AI Tutors Struggle with Discretion

TutorMoments, a framework for evaluating AI in education, shows that current AI tutors tend to over-help students, failing to match human-level tutoring judgment. A new dataset and evaluation tools have been released.

Understanding when to assist learners or let them struggle is crucial in education. AI tutors need to adapt better to improve student learning outcomes.

Educational professionals, AI developers, and students who rely on AI-driven tutoring systems are key stakeholders impacted by these findings.

While the dataset shows models currently over-support students, uncertainty remains about how improvements can align AI with human judgment.