Winning the Race: Ensuring Generative AI does not Outpace the Science of Learning

Abstract

With 86% of students in 2024 using generative artificial intelligence for their studies, the effect of AI on learning has received increased attention in educational spheres and policy conversations.1 The immediate implications are profound for student learning; however, there is limited empirical evidence on AI’s impact on learning and how edtech companies are designing for AI risks. Thus, this paper aims to fill a gap in the literature. Through interviews with professionals from educational AI companies, this thesis examines how principles of learning science are understood and operationalized by designers. By understanding how developers account for learning science in the creation of educational tools, this thesis aims to evaluate whether current educational AI tools support or undermine learning, providing policy implications for regulating educational AI. Throughout history, prominent fundamentals of learning have emerged as essential to student development. Two specific principles, memory and motivation, are likely to see substantial disruption from generative AI, both positively and negatively. Given their critical role in learning outcomes, this thesis focuses on these two principles by outlining key frameworks and examining how educational AI tools recognize and operationalize them in practice. The principle of memory explores how children retain information, modulated by attention, working memory, cognitive load, and scaffolding practices. The principle of motivation explores how children stay engaged through intrinsic value, belonging, productive struggle, and self evaluation. Through these two lenses, this thesis analyzes educational AI tools, providing insights into pitfalls and benefits. Interviews were conducted with seven professionals from education technology companies. I hypothesize that AI professionals incorporate principles of learning science into design but provide limited evidence on the effectiveness of the tools they develop. Through comprehensive analysis, the results supported this hypothesis, revealing that edtech companies recognize the importance of designing AI tools using research-backed learning science, including principles of memory and motivation. Most participants demonstrated specific design choices grounded in learning science principles. However, companies do not yet have the data or learning outcomes to demonstrate the effectiveness of their AI tools. Given these results, my policy recommendations aim to hold companies accountable by monitoring product use and designing tools grounded in learning science. Further, digital literacy programs aim to inform students, teachers, and districts of the best practices regarding responsible AI use.