Intelligent Tutorial Systems
Book: HYPERMEDIA LEARNING SYSTEMSTHEORY – DIDACTICS – DESIGN
Prof. Dr. Rolf Schulmeister
Talk by:Martin Homik
For ActiveMath Lesegruppe:08.06.2005
Overview
ITS
Components Expert Systems Adaptivity
Knowledge ModelExpert Model
Learner ModelDiagnose Model
Tutor ModelPedagogical Model
Interface
Intelligent TS:• Agent Theory: perceive, reason, act• Goal: minimize gap between expert and
learner• Flexible and adaptive
Components[Barr/Feigenbaum]
Knowledge ModelExpert Model
Learner ModelDiagnosis Model
Tutor ModelPedagogical Model
Interface
ITS
Knowledge Domain(Expert Model)
Knowledge Domain
Declarative Procedural Heuristic
KnowledgeDesignModels
Black Box Glass BoxSemantic
Nets
Knowledge Domain(Expert Model)
• Declarative knowledge defines:– Terms from knowledge domain by their attributes– Relationship of the terms (by frames/inheritance)
• Procedural knowledge consists of:– Arguments/rules which help in solving problems
• Heuristic knowledge consists of:– Experience/problem-solving knowledge of experts– … not confined to particular contents
[Winograd (1975)]
Knowledge Domain Design Models
• Black box:– No reproduction of human intelligence– Example: SOPHIE I [Brown, Burton (1974)]
• Glass box [Goldstein/Papert (1977)] :– KD: modelled in the form of an expert system– Transparent: reproduces the problem-solving
behaviour of human experts
• Semantic Nets [Jonassen (1992)]:– Nodes with patterns– Typed relations
Student Model
Declarative Procedural
Knowledge Models
Subset(Overlay)
DeviationBuggy/Perturbation
Functions
Corrective Elaborating
Strategic Diagnostic
Predictive Evaluative
Student Model(Learner Model, Diagnosis Model)
Kinds of student model
• Subset model (overlay model):– The parts of the expert knowledge which the
student has done are ticked off • Deviation model
– Analyse student’s answers, conclude by way of inference what has been understood,
– … not »explaining« his learning behaviour.
Functions of SMs [Self (1988)]
• Corrective function:– find/correct student’s mistakes– follow the learner’s train of thought step by
step • Elaborating function:
– intervene if learner’s knowledge correct but incomplete
– compare the expert model with the learner’s current state of knowledge and suggest actions
• Strategic function:– change of the methodical level– provision of other learning strategies
Functions of SMs [Self (1988)]
• Diagnostic function:– find out the learner’s ideas– the system analyses the SM by itself
• Predictive function:– simulate the learner – … make predictions about behaviour
• Evaluative function:– reconstruct the learner’s learning process
Criticism on Student Models
“These approaches both imply a very simplistic model of the learning process (not far removed from rote learning), which takes no account of the rich range of learning styles and capabilities for which there is psychological evidence.“ [Elsom-Cook]
Criticism: Diagnosis Function (Models)
• Mostly: in the sense of bug recognition– Provide fix set of bugs– Register bugs in the course of the program and
then machine learn [Ohlsson and Langley (1988)]
– Problem: compound bugs, noise [see also Hennecke]
• “Neither the bug library technique nor the machine learning approach is currently used extensively in instructional computing systems” [Ohlsson (1993)].
Criticism: Learning Behaviour Gaps
• Wish: evaluate the psychological plausibility of a solution or mistake, but “there are disappointingly few psychologocal principles that can be used for that purpose.” [Ohlson/Langley (1988)]
• Individual learning styles and strategies researched by psychology play a minor role
Tutor Model(Pedagogical Model)
• Presentation of learning materials:– “What, when, how?”
