The local governance of public security in France: fragmentation and new orientations
3. El modelo mixto de las estrategias locales
3.2. Una diferenciación de los enfoques locales con respecto al
One of the distinguishing features of ILEs is the student model; ‘the component that represents the student’s current state of knowledge’ [VanLehn, 1988]. The [Michie, Paterson & Hayes-Michie, 1989] system did not include a student model. Although [Palthepu, Greer & McCalla, 1991] include a student model in their system description they question its utility. In the analogy with peer tutoring however, a student model seems perverse: a tutee generally can’t model the knowledge of their tutor. The state of the student’s knowledge is reflected in the model built up by the ISS and so a separate student model would be redundant.
The interface between the control strategy and the learnt model of an ISS can be considered to consist of two parts: actions that maintain the model and those which assess its contents. Figure 4.1 shows a schematic diagram of the theoretical architecture of an ISS with an optional domain model. The components are:
· learnt model:23 the model the system has acquired from the learner
merged with the ‘seeded’ initial model (if present).
· conceptual syntax: the restrictions on the contents of the learnt model.
· dialogue strategy: control of the system’s behaviour, including when to update the learnt model. Achieved via conventional programming constructs: variables, conditional tests, loops, etc.
· model maintenance functions: that modify the learnt model; such as adding and deleting beliefs acquired from the learner.
· model access functions: the mechanisms with which the dialogue strategy tests the state of the learnt model and the domain model (if present). These are read-only functions which leave the learnt model unaltered.
· interface functions: communication with the user (dialogs, menus, prompts, graphics etc).
· domain model: (optional) a model of expert domain knowledge in the domain in which the system is being used.
The outline input-output cycle is: 1) learner provides an input
2) input checked against the current conceptual syntax
3) dialogue (or learning) strategy takes the checked input and, accessing the learnt model (and optionally the domain model), determines the next action, including, for example, a combination of:
· modifying the learnt model
· modifying the conceptual syntax
· asking the learner a question (which could be confirmatory,
exploratory, open or closed) or some other interface actionHUMAN LEARNER ( TUTOR )
INTELLIGENT
STUDENT
SYSTEM
( TUTEE )
DIALOGUE STRATEGY CONCEPTUAL SYNTAX DOMAINMODEL LEARNT MODEL
MODEL MAINTENANCE MODEL ACCESS FUNCTIONS MODEL ACCESS FUNCTIONS INTERFACE LEARNER'S MODEL INTERFACE FUNCTIONS PROGRAMMING FUNCTIONS
Figure 4.1 Theoretical Architecture of an ISS
4.5 Synthesis
The peer tutoring research suggests that a learner can achieve significant educational benefits through interacting with another learner who is less knowledgeable about a domain. Similarly, an ISS has less knowledge than the learner. This runs counter to the prevailing ILE philosophy of building intelligent systems which are superior to the learner in domain knowledge and have accompanying pedagogical
knowledge to mediate the interaction. The ILE community has been fixated on the conventional teacher-student relationship which has resulted in attempts to build systems which require large amounts of knowledge engineering. Whilst this is not explicitly incorrect the focus on a ITS ‘communication’ approach [Wenger, 1987] has been, at least partially, a misdirection of the research effort. The input (computational complexity) to output (educational effects) ratio of an ISS approach is considerably better than that of the ITS.
There are two main reasons for the simpler architecture of the ISS: 1) the student model and the domain model of an ITS are merged into the single ISS learnt model. Although the theoretical architecture of the ISS allows a full domain model it is expected that ISSs can function effectively without a fully specified model of the domain. To the extent that ISSs can work with partially full domain models then ISSs are simpler than ITSs.
2) the dialogue strategy of a ISS is based on learning rather than teaching. Teaching effects emerge from placing the human learner in the role of a teacher; they do not have to be explicitly coded in the dialogue strategy. A strategy that teaches is inherently more complex than one that learns because it has to combine information from two models (a student model and a domain model) whereas a learning strategy only reasons about one model.
The central argument so far can be summarised as:
· ILE research has been biased towards ITSs in domains which can be
adequately modelled with single-viewpoints.· The problems of multiple viewpoints will become increasingly
apparent as ILEs move into new domains.· These problems can be dealt with through improved reasoning
techniques or designing new ILEs with knowledge requirements that inherently reduce viewpoint-based difficulties – however any ILE must be ‘viewpoint-aware’.· An ISS is an example of an ILE that attempts to minimise the
problems of viewpoints through system design.· The ISS concept is grounded in current educational practice, peer
tutoring, but has only ever been explored once – in a statistical manner in a procedural domain.· ISSs have the potential for providing a new form of ILE interaction;
complementary to, and computationally simpler than, ITSs.The following empirical research questions arise from the above points:
1) can an ISS be used as a learning by teaching tool?
2) does it replicate any of the effects of human-human peer tutoring? 3) what domain-related knowledge does an ISS require?
4) do learners find ISS interaction useful/interesting?
5) can ISS results be as general as those from human-human peer tutoring?
6) do ISS activities require a concrete task other than teaching? 7) does an ISS require an additional student model?
8) what type of learning should an ISS perform?
Only one experimental study [Michie, Paterson & Hayes-Michie, 1989] has ever attempted to address any of these questions (in an atypical domain) and there are sound educational, computational and economic reasons for further investigation of ISSs.
The scope of these questions is clearly greater than the scope of this thesis. Questions 2 and 5 require a greater level of detail and precision in studies of human-human peer tutoring [Kennedy, 1990] before they can be answered in full. Although some aspects of question 2 may be answered relatively swiftly question 5 implies many studies of ISSs covering a diverse set of domains. Similarly, question 8 calls for numerous controlled experiments with ISSs containing different dialogue strategies.
Questions 1, 3, 4, 6 and 7 lend themselves to examination within the resource limitations of this research. Although unequivocal answers are unlikely to emerge from a single study it is with these five questions that the study will be examined in section 7.5. The research therefore requires an ISS to be designed, implemented and tested.
This Chapter has provided the theoretical framework for an investigation into ISSs – the following Chapter describes the design and implementation of an ISS in the domain of economics.