Intelligent Learning Environments

This section reviews ILE work on viewpoints which has not been implemented. The intangibility of this research arises from several sources. Some work is in the form of a review, other authors provide suggestions or speculations and others simply fail to indicate whether any implementation has been attempted.

3.2.1 Wenger: Interpretative Contexts

[Wenger, 1987] considers viewpoints to be necessary ‘to place misconceptions in a broader conceptual context.’ A viewpoint is defined as an interpretative context determined by its kernel and its scope.

The kernel of a viewpoint consists of a variable number of keys, which together define the interpretative context referred to as a viewpoint. Keys can be prior decisions, correct or incorrect beliefs, analogies, or assumptions that either explicitly belong to the model or must be inferred as underlying it. The scope of a viewpoint delineates its foreseeable area of relevance. Note that the exact scope of a viewpoint is rarely precisely defined a priori since viewpoints are likely to have obscure ramifications. “Foreseeable implies that the scope of a viewpoint is some minimal area of probable applicability.” In sum, a viewpoint is an interpretative context whose kernel contains critical keys to the proper understanding of entities within its scope.

[Wenger, 1987] pp 355

· situation specific: problems can be viewed and solved in different ways.

· domain specific: domains can be viewed in different ways. · background: the assumptions of a learner’s worldview.

In addition [Wenger, 1987] briefly outlines composite viewpoints formed by combining the kernels of two viewpoints. He also states that viewpoints can be compatible or competing but gives no examples or justification.

The usefulness of viewpoints in an instructional setting is decomposed into:

· optimisation of diagnosis: understanding the student's input

· optimisation of didactic steps: acting in line with the student's viewpoint

· instructional tools: introducing new viewpoints on a subject

· instructional targets: using the attainment of a new viewpoint as a goal

Wenger gives no indications of the possible links between the three levels of viewpoint application – or indeed how/if reasoning might differ between the levels. The reason for this is that the levels lack clear boundaries – quite where a situation becomes a domain or a domain becomes a worldview is unclear. The differences between seeing ‘a problem in multiple ways’, using ‘different primitives’ and different ‘worldviews’ are linguistic rather than semantic. The division is equivalent to picking out three levels of an inheritance hierarchy in preference to the others. Wenger’s proto-typology of viewpoints merely serves to illustrate that viewpoints are relevant at all levels of knowledge.

The remarks on composite viewpoints are too vague to be useful. The uses of viewpoints, above, are orthogonal to those noted in [Moyse & Elsom-Cook, 1992] (see 2.2.1, pg 24) but address the same underlying issues.

3.2.2 Self: Belief Sets

[Self, 1992] regards a viewpoint as a set of beliefs held by some agent.

A belief is a dispositional state. A belief (and hence a viewpoint) is held by some agent. ... dispositional is intended to indicate that possession of the belief disposes the agent to behave in an certain manner but does not guarantee it.

viewpoint

uni-agent multi-agent

incremental disparate independent inter-dependent Figure 3.1 Classification of Viewpoints from [Self, 1992]

[Self, 1992]

Applying a viewpoint yields a view. The discussion of viewpoints takes place in the context of the classification shown in Figure 3.1. The major distinction is between viewpoints held by one agent and those held by two or more agents.

Viewpoints held by one agent vary from each other by differing amounts, those close to each are called incremental viewpoints, those significantly different, disparate. These distinctions represent fuzzy areas on a viewpoint spectrum rather than discrete alternatives. An example of incremental viewpoints are the successive states of solving a state-space problem such as the missionaries and cannibals. In contrast disparate viewpoints involve changes in perspective that hopefully make the problem easier to solve. Different representations of the problem can have significant influences on the ease of solution, e.g. the mutilated chessboard.

The main complication that arises in multi-agent scenarios is that agents can maintain viewpoints about other agents and consequently viewpoints about the viewpoints of other agents, etc. Blackboard systems [Hayes-Roth, 1985] are offered as an example of independent viewpoints where each knowledge source acts without reference to the others in order to solve the problem. In contrast distributed problem-solving also occurs when agents do reason about each other, thus creating inter-dependent viewpoints (e.g. see Viewgen in section 2.3.1).

3.2.3 Other ILE Work

Using resolution in different viewpoints has been discussed in the context of diagnosing a student’s misconceptions in an ITS [Costa,

Duchenoy & Kodratoff, 1988]. It is based on the idea that misunderstandings can be caused by ‘teacher and learner working with different contexts.’ The system must therefore have partitions in its knowledge base and establish which partition, or context, the interaction takes place in and which the student thinks it takes place in. Costa’s example shows that by locating the correct context, through resolution, the student can be more accurately diagnosed.

[Finch, 1988] regards viewpoints as filters over a domain model which are selected to match an explanation of a domain concept to the student’s current pedagogical needs. This is similar to the overlay learner model (section 1.3.2). Multi-expert knowledge acquisition is recommended as a method of obtaining the different viewpoints to be drawn over the domain.