Measuring and Transforming the Believability of Embodied Agents Jeremy Bailenson Department of Communication Stanford University Virtual Human Interaction Lab
Overview Metrics of Copresence (Believability): Theories, terms, empirical examples Quick Revisit to the Uncanny Valley Strategic Transformations of Agents: –“Augmented Social Interaction”
Copresence What is it? –Is that virtual human perceived as if it were a physical human? Is it measurable? Synonyms: Social Presence, Believability, Engagement, Rapport, Interactional Synchrony
Current Methodology: Questionnaires Questionnaires 95 percent of copresence research Problems with Questionnaires: –Ambiguity: "How much did it seem as if you and the other people both left the places where you were and went to a new place?“ -Demand Characteristics -Implicit/subconscious
How do you measure copresence? Questionnaires (95%) cheap and easy Open ended interviews Physiologically (brain activity, heart arousal, skin conductance) Memory for objects in VR (compared to physical space) Presence as absence (in VR does the physical world disappear) Behavior (do in VR as people do in physical space: nonverbal behaviors, learning, physical reactions, etc.) Today: Proxemic behavior, Eye Gaze, Disclosure, Turing Tests, Learning
Social Presence and Proxemics
Proxemics Background Studied extensively in psychology and anthropology since 1950’s Equilibrium Theory (Argyle & Dean, 1969) –NVB’s trade off –Predictions in regards to proxemics Do proxemics patterns hold true with virtual people?
Sample Proxemics Task
Sample Data
Equilibrium: Personal Space and Gaze
In the “Real World”: Second Life
Personal Disclosure (verbal and nonverbal) Show movie
Eye Gaze as Copresence Proxy
TSI: Detection: Nonverbal Turing Test
Learning
Uncanny Valley Revisited StaticLowMediumhigh BlockN >20 BearN >20 HumanN >20 Head Movement Realism Measures: Subjective Ratings Gaze/Head Movements Memory Proxemics
Transforming Agents to be Effective (Believable?)
Collaborative Virtual Environments
Transformed Social Interaction (TSI) Actual Behavior Strategic Filter Transformed Behavior
3 Dimensions of TSI Transforming Self Representation Transforming Social-Sensory Abilities Transforming Social Context
TSI: Augmented Gaze Gaze is powerful: –Learning (Sherwood, 1987) –Persuasion (Morton, 1980) –Physiological Arousal (Wellens, 1987) –Shaping the structure of a conversation (Kendon, 1987; Argyle, 1988)
TSI: Augmented Gaze
TSI: Digital Chameleon
Persuasive passage read by Agent 60 subjects –Mimic (4s lag) –Recording of other subject
TSI: Digital Chameleon
TSI: Facial Identity Capture
TSI: Identity Capture Similarity among people results in: –Attraction (Shanteau & Nagy, 1979) –More Persuasion (Chaiken, 1979) –More purchases (Brock, 1965) –More altruistic helping behavior (Dovidio, 1984) –Trust (DeBruine, 2002)
Facial Identity Capture: High Info, Familiar Target National random sample (N = 200) 1 Week before presidential election Viewed candidate photos while evaluating Bush and Kerry 3 groups of subjects –No morph –Bush Self, Other Kerry –Kerry Self, Other Bush
TSI: Facial Identity Capture
The Virtual Mirror
The Proteus Effect
Learning: Augmented Social Perception T
Learning: Transformed Proximity
Learning: Virtual Knockout
Ethics
Collaborators Megan Miller Andrew Orin Nicole Lundblad Julia Hu Claire Carlson Aaron Sullivan Boyko Kakaradov Hassan Adubu Stanford Graduate Students/ Post Docs Nick Yee Dan Merget Manos Pontikakis Kayur Patel Robby Ratan Hunter Gehlbach Stanford Faculty Shanto Iyengar Cliff Nass Roy Pea Byron Reeves Dan Schwartz UCSB Faculty/ Post Docs Andy Beall Jim Blascovich Jack Loomis Matthew Turk Rosanna Guadagno Thank you! Virtual Human Interaction Lab Josh Ainslie Adrian De La Mora Jon Shih Jaireh Tecarro Sam Warburg Kathryn Rickertsen Jerry Yu Stanford Undergraduates Berkeley Faculty Ruzena Bajcsy Jaron Lanier
Applications Learning Communication Technology Advertising Politics
Face to Face TSI?