3-Read Friday #110
Responsible AI, LLMs Mimicking Students, and Screen Time in Schools
This week’s 3-Read Friday features selections from two of my Eedi colleagues: Simon Woodhead, our Chief Data Scientist, and Bibi Groot, our Chief Impact Officer. Oh, and one from me.
1. Responsible Artificial Intelligence for Test Equity and Quality: The Duolingo English Test as a Case Study by Jill Burstein et al
From Simon: Eedi has always been dedicated to novel forms of assessment that elicit students' misconceptions. We started with diagnostic MCQs, collected students' explanations and confidence levels, and are now collecting their dialogues with a tutor. Normally you would turn to psychometrics to map these student behaviours to cognitive insights, but traditional psychometric methods can only be applied to simple assessment types. In this article, Duolingo uncovers its novel approach to reliable, validated assessments outside the traditional setting.
2. Large Language Models as Students Who Think Aloud: Overly Coherent, Verbose, and Confident by Conrad Borchers, Jill-Jênn Vie, and Roger Azevedo
From Bibi: A paper I found eye-opening: Borchers, Vie and Azevedo asked GPT-4.1 to simulate novice learners thinking aloud, and the LLM simulations of "learner utterances" were much too long, confident, and coherent (in comparison to think-aloud data from real students). LLMs are trained on expert-authored text; they don't naturally grasp messy, emotional, half-formed reasoning. For anyone building AI tutoring, this matters: if the model can't faithfully represent how a struggling student thinks, it might fade scaffolding too prematurely. Super interesting design problem.
3. A screen isn’t human. Neither is a book by Zach Groshell
Zach tackles the divisive topic of screen times in schools with his trademark nuance. He argues that banning screens wholesale dodges the harder question of what students are actually doing on them, and risks discarding scalable, well-designed instructional tools because activity-based teaching has misused them.
Have a great weekend!
Craig
🏃🏻♂️Before you go, have you… 🏃🏻♂️
… checked out my brand-new book series: The Tips for Teachers guides to…
And checked out my all-new, ad-free, Mr Barton Maths website, including my new Topics page.
The LLMs mimicking students piece is one teachers need to sit with seriously. If a model can produce output that looks like student thinking, what exactly are we assessing? That question alone is forcing a rethink of what assignments are actually measuring. And that rethink is long overdue and healthy.