CAPÍTULO II.2: REVISIÓN DE PROGRAMAS DE DEPORTE ESCOLAR
FASE 6 Diseño de instrumentos de seguimiento y evaluación
6. Deporte escolar en la Comunidad Autónoma de Castilla-La Mancha
A central goal of this project has been to examine ways in which the production of computer learning environments may be facilitated. It was envisaged that ideas from knowledge-based systems might provide the key to achieving this goal. As a prelude to the investigation, research into the role of knowledge in learning and how knowledge can be represented on the computer has been reviewed.
As a result of this review it was determined that faithful models of reality are highly desirable for producing environments from which the user could learn. More specifically, designers should aim for cognitive fidelity: taking into account not only the ability of the users to understand how the system operates, but trying to model closely their problem solving behaviour. Providing such a model simplifies the tasks of adapting the software to different requirements and including suitable feedback.
A further important consideration is that methods used for designing such systems should be accessible to non-computer specialists. Teaching and domain experts should be closely involved in the development of software and, to achieve this, it is desirable to have schemes that can be understood by them. At the same time, a degree of rigour is essential to remove ambiguity and to allow programs to be generated as directly as possible from the specifications.
Based on these ideas, a framework was outlined for using high-level primitives as building blocks for the development of interactive learning environments. The use of such primitives attempts to bridge the gap between the teacher and domain expert on one hand and the system developer on the other.
To determine whether such a framework was feasible, suitable primitives needed to be investigated. It was clear from the initial analysis that there would be no single tool that could be used in every kind of teaching situation, since domains and domain types vary greatly. The strategy was, therefore, adopted of considering two very different kinds of domains and determining appropriate techniques for system development for each.
8.2.1 Procedural tutors
Firstly, procedural domains were considered. In this kind of domain, the aim is to model the behaviour of devices and how people interact with them. Here, the user needs to acquire an understanding of how a device works and also how to carry out various tasks within the system.
It is important in domains of this type for users to be able to try different alternatives and see the impact. Obviously, the model must operate, as closely as possible, in the same way as the real device. Also, if something unexpected happens, the system should be able to help the user find out why.
In addition to exploration, such models can be used for setting tasks for learners. A
goal may be achieved by trial and error but research suggests that constructive guidance can improve students' understanding of tasks and speed up their learning. Guidance is particularly beneficial if it includes feedback when users have gone wrong, pointing out their mistake and suggesting alternative steps that might be taken to rectify matters. To be able to simulate the effect of users' actions in the domain, a state-space model was adopted. The use of transition nets was investigated for domain and task representation but was found to be inefficient and opaque. Alternative approaches based on planning formalisms used in artificial intelligence proved to be more efficient. Also, the suitability of the planning methods for denoting cause-effect relationships makes them much better for providing explanations and general guidance.
TWEAK, a Strips-style planner, provided some improvement over the transition net, but the most effective approach was found to be Tenenberg's more elaborate Strips
planning model. This separates out the static conditions of the system from the dynamic ones. Such a formalism adds both to efficiency (since fewer dynamic variables have to be manipulated) and to clarity (since we distinguish what is susceptible to change from what is not). Tenenberg's formulation has been shown to be theoretically sound and so there are no problems about possible inconsistencies arising.
These methods were adapted in order to design learning environments which could provide feedback. Programs written to simulate several systems have demonstrated the feasibility of these techniques. Systems described in the text have included VCR simulators and models for assembling and dismantling a gearing mechanism.
The planning methods investigated were found particularly suitable for separating out the domain and task considerations. Both for system development and for teaching purposes it is seen as desirable to be able to do this. Steps in each individual task can be represented as an overlay on the domain description. To specify these steps some further notation was needed. A suitable scheme, originally used by Mark Drummond as a controller for automated planning systems, was adapted for task description purposes. His concept of situated control rules allows guidance to be given to students at appropriate points in their attempts to achieve specific goals. Programs based on the modified Drummond scheme demonstrate the feasibility and appropriateness of the method both for the VCR and gear mechanism problems.
Although the primitives used in procedural domains were found to be appropriate for program development, they are not particularly suitable for high-level design or for communication to non-computing collaborators. A more transparent notation for communicating domain and task descriptions was considered desirable.
A pictorial notation called the plan net, based on a variant of the petri-net, is proposed for domain description. Using plan nets, cause-effect flow can be represented in a direct fashion. It is suitable for designing systems on paper, or straight onto the computer. The latter is possible since plan nets can be easily converted to the action tables used in Strips-like systems.
A design scheme that incorporates all of the above ideas has been proposed and has been elaborated in Figure 5.22. Complete programs that follow this scheme have not been written but examples of action tables, pictorial design elements and programs for all of the key components of the system have been produced.
8.2.2 Knowledge-based simulation in ill-structured domains
As a contrast to the procedural-style tutors, an examination was undertaken of ill structured problems, ones where solutions are not clear-cut. Modelling of situations involving people was seen to be of particular concern. It is perceived that dealing with such domains is of increasing theoretical and practical significance.
Analysis has indicated that knowledge-based simulation seems to provide a suitable framework in this case. As before, it is important to attain a level of cognitive fidelity. In order to do this, it was proposed that more than a superficial knowledge of human behaviour patterns would need to be included.
As with procedural tutors, it was envisaged that suitable knowledge-level representations were probably already available. An attempt was made to use the CYC model for events, their causes and effects. This was found to have some value but to be generally insufficient to include information about people's motives, plans, goals and actions. An alternative notation, based on the techniques that Schank and Abelson developed for natural language understanding, was found to be appropriate for behaviour simulation.
These techniques have been tested for specific domains. The CYC approach was used in the design of PO PIT, a scheme for demonstrating problem-oriented policing to novice police officers. The behaviour simulation method was applied to the problem of modelling domestic violence situations and the actions of the people involved.
Programming the domestic violence simulation is problematic since various other technical difficulties, such as the processing of natural language, would have to be considered. To avoid getting sidetracked into dealing with problems not directly related to the project goals, it was decided to test out the proposed scheme by using the Wizard of Oz approach. This involves the user interacting with the computer in the same way they might interact with a software package, but the processing of user input and the generation of responses is carried out by a human 'Wizard' located at another terminal. The Wizard follows as closely as possible the interactive scheme that has been produced, but may need to extemporize at various points. Results from this experiment suggest that the basic approach is sound and could be implemented using the techniques presented in Chapter 6.