Computational Models of Learning: Deepening Care and Carefulness in AI in Education

Abstract

The field of Artificial Intelligence in Education (AIED) cares by supporting the needs of learners with technology, and does so carefully by leveraging a broad set of methodologies to understand learners and instruction. Recent trends in AIED do not always live up to these values, for instance, projects that simply fit data-driven models without quantifying their real world impact. This work discusses opportunities to deepen careful and caring AIED research by developing theories of instructional design using computational models of learning. A narrow set of advances have furthered this effort with simulations of inductive and abductive learning that explain how knowledge can be acquired from experience, initially produce mistakes, and become refined to mastery. In addition to being theoretically grounded, explainable, and empirically aligned with patterns in human data, these systems show practical interactive task learning capabilities that can be leveraged in tools that interactively learn from natural tutoring interactions. These efforts present a dramatically different perspective on machine-learning in AIED than the current trends of data-driven prediction.