Conversational Agents and Learning Outcomes: An Experimental Investigation

Abstract: An experimental approach was used to compare two types of web interfaces (text-based vs. conversation-based) to content on the life and theories of Jean Piaget. The content in each interface was identical with the exception of third- vs. first-person references. Fifty-nine students in psychology first completed a pretest of Piagetian knowledge and then were randomly assigned to one of the two interfaces. After 20 minutes of review/conversation, students completed a 35-item exam designed to measure knowledge retention and a questionnaire to measure their perceptions of the assigned interface. Contrary to expectations, the text-based interface was rated significantly higher on measures of enjoyment and utility and led to better learning outcomes in comparison to the conversational agent. Altogether, the findings indicate that the use of conversational agents in distance education needs to be carefully matched to the learning goals and outcomes. The use of conversational agents in distance education falls under the broader category of pedagogical agents, or the design of computer software that is autonomous, interactive, anthropomorphized, and directed towards educational goals and outcomes The design of pedagogical agents is guided by a number of theoretical frameworks drawn from different disciplines. For example, the work of Graesser and colleagues on AutoTutor, a conversational intelligent tutor system (see Graesser, Wiemer-Hastings, Wiemer-Hastings, Kreuz, & Tutoring Research Group 1999), is base