Demonstrating Actions

An important part of our project is testing the degree to which concepts about agents affect how people behave when interacting with them, and how these behavioral variations affect how people learn from these interactions. This project, led by Vanderbilt graduate student Jonathan Herberg, is based on research demonstrating that people often learn best when they must teach the new procedure to someone else. So, in this experiment our goal was to test the degree to which teaching a person might induce better learning than teaching a computer. We initially predicted better learning in the human-audience condition on the assumption that demonstrating actions to a human audience would lead to better learning because the human demonstration would induce deeper goal-focused reasoning about the task. To test this hypothesis, we asked participants to learn how to solve a simple three-ring version of the Towers of Hanoi task. Then, they demonstrated the solution to a computer or a human “audience” (again, just a picture). Finally, the audience was removed, and participants were simply asked to solve a more difficult 4-ring version of the task. To everyone’s surprise, in two separate experiments we found that participants were significantly worse at solving the generalization task after showing the more basic task to a human audience. In a series of analyses, we found that the production of manual social cues (e.g. pointing) moderated the effect. So, as a follow-up, we ran an experiment eliminating the alternative hypothesis that simple spatially-oriented movements were sufficient to interfere with problem solving by asking subjects in a nondemonstrating condition to simply point to a target with similar frequency to the previous experiment. We observed that these simple points were not sufficient to lessen generalization, suggesting that the interference came from a specifically social set of gestures.

This research suggests that in some cases, at least, learning  by teaching may be improved when the audience is NOT human! Future research will explore the generality of this finding. For example, it is possible that computer audiences are better only when tasks do not have a social component, or only for manual tasks for which gesturing is likely. In any case, though, this research makes clear the importance of carefully considering the relative costs and benefits of having a human audience in learning-by-teaching settings.

References

Herberg, J.S., Levin, D.T., & Saylor, M, M. (2012). Social audiences can disrupt learning by teaching. Journal of Experimental Social Psychology, 48(1), 213-219.

See also:

Herberg, J. S., Saylor, M. M., Ratanaswasd, P., Levin, D. T., Levin, D. T., & Wilkes, D. M. (2008). Audience-contingent variation in action demonstrations for humans and computers. Cognitive Science, 32, 1-19.