"It's To, Not Too!": The Impact of Robot Errors on Children's Learning in a Learning-By-Teaching Paradigm

Abstract

Access to quality education is widely acknowledged as a key factor to address the problems we face globally as a society. Teacher resources, however, continue to dwindle due to lack of funding and increasing demands for more personalized education, making it difficult to find the time to help each student succeed. As such, the introduction of educational tools, such as social robots for extra 1-on-1 help, has been shown to have comparable effects to being taught by human tutors, which may help alleviate the burden on teachers. In this thesis, we present an experiment investigating the design and utilization of strategic robot errors in a learning-by-teaching scenario to improve children’s reading ability. The experiment tested three different conditions to help differentiate the best strategy for developing robot errors: targeted mistakes designed to engage the zone of proximal development, the challenge level of problem-solving for optimal learning, simple mistakes which are easy and obvious to identify requiring little effort on the part of the tutor, and no mistakes which acted as a baseline. While we did not find significant results regarding the cognitive learning efficacy of different mistakes, post-hoc analysis was performed indicating that certain mistakes impact affective characteristics that contribute to learning such as attention and self-efficacy. The implications of this research for robot design, research and implementation and broader applications in society are discussed, as the use of mistakes to influence affective learning outcomes may be effective at overcoming other known shortfalls of the technology. In addition, recommendations in regard to improving experimental methodology for future studies using robot tutees and future research directions for robot error design are explored.