LC0cA1 Introduction (with Yazdani) ^

The Possible Synergy Between
Intelligent Tutoring Systems
and Computer-based Microworlds

Robert Lawler and Masoud Yazdani

Comparison of ITS and CBM
In AI research, there are currently two different approaches to the use of computation for education. These can be characterized by comparing an emphasis on the use of intelligent computer aided instruction with an emphasis on computer facilities for familiarizing students with important representations and models. The former were advanced in the literature by Intelligent Tutoring Systems (Sleeman and Brown (eds.), 1982 Academic Press) and Learning and Teaching with Computers (O’Shea and Self, 1983, Prentice-Hall); the latter by arguments of Mindstorms (Papert, 1980, Basic Books) and “Designing Computer-Based Microworlds” (Lawler in New Horizons in Educational Computing, Yazdani (ed.) 1984, Ellis Horwood). Both these points of view were advocated by attendees at the conference. The dimensions in which their emphases can be contrasted most profitably are their Goals, their Strengths, and their Limitations.

Dimensions of Contrast
Intelligent Tutoring Systems have as a central objective communicating some content through computer facilities which may employ mechanized domain-specific expertise, error analysis, and user-knowledge modelling. Their user-oriented intelligence is controlled by mechanized instructional strategies which present problems and then test for understanding of the content. The more ambitious ITS aim to communicate strategies, generally conceived of as a collection of methods for problem solving, as well as domain specific content.
The strengths of ITS are their good definition and their completeness in the following senses. An ITS will have a well articulated curriculum embodied in its domain expertise and an explicit theory of instruction represented by its tutoring strategies. This completeness permits an ITS to package existing expertise and focus on the novelty, which is the use of mechanically embodied sets of rules as a tool for instruction. Because an ITS can be well defined for a given curriculum, the achievement of its goals can be unambiguously evaluated, in principle. The weaknesses set against these considerable strengths are inadequate complexity in models of what the user knows, how the user learns new knowledge, and what could be an appropriate instructional theory for complex minds.
In contrast with ITS, computer-based microworlds have the differently focussed aim of leading the students to powerful, authentic knowledge. This is partly a response to the observation of Piaget:

If we desire to form individuals capable of inventive thought and of helping the society of tomorrow to achieve progress, then it is clear that an education which is an active discovery of reality is superior to one that consists merely in providing the young with ready-made wills to will with and ready-made truths to known with. [1]

Authentic knowledge means what is to be learned should not merely in ‘ added to the knowledge base’ but rather assimilated into the person’s pre-existing system of knowledges and even more, should be freely expressed from internal motives when appropriate. The emphasis leads to a focus less on significant domain knowledge than on the activity of learner because it must be HIS internal action that integrates new knowledge to old.
The essential strength of such exploratory learning environments is that they CAN provide individuals with simple, concrete models of important things, ideas, and their relationships.[2] For example, as compared to arithmetic drill, which may enhance the memory of specific sums, model based instructional systems can provide people with new ways of looking at the world. When it works, this is a very powerful result.[3] To the extent that such enhanced understanding provides the person more power over his own mind and life, the knowledge is its own reward.

The fundamental limitations of microworlds are derive from their theoretical commitments. Establishing the impact of their use is very difficult because of the nature of the effects sought and the complexity ascribed to the human minds within which those effects take place. To the extent that their design embodies commitments to using the computer more as a medium than a preprogrammed tool, taking seriously the complexity of the learner, and trying to draw connections between new content and prior idiosyncratic experiences, implementing computer microworlds leads to major difficulties because:

  • computation is a protean, imperfectly understood medium of communication;
  • theories of how individuals learn through experience are inadequate; [4]
  • there exists no adequate model-based epistemology of instruction.
    At a more pragmatic level, there are two main problems. The program doesn’t know what the student is doing. The problem of giving guidance while permitting freedom of exploration has not been solved in any general way. Some of these problems can be addressed now.

    New Directions
    The core problem ITS now confront seems to be the complexity of actual users of the instructional systems. A short example will illustrate the point. In their famous work on children’s arithmetic, Brown and Burton [5] label ‘bugs’ children’s deviations from standard arithmetical procedures. In performing the sum below, the child exhibits a non-standard algorithm for performing an addition in the vertical form, but a description of the performance as a bug would misrepresent significantly the child’s functioning knowledge and the sorts of inference employed: [6]





    The child reasoned as follows:
    ‘Seven and five are twelve, but two numbers won’t fit underneath them. The biggest number we can put there is a nine. But that leaves us three left over, so let’s carry three to the next place where we can add them with the numbers there. Three, six, nine. Ninety nine must be the answer.’

    The child’s logic is a deviation from the standard addition algorithm. Characterizing the performance as a ‘bug’, however, would miss the strong use of a conservation principle, the child’s commitment to what she knows and understands well, and the inventiveness shown in her trying to cope with a problem beyond her capacity. These are primary values we want to support. One wants to support a child but not development of a ‘wrong theory’. Beyond granting the complexity of the child’s mind, we must recognize the complexity of the instructional situation.

