1. Intelligent Tutoring Systems
Prof. Dr. Mohamed M. El Hadi
Sadat Academy for Management Sciences
M. M. El Hadi

2. OVERVIEW Introduction: Concepts, Objectives and Main Topics.
Learning Scenarios and Knowledge Representation
Factors Influencing ITS
ITS Conventional Model and Main Components
Case Based Reasoning and ITS
Perfect Teacher and ITS
Development Process to Create ITS
M. M. El Hadi

3. 1. INTRODUCTION: What Are Intelligent Tutoring Systems?
System that provides personalized tutoring by:
Generating problem solutions automatically
Representing the learner’s knowledge acquisition processes
Diagnosing learner’s activities
Providing advices and feedback
M. M. El Hadi

4. Intelligent Tutoring Systems
Traditional CAI
Fully specified presentation text
Canned questions and associated answers
Lack the ability to adapt to students
ICAI: intelligent computer-aided instruction
Reasoning
Rich representation of domain
User modeling
Communication of information structures
M. M. El Hadi

5. ITS - Objectives Practising environment (learn by doing)
Provide useful feedback on a student’s answer to a problem
Model the content and the student
Allow to make inferences about a student’s knowledge in order to adapt the content.
M. M. El Hadi

6. ITS Main Topics Learning Scenarios Domain Knowledge Representation
Student Modeling
Student Diagnosis
Problem Generation
User Interface
M. M. El Hadi

7. 2. LEARNING SCENARIOS AND KNOWLEDGE REPRESENTATION Learning Senarios
The situation in which the student’s learning is to take place
Coaching: offer a student advice and guide him when misdirected
Gaming environment: combine both coaching and discovering learning
Socratic teaching method
Simulation-base training
Discovery learning
M. M. El Hadi

8. Knowledge Representation
Knowledge is the key to intelligent behavior
The form in which we store the knowledge is crucial to our abilities to use it
No general form suitable for all knowledge
Challenge
determine the type of knowledge required, and suitable representation for that knowledge, to support teaching particular subjects
M. M. El Hadi

9. Script Representation
WHY, a Socratic tutoring system
Test student’s understanding of the major casual factors involved in rainfall
Require a representation with different levels of abstraction
Script
Nodes represent processes and events, links represent such relations as X enables Y or X causes Y
Each node have a hierarchically-embedded subscript
Roles are bound to geographic or meteorological entities in a particular case
M. M. El Hadi

10. Semantic Network SCHOLAR Semantic network
A mixed-initiative, fact-oriented system
Requires a highly-structured data base in which concepts and facts are connected along many dimensions
Semantic network
Nodes and links represent objects and properties
Generate questions, answers, errors and branching information from the semantic network of knowledge
Support flexible query and reasoning
M. M. El Hadi

11. Knowledge Representation Techniques
SCHOLAR
A mixed-initiative, fact-oriented system
Requires a highly-structured data base in which concepts and facts are connected along many dimensions
Semantic network
Nodes and links represent objects and properties
Generate questions, answers, errors and branching information from the semantic network of knowledge
Support flexible query and reasoning
M. M. El Hadi

12. 3. FACTORS INFLUENCING ITS Student Modeling
Overlay Modeling
student’s knowledge is viewed in terms of the tutor’s domain knowledge
Several approaches
Semantic net with nodes and links are added as they are taught
Stars with the expert knowledge base and annotates deviations that are subsequently discovered
Skill modeler: student modeled by the set of skills he has mastered
M. M. El Hadi

13. Buggy Model
Fact: the novice’s error can not be explained by the expert’s knowledge
Buggy model employs both correct and “buggy” rules
To understand an error, a combination of these correct and buggy rules has to be found to produce the same incorrect answer
M. M. El Hadi

14. Student Diagnosis Buggy model
Procedural networks: partially-ordered sequences of operations
Answer is evaluated by search for a path through this network of skills
Problem:
The number of paths grows exponentially
Require an explicit enumeration of bugs
M. M. El Hadi

15. Student Diagnosis (Cont.)
Error taxonomy
The knowledge of the types of misconceptions in a particular domain
Object-oriented approach
Each knowledge class inherits diagnostic capabilities from a particular Diagnoser class
M. M. El Hadi

16. Problem Generation A tree-structured decision process Semantic net
Each level represents another decision on what to include in the problem
Each branch represents one alternatives
The branches can be augmented with probabilities
Semantic net
Encode the types of objects and relevant attributes of these objects
A generative procedure fill in the particulars of the problem
M. M. El Hadi

17. Problem Generation (Cont.)
Problem generation, expert problem solving and student diagnosis can be viewed as a set of constraints on their solution
We can evaluate student answers by checking that al constraints are satisfied
Give student feedback on wrong answers by telling him which constraints he failed to satisfy
M. M. El Hadi

