Pedagogical Agent Design for Distributed Collaborative Learning

1
Pedagogical Agent Design for Distributed
Collaborative Learning
Half day tutorial at ICCE 2003, Hong Kong,
December 2, 2003

• Weiqin Chen and Anders Mørch
• Dept. of Information Science, University of
  Bergen
• InterMedia, University of Oslo

2
Outline

• Software Agents
• Agents in education
• Pedagogical agents in distributed collaborative
  learning
• A case study with demo
• Lessons learned

3
Software Agents
4
Outline

• Definition
• Classification
• Brief history
• Research Questions
• Tools and Platforms
• Agent Applications

5
What is an Agent? (1)

• An over-used term
• autonomous agents, software agents, intelligent
  agents, interface agents, virtual agents,
  information agents, mobile agents
• No commonly accepted notion

6
What is an Agent (2)

• computer programs that simulate a human
  relationship by doing something that another
  person could do for you.
  T. Selker (1994)
• persistent software entity dedicated to a
  specific purpose.
  D. smith, et al. (1994)
• situated in some environment, that is capable of
  flexible autonomous action in order to meet its
  design objectives.
• N. Jennings M. Wooldridge (1998)
• Other definitions in S.Franklin A. Graesser
  (1997) http//www.msci.memphis.edu/franklin/Agen
  tProg.html

7
What is an Agent? (3)

• Weak Notion of Agency
• Autonomy
• Social ability
• Reactivity
• Pro-activeness
• Strong Notion of Agency
• Belief, desire, intention

8
What is an Agent? (4)

• Two extreme views
• Agents as essentially conscious, cognitive
  entities that have feelings, perceptions, and
  emotions just like human
• Agents as automata and behave as they are
  designed and programmed

9
Agent Taxonomy
Franklin and Graessers (1997) agent taxonomy
10
Agent Typology
Typology based on Nwanas (Nwana, 1996) primary
attribute dimension
11
Multiagent Systems (MAS)

• A loosely coupled network of problem solvers that
• work together to solve problems that are beyond
  the individual capabilities or knowledge of each
  problem solver.
• Characteristics
• Limited viewpoint
• No global system control
• Decentralized data
• Asynchronous computation

12
Origin of Software Agents

• The idea of an agent originated with John
  McCarthy in the mid-1950s and the term was
  coined by Oliver G. Selfridge a few years later,
  when they were both at the Massachusetts
  Institute of Technology. They had in view a
  system that, when given a goal, could carry out
  the details of the appropriate computer
  operations and could ask for and receive advice,
  offered in human terms, when it was stuck. An
  agent would be a soft robot living and doing
  its business within the computers world.
• -Alan Kay
• Computer Software, 1984

13
Brief History(1)

• 1994
• CACM special issue on agents
• P. Maes, Agents that reduce work and information
  overload
• D. Norman, et al. How might people interact with
  agents
• ATAL (Workshop on Agent Theories, Architectures,
  and Languages)
• 1995
• Intelligent agents Theory and Practice by M.
  Wooldridge N. Jennings
• ICMAS (Int. Conf. on Multi-Agent Systems)

14
Brief History (2)

• 1996
• PAAM (Int. Conf. on Practical Application of
  Intelligent Agents and Multi-Agent Technology)
• FIPA (Foundation for Intelligent Physical Agents)
• 1997
• J. M. Bradshaw, Software agents
• AA (Int. Conf. on Autonomous Agents)
• 1998
• JAMAS (Journal of Autonomous Multi-Agent Systems)
• 2002
• AAMAS (joint of ICMAS, AA ATAL)

15
Agent-Enabled System Architecture (Bradshaw, 1997)
16
Design Issues (1)

• AgentUser
• Control -who takes control?
• Understanding
• -how to make agents understandable /trustworthy?
• Personification
• -how to present agent?
• Distraction -how to minimize distraction?
• User modelling
• -how to model emotion, intention, social
  behaviours, etc?
• Privacy
• -how to protect privacy?

