Download Learning companions 1

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Machine learning wikipedia , lookup

Human–computer interaction wikipedia , lookup

Human-Computer Interaction Institute wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Concept learning wikipedia , lookup

Expert system wikipedia , lookup

Adaptive collaborative control wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Intelligence explosion wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Transcript
Learning companions 1
CPI 494, Kurt VanLehn
March 26, 2009
3 major dimensions of LCs
(LC = learning companions)
• Kim, Y. (2007). Desirable characteristics of
learning companions. International Journal of
Artificial Intelligence and Education, 17, 371388.
• Interviews suggest 3 factors
– Competency of the LC
– Personality of the LC: friendly vs. neutral
– Interaction control: LC vs. human has the
conversational initiative
Why now?
• First try to find out what kind of LC is best
• Then test efficacy vs. ITS vs. baseline
Preference for strong vs. weak LCs
when have a choice
• Hietala, P., & Niemirepo, T. (1998). The
competence of learning companion agents.
International Journal of Artificial Intelligence
and Education, 9, 178-192.
Types of LCs
• Strong LC
– One boy, one girl
– Never make mistakes
– Confident
• Weak LC
– One boy, one girl
– Often make mistakes, especially at beginning
– More hesitant
Human student’s interface
Experimental method
•
•
•
•
•
•
•
•
Students can switch LC at any time
13 year old
Learning how to solve equations
ONLY 14 SUBJECTS!
Split on IQ
Also split on introversion vs. extroversion
6 sessions of 30 minutes
Pretest & posttest
Prefer weak LC at beginning and strong
LC at end.
• Weak prefer weak
• Strong prefer strong
Achievement tests
• All students learned
• No conditions, so no comparison
Expert vs. Motivator vs. Mentor
• Baylor, A. L., & Kim, Y. (2005). Simulating
instructional roles through pedagogical
agents. International Journal of Artificial
Intelligence and Education, 15, 95-115.
Types of LCs
• Expert – knowledgeable
• Motivator – supportive
• Mentor – both knowledgeable & supportive
Implementation
• Animated facial, head & hand gestures
– Expert looked like professor
– Motivator & mentor looked like college students
• More animated gestures
• Voice
– Expert was monotone, authoritative, formal
– Motivator was enthusiastic, energetic, colloquial
– Mentor was in between expert & motivator
Results: Self-reporting
• Facilitating learning
– Expert best, but only in longer study
• Credibility
– Motivator < Mentor < Expert
• Human-like
– Expert < Mentor < Motivator
• Engaging
– Expert < Mentor < Motivator
– But not in longer study
Outcomes
• Self-efficacy question “How confident are you
that you can write a lesson plan?”
– Expert < Motivator < Mentor
• Domain interest “what do you think about
instructional planning?”
– NS
• Designing a new lesson
– Motivator < Expert < Mentor
Is there an ATI?
• Kim, Y. (2007). Desirable characteristics of
learning companions. International Journal of
Artificial Intelligence and Education, 17, 371388.
• Do strong students prefer strong LC?
• Do weak students prefer weak LC?
• Any differences in learning?
Task domain
• Instructional design: Plan & implement supply &
demand lesson
Experiment design
• College students in instructional design class
• Two instructional factors
– Interaction control:
• LC provides info without being asked
• LC provides info only when asked
– LC competency
• Strong LC presents complete, accurate info with
confidence
• Weak LC presents incomplete but accurate info; more
tentative
Strong vs. weak LC
Results: Strong vs. weak LC
High GPA humans
Low GPA humans
Designing a new lesson
NS
NS
Recall of ideas from training
Strong LC
Weak LC
Which LC seems more valuable for learning?
Strong LC
NS
Which LC produces higher self-efficacy?
Strong LC
Weak LC
Results: LC vs. Human initiation of info
presentation
High GPA humans
Low GPA humans
Designing a new lesson
NS
NS
Recall of ideas from training
NS
NS
Which LC seems more valuable for learning?
LC control
Human control
Which LC produces higher self-efficacy?
LC control
LC control
From your reading of Chou, Chan & Lin
(2003)
• What do these roles mean?
Modes/roles of human & learning
companion (LC) learners
•
•
•
•
•
•
•
•
•
•
Human edits LC’s knowledge (e.g., Betty’s Brain)
LC solves & human gives immediate feedback, hints
Human solves & LC gives immediate feedback, hints
Human & LC solve separately, then compare (competition)
On each step, human & LC negotiate who will do it & what will be
done (collaboration)
Human is reaching mastery & LC challenges them with strongly
asserted, but wrong opinions – trip them up
Human solves problem while delating simple stuff to the LC is
limited assistant
Teach with conventional teaching then see if agent has learned it
LC provides motivation only
Source of answers to questions & other content
Competence of LC
• Strong
– Assertive vs.
– May reject human’s advice
• Weak
– Just incomplete vs.
– Troublemaker: Sometimes gives bad suggestions
Personality of the LC
• Neutral, unemotional
• Authority on the subject
• Enthusiastic & empathetic