Intelligent Tutoring Systems

The DynaLearn Approach

Wouter Beek, Bert Bredeweg me@wouterbeek.com, B.Bredeweg@uva.nl

Informatics Institute University of Amsterdam

The Netherlands

This work is co-funded by the EC within FP7, Project no. 231526, http://www.DynaLearn.eu

Problem statement

• Worrying decline in science curricula – Less students sign up; more students drop out.

• Main reasons: – Lack of engagement and motivation in science teaching

(Osborne et al. 2003). – Teaching practice involves surface knowledge in terms of

formulas and uninterpreted numeric data. – Lack of interactive tools to construct conceptual

knowledge.

Conceptual science education

Having learners acquiring conceptual knowledge of system’s behaviour: • Deep knowledge in terms of the concepts that are involved. • Learn basic principles that can be carried over to other

problem instances. • Learn to adequately explain and predict the behaviour of

systems to utilise their functioning for human benefit. • A prerequisite for working with numerical models and

equations. • Communicate insights to the general public.

Knowledge construction

• Develop an interactive learning environment that allows learners to construct their conceptual system knowledge.

• Characteristics: – Accommodate the true nature of conceptual knowledge. – Automate feedback for open-ended construction tasks. – React to the individual knowledge needs of learners. – Applied to the interdisciplinary curriculum of

environmental science. – Be engaging by using personified agent technology.

LEARNING BY CONCEPTUAL MODELLING

Part I

Learning by Conceptual Modelling

• Modelling is fundamental to human cognition and scientific inquiry (Schwarz & White 2005)

• Simulations mimic the behaviour of real-world systems.

• Conceptual Reasoning captures the human interpretation of reality: – Couched in the appropriate vocabulary. – Remove numerical ‘overhead’. – Provides handles to automate interaction.

Quantitative & Qualitative Knowledge

– An increase (or decrease) in Force causes an increase (or

decrease) in Acceleration – An increase (or decrease) in Mass causes an decrease (or

increase) in Acceleration – An increase (or decrease) in Acceleration causes a

decrease (or increase) in Mass.

m F F = m * a

Explicitizing the semantics of the domain

• Scope: Which aspects of the system should be included in the model? (relevant/irrelevant)

• Granularity: What is the level of detail that should be modeled?

• Compositionality: How must knowledge be put in modules in order to allow knowledge reuse?

• Conditionality: Under what conditions do certain knowledge modules apply?

engine

1 2 3

5

4

6

build

1

2

3

5

4

6

simulate

5 4

6

1

5

6

2 3 4

6

1

5

State graph Value history Transition history Quantity values

Equation history

Dependencies Causal view Model fragments

Causal relations Derivative values

Magnitude values Quantity spaces

Causal influence Correspondences (In)equalities External factors Calculi

Conditionals Knowledge library

Hierarchies Multiple scenarios

Example: Population dynamics

• How do populations in general behave? • What processes determine their behaviour?

• Issues:

– Size (number of individuals) – Birth / Natality – Death / Mortality

Constructing knowledge (1)

Generic class

Specific instance

Quantity

Possible values Derivative

(direction of change)

Current value

Constructing knowledge (2) Influence: The amount of Birth increases Number of

Proportionality: Changes in Number of determine changes in Birth

Constructing knowledge (3)

Negative influence Positive influence

Simulation results

GROUNDING, FEEDBACK, RECOMMENDATIONS

Part II

Grounding

http://dbpedia.org/resource/Mortality_rate http://dbpedia.org/resource/Population

http://dbpedia.org/resource/Size

Expert/teacher Student

grounding

Semantic repository

Feedback & Recommendations

feedback

recommendations

Student

Expert

Community of users

e.g., “You can complete your model with a P+ proportionality”

e.g., “Users who modelled death also modelled birth”

DIAGNOSIS & REPAIR Part III

Initial OBS

Blueprint

Device

Component Library

Diagnose

CCM/SD

Build CCM

Diagnoses

OBS

Probe point Perform

Read

Observe

Measuring result

Measuring action

Inputs

Outputs

Repair

Initial OBS

Outputs

Inputs

Device Student

Blueprint Component Library

Repair Diagnose

Expectation

Build CCM

Diagnoses

Modelling Goals

QR Sim QR Model Simulate

Model

Inspect

OBS

Probe point Perform

Read

Observe

Measuring result

Measuring action

CCM/SD

Question Ask

Responds Answer

Communicate

Automatic Repair

Example

I expect Free Space to be Low.

What should be the value of Inhabited Space in state 2?

Inhabited Space should be High there.

Then this directed correspondence cannot be right.

MOTIVATION & ENGAGEMENT

Part IV

Character roles

• Feedback & Recommendations • Diagnosis & Repair • Causal Explanation (why?) • Teachable agent • What is? • How to? • Quiz

Teachable agent

Biswas, et.al (Betty's Brain)

Causal Explanation (why?)

Concluding remarks

• Problem statement and Context – Communicative interaction (science education)

• Knowledge representation and Reasoning – Qualitative system dynamics / Conceptual knowledge

• Progressive learning spaces (6 spaces) • Feedback for Reflective thought

– Semantic Web techniques (model ingredients) – Consistency-based Diagnosis (simulation results)

• Inducing Motivation (virtual characters & modes)

http://www.DynaLearn.eu

Project Partners

UVA (Netherlands) TAU (Israel)

UPM (Spain) UH (UK)

UAU (Germany) IBER (Bulgaria)

FUB (Brazil) BOKU (Austria)

http://www.DynaLearn.eu