The Illusion of Learning: Toward Growth-Centered AI — David Barron, Virginia Tech

"The Illusion of Learning: Toward Growth-Centered AI" This paper argues that educational systems in the age of AI must become behaviorally aware: able to see not just what learners say, but what they do. Struggle is the forcing function of learning, and human tutors calibrate their interventions on observable student effort. However, Large Language Models are fundamentally blind to the behaviors that signal struggle, providing the same response to a student who has tried everything and one who has tried nothing. Prior work shows learners who engage with educational content primarily through AI interaction not only learn less but falsely self-perceive their understanding as higher than it is. In turn, we are creating a generation of learners who do not know they are not learning. To combat this, we must create systems that reward students for what they did, not what they said. To demonstrate the feasibility of behavioral awareness, we deployed an AI-assisted programming environment across eight introductory courses, capturing 882,467 behavioral events from 664 students. We propose a four-layer framework for behavioral awareness and outline a roadmap toward growth-centered AI that responds not just to what learners explicitly say, but to what they implicitly need.