18. User Interface Text generation in tutoring systems
Most avoid true natural language mechanisms
SCHOLAR incorporate rich natural language in two distinct levels: semantic and syntactic
M. M. El Hadi

19. User Interface (Cont.) Natural language parsing
Rich natural language facilities
Semantic grammars: look for understandable fragments in the input
Using graphical or menu-based input
M. M. El Hadi

20. 4. ITS CONVENTIONAL MODEL AND MAIN COMPONENTS
M. M. El Hadi

21. The Three Main Components of an ITS
The Student Model
The Pedagogical or Tutor Model
The Domain Knowledge
M. M. El Hadi

22. ITSs and Their Interaction
M. M. El Hadi

23. How long should we keep the information?
The Student Model
Keeps track of all information related to the learner :
Description of student behavior with regard to a specific problem
Performance concerning the material being taught
Misconceptions
Knowledge gap
How long should we keep the information?
M. M. El Hadi

24. The Tutor Model Information about the teaching process:
When to review ?
When to present new topics?
What topics to teach?
Get input from the Learner model to make its decision to reflect the differing needs of each student.
M. M. El Hadi

25. The Domain Knowledge Contains the information the tutor is teaching
Most important part of the ITS
Issues:
How to represent knowledge so it easily scales up to large domain?
How to represent domain knowledge other than facts and
M. M. El Hadi

26. 5. CASE-BASED REASONING (CBR) AND CBITS
To represent the Student model and Domain Knowledge
There are different sources to obtain cases:
Produced by the learner himself
Experience from other learner
On-demand case generation
Predefined cases given by human tutors
M. M. El Hadi

27. Concepts of CBITS Where CBR technique become useful ?
During the Problem Solving phase : Find similar problem solved in the past to provide learner with past experience feedback.
Case-Based Adaptation
Case-Base Teaching
M. M. El Hadi

28. Case Based Adaptation Where CBR technique become useful ?
During the Problem Solving phase : Find similar problem solved in the past to provide learner with past experience feedback.
Case-Based Adaptation
Case-Base Teaching
M. M. El Hadi

29. Case-Based Teaching (Cont.)
Main goal is to provide learners with useful information (in order to understand new topics and to help during the problem solving phase).
Case-Based Teaching system are either:
Static (use given case base)
Adaptive (learn new case from learner experience)
M. M. El Hadi

30. Different Types of Case-Based Reasoning
Different type of CBR methods:
Classification Approach (used to provide help on well known pre- analyzed cases)
Problem Solving Approach (to diagnose solution proposed by the learner and to identify the problem solving path used)
Planning Approach (to support planning in the system)
M. M. El Hadi

31. Case Representation
As a Complete case: Problem definition + detailed solution
As Partial Case (Snippet) : Subgoals of problems + solution within different contexts
M. M. El Hadi

32. CBITS In Real Life But Why?
CBITS have been used in many different areas:
Biology : INVISSIBLE (under construction)
Physics : ANDES
Math : ActiveMATH
Jurisprudence
Economics
The most popular ones are:
Programming : ELM-Art, SQL-Tutor,
Chess : CACHET
But Why?
M. M. El Hadi

33. Further Work of CBITS Reduce development time and cost:
Using Authoring tools (API that would simplify programmer’s task to represent knowledge and teaching strategies)
Using Modularity of the student, tutor and domain models for future reuse
Collaborative Learning
Allowing student to interact (help) with each other while learning with an ITS
But problem concerning modeling student knowledge and defining teaching strategies
M. M. El Hadi

34. SQL-Tutor
Developed in 1996 by Dr. Mitrovic from University of Canterbury, New-Zeland.
Provide a good “on-hand” practice to student discovering SQL
Teaching with example and built-in Database relations.
Useful feedback is given by the system
M. M. El Hadi

35. SQL-Tutor (Cont.)
M. M. El Hadi

36. Current State of Intelligent Tutoring
ITS failed to recognised the fact that knowledge has contextual component (everything does not work everywhere).
Natural intelligence of student is ignored.
ITS attempted to replace human teacher!
Attempts to create a perfect teacher rather than a teacher’s tool.
M. M. El Hadi

37. 6. PERFECT TEACHER AND ITS
Different roles of teacher:
ITS designer teacher
ITS implementer teacher
(personality attributes, styles, preferences …)
Many implementer teachers distrust ITS as employing beliefs of the ‘designer teacher
M. M. El Hadi

38. Tutoring software contains attributes from designer teacher.
Participants of ITS
ITS is a joint cognitive system (Dalal & Kasper, 1994) involving:
a tutoring software
a student, and
an implementing teacher.
Tutoring software contains attributes from designer teacher.
M. M. El Hadi

39. Context of ITS
Besides the interactional context, the environmental and objectival contexts are important for any educational system.
M. M. El Hadi

