base (KB) is the program’s store of task specific knowledge, the IE is an interpreter (or control structure) that uses the KB to solve problems at hand. The KB is accessible to the extent that additional expertise would enhance the KB of a program. The responsibilities of an inspecting team of health and environmental monitoring can be compared with those of other diagnostic tasks. The hazard potential of chemical dump-sites, emission sources of air and water pollution is composed of the exposure potential of the population at large and of a set of structural parameters on toxicants (e.g., the degree of toxicity, biodegradability, accumulation, etc.). Intelligent monitoring of such sites involves quick and efficient diagnosis and possible treatment or ‘action- planning’ in case of emergencies, or accidents.
If diagnosis and action-planning are effective, most incidents can be terminated without serious consequences, also preventive steps can be taken against newly produced chemicals as to their health and environmental impacts. Such a preventive scheme is envisaged by the OECD’s Minimum Premarket Data (MPD) Hazard Assessment Scheme.
The purpose of HAZARD is to monitor a hazardous waste facility (including a rad-waste facility), to detect deviations from normal operating conditions, to determine the significance of the situation, and to recommend an appropriate action. It performs these tasks by operating on a large KB, such as one derived from a chemical information system, with an IE that reasons recursively in a basic identification system. HAZARD is similar to the interactive structure of MYCIN’S rule-based consultation and explanation system (Shortliffe, 1977), however, unlike MYCIN, it belongs to the category of second generation ES-Technology (EST) or deep systems (Hart, 1982), or, systems of high structural complexity (Gottinger, 1983). It reasons explicitly on the basis of causal rules of judgement and model-based assessment as derived from the impact-modelling of ‘potential hazards’ (see enclosure: ‘Analytical Treatment of Potential Hazards…’).
HAZARD’S KB contains two types of knowledge: function-oriented knowledge and event-oriented knowledge. Function-oriented knowledge concerns the composition of environmental chemicals at a particular source and how they interact synergistically or antagonistically over time. Event-oriented knowledge describes expected acute or long-run consequences if certain threshold limits on certain parameters are exceeded. Event-oriented knowledge is represented by ‘if-then’ rules. HAZARD is to be implemented in LISP.
KeywordsExpert System Inference Engine Normal Operating Condition Waste Facility Expert System Application
Unable to display preview. Download preview PDF.
- Shortliffe, E.H.: Computer-Based Medical Consultations: MYCIN, Elsevier, North Holland: New York, Amsterdam 1976.Google Scholar
- Hart, P.E.: ‘Directions for AI in the Eighties’. Sigart Newsletter 79, 1982, 11–16.Google Scholar
- Gottinger, H.W.: Elements of Statistical Analysis. De Gruyter: Berlin, New York 1980.Google Scholar