MC-CPO: Mastery-Conditioned Constrained Policy Optimization for Pedagogically Safe Intelligent Tutoring Systems
Researchers have developed a new reinforcement learning framework called Mastery-Conditioned Constrained Policy Optimisation (MC-CPO) to improve intelligent tutoring systems. Analysis of over 21 million student interactions revealed that engagement signals often do not correlate with actual knowledge acquisition, occurring in 26.5% of interactions on one platform and 3.1% on another. MC-CPO addresses this by linking instructional actions to learner mastery, ensuring prerequisite knowledge is met before new concepts are introduced, thereby enhancing pedagogical safety and increasing mastery gains by up to 54.0% compared to existing methods. AI
IMPACT Enhances AI-driven educational tools by ensuring learning progress is tied to genuine mastery, not just engagement metrics.