Universities Must Measure Learning With and Without AI Assistance
Assessing student performance requires evaluating both AI-assisted capabilities and independent cognitive mastery for accurate results.
1The Passive AI Risk
Marc Schmitt highlights that the primary danger of artificial intelligence in education is passive usage. Universities must shift their pedagogical focus to teach students how to actively spar with AI models. Educators should prioritize assessing how effectively students can challenge and evaluate these models rather than merely accepting generated outputs as truth.
2The Performance Divergence Gap
Measuring student performance solely during AI-assisted sessions creates a misleading picture of actual competence. Research indicates that performance levels can diverge significantly once the AI partner is removed. Evaluating students only while they have access to digital tutors fails to capture their true, independent mastery of the subject matter being studied.
3Evidence from Mathematics Trials
A randomized trial involving secondary-school mathematics students demonstrated this discrepancy clearly. While unrestricted chatbot access significantly raised practice scores during sessions, it simultaneously lowered performance on subsequent exams taken without AI assistance. This evidence proves that high scores during assisted practice do not necessarily translate into genuine, long-term knowledge retention for students.
4Redefining Success Metrics
If an evaluation process stops at the assisted session, it incorrectly labels a tutor as a success. Institutions must implement a dual-measurement strategy that captures both assisted and unassisted performance. This ensures that the educational system validates the student's ability to function independently, rather than just their ability to prompt a model.
5Pedagogical Shift Requirements
Universities need to integrate rigorous testing protocols that account for the presence of AI tools. By requiring students to demonstrate skills without digital aid, educators can distinguish between tool-assisted output and internal cognitive development. This structural change is essential to maintain academic standards in an era where AI is increasingly ubiquitous.
6The Sparring Partner Model
Viewing AI as a sparring partner in the ring provides a useful framework for modern education. Students must learn to engage with these models critically, treating them as collaborators rather than sources of final answers. This approach fosters deeper analytical thinking and prepares students for professional environments where AI interaction is standard.
7Long-term Knowledge Retention
The ultimate goal of higher education remains the cultivation of independent expertise. Relying on AI for practice can create a false sense of security regarding one's own knowledge. Universities must ensure that students develop the foundational skills necessary to perform effectively even when advanced technological support systems are completely unavailable to them.
8Future Assessment Strategies
Moving forward, academic institutions must adopt comprehensive assessment frameworks that measure learning in both assisted and unassisted environments. This balanced approach protects the integrity of degrees and ensures graduates possess the core competencies required in their fields. Without these dual metrics, the true impact of AI on human learning remains dangerously obscured.
Original article: Nature — Universities Must Measure Learning With and Without AI Assistance