The computer as a constructorium: Tools for observing one's own learning

This chapter explores a particular form of the "transparency" concept: designing styles of interaction that explicitly reveal the pedagogical foundations of the ECS. Reflection tools and activities can help the learner to internalize an active model of learning. This model would emphasize the active, heuristic and negotiated nature of learning, by opposition to the schoolmade passive idea of what learning is. The system's model of learning cannot be accessed in an abstract way but only through the instantiation of this model by the learner's behaviour. DILLENBOURG P. (1992) The computer as a constructorium: Tools for observing one's own learning. In M. Elsom-Cook and R. Moyse (Eds), Knowledge Negotiation.(pp. 185-198) London: Academic Press. Dillenbourg 2 The Constructorium 1. The pedagogic facet of knowledge negotiation Knowledge negotiation (KN) defines a particular style of interaction between a learner and an educational computing system (ECS). The choice of a particular type of interaction is a complex decision process based on knowledge about the domain, the learner and the pedagogical methods. The KN approach has its specificity with respect to each of these three components. We briefly describe the two first, the domain and student models, and develop this chapter around the third one, the pedagogical approach. If we consider the domain model, KN tackles philosophical issues about the very nature of knowledge : the Platonic view of knowledge, that underlies the widespread idea of expertise, is progressively abandonned for a view of knowledge as the temporary product of social processes (Carley, 1986). Concerning the student model, KN addresses some major criticisms made against current cognitive diagnosis techniques. It especially questions the possibility of representing the student knowledge as a perturbed copy of the expert's knowledge. Representing the learner's knowledge as a collection of partial models, potentially completely different from the expert's model, sometimes incompatible, selected and applied according to the context, finds instead a growing support in human psychology and artificial intelligence (Richard,1990). This chapter will concentrate on the third facet of KN, i.e. the relationship between the KN style of interaction and the pedagogical decisions. The pedagogic approach determines the features of the learning interaction. In the particular case, the KN style is based on a constructivist view of learning, which implies that the learner tests her own representations in the learning environment. These statements reflect the designer's view : pedagogical decisions lead to some particular design. The learner observes indeed this relationship from the opposite point of view : the interaction reifies the underlying pedagogical approach and hence conveys the designer's theory of learning. This speculative chapter investigates Dillenbourg 3 The Constructorium a copernician move towards the learner's position. We explore a particular form of the "transparency" concept (Wenger, 1987; Brown, 1990) : designing styles of interaction that explicitly reveal the pedagogical foundations of the ECS. 2 . The learner's model of learning. Children elaborate concepts that describe mental activities, such as knowing or remembering. For early children, these concepts often do not correctly match our understanding of these words (Wellman, 1985). For instance, pre-school children seem to consider that they "know" something when they give a correct answer, even if this answer results from guessing right (Misciones et al., quoted by Wellman, 1985). More globally, children tend to define mental activities by some associated external behaviour, they understand mental verbs as referring to observable behaviours. We are concerned by a particular mental activity concept : learning. Learners have some representation of the meaning of "learning", here referred to as a "model of learning". Like other similar concepts, the childs representation of learning learning finds its roots in the observable activities labelled or considered as learning activities, i.e. mainly in school activities. Therefore, we must compare it with the representations that students elaborate from educational interactions. By the terms "model of educational interaction", we refer to the student's representation of any interaction which explicitly aims to promote learning for at least one of the subjects in interaction. The role of this model appeared in educational literature under the concept of "didactic contract" (Brousseau, 1980) : students and teachers tend to behave according to implicit rules they induced from previous classroom experiences. The important assumption here is that these interactions are internalized into some representation of learning: "The representations constructed by pupils of knowledge and the causes of success and failure in obtaining this knowledge are determined by the interactional and didactic context of the classroom" (Schubaueur-Leoni and Bell, 1987). As Dillenbourg 4 The Constructorium Lochhead pointed out, passive school experiences potentially generate a baneful representation of learning : "Students tend to view learning as a passive experience in which one absorbs knowledge or copies fact into memory. Little of what they do in schools leads them to question that perspective." (Lochhead, 1985). University teachers may confirm this statement for students that have spent around 12 years in schools. The mechanisms by which educational interactions are internalised into some model of learning will be described in section 3.3. The relationship between the learning model and the educational interactions model is not straightforward. Our representation of any event is closely associated with the context in which the event occurred (Tiberghien, 1986). Learners probably have context-related submodels of learning interactions, and subsequently, contex-related models of learning. Research on mental models confirms this complexity. Humans do not handle a simple unitary model of the task performed, but instead some "distributed" model, i.e. a complex structure of partial models (DiSessa,1986). In "learning a poem", "learning to drive a car", "learning a foreign language", "learning to be happy", the word "learning" refers indeed to processes we perceive as quite different to each other. These differences result from the object learned (data, a complex skill,...), the domain (biology, mathematics, sport,....), the context (in/out school, with/without adults, ...) and the learner. 3. The "iconoclastic" principle The adjective "iconoclastic" describes actions that lead to breaking an image. In our case, this image is the student's representation of learning. Our iconoclastic goal is to break the learner's passive model of learning. The iconoclastic principle is articulated around the link between the interaction model and the learning model : if this link has been used for creating some model of learning, it can be reused in order to break this model. The rest of this chapter investigates how to implement this principle in an educational computing system. Dillenbourg 5 The Constructorium This implementation is articulated around four stages: • The transfer: the iconoclastic principle is only applicable if there exists a transfer mechanism from classroom situations to learner-computer settings; this transfer mechanism generates the learner's expectation towards the educational interaction she will participate in; • The conflict : in order to deserve its iconoclastic title, an ECS must contradict these expectations by proposing a different interaction style; • The constructorium : the ECS must offer activities that will lead the learner to internalise a better model of learning; • The reverse transfer : if learners would bring their new model back in the classroom, then ECSs would get some potential to modify school practices, i.e. ECS would be considered as agents of innovation. These four stages form a continuous and recurrent process, but they are now described separately and sequentially for didactic purposes. 3.1. The transfer from classroom to computer contexts. There is an apparent contradiction in our discourse. We described mental models as partitioned in context-related subsets and we now postulate that students transfer the classroom-made model to settings where they engage in a dialogue with a computer. This contradiction may be overcome by analysing more closely the differences between the two contexts, the classroom and the workstation, and between the respective sub-models of learning. The context of learning activity is defined by various factors. Replacing the blackboard by a keyboard modifies one of these factors. Replacing the teacher by a computer and suppressing the other learners constitute more important changes. But on the other hand some factors do not change. The "scholastic" label of the activity often remains. Many scholastic concepts contribute to make the two contexts closer: exercises, errors, scores, tests, definitions, ... And finally, the distribution of educational roles is not modified : with Dillenbourg 6 The Constructorium most ECS, there is still an ignorant agent that has to learn and a knowledgeable agent that has to "communicate" (in Wenger's sense) its knowledge. These similarities between learning in a classroom and learning with a computerized teacher (sometimes in a classroom) mean that there is some overlap between the learner's respective models of learning interaction. The extent of this overlap determines the transfer process. For instance, device-dependent features of behaviour will probably not be transferred : most learners know for instance that they cannot communicate with a computer by using the same language as they use with a teacher. But they may transfer more device-independent pieces of knowledge such as the necessity of systematic practice to reinforce newly