• No results found

Chapter 2 Learning by Computer

2.5 Intelligent Tutoring Systems

2.6.5 LRDC software

A significant amount of software has been developed at LRDC either to test ideas on knowledge acquisition and learning or, under contract, to supply particular needs. A package that comes into the first category is Smithtown (Shute and Glaser, 1 990; Shute and Glaser, 1 99 1 ) . This program provides a 'laboratory' for teaching students the elements of economics. In a simulated environment various products can be bought and sold. Prices vary according to such factors as availability and demand. By experimentation the student has to determine what the relationships are between these factors. Although the subject is economics, the real purpose of the investigation is to look at how effectively people can learn by using an experimental approach and what facilities can be provided to aid this process.

In this guided discovery environment, students are encouraged to generate and test hypotheses, observe, record and organise test results, modify hypotheses in accordance with the results, and induce regularities and laws. Thus the focus is on developing effective weak problem solving methods in an unknown domain . Some o f the experiments with students (Shute and Glaser, 1990) suggest that appropriate use of weak methods is a better predictor of success than general intelligence.

As has already been noted, self-regulation is an important weak problem solving method. A facility for encouraging this process has been added to Srnithtown (Raghavan et al. , 1 99 1 ). DARN (Discovery And Reflection Notation) allows the student to view activities in the laboratory f�m three different perspectives: the Student View, the Plan View and the Expert View. The Student View traces the actions that the student has taken, presenting this as a search path. The Plan View relates what the student has done to the plans they have proposed, indicating which actions are and are not consistent with that plan. The Expert View illustrates what the expert would have discovered, given the same data.

Other laboratory-style environments that have been developed at LRDC include Voltaville (Schauble et al. , 1 99 1 ) where the subject is electrical circuits, and REFRACT (Reimann et al. , 1 988) which is a discovery environment for geometrical optics. Schauble and his colleagues look at the impact of effective models of the domain on the learning ability of students. Reimann's group examine the effect of inductive learning.

All the previously mentioned programs have been written to investigate ideas about learning. The Sherlock project (Lesgold et al. , 1 988) has put many of these ideas into practice. The American Air Force use a test station to locate problems in electronic navigation equipment for F- 15 aircraft. If the test station is functioning properly then the faults can be isolated quickly. Unfortunately, the test station, a very complex piece of equipment, quite frequently breaks down. Technicians have to be able to track down faults in this test station, a skill that can take up to 5 years to develop. To fast track this learning process the Air Force has tried various means, principally giving the apprentices a detailed course on how the test station works. As has been observed by Glaser and his eo-researchers (Glaser et al. , 1 985), this approach is not very effective. The best way of teaching problem solving is to set up faults in the test station and get the student to locate them. Because of costs and the possibilities of damaging the test station this method is impracticable.

To get over this problem Sherlock was developed. Sherlock provides a coached practice environment that simulates the activity of the test station including the effect of many

different faults. It is easy to set up the system to demonstrate the effect of any of these faults so the tutor can focus on a particular problem, perhaps one that the user has never seen before or one that slhe has previously had trouble diagnosing. As noted by Lajoie and Lesgold ( 1 989) the help that Sherlock provides can be seen as a form of cognitive apprenticeship (Brown et al. , 1 989) where the system initially gives the student a significant amount of help in problem solving but gradually withdraws that assistance as the student improves. To achieve this fading, Sherlock stores a model of the student's current level of competence.

Not only was Sherlock developed by workers with a deep understanding of all aspects of learning and teaching but the original version was also a very sophisticated piece of software. It was highly successful in its aim of providing the Air Force with a system that could accelerate the teaching of novices. Lajoie and Lesgold ( 1 989) observe that experiments have shown that after twenty to twenty-five hours using Sherlock, less experienced technicians could achieve a comparable trouble shooting ability to colleagues with four or more years on-the-job experience.

B uilding on the experience of the initial version of the package, researchers have developed Sherlock II (Lesgold et al. , 1 990; Katz et al. , 1 992a; Katz et al. , 1 992b; Lesgold et al. , 1 992a). The new version incorporates a deeper model of the device and expert knowledge, a sophisticated interface with video graphics, uses fuzzy logic for student modelling, and plans appropriate exercises for the student to attempt.