Search

This report introduces archivists to the potential of expert systems for improving archives administration and alerts them to ways in which they can expect intelligent technologies to impact federal record-keeping systems and scholarly research methods. The report introduces the topic by describing expert systems used in three Fortune 500 companies. It then defines expert systems, distinguishes them from conventional programs, and presents the capabilities of the technology together with examples of suitable applications. Discussion of the building of an expert system application begins with a short history of the evolution of the technology, followed by a detailed account of knowledge engineering, i.e., the process used to develop an expert system. Descriptions of several expert systems applications in the federal government highlight applications in the Internal Revenue Services, the Social Security Administration, and the Office of Management and Budget (Executive Office of the President). A report on the library profession's emerging use of this technology focuses largely on the three national libraries of the federal government: the National Library of Medicine, the National Agricultural Library, and the Library of Congress. The discussion of recent advances in expert systems technology that concludes the paper examines limitations of the technology, identifies likely frontiers for further research and development, and considers the implications of the technology for archives administration. A list of sources and related bibliographies is appended. (MAB)

The objective of the current research program was to develop a prototype knowledge acquisition tool that could enable the investigation of knowledge acquisition issues for instructional systems. Our approach to accomplishing this objective entailed the identification of the types of knowledge to be acquired, the evaluation of methods of acquiring such knowledge, the specification of appropriate interaction metaphors for the direct acquisition of knowledge from multiple human sources, and the iterative development of an Artificial Intelligence in Training (AIT) prototype that operationalized the findings of the first three tasks. The AIT prototype demonstrated critical characteristics of a knowledge acquisition environment designed to acquire multiple types of knowledge using a simulated domain environment in which multiple, animated views of knowledge are completely introspective and manipulatable. In addition, strategic knowledge construction was demonstrated based on observations of user actions. We described sample knowledge acquisition scenarios to characterize the functionality supported by the AIT prototype. The following areas for additional research were recommended: acquisition of instructional heuristics, dynamic construction of justification knowledge, use of higher level problem representations such as the goal-action hierarchy, and most importantly, the veridicality of the expertise captured using these methods. (Contains 34 figures, 3 tables, an appendix of instructions for software maintenance, and 26 references.) (Author)

This paper presents a model for an integrated system used for job-aiding, training, and performance assessment for persons who maintain systems of various types. The model is driven by updatable job aids, by integrated human-machine heuristics, and by an expanding matrix of maintenance activities. The model uses the job-aiding base, updated by computer networks and retrieval systems. In the model, this job-aiding system is part of an expert system. All inputs and outputs are envisioned to be in natural, human languages presented in a user-friendly series of displays and menus. The model also provides for training and performance assessment. To create training modules, the computer subsystem implements the appropriate job aid by presenting it in a training frame. To create a performance assessment battery, the computer subsystem presents the job aid after filtering it through a linguistic transformation that turns it into a case study or, if appropriate, a series of questions. The paper includes 52 references. (Author/KC)

The first of two reports in this document identifies research topics relevant to the introduction of information technology (IT) into schools. A discussion of general issues in IT points out the need for detailed empirical evaluation for the implementation of IT; the need to overcome barriers to innovation; and the need to support and encourage the growth of research teams with a wide mixture of skills. Immediate needs for the facilitation of constructive innovation are identified as robust software; guidelines for managing IT in schools and for local education agencies (LEAs) on the introduction of IT; and materials for inservice teacher education. The second report describes individual research projects supported by the Economic and Social Research Council in the United Kingdom. These projects address several facets of knowledge-based systems, student interaction with computers, penetration and effectiveness of IT in teacher education, students' intuitive ideas about science, use of microcomputers in primary schools, expert system tools in education, group interactive processes and pupil understanding, social interaction and the child's representation of computing devices, novices' problems with the Prolog authoring package, a knowledge-based approach to computer assisted learning, software design for teaching arithmetic, developing microworlds for secondary mathematics, and policies and practices in the integration of IT into the curriculum. References are provided for both reports. (DJR)