17
Design Issues (2)

• Agent--Other Agents
• How to find other agents?
• How to model other agents?
• How to communicate with other agents (language,
  ontology)?
• KQML (Finn Labrou, 1997) KIF (Genesereth
  Fikes, 1992), FIPA ACL FIPA SL
• Ontology (Gruber, 1993)
• How to cooperate/negotiate with other agents?
• AI techniques (inference, planning, logic,
  constraint satisfaction)

18
Design Issues (3)

• AgentLegacy software
• Three possible solutions (Genesereth Ketchpel,
  1994)
• Rewriting the software
• Transducer (interpreter)
• Wrapper
• Another suggestion
• Web Services
• Scalability, stability and performance

19
Tools and Platforms

• OOA (Open Agent Architecture) by SRI
  Internationals AI Center
• http//www.ai.sri.com/oaa/
• JATLite (Java Agent Template, Lite) by Center for
  Design Research, Stanford Univ.
• http//cdr.stanford.edu/ABE/JavaAgent.html
• ZEUS by British Telecommunications (BT)
• http//more.btexact.com/projects/agents/zeus/inde
  x.htm
• JADE (Java Agent Development Framework) by
  Telecom Lab, Italia (TILAB)
• http//sharon.cselt.it/projects/jade/

20
Agent Applications

• Workflow management
• Network management
• Air-traffic control
• Business process engineering
• Entertainment

• Personal assistants
• E-mail filtering
• Information management
• Data mining
• E-commerce
• Education

21
Agents in Education
22
Outline

• Roles of agents from CAI to CSCL
• Spectrum of educational systems
• Positioning the agent component
• Some examples
• Intelligent tutoring systems
• Domain-oriented design environments
• Collaborative learning environments

23
Roles for agents CAI to CSCL
Computer-Assisted Instruction
24
Educational systems paradigms

• Computer Aided Instruction (CAI)
• Intelligent Tutoring Systems (ITS)
• Microworlds (MW)
• Guided Discovery and Critiquing (GDC)
• Knowledge Building Environments (CSCL)
• Note the paradigms are continually evolving and
  mutually influencing each other

25
Spectrum system-user control

• Positioning educational systems along a line of
  increasing order of their enabling user control,
  or alternatively allowing predefined
  instructional sequences
• On the left Instructional systems and ITS
• One the right open (constructionist and
  constructivist) learning environments
• In between Guided discovery and critiquing

26
Distribution of (human-computer) control in
educational systems
CAI
ITS
GDC
MW
CSCL
On left Systems supporting well-defined
instruction On right Systems allowing
user-defined interaction Note ! Comparison leaves
out important variables ..
27
Omission 1 number of users

• First and second generation systems (CAI ITS
  GDC MW) were primarily built for single users
• The field of CSCW had not yet matured
• Third generation systems (CSCL) are multiuser,
  since the focus now is on how to support
  collaborative learning

28
Omission 2 type of artefact

• First generation systems (CAI, ITS) tended to
  favour behavioural and mental aspects of learning
  (psychology)
• Second generation systems (MW, GDC) put more
  emphasis on the physical aspects of learning
  (learning by doing)
• Third generation systems (CSCL) tend to favour
  conceptual aspects of learning (learning to
  reason)

29
Positioning the agent component

• Agents can support part of the system
  functionality of a learning environment
• Agents can also support part of the user work
  in a learning environment
• Agents are positioned somewhere in-between hard
  coded (programmed) functionality and informal
  rules to guide user interaction and social
  conduct

30
1st Gen. Tutors and Coaches

• Expert systems for teaching and learning
• Works best in well-defined domains (e.g. physics,
  computer programming)
• Instructional planner
• High-level goals and strategies
• Individual student model
• Many opportunities for agent (coach) interaction
• Few opportunities for learning by discovery

31
Grace A Tutor for Cobol
32
Andes ITS with coaching
33
Microworlds

• Microworlds are not directly associated with
  agents, but have been important inspiration and
  platform for agent integration
• Microworlds define domain-specific worlds users
  can freely explore to build artefacts of the own
  creation and learn as a by-product
• Microworlds support constructionist learning or
  learning by doing (Harel Papert, 1991)

34
2nd gen. Guided discovery

• Combining open learning environments with teacher
  guidance
• Conceptual models and well-defined tasks to be
  discovered
• Learners construct knowledge themselves by being
  engaged philosophy is just try it
• Teacher as facilitator standing behind the
  shoulder to encourage, challenge, and direct
• Goal of teacher stimulate students critical
  thinking skills

