Relational AI: Creating long-term interpersonal interaction, rapport, and relationships with social robots

Committee:
Dr. Cynthia Breazeal, Associate Professor of Media Arts and Sciences, MIT Media Lab
Dr. Paul Harris, Professor of Education, Harvard Graduate School of Education
Dr. Rosalind Picard, Professor of Media Arts and Sciences, MIT Media Lab

Abstract
Literacy, language, and interpersonal skills are some of the most important skills any child will learn, as they can greatly impact children's later educational and life success. However, many children do not receive sufficient support, instruction, or practice in developing these crucial skills. Many of the proposed technological interventions that address this need engage children passively and target older children, despite the fact that many early language and interpersonal interventions are more effective when they target young children aged 3-6 years using active, dialogic, social methods. 

In this thesis, I examine the use of social robots as a technology to support preschool children in learning early literacy and language skills. I hypothesize that a key aspect of why social robots can benefit children's learning is their nature as a relational technology—that is, a technology that can build long-term, social-emotional relationships with users. Thus, through a series of empirical child-robot interaction studies, this thesis first establishes the role of social robots as relational technologies. I demonstrate their capabilities as learning companions for young children that afford opportunities for social engagement and reciprocal interaction, particularly peer-to-peer modeling. I discuss how we can understand children's conceptualizations of social robots as relational agents and measure children's relationships over time. 

Second, I introduce the term relational AI to refer to autonomous relational technologies that change through time. I develop a computational relational AI system to examine how using relational AI in a social robot can impact longitudinal child-robot learning interactions. Through testing the system in a longitudinal study, I explore connections between children's relationship and rapport with the robot and their engagement and learning. I show that relational AI is a new, powerful educational tool, unlike any other existing technology, that we can leverage to support children's early education and development.