| Notes |
PEDAGOGICAL AGENTS
PANEL
Bill Swartout
USC Institute for Creative Technologies
BACKGROUND
• BASIC Instructional Program (BIP) with Avron Barr
• PI for Ada and Grace, Coach Mike (with Chad Lane, David Traum,
Diane Piepol & Jacki Morie)
PERSONAL ASSISTANT FOR LIFE
LONG LEARNING (PAL3)
• Project Goal: Create a personal agent that can mentor a sailor (ultimately)
throughout his career
• Uses Learning Record to understand context
• What sailor knows
• Where he is going and what will be expected
• ONR funded
• Collaboration with Memphis (Art Graesser) and Kurt Van Lehn (ASU)
WHAT’S NOVEL?
• Life Long Learning Model for transition between learning environments
• Very few current systems use any persistent model at all
• Of those that do, limited to single task domain over a semester (or so)
• Even fewer (none?) accompany the learner into the workforce
• Use of embodied, pedagogical agents on a persistent, extended basis
• Address issues in decay of knowledge and skills
• Largely uncharted waters for tutoring systems
THINKING ABOUT TODAY
IMPROVING PEDAGOGICAL
AGENTS
• Three encouraging trends
• Richard Mayer’s principles: methodical exploration of space of capabilities
• Noah Schroeder’s meta analysis
• Michelle Perry & James Lester: improvements suggested by studying how people
interact
CONTEXT
• Critical in understanding results of studies (but often implicit)
• Richard Mayer’s Image Principle vs. Embodiment Principle
• Failure to understand context can lead to over-generalization
• Often not dealing with monotonic space
• Example: Uncanny Valley
• Published papers could be improved to provide more context for meta
analysis
NEXT STEPS
• Wizard of Oz experiments
• Exploit studies of master teachers
• Related to Samantha’s talk: Don’t just think of pedagogical agents as
approximate substitutes, instead ask how they can do better than people:
example SimSensei
SIMSENSEI
• Virtual “Intake Nurse” for PTSD and Depression
• Computer vision
• Voice tracking
• Speech recognition
Rizzo, Morency, Bolas, Forbell, Gratch, Hartholt, Marsella, Traum
11
SimSensei video removed
SIMSENSEI: SUBJECT TESTING
• Non-verbal communication (gestures, head nods, etc) build rapport,
encourage conversation
• Because SimSensei is computer-based, subjects do not fear judgment
• Potentially creates a “sweet spot”
• How could this idea be exploited in learning systems to combat
• Gender bias
• Math phobia
• Poor self image
• Etc |