Main Article Content
Abstract
Artificial intelligence (AI) tutors and chatbots can now give students instant feedback on every step of an algebra problem. This capability promises personalised support at scale, but it also creates a dilemma. Research in mathematics education shows that productive struggle, sustained effort on problems just beyond a learner's current competence, is a key driver of conceptual understanding. Feedback that arrives too quickly, too often or too explicitly may remove the struggle that makes learning durable, while feedback that arrives too late may leave students frustrated and stuck. This theoretical paper examines the impact of real-time AI feedback on students' productive struggle and conceptual understanding in algebra. Drawing on Vygotsky's zone of proximal development, the assistance dilemma, desirable difficulties and productive failure, it reviews evidence from intelligent tutoring systems, feedback research and recent studies of generative AI in mathematics. The review indicates that the impact of real-time AI feedback depends less on its presence than on its design: process-oriented, graduated and metacognitive feedback tends to preserve productive struggle and build understanding, while immediate answer-giving tends to encourage dependence, gaming behaviour and procedural rather than conceptual learning. The paper proposes a framework for calibrated AI feedback in algebra built on five principles: wait before helping, escalate hints gradually, target reasoning rather than answers, detect unproductive struggle, and prompt reflection. Implicat