35
2nd gen. Critiquing systems

• Computational approach to guided discovery by
  integrating Microworlds with ITS rule-bases
• Conceptual foundation in Donald Schöns theory of
  expert everyday knowledge building
• Learning by doing (physical activity)
• Learning by reflection (mental/conceptual aspect)
• Application domains
• Learning-on-demand
• Design (architecture, network design, lunar
  habitat)
• The computer-based critic systems are referred to
  as Integrated Design Environments

36
Janus The doing face
37
Janus The argumentation face
38
Janus Critics

• Critics are intelligent interface agents
• Linking doing and conceptualization, and
  correspondingly construction and argumentation
• Breakdowns in construction may create new
  (unanticipated) learning opportunities
• Making students pause and reflect
• Abstract concepts are presented to students in a
  context when it is meaningful for them

39
Comparing Tutors Critics
40
CoPAS Simulating CSCL Agents
41
Pedagogical Agents in Distributed Collaborative
Learning Environments
42
General definition

• Pedagogical agent definition adopted from Johnson
  et al., (1999)
• Pedagogical agents can be autonomous and/or
  interface agents that support human learning in
  the context of an interactive learning
  environment.

43
3rd Gen. Agents for CSCL

• Agents as online facilitators in CSCW and CSCL
  environments
• Pedagogical agents operate in the context of
  collaboration systems, such as groupware
• Facilitating communication and coordination among
  collaborating peers
• In our case facilitating knowledge building and
  progressive inquiry (to be discussed)

44
CSCW

• Computer Supported Cooperative Work
• CS-part focus on groupware, knowledge management
  other collaboration systems
• Technical issues include distribution, document
  sharing, coordination, awareness
• CW-part address social aspects of using the
  systems by empirical (usually field) studies
• Also important are conceptual approaches, such as
  coordination theory and languages

45
CSCL

• Computer Supported Collaborative Learning
• Educational CSCW applications for teaching and
  learning (school or workplace)
• Broad and multifaceted conceptual foundation,
  which includes
• Socio-cultural perspectives
• Situated learning
• Distributed cognition

46
Knowledge Building

• A technique for collaborative learning
• Students learn by talking (reasoning aloud) for
  the purpose of developing explanations
• Formulating research questions, answering them
  independently, and finding support
• Structured as a discussion with message
  categories modelled after scientific discourse
• Computer supported by environments such as
• CSILE and Knowledge Forum
• Fle3

47
Agents as facilitators

• Monitor participation in KB discussion and
  provide advice to the participants
• Encourage non-active students to be more active
• Suggest what messages to reply to and who should
  be doing so
• Suggest what category to choose for the next
  message to be posted
• Suggest when messages do not follow the
  scientific method of knowledge building, etc.

48
Conceptual design

• Agents for CSCL can be designed from different
  perspectives
• Technological-based design
• Theory-based design
• Empirical-based design
• These perspectives can be combined

49
Technological-based design

• Build agent technology from
• Scratch (e.g. java, python)
• Existing agent development environments, such as
  JADE, ZEUS and Microsoft Agents
• Build agent systems by integrating them with
  existing educational systems
• Open source systems (e.g. Fle3)
• Other systems (e.g. Teamwave Workplace)
• Combinations of the above is possible

50
Theory-based design

• Coordination theory
• Conceptual models of collaboration
• Patterns of collaborative interaction, such as
  genuine interdependence
• sharing information
• mindful engagement
• joint construction of ideas

51
Empirical-based design

• This has been the main perspective in DoCTA
• Example of a finding that can drive design
• Three kinds of postings in the forum
• knowledge building proper
• meta commenting (discussion of the KB process)
• social talk and chatting
• Design implications
• help students with choosing KB categories
• off-load some of meta commenting to agents

52
Example of empirical data that have been used to
feed design
Table 2 One the left Knowledge-building thread
in FLE with notes coded according to Knowledge
Building proper (KB), Meta commenting (MC), and
Social talk (ST). On the right The postings of
the pilot group. From Mørch, Dolonen Omdahl
(2003).
53
Design space for pedagogical agents

• Technological and conceptual dimensions providing
  guidance (questions, possibilities, constraints)
  for design
• presentation
• intervention
• task
• pedagogy

