Knowledge that Transforms

Rule-based expert systems deal with inexact reasoning through a variety of quasi-probabilistic methods, including the widely used subjective Bayesian (SB) and certainty factors (CF) models, versions of which are implemented in many commercial expert system shells. Previous research established that under certain independence assumptions, SB and CF are ordinally compatible: when used to compute the beliefs in several hypotheses of interest under the same set of circumstances, the hypothesis that will attain the highest posterior probability will also attain the highest certainty factor, etc. This monotonicity is important in the context of expert systems, where most inference-engines and explanation facilities are designed to utilize relative scales of posterior beliefs, making little or no use of their absolute magnitudes. This research extends the comparative analysis of SB and CF to the field, where subjective degrees of belief and different elicitation procedures are likely to complicate their analytic similarity and impact their actual validity. In particular, we describe an experiment in which CF was shown to dominate SB in terms of several validity criteria, a finding which we attribute to parsimony and robustness considerations. The paper is relevant to (i) practitioners who use belief languages in rule-based systems, and (ii) researchers who seek a methodology to investigate the validity of other belief languages in controlled experiments.

In this paper, we consider the problem of generating effective information-gathering, communication, and decision-making (ICD) strategies for a distributed expert problem-solving (DEPS) system. We focus on the special case of a dual-processor DEPS system and present a decision-theoretic model that enables the characterization of feasible, efficient, and optimal ICD strategies. In view of the tremendous amount of computing needed to generate optimal strategies for problems of practical size, we develop useful heuristic procedures for constructing high-quality efficient ICD strategies. We illustrate the use of the model and the solution procedure through an example.

One of the major unsolved problems in knowledge acquisition is reconciling knowledge originating from different sources. This paper proposes a technique for reconciling knowledge in two independent knowledge bases, describes a working program built to implement that technique, and discusses an exploratory study for validating the technique. The technique is based on the use of heuristics for identifying and resolving discrepancies between the knowledge bases. Each heuristic developed provides detection and resolution procedures for a distinct variety of discrepancy in the knowledge bases. Sample discrepancies include using synonyms for the same term, conflicting rules, and extra reasoning steps. Discrepancies are detected and resolved through the use of circumstantial evidence available from the knowledge bases themselves and by asking sharply focussed questions to the experts responsible for the knowledge bases. The technique was tested on two independently developed knowledge bases designed to aid novice statisticians in diagnosing problems in linear regression models. The heuristics located a significant number of the discrepancies between the knowledge bases and assisted the experts in creating a consensus knowledge base for diagnosing multicollinearity problems. We argue that the task of identifying discrepancies between independent bodies of knowledge is an inevitable part of any large knowledge acquisition effort. Hence the heuristics developed in this work are applicable even when knowledge acquisition is not done by reconciling two complete knowledge bases. We also suggest that our approach can be extended to other knowledge representations such as frames and database schemas, and speculate about its potential application to other domains involving the reconciliation of knowledge, such as requirements determination, negotiation, and design.

One fundamental requirement in the expected utility model is that the preferences of rational persons should be independent of problem description. Yet an extensive body of research in descriptive decision theory indicates precisely the opposite: when the same problem is cast in two different but normatively equivalent “frames,” people tend to change their preferences in a systematic and predictable way. In particular, alternative frames of the same decision-tree are likely to invoke different sets of heuristics, biases and risk-attitudes in the user's mind. The paper presents a modeling environment in which decision-trees are cast as attributed-graphs, and reframing operations on trees are implemented as graph-grammar productions. In addition to the basic functions of creating and analyzing decision-trees, the environment offers a natural way to define a host of “debiasing mechanisms” using graphical programming techniques. Some of these mechanisms have appeared in the decision theory literature, whereas others were directly inspired by the novel use of graph-grammars in modeling decision problems. The modeling environment was constructed using NETWORKS, a new model management system based on a graph-grammar formalism. Thus, a second objective of the paper is to illustrate how a general-purpose modeling environment can be used to produce, with relatively little effort, a specialized decision support system for problems that have a strong graphical orientation.