40. Teacher and ITS
M. M. El Hadi

41. Teacher as an Implementer of ITS
provides context (background, culture, policies..)
selects and schedules other educational technologies
manages the curriculum, and
oversees the learning progression.
In the ensuing power relationship, tutor’s preferences may be more important than the learning styles of a student!
M. M. El Hadi

42. Teacher as an Implementer of ITS (Cont)
Different teaching styles may result in points of divergence within the joint cognitive learning.
Clark (in press) notes:
Instructional methods, not the media cause learning.
Human brain can be overloaded by technologically delivered sensory output.
However, the situation is not so straight forward!
M. M. El Hadi

43. Teacher as an Implementer of ITS (Cont)
Novice learners may benefit from richer content, but may also get distracted without directed learning.
Different teachers would constrain the learning process in different ways reflecting their teaching styles!
Where does the role of teacher fit in overall environmental context of ITS?
M. M. El Hadi

44. Environmental Context of ITS
M. M. El Hadi

45. Teacher as an Environmental Context
Implementer Teacher provides power relationship.
Preferences of a teacher prove to be more important than learning styles of students.
Human teacher plays very important role in the acceptance of a tutoring system.
Designer Teacher needs to take into account of implementer teacher’s preferences!
M. M. El Hadi

46. Modeling Human Teacher
Human teachers may have:
different personalities
different teaching styles (born out of their traditional, progressive or vocational outlook and their own learning style)
It is not possible to envisage all the preferences of implementer teacher at design time.
M. M. El Hadi

47. Human Teacher Model
We recommend a re-configurable human teacher model to be incorporated in the design of ITS:
to recognise the different teaching styles
to put on record the teaching style(s) adopted in the design, and
enable manual or automatic adaptation to suit the implementing teacher
M. M. El Hadi

48. Why Explicit Record of Designer’s Teaching Style (s)
Better understanding of designer’s rationale by implementing teacher
Help in dealing with the cognitive dissonance arising from any differences in teaching styles
M. M. El Hadi

49. Why Explicit Record of Designer’s Teaching Style (s) Cont.
Clear rationale behind adopted teaching strategy may also help in the student learning in less adaptive systems
Easier understanding of representations which are difficult due to cultural differences
If designer’s teaching style is unproductive in a culture, the system may be localised
M. M. El Hadi

50. ITS Incremental Growth
ITS, in their current stage, cannot replace all the functions of human teacher.
Efforts should be on increasing productivity (just like initial word processors for steno-typists).
ITS designers should treat human teacher as their target user.
Human teacher model is next logical approach in that direction.
M. M. El Hadi

51. 7. DEVELOPMENT PROCESS TO CREAT ITS
Interface Learning Environment and ITS
Select Domain
Choose ITS
Model
Choose Learning
Environment
Deploy ITS
Validate ITS
Create Problems
Author ITS
Evaluate ITS
M. M. El Hadi

52. Main Stages of ITS Development Process
Phase 1: Select Domain:
- Choose domain where problem solving plays a major role.
- Building an ITS requires high investment of efficient developers and their expertise.
Phase 2: Choose ITS Model
- Most popular models are cognitive tutors (Model Tracing, Knowledge Tracing) [Developed by Carnegie Mellon University, Pittsburgh, PA] and Constraint Based Tutor [Developed by Conterbury University, New Zeland]
- Choose CT to teach procedural skills.
- Choose CBT to teach open-ended domains.
M. M. El Hadi

53. Main Stages of ITS Development Process (Cont.)
Stage 3: Choose Learning Environment:
- Student interface has to be simple and adapted to the domain (reduce memory load, motivate, etc.)
- A lot of learning environments exist already and can be reused)
Stage 4: Interface Learning Environment (LE) and ITS:
- Convert solution from learning environment into format required by ITS.
- Possibly integrate LE into ITS to make it Web accessible.
- Convert student solution to the XML web language format required by ITS.
- Convert Java language application to Java applet required by Web
M. M. El Hadi

54. Main Stages of ITS Development Process (cont.)
Stage 5: Author ITS:
- Existing ITS Examples:
* WETAS: a development environment authoring tool done through text files. WETAS is a domain model (knowledge base)
* ASPIRE: a deployment environment authoring tool (ASPIRE Tutor). It guides author in the development process and considered a semi-automated process for building the domain model.
- Constraint Based Tutor Model domain with constraints
* IF X (relevance condition) is true
Then Y (satisfaction condition) has to be also true
- Feedback generated when:
* Relevance condition is true
* satisfaction condition is false
M. M. El Hadi

55. Main Stages of ITS Development Process (cont.)
Stage 6: Validate ITS:
- Test the knowledge base (constraints)
- Test if generated feedback is appropriate.
Stage 7: Deploy ITS:
- Make ITS available to students
Stage 7: Evaluate ITS:
- Identify aspects to be evaluated (example: effectiveness of the feedback).
- Use real students learning the domain
- Pre- and post – test
- Comparison group (uses full version of ITS)
- Control group (uses version with no feedback).
M. M. El Hadi