This paper provides an overview of Student Modeling techniques that have been employed in Intelligent Languages Tutoring Systems (ILTSs) over the past decade. It further discusses the Student Model of the English-as-a-Second-Languages (ESL) Tutor, an ILTS for ESL. The Student Model is based on student subject matter performance and provides feedback and remedial exercises suited to learner expertise. The paper further reports on a study in which the extent to which the Student Model addresses the need for an individualized languages learning environment is determined. (Contains 14 references, 1 table, and 2 figures.) (Author)

A framework for designing intelligent assistance in a discovery learning environment is proposed in this paper. The process of discovery learning is analyzed and the required functions for intelligent assistance are discussed. A flexible simulator that fits any type of discovery learning is necessary. The problem solvers that perform fundamental tasks of discovery are also needed: hypothesis generator and experiment designer. In particular, the paper focuses on another important function: evaluator of the effectiveness of counterexamples. Counterexamples have many clues for learning, but a learner often feels difficulty in utilizing them. This paper proposes the method of evaluating counterexamples from two educational viewpoints that ask the following: "Does it suggest the occurrence of error clearly?" (visibility), and "Does it suggest the cause of error?" (suggestiveness). Some case studies are presented to illustrate these functions, followed by a description of the framework. (Contains 12 references, 5 figures, and 4 tables.) (Author/AEF)

This paper looks at learner-formulated questions in technology-supported learning applications. Traditionally, technology-supported learning applications request input from the learner. The learner's response is used to assess the knowledge of the learner, to define a navigation path through the material or to construct a learner model. The objective is to add another form of interaction between learner and system, where the learner can pose questions to the system in similar fashion as to a human tutor. The paper discusses ways of dealing with these learner-formulated questions, question formats and existing approaches. It then introduces an approach for learner-formulated questions that is based on the Flexible Structured Coding Language (FSCL). Two specific approaches are presented: the syntax-based and the semantic-based. After a discussion of these approaches, the paper concludes with an outline of future work. (Contains 12 references.) (Author/AEF)

Adopting perspectives based on applications of artificial intelligence proven in industry, this paper discusses methodological strategies and issues that underlie the development of such software environments. The general concept of an expert system is discussed in the context of its relevance to the problem of increasing the accessibility of expert assistance to research practitioners. Important methodological issues and development strategies that underlie the construction of such systems are discussed. Some illustrative examples addressing the question of representation of research expertise (versus textbook-based information) are presented. Finally, implications and priorities for the longitudinal development of practitioner-supporting expert systems for the representation of the research knowledge-base and the design of knowledge-based instructional environments for improving graduate training in research are noted. Nine figures illustrate the discussion. (Contains 25 references.) (SLD)

This paper describes the Guided Approach to Instructional Design Advising (GAIDA), an automated instructional design tool that incorporates techniques of artificial intelligence. GAIDA was developed by the U.S. Air Force Armstrong Laboratory to facilitate the planning and production of interactive courseware and computer-based training materials. The tool is a case-based system that incorporates a short exposition of Gagne's nine events of instruction and four complete examples of applying the nine events to identification of naval insignia, classifications of electronic resistors, checklist procedure for the F-16 Gatling gun, and procedure for testing a patient's respiratory capacity using a spirometer. The paper covers (1) the issues involved in developing an automated instructional design advisor; (2) the results of Gagne's evaluation of the initial GAIDA case (the checklist for the F-16 gun); (3) a description of how GAIDA has been redesigned; and (4) how GAIDA will be used in future instructional design research and development projects. (Contains 18 references.) (KRN)