The value of mathematical modeling and analysis in the decision support context is well recognized. However, the complex and evolutionary nature of the modeling process has limited its widespread use. In this paper, we describe our work on knowledge-based tools which support the formulation and revision of mathematical programming models. In contrast to previous work on this topic, we base our work on an indepth empirical investigation of experienced modelers and present three results: (a) a model of the modeling process of experienced modelers derived using concurrent verbal protocol analysis. Our analysis indicates that modeling is a synthetic process that relates specific features found in the problem to its mathematical model. These relationships, which are seldom articulated by modelers, are also used to revise models. (b) an implementation of a modeling support system called MODFORM based on this observationally derived model, and (c) the results of a preliminary experiment which indicates that users of MODFORM build models comparable to those formulated by experts. We use the formulation of mathematical programming models of production planning problems illustratively throughout the paper.

There is a growing interest in the use of induction to develop a special class of expert systems known as inductive expert systems. Existing approaches to develop inductive expert systems do not attempt to maximize system value and may therefore be of limited use to firms. We present an induction algorithm that seeks to develop inductive expert systems that maximize value. The task of developing an inductive expert system is looked upon as one of developing an optimal sequential information acquisition strategy. Information is acquired to reduce uncertainty only if the benefits gained from acquiring the information exceed its cost. Existing approaches ignore the costs and benefits of acquiring information. We compare the systems developed by our algorithm with those developed by the popular ID3 algorithm. In addition, we present results from an extensive set of experiments that indicate that our algorithm will result in more valuable systems than the ID3 algorithm and the ID3 algorithm with pessimistic pruning.

Microcomputer and work-station technology is the latest wave in computing technology to influence day-to-day operations in business and government organization. Does sector affect adoption of this new information technology? If so, how? Utilizing the data from a large comparative national survey of data processing organizations, this proposition was examined. The results confirm that after controlling for other factors such as organizational size, experience with computer technology, current investment in computer technology, procurement practices, and the task environment of the organization, the sector an organization operates within has a major differential effect on adoption of microcomputer technology. Public organizations have more microcomputers per employee, a result that is potentially due to a more information intensive task environment and the potential use of microcomputer technology as a side payment in lieu of salary. The latter factor derives from lower wage rates faced by public employees.

A major tenet in organizational behavior literature is that feedback improves performance. If feedback is thought to improve performance, then individuals should actively seek feedback in their work. Yet, surprisingly, individuals seldom seek feedback perhaps because of face-loss costs of obtaining feedback face-to-face. Furthermore, in cases where the giver is perceived to be in a bad mood, individuals may be even more reluctant to seek feedback if they believe seeking feedback risks the giver's wrath and a negative evaluation. In this paper, we explain how information technology can be designed to mediate feedback communication and deliver feedback that promotes feedback seeking. In a laboratory experiment, the effects of information technology and the perceived mood of the feedback giver on the behavior of feedback seekers are examined. The results showed that individuals in both the computer-mediated feedback environment and the computer-generated feedback environment sought feedback more frequently than individuals in the face-to-face feedback environment. In addition, individuals sought feedback more frequently from a giver who was perceived to be in a good mood than from a giver who was perceived to be in a bad mood.

Seventeen major types of information systems are identified and defined by vectors of their attributes and functions. These systems are then classified by numerical methods. The quantitative analysis is interpreted in terms of the development history of information system types. Two major findings are that the numerical classification autonomously follows the chronological appearance of system types and that, along the time line, systems have followed two major paths of development; these have been termed the applied artificial intelligence path and the human interface path. The development of new types of systems is considered within the framework of a theory of technological evolution. It is shown that newer types of systems result from gradual accretion of new technologies on one hand, and loss of older ones on the other. Conclusions are drawn concerning the value of taxonomy in studying information systems, in suggesting possible research directions, and the desirability of rationalizing research efforts within the IS discipline.