Higher Marks, Lower Learning? AI's Favourite Magic Trick

Higher Marks, Lower Learning? AI's Favourite Magic Trick
This week, Ray and Dan unpack a bumper collection of AI in education news and research, all pointing to one critical question: Is AI helping students learn…or just helping them score higher?
The episode begins with a striking case from Brown University, where a take-home exam averaged 96%, only for the same cohort to average 48% in a supervised final. From there, Ray and Dan explore what the latest research says about AI-assisted assessment, automated marking, AI tutors, critical thinking, and the growing importance of student agency.
They also discuss:
• New evidence on AI marking compared with human examiners
• Research showing how guided AI tutoring can improve maths learning
• Why critical thinking is the biggest predictor of successful AI use
• The risks of "never-skilling" when students rely too heavily on AI
• AI detectors: what they can (and can't) reliably tell us
• New Australian data on AI's impact on employment
• The latest Digital Education Council survey of university students
• Online safety and the growing challenge of AI-assisted image abuse
As always, the episode is packed with practical insights, research summaries, and plenty of discussion about what all this means for educators, schools and universities.
Links - News
• Voice got an upgrade openai.com/index/introducing-gpt-live
• Inside Higher Ed article: Brown Professor Suspects Majority of His Class Used AI to Cheat insidehighered.com/news/faculty/learning-asses…
• Digital Education Council AI in Higher Education Global Survey digitaleducationcouncil.com/resource-library-i…
• One in 25 teens affected by AI-assisted online abuse, study finds. Read Lee Barrett's opinion: linkedin.com/posts/lee-barrett-a7624923b_boys-…
• AI and employment in Australia - report from Department of Education and Workplace Relations dewr.gov.au/download/17666/ai-and-employment-a…

Links - Research
• AI in University Assessment: Evaluating the Opportunities and Risks of Automated Marking emotional-cognition.psychol.cam.ac.uk/sites/de…
• LLM Performance on a Real, Double-Marked GCSE Benchmark arxiv.org/abs/2606.24973
• Teaching with Gemini: Measuring the impact of Guided Learning on student mathematics progress in Sierra Leone deepmind.google/blog/measuring-the-impact-of-l… storage.googleapis.com/deepmind-media/LearnLM/…
• Access is Not Enough: Human Support Improves Engagement with AI Tutoring nssa.stanford.edu/studies/access-not-enough-hu…
• The Generative AI Learning Penalty: Evidence from Chinese Secondary Education papers.ssrn.com/sol3/papers.cfm?abstract_id=68…
• The agency gap: perceived human AI agency, reflection and generative AI learning across UK and China based higher education contexts tandfonline.com/doi/full/10.1080/03075079.2026…
• How Do Autonomy, Critical Thinking, and Artificial Intelligence Value Perception Affect University Student Self-regulated Learning? link.springer.com/article/10.1007/s11528-026-0…
• The cognitive impact of ChatGPT in higher education: A systematic review of critical and creative thinking outcomes sciencedirect.com/science/article/pii/S2666920…
• AI-induced never-skilling in medical education doi.org/10.1038/s41591-026-04438-y (Paywalled)
• Who wrote this? Evaluating the reliability of AI detection tools in higher education link.springer.com/article/10.1007/s40979-026-0…
.