• Simulates the decision behaviour of a teacher– Referring to pedagogical intervention– Generates appropriate instructions – Basis: difference between expert and student
model
Tutorial Strategies
• Socratic dialogue– Questions encourage analysis of learner’s
mistakes
• Coaching– Problems and activities for exercising skills– Trying out solutions to problems– Feedback provision
• In summary:– Tutor model follows rather the instructional
approach than the concept of discovery learning or the cognitive tool
Tutorial Gaps
• Everyday reasoning of the teacher• … his assumptions about the learning
process of the pupil or student, • .. his knowledge of the situation structure
and rules of interaction• Passive student:
– “the assumption of a given task and given expertise puts students in a passive role with respect to finding their own problems and developing their own expertise” [Bredo (1993)]
– Solution: Assembly Tool
Interface Interactions
• Socratic Dialogue– Ask questions and reason on answers
• Coaching– Analyse help requests
• Learning by Doing– System demands tasks; difference reasoning
• Learning while Doing– Tutor stays in the background– Provides occasionally help
Criticism
• Current TS are either directive or non-directive
• … but not both yet. [Elsom-Cook, 1988]– … it is by no means always the case that the
dialogue is truly Socratic”
• [Mandl/Horn] distinguish between:– Guided learning or instruction as aim
([Anderson/Reiser (1985)])– Microworld concept, with discovery learning as
aim ([Shute/Glaser et al (1989)])
Smithtown
LISP Tutor
Interface Types [Kearsley (1987)]
• Socratic dialogue• Coaching• Debugging• Microworld• Explanatory expert systems
complete control freedom of learning.
SocraticDialogue Microworld
Elsom-Cook Continuum
Natural Language Behaviour
• Simulates teacher [Mandl/Hron (1990)] • “…approach a natural language dialogue”
[Mandl/Hron (1990)] • Necessary feature of a tutorial system
[Spada/Opwis (1985)]• Linguistic Research today is much more
advanced, but “effective communication requires looking beyond the words that are spoken and determining what the tutor and student should be communicating about” [Woolf’s (1987)]
DiBi
KAVIS SUOMO
Systems or Prototypes?
• “…most systems focus on the development of only a single component of what would constitute a fully usable system” [Kearsley (1987)].
=> Systems are rather prototypes
Operationalisation of Concepts?
• Learning bases on a concept of behaviour– domain model: model of concepts (behavioural objectives) – student model: model of the student’s behavioural sequences
• In contrast to behaviourism: ITS attempts to define cognitive concepts for the domain. – Cannot avoid an operationalisation of these concepts as
behavioural objectives, if a comparison of student model and knowledge domain are to be possible
• Psychological theories [Pask, Saljö, Martin, Entwistle]:– help the educator to design and understand – cannot be operationalised in the sense of ITS– resist any reduction to if-when rules
Operationalisation of Concepts?
Problems:• Concept of understanding• Cognitive concepts [Dillenbourg/Self (1992)]
– True cognitive concepts do not exist as yet– “most of the work on learner modelling has been
concentrated on the […] behavior <–> behavioral knowledge mapping, with a relative neglect of the conceptual knowledge component”
• works of Resnick, Chabay, Larkin, Merrill, Ohlsson and others: Cognition is used in the sense of »cognitive science«
• “The major problems facing ITS design at present stem from a lack of applicable models of human learning” [Tompsett (1992), 98]
Lack of Success?
• Commercial failure [McCalla (1992a)]• Prototypes bound: particular knowledge domain
– Change of domain: effort of developing [Schulm.]– … CBT had more success than ITS [Duchastel (1992c)]
• Theoretical problems of ITS [Woolf (1987)] • Clancey (1989):
– His programs are not being used– “The effect is that our technological goals–exploring the
space of what computers can do for instruction–dominated over our educational goals.”
