#learning

Since 2023, nearly every educational AI tutor has adopted the same design philosophy: withhold the direct answer and offer a guiding question in return. Khanmigo, ChatGPT's study mode and Gemini's Guided Learning all rely on this approach. The underlying pedagogy is sound, yet two recent trials suggest we have been focusing on the wrong thing entirely.

Smart glasses can already read a live transcript, generate an answer, and float it an inch above the wearer's own eyeline, invisible to anyone looking at their face. Nick McIntosh put it bluntly on LinkedIn: "smart glasses have quietly dissolved the one assessment format everyone thought was AI-proof" (McIntosh, 2026). The format institutions dug up specifically because it felt unhackable lasted, in practice, about as long as it took a wearables company to ship a product.

Earlier today I attended a webinar organised by the Community for Educational Innovation (a European Commission initiative) titled Educating to Thrive in the Digital World. One of the interventions was by Julian Estevez, Professor at the University of the Basque Country, who posed a deceptively simple question: Personalised education with AI — a myth? His presentation was brief introduction for a full argument, and it was the kind of thing that immediately starts pulling threads.

Conversations about educational technology often orbit around efficiency, but the rapid rise of generative AI has forced universities into a long-overdue reckoning with a much deeper question: what exactly are we doing when we provide feedback? If feedback is merely the transfer of corrective information, then large language models have already won. They can parse essays, spot logical flaws, and debug code with astonishing speed. However, reducing feedback to a glorified diagnostic tool misses the fundamental reality of how university students actually learn.

Three years in. That's where we are now with generative AI in higher education. ChatGPT's arrival in late 2022 feels like both yesterday and a lifetime ago. The initial panic ("How do we AI-proof assessment?") has given way to something more interesting, more nuanced, and dare I say it, more hopeful.

A constant theme in my recent conversations with teachers, faculty teams and educational developers is the challenge of Generative AI (GenAI) and assessment. There's a palpable sense of pressure in these meetings, an anxious search for a definitive "solution." It’s a feeling I’m sure many in higher education will recognise, as institutions everywhere scramble for policy.

There is a Japanese word, 'tsundoku,' for the habit of buying books and letting them pile up unread. My pile was digital: a long list of blogs and articles on teaching, learning, generative AI... The pile grew. So did the feeling that I was falling behind.

The discussion about AI in education often gets bogged down in practical questions: are you allowed to use it, and how do we then ensure ownership and reliability? While relevant, these questions stem from the outdated idea that technology is a neutral tool, separate from the learning process itself.