    The dilemma is ours: we want to support the child’s commitment to his authentic point of view and its creative application; at the same time we want to modify that point of view. Can we circumvent this barrier? In the instructional situation, the teacher’s great strength is getting around the inutility of materials which are not helping the individual child. In effect, the teacher is able to back up to a more global perspective and switch instructional contexts to another one, hopefully a more productive one. An intelligent human tutor might try, at such a juncture, to probe the child’s representation schemes and processes of reasoning. Following such a model in a mechanized system would require developing a more comprehensive and well articulated epistemology of instruction,[7] for we must ask how to judge which representation could be more suitable for use by a specific student who does not yet understand a particular idea ?

    Consider an approach for circumventing our dilemma, as follows. Assume a system which is capable of casting its range of problems and knowledge in some set of different representations. Recognize that different representations of problems promote into salience different aspects of things and their relations.[8] If the student produces some response which the system judges a misstep according to domain knowledge, the diagnosis should procede by trying to determine how the child may be thinking about the problem. It is not enough to say the student has impoverished knowledge of a domain, even if he does not show knowledge as applicable within the surface representation being used by the system. If one hopes to develop an adequate instructional system one must ask if the student has enough local knowledge in some alternate representation which could be extended analogously and applied to enrich his capability within the surface representation being used by the system for the given problem. Shifting contexts to some alternative microworld, where the student could work or play with a different representation, and engage in exploratory familiarization with it, would permit his later, more effective return to the primary domain representation. With an extended repertoire of alternative knowledges available as the basis for analogical extensions, the student will be much better able to construct for himself an adequate specification of what’s what and what follows for solving problems in the primary representation of the domain.

    For example, in the problem above where 35 and 37 are summed to 99, the child’s justifications show that for her the reality behind the things she was manipulating is one where counting is salient and grouping is not being considered. For her to understand carrying would require familiarity with the interrelations of groupings and values (such as one might develop through play with a money-based representation scheme), familiarity with groupings where a uniform multiplicative factor is salient (as one would encounter in playing with Dienes blocks with their unit blocks, rows of 10, and flats of 100), and finally experiences which would lead to unification of those separately mastered schemes.

    There will surely occur cases where the student uses representations beyond the current scope of the system. What should the system do when it determines that it does not, and even more cannot, understand what the student is thinking? The system should have available an extensible repertoire of representations. No system should be called intelligent which is too rigid to learn. While people learn more from experience than from being told, for computers the opposite is more nearly true. Computers can learn from experts today; for use in education, an ideal instructional system should be able to learn new representations applying to its domain of primary expertise even from non-experts, as when the students are introducing some alternative way of thinking — however non-standard that way may be.

    One outcome of the AISB conference on AI and Education has been provoking realization of ways in which the agendas of researchers with disparate programs can support the achievment of their common aim, the use of new technology to enhance both the quantity and quality of knowledge in human minds. If ITS could be opened up to more complex understanding of what students bring to problem solving and if model based instructional systems could begin to make use of the sorts of techniques developed within the ITS community, a synergy could be possible that would permit the strengthening of both. Such an endeavor is now being advocated by Feurzeig under the banner of ‘intelligent microworlds’. His paper in this volume appears to be a point of departure for such an effort. Even if we imagine some ideal instructional system which fuses the best of the ITS and Computer-based microworld paradigms, we must still admit that significant problems remain. But a second outcome of this conference, witnessed by the articles in this volume, is the hope that advances to cope with those problems are within reach. Specifically, Feurzeig and the Lawlers show new approaches to rethinking curriculum in the light of what computational facilities make possible. The work of Carley and Drescher expand the objective of learning theories to broader coverage and greater technical depth. Ohllson’s analysis of ITS and DiSessa’s observations and arguments point the way to an epistemologically deepened program for curriculum materials design. There is much to be done, but the prospects are promising.

    1 The Science of Education and the Psychology of the Child. The Viking Press, 1971.
    2 ‘Concrete’ in this sense refers not to physical manipulablity but to having a basis in personal experience. Even ‘virtual’ experiences with artificial worlds can be concrete. DiSessa’s article illuminates this old distinction.
    3 For an example of such an outcome, see “Extending a Powerful Idea” in Computers and Cognition: Reflections on Logo, Lawler, DuBoulay, Hughes and MacLeod; Ellis Horwood Ltd., 1986.
    4 Knowledge based systems, with rules stored separately from the processes which use them, have proven their utility because their very lack of organization permits the addition of new rules without reprogramming; if, on the other hand, the organization and reorganization of knowledge in the mind is THE central need in effective education, thinking of learning as ‘adding another rule to the database’ may be counterproductive.
    5 “A diagnostic model for procedural bugs in basic mathematical skills”. Cognitive Science, 2, 155-192.
    6 From Computer Experience and Cognitive Development, Lawler, 1985.
    7 Such a direction of research is suggested by Ohllson’s article as the central challenge of future work in this area.
    8 If addition is taken as an example, then a unit-based representation would emphasize counting; a coin-based representation would emphasize grouping; a vertical-form problem representation would emphasize the interaction of separable component sums.

  • Print Friendly, PDF & Email