Ed & Tech must-reads 030226

AI detectors, suing your uni and "playing" The Scream

screenshot of Munch's The Scream with a man as playable character

Heads we win, tails you lose: AI detectors in education from Journal of Higher Education Policy and Management

By this stage, I doubt there are many must-readers who feel strongly that AI detectors are a good thing (though I do appreciate the simplicity that they would bring if they worked) but I thought I’d share this article anyway because it is comprehensive and recent. A cavalcade of academic integrity notables from 7 educational institutions on the east coast of Australia come together to methodically explain how so-called AI detectors work, problematic elements of them (unverifiable probabilistic estimates, mutually exclusive linguistic markers), evidence used in GenAI academic integrity investigations, and issues with security and fairness and false positives. One small quibble I have is whether this should be classified as a research article, given that it has no methods and doesn’t collect data. I guess it is a decent lit review, even if that is not overtly stated.

I must say, I wouldn’t run a university these days if you paid me. A lot. In a nutshell, a nursing student asked a question on Facebook about an exam question after taking an exam - a few days before passing a supplemental exam after failing the first time with a 78%. The student says that her institution, the Australian Catholic University, has failed her, withdrawn credit for a 4 