This paper discusses issues related to the design of software tools that support learners in their participation in network-based learning activities. To guide the development and use of a new class of educationally-oriented network tools, this paper proposes a cognitively-based, distributed network learning framework (DNLF). This framework has three main aspects: (1) network mediators and the flow of information and knowledge; (2) networks and cognitive theories of learning; and (3) the human-network interface. As an example of an application of the DNLF, an ongoing research and development project is described that involves an electronic mail program called The Message Assistant is described. The Message Assistant is designed to promote higher order learning goals as a part of instructional activities conducted over distributed educational networks. In addition to the standard electronic mail features such as creating, sending, receiving, and reading messages, this program includes a user-defined incremental expert system and hypertextual linking functions to assist network users in their evaluation, organization, and distribution of network information and knowledge. Research using the DNLF can provide insights into important aspects of electronic educational networks and help guide the design of tools to better support learning in these rapidly evolving network environments. Six figures illustrate The Message Assistant screen displays. (Contains 14 references.) (Author/ALF)

This paper presents framework specifications for an instructional systems development (ISD) expert system. The goal of the proposed ISD expert system is to improve the means by which educators design, produce, and evaluate the instructional development process. In the past several decades, research and theory development in the fields of instructional technology and cognitive science has advanced the knowledge base for instructional design theory such that learning and thinking can be significantly improved by direct instructional intervention. Unfortunately, these advancements have increased the complexity of employing instructional design theory, making instructional development both costly and time consuming. It is proposed that through the application of expert system methods, it is now possible to develop an intelligent computer-based ISD expert system that will enable educators to employ instructional design theory for curricular and instructional development. Presented in this paper is a framework for the development of an ISD expert system that will assist both experienced and inexperienced instructional developers in applying advanced instructional design theory. Diagrams of three generations of the proposed model are appended. (38 references) (Author/BBM)

The use of intelligent software agents within computer mediated learning environments has become an important focus of research and development in both AI and educational contexts. Some of the roles envisaged and implemented for these electronic entities involve direct interactions with students, participating in the "social" dimension of the classroom that is of such importance in contemporary pedagogical theory. Others contribute to the many background tasks that support the teaching/learning process. Each type of activity raises its own special challenges in relation to the capabilities of the software and to our understandings of teaching and learning. Through discussion of both theoretical perspectives and practical examples, this paper explores a selection of these issues. (Contains 13 references.) (Author)

This paper describes preliminary work on "Multiple Intelligent Mentors Instructing Collaboratively" (MIMIC), an intelligent Web-based agent environment for learning instructional design. The focus is on developing theoretical foundations of instructional design and instructional theory that form the foundation for systems development. In the proposed intelligent learning environment (ILE), intelligent agents will serve as mentors and learner companions in support of pre-service teachers (PSTs) in learning instructional design, a key aspect of their professional preparation. This environment also promises to promote higher-order thinking, stressing learning by reflection and metacognition. The paper discusses: the features of MIMIC; overall framework; motivational components; cognitive considerations for agent-based learning environments; and future theoretical development. An example is provided to illustrate the interactions that take place with the MIMIC system. (Contains 25 references.) (AEF)

This paper describes a system that provides faculty with automatic testing and grading online facilities, as well as course administration resources, using the World Wide Web as a common and ubiquitous interface. The system was built using the Common Lisp HTTP Server from MIT and its functionality to generate HTML "on the fly," which offers the appropriate level of interactivity and flexibility needed to design such systems. It allows instructors to create exams and to manage them online, and it includes automatic grading and immediate feedback to students. Also, the system allows for automated scores bookkeeping: students may check their grades at any time and so may their instructor. It is argued that the system might be a useful framework for enhancing Web-centered education, and that the CL-HTTP server is an ideal instrument for easily achieving this sort of interactive and automatic management. (Author/AEF)

This paper develops a new conception of organizational knowledge management as "knowledge ecology." The Socrates electronic learning environment is described as an example of a knowledge ecosystem for education and training. The Socrates program was developed as a knowledge management system, viewing classrooms as organizations for lifelong learning. Socrates allows even novice Web users to quickly (in a day) create instructional Web sites with integrated knowledge resources and Web training tools. Based on its early experience, several implementation barriers are identified. Future research needs for developing and implementing organizational knowledge ecologies are discussed. (Author/AEF)