54
Presentation dimension

• How an agent should present itself to the user
• Computational technique Separate window,
  overlapping window, pop-up box, animated
  character, etc.
• How to present information Text, speech,
  graphics, body language simulation, etc.
• Examples (Office Assistant, STEVE, text pop-up in
  CoPAS)

55
Intervention dimension

• When an agent should present information to the
  user (a timing issue)
• Analogy with thermostat When a certain
  environmental variable has reached a threshold
  value, an action is taken
• Intervention strategies to be decided
• degree of immediacy (how soon)
• degree of repetition (how often)
• degree of intrusiveness (block or superimpose)
• degree of eagerness (how important)

56
Task dimension

• Interacting with a learning environment w/agents
  is radically different from interaction with the
  same environment without agents
• Different tasks may require different agents
• Well-defined tasks (eg. physics) are different
  from
• Ill-defined tasks (e.g. city planning)
• Agents can help to simplify the task
• Agents can make the task harder to complete
• Agents can create breakdowns in task performance,
  causing problem restructuring

57
Pedagogy dimension (CSCL)

• Agents as conceptual awareness mechanism,
  coordinating multiple knowledge sources (humans
  and online resources)
• Serving as missing link in distributed settings
• A new person just logged on need to be updated
• Informing teachers about student activity
• Agents embodying collaboration principles
• Division of labour
• Participation and coordination of joint work
• About scientific discourse (knowledge building)

58
Balancing the dimensions

• Making design decisions for agents based on
  choosing values for each of the four dimensions
• Do we need to take all of them into account in
  any one design, or is a subset sufficient? and
  are there other dimensions that could be included
  as well?
• Need also to balance human-agent distribution
• We give example of the design decisions chosen
  for a specific project DoCTA-NSS

59
A Case Study with Demo
60
Outline

• Background DoCTA-NSS
• Pilot Study
• FLE3
• Assistant in FLE3

61
DoCTA-NSS

• DoCTA DoCTA-NSS
• Design and use of Collaborative
  Telelearning Artefacts Natural Science Studios
• Goal
• study social, cultural and pedagogical
  aspects of artefacts in collaborative
  telelearning process and apply the findings to
  the design of environments
• Pilot study
• gen-ethics scenario

62
DoCTA NSS The physical setting
Bergen
Fle3 Server
Fle3
Internet Connection
Fle3
Online Resources
Online Resources
Mindmap
Mindmap
Hovseter local group user interface
KB Activity Overview
Sandgotna local group user interface
Student Coach
Instructor
Curriculum Manager
Email Notification
Kirstens (Instructor) Workspace
Subject Matter Expert
Activity Organiizer
63
FLE3

• Developed at UIAH Media Lab, University of Art
  and Design Helsinki http//mlab.uiah.fi/
• Builds on many years of research on networked
  learning http//www.helsinki.fi/science/networkedl
  earning/eng/index.html

64
FLE
Fle2 is design to support problem based learning
(PBL) and inquiry learning. Fle2 helps students
and teachers to engage in coordinated efforts to
solve problems and build knowledge together.
(Muukkonen, H Hakkarainen K. Leinonen T.
2000).
65
Progressive Inquiry
66
FLE3 Interface
From FLE3 website
67
Agent Design Issues

• Findings of pilot study
• Students have difficulties in choosing categories
• Instructors have difficulties in following the
  collaboration and give advice
• Facilitating Collaborative Knowledge Building
• Awareness
• Presenting advice and explanation
• Learning from feedback

68
Functionalities

• As an observer
• Collect information
• Participant, activity, timestamp
• Last log on, last contribution (for each
  participant)
• Compute statistics
• Present statistics--chart
• As an advisor
• Present updates, statistics
• Advice instructor on possible problems and
  sending messages to students
• Advice students on the use of categories

69
Assistant in FLE3
70
Student interface
71
Instructor interface
72
Technical Details

• Database
• MySQL
• Learning algorithm
• CN2 (Clark Niblett, 1989)
• Knowledge representation
• RuleML

73
Demo
Http//nattergal.ifi.uib.no8088/FLE
74
Lessons Learned

• Scalability
• from single user to multi user systems
• from well defined to ill defined domains
• CSCL needs more than (conceptual) knowledge
  building (what about domain-specific
  microworlds?)
• Instantiating various design dimensions
• Importance of understanding collaboration
• Integration of agents with human facilitation
• A full scale field study