– … constructivist paradigm of »situated cognition«: “researchers must participate in the community they wish to influence”
GUIDON
GEO Tutor
Expert Systems
Expert Systems
Knowledge Base Inference Application
Facts
Rules
Strategies
Forward
Backward
Ask
Interpreter
Explanatory Component
Tutorial Decisions
ZEERA STAT-EXPERTGUIDON
Expert Systems
• Knowledge base (facts, rules, strategies)– Expert knowledge– Usually in logical notation– If-then rules
• Inference (forward/backward reasoning, ask)
• Applications:– Interpreter– Explanatory component – Tutorial decision about didactic strategies
Expert System vs. ITS
• … do not strive to simulate human reasoning or problem-solving
• … one cannot learn anything from expert systems, since expert systems merely acquire the necessary data by asking the users for information, and then draw their conclusions from them independently and ‘invisibly’.
• Clancey ( → ):– “… it cannot explain why a particular rule is correct, and
it cannot explain the strategy behind the design of its goal structure […] At a certain level, MYCIN is aphasic – able to perform, but unable to talk about what it knows”.
MYCIN GUIDON
Adaptivity
• Adaptive tutorial strategies:– Precond.: student model with diagnostic
functions– Determine the learner’s current level and history– Transmit this information to tutor
• Problem: Adapt to something that has not yet been fully researched by science– Bastien: concentrate on IUI– Presuppose a mental or cognitive model of user
thought processes
Adaption and Hypermedia Systems
• “Hypermedia is a non-pedagogical technology […] which must count on the student’s own intelligence for learning guidance.” [Duchastel]
• “Didactics […] are essentially goal-directed processes.” [Duchastel]
• Hypermedia ITS “provoke the student into browsing” [Duchastel]
• Schulmeister: – ITS supporting self-guided learning are still valuable
pedagogical tools– … better than giving expository instruction – … didactics should support open, exploratory,
constructive learning situations
Planned Adaptivity
• Instructional adaptivity [Duchastel]:– Not individuum-oriented– Adaption ≈ pedagogical knowledge– Adaptation to pre-imagined types of learners
invested into the program design by the designer
• Hermeneutic adaptivity [Schulmeister]:– Individuum-oriented– Learner furnishes the interpretative and the
“subject gives way“ – ITS pedagogics cannot do other than plan
adaptivity.
Planned Adaptivity
Good adaptivity → (different) learner parameters
Problems:• Combinatoric explosion• Logical consistency (if too many
parameters)• Internal consistency (parameters overlap)
Microadaption
• Adaptation to student models by different strategies in the instructional system
• Example [VanLehn (1991)]:– Explanation-based learning
• assumes complete mastery of the domain • presupposes stored knowledge can be accessed/applied
– Similarity-based learning→ Able to change the level of explanation
Problems:• Inadequate reproduction of learning processes• Cannot react to individual problems (are not
recognized by the diagnostic component)
Sierra
Adaptivity through Teaching Methods
• Not as teachers do …• … multitude of teaching methods/analyse a
multitude of learner variables:– Drill & practice – Tutorials with exercises– Interactive construction– Socratic dialogues– Exploratory learning environments
• Limits to the modification of didactic strategies: – Restricted to the knowledge domain– Restricted to observable learner behaviour
Summary
ITS
Components Expert Systems Adaptivity
Knowledge ModelExpert Model
Learner ModelDiagnose Model
Tutor ModelPedagogical Model
Interface
Understanding
• Simon and Hayes (1976):– Solving the logic of a problem, e.g. understanding the
operative structure of fractions.
• Greeno and Riley (1987):– Exclusively grasping concepts of natural science
• Why self-limitation to simple cognition levels?– “The major problems facing ITS design at present stem
from a lack of applicable models of human learning” [Tompsett (1992), 98]
• … starting point is the observation that students approach scientific problems in different ways than experts, and whose aim it is to approximate the student’s knowledge model to that of the expert.
ITS
Expert Systems and ITS
• ECAL: Elsom-Cook, O’Malley(1990)– CAL <-> ITS
• BIOMEC: Giardana (1992)– Allow discovery learning and apprenticeship– Dynamic: links between expert and student knowledge
• Physics Tutor: Jonassen, Wang– ITS with an expert system and a hypertext– Explore the practicability and generalizability of the ITS
concept
Authoring Systems
Simulators Constructive Learning Situations
Adaptivity and Control
• … control over learner?