AI ACES GRADES—BUT STUDENTS MAY NOT LEARN

An experiment at Bocconi University tested GPT-4o's impact on student performance across 1,053 participants working on marketing assignments. The AI tool improved grades by approximately 0.9 points on a five-point scale—a significant jump that would typically indicate stronger mastery. However, the study's critical limitation reveals a troubling gap: researchers did not assess whether students gained genuine understanding or developed skills independently. The grade improvement alone cannot confirm learning occurred. This distinction matters because other research suggests that relying on AI to complete academic work without independent thinking carries long-term costs. Students may earn better marks in the short term while failing to build foundational knowledge, critical thinking, or problem-solving abilities they'll need beyond the classroom. The finding highlights a paradox in AI-assisted education: the skills most valued by grading systems—polished writing, structured arguments, comprehensive coverage—are precisely the outputs AI generates most convincingly. A well-organized essay or marketing proposal can satisfy grading rubrics without reflecting the student's understanding or analytical depth. Educators face a practical challenge. Traditional assessments reward the deliverables AI produces best, making it difficult to distinguish between student competence and AI capability. Schools using conventional grading methods may inadvertently incentivize AI use while undermining actual skill development. The Bocconi study suggests the education sector needs to rethink how it measures learning. Assessments focused on process—showing work, explaining reasoning, defending choices—may better capture genuine understanding than those emphasizing final output quality. As AI tools become standard in academic and professional settings, the question shifts from whether students should use them to how education systems can ensure AI augments learning rather than replacing it.