Daniel Schwartz, Professor, Stanford University

Over the past years, we have been developing a computer learning technology called a Teachable Agent. The work leverages the common wisdom that people "really" learn when they have to teach. After a few years of positive results, we decided to get a better look at the mechanisms that drive learning in social interaction. Most cognitively inspired theories of learning point to the useful questions and ideas that arise in social interaction. But, can we really reduce the learning benefits of social interaction to a question of information flow? In this talk, I will present "laboratory" style data that tightly controls the types of information that arise in social interaction. I will show that learning from social interaction cannot be wholly ascribed to the different types of information that arise. For example, in a wizard of oz study, people thought they were interacting with a person or a computer, and information exchange was held constant across conditions. People learned more when they thought it was a person; people were more aroused when they thought it was a person; and finally level of arousal was correlated with learning.