This paper presents a model of an intelligent computer network that provides real-time evaluation of students' performance by incorporating intelligence into the application layer protocol. Specially designed drills allow students to independently solve a number of problems based on current lecture material; students are switched to the most appropriate assignment level (regular, advanced, or simplified) based on computer evaluation of performance. The state of each network is defined as a function of a student's performance during the course of the drill. The analysis of a state of the nodes in real-time permits algorithmic reconfiguration of the network resources. As a consequence of this reconfiguration, the instructor will monitor only those students who are behind the schedule. This, in turn, allows the instructor to concentrate on the analysis of the students' problems and the ability to provide real-time help rather than attempting to identify those students who are experiencing difficulty. Figures present the matrix of student performance, graphs of a students' performance and overall class performance, and the algorithm of real-time performance analysis. (AEF)

Information retrieval on the World Wide Web has a major obstacle: although data is abundant, it is unlabeled and randomly indexed. This paper discusses the implementation of a consultative Web search engine that minimizes the expertise level that is required from a user to accomplish an advanced search session. The system takes advantage of the meta-knowledge (Selection Routine), used by expert librarian searchers and applies it to a heterogeneous search space such as CD-ROM databases and Web-based environments acting as an intermediary expert system. Topics discussed include: (1) the Selection Routine, including the three basic stages of a typical online search--definition of query structure, selection of search keys, and feedback review; (2) system description, including the Web-based interface, spell checker, consultative core, and retrieval component, as well as examples of the Selection Routine and Metaknowledge rule sets; and (3) implementation issues. Three figures present components of an online search, the decision tree of the Selection Routine, and system description. (AEF)

The Electronic Homework assistant system is composed of two components: the Computer Tutor and the Homework Administrator. The Computer Tutor is an intelligent tutoring system that can provide personal assistance like supplying hints, checking errors, providing remediation and prioritizing problems. The Homework Administrator is a teacher's assistant in assigning and marking the homework, and summarizing errors for teachers' reference. Both components of Electronic Homework are composed of several modules. Different types of knowledge, mainly obtained from human experts, are incorporated into the Computer Tutor, and the pieces of knowledge, in the form of rules, are stored in the following separate modules: the expert module, which contains the knowledge that the system imparts to the student; the student module, which stores the knowledge of the students, both correct and incorrect, in the form of rules; the tutoring module, which contains the knowledge on how and when to help students correct their errors; and the communication module, which deals with the interaction between the user and the computer tutor. The Homework Administrator is responsible for handling routines normally done by teachers, which include ordering problems, collecting students' errors and marking students' work. The effect of using Electronic Homework was investigated among students of different academic abilities; in one school, students using the system performed better on less abstract problems. A major shortcoming of Electronic Homework is that it reacts very slowly in some situations. (Contains 17 references.) (AEF)

Motivated by the brain, neural networks are a right-brained approach to artificial intelligence that is used to recognize patterns based on previous training. In practice, one would not program an expert system to recognize a pattern and one would not train a neural network to make decisions from rules; but one could combine the best features of each in solving a problem. One of the simplest and most common neural network models is the fully connected three layer model that consists of the input layer, the hidden layer, and the output layer. Instead of programming a neural network, the neural network is trained by presenting a history of inputs and outputs to the network. Training can be difficult and time-consuming, but after training, the neural network can quickly recognize patterns. One of the easiest places in which to integrate neural networks into the curriculum is a follow-on to the study of regression. The goal of regression is to determine a functional relationship between a dependent variable and one or more independent variables. Neural network software is becoming more available and more affordable. An annotated bibliography of selected literature on neural networks and a brief history of neural networks follows the discussion in this paper. (Contains 11 references.) (AEF)

