Ed/Tech must-reads 120825

AI industry horrified to face largest copyright class action ever certified from Ars Technica

This one goes a little deep into the legal weeds and is full of ifs and coulds but it is nonetheless worth paying attention to for the sheer scope of the possible consequences of stealing/training LLMs on other peoples’ intellectual property. Anthropic (Claude) is currently asking an appeals court to refuse to hear a class action of over 7 million claimants. (And at $150,000 per claimant if the case is successful, that’s more than $1T - nearly enough to buy a house in Melbourne, Australia). The article goes into more serious examination of the claims and merits of the case - does every claimant need to prove ownership of their work? - and highlights that money is probably once again going to be the ultimate arbiter.

Delving Into the Psychology of Machines: Exploring the Structure of Self-Regulated Learning via LLM-Generated Survey Responses from Computers in Human Behavior

The title of this article belies the seriousness of the ideas being explored - namely whether any GenAI LLMs have the capacity to “reliably simulate survey responses at scale, they could be used to test interventions, refine theoretical models, augment sparse datasets, and represent hard-to-reach populations.” It is the last of these that strikes me as the worst - if we consider that most GenAI LLMs effectively reproduce the most statistically probably responses based on what they have been trained on, how can we assume there is any worthwhile data about ‘hard-to-reach’ people in the system. The authors acknowledge this issue - but decided to crack on anyway.

Interactive Oral Assessments: Charting the Course While Holding the Line from ASCILITE TELall blog

Meanwhile, day to day educators seeking to mitigate the impact of students outsourcing their assessments to Clippy continue to return to (not unreasonable) teaching and assessment approaches of days gone by. Danielle Logan-Fleming (Torrens Uni) shares insights from recent work on the Horizons report exploring the rise of Interactive Oral Assessments. This post provides some solid detail about how and why this is being rolled out and how educators and institutions are addressing some of the perennial issues like managing them at scale. She also shares links to a rich collection of institutional resources for further reading.

Vendor Lock-in and Google Gemini from Ben Williamson (LinkedIn)

The challenges of vendors keen to preserve their commercial relationships with institutions by any means necessary have long been joked about in the operational side of the ed tech world (see the cover image of this post). Ben Williamson (Uni Edinburgh) spells it out clearly - via a discussion of a 2024 paper from Cone & Lai about the different ways they found that Google tends to lock institutions into their tech ecosystems. These range from the data used and captured to regulations, discourse and time. To this, he adds a pedagogical lock-in, where support resources about teaching using Google’s Gemini AI platform informed by learning science invariably seem to end with sticking to the product.