With the release of ShockWave, MacroMedia Director animations can now be incorporated directly into Web pages to provide high quality animation and interactivity, to support, for example, tutorial style questions and instantaneous feedback. This paper looks at the application of this technique in the translation of a traditional computer-based tutorial, Ariadne's Thread, into an online version. A significant change was evident in its development and implementation processes and impact on the student learning and departmental infrastructure. This change was seeded in part by the increased modularity and open delivery of online Web materials, and in part by the development of specialized interactive learning devices and support tools. In particular, the shift from authoring to writing tools has moved the locus of project control from the technical to educational, involving teachers more directly in development and implementation. A simultaneous shift from a development-centered project approach to one that is implementation-focused, has increased the potential of responding to learning needs on an ongoing basis. Five figures illustrate the program. (AEF)

Many students find climate classification laborious and time-consuming, and through their lack of repetition fail to grasp the details of classification. This paper describes an expert system for climate classification that is being developed at Middle Tennessee State University. Topics include: (1) an introduction to the nature of classification, Koppen's classification system, and climographs; (2) the advantages and disadvantages of computer-based classification; (3) the rationale for using the C programming language and a DOS-based program; (4) the program design, the modules of software development, and future interface and graphics enhancements; (5) the use of the expert system in both classroom and independent instruction and the project evaluation. Three figures depict: flowcharts for Koppen classification system; a summary of climate subcategories and climate codes within the Koppen system; and a sample climograph. (AEF)

This paper reports the results of using a three-layer backpropagation artificial neural network to predict item difficulty in a reading comprehension test. Two network structures were developed, one with and one without a sigmoid function in the output processing unit. The data set, which consisted of a table of coded test items and corresponding item difficulties, was partitioned into a training set and a test set in order to train and test the neural networks. To demonstrate the consistency of the neural networks in predicting item difficulty, the training and test sets were repeated four times starting with a new set of initial weights. Additionally, the training and testing runs were repeated by switching the training set and the test set. The mean squared error values between the actual and predicted item difficulty demonstrated the consistency of the neural networks in predicting item difficulty for the multiple training and test runs. Significant correlations were obtained between the actual and predicted item difficulties and the Kruskal-Wallis test, indicating no significant difference in the ranks of actual and predicted values. (Author/MDM)

As a person learns, his problem solving ability improves and one reason for this is the increased acquisition of "macro-rules" which make problem solving more efficient. An intelligent computer assisted learning (ICAI) system is being developed which automatically acquires the useful knowledge from the domain experts; as experts give the learning system instructions of how to solve given problems, the system extracts the strategic knowledge of problem solving from the instructions by generalizing them. The educational system then teaches and coaches students based on the acquired strategic knowledge by making a student give instructions instead of an expert, thus extracting a student's strategic knowledge. If macro-rules are prepared for various states, then the system is able to give an explanation of problem solving for various similar problems. The learning system described in this paper acquires strategic knowledge according to the model of explanation-based learning; the domain is physics at the high school level. Knowledge was prepared for the domain theory in terms of the knowledge: (1) for understanding the structure of a problem; (2) for inducing the appropriate formula from the problem structure; (3) for solving an equation; (4) for solving plane trigonometry; and (5) of a dictionary on the concepts/technical terms. The system makes the knowledge processing according to instructions on: attributes of objects; on defining attribute-value of an object as a variable; on a causal relationship among objects; and on equations. (AEF)

This paper describes RaPiD, a computer-aided assistant for the design of dental prostheses called removable partial dentures. The user manipulates icons directly to indicate the desired design solution to a given clinical situation. A developing design is represented as a logic database of components in a design; expert rules are applied as integrity constraints governing valid database transactions/design alterations. RaPiD has two modes of operation: manual mode designed for educational use (at either "student" or "expert" level), and automatic, which is intended for dentists in practice. In automatic mode, the dentist enters key clinical information, after which the system takes over and completes the design. Contravention of design rules is presented to the user in a critiquing style. The critiquing style strategies form the basis for the system's use in undergraduate and graduate dental education. Critiquing strategies used include: (1) a critique is issued only when the user has completed the proposed alteration; (2) a critique is issued immediately upon the user's radical misuse of a tool; (3) critiquing dynamically without negotiation with the user; (4) critiquing requested by the user upon completion of a design session, or at certain other stages in the design process; and (5) optional critiquing, requested by the user who wishes to compare his design with that which would have been produced independently by the system. An assessment of the educational effectiveness is planned. Expanding the range of design rules in RaPiD so that it becomes comprehensive is a current priority, as is the introduction of critiquing strategies (4) and (5), as well as completion of the automatic mode of operation. The extensive use and testing of RaPID, already carried out, show it to be a versatile and robust knowledge-based system. (Contains 14 references.) (Author/MAS)

Hypertext is well-suited for educational applications where open learning and knowledge exploration is desired. In such applications, principles of good hypertext interface design should be employed to avoid navigational problems so as to maximize learning. Interface design, however, may also directly enhance acquisition of a particular knowledge structure. An interface that is based upon an expert's structural knowledge map may assist novice learners in developing a more sophisticated knowledge structure more closely resembling that of an expert. Pathfinder networks have been used successfully to differentiate expert and novice knowledge structures. The procedures described in this paper were used to design a structured interface for use in a research study currently in progress. Two microcomputer programs were used to develop the interface: The Knowledge Network Orientation Tool (KNOT) was used to create the pathfinder network, and Asymetrix Toolbook for the hypertext application. (Contains 19 references.) (Author/JLB)

This paper describes the process used to design and evaluate a computer-based tool that guides the development of lesson plans. The tool, PLANalyst, contains a small expert system with a human face that analyzes lessons and gives specific feedback for improvement. The process used to create PLANalyst can be generalized to the development of tools to support any complex intellectual task. PLANalyst runs on Macintosh computers and was constructed with HyperCard. The tool allows easy creation and editing of lessons and provides printed out lesson plans that encourage lessons based on sound instructional principles and realistic estimation of time. It also provides a common language and a tool for team teaching and instructional development. The model on which it is based most closely resembles the nine events of instruction of R. M. Gagne and others. Preparation screens and event screens are illustrated, and the expert system is described. Field tests with close to 200 college students have supported the utility of the tool. The final version, still in development, will incorporate a database of useful teaching techniques drawn from ERIC citations. Six figures illustrate the screens. Two appendixes contain event categories and an evaluation form. (Contains 7 references.) (SLD)

Console-operation skill is procedural knowledge of control panel actions. The console operator must select appropriate sequences of steps (e.g., setting dials, pressing buttons, etc.) as mandated by desired goals and subgoals. A computer-based training (CBT) environment can simulate an environment where knowledge acquisition is efficiently and effectively achieved. In the training of console-operation skill, a CBT environment can simulate control-panel responses and device actions while providing instructional support. The instructional support delivered during practice consists of a set of intervention strategies. This paper discusses an ongoing investigation of CBT approaches in the acquisition of console-operation skill. Two variables of the intervention include the content of feedback and the timing of feedback. Preliminary results from a laboratory experiment suggest that feedback which elaborates on the goal structure of the console-operation skill is more effective than feedback that simply corrects errors. (Contains 42 references.) (Author/JLB)

This paper describes the use of an automated performance support system, the Mental Retardation-Expert (MR-E), in the training of 16 student clinicians to think "functionally" about undesirable behaviors of persons with mental retardation or developmental disabilities (MR/DD). The MR-E models human expertise in behavioral diagnosis and treatment, expanded and supported by the current scientific literature. The MR-E assists in the development of functionally based treatment protocols for individuals with MR/DD who engage in significant disruptive behaviors. The MR-E also provides an interactive process of case consultation and a basic understanding of the relationship between functional hypotheses and treatment. A pretest versus posttest strategy was applied to the environmental group (N=16) and control group (N=7), producing results that indicated that both training and interactive experience with the MR-E lead inexperienced clinicians to accept its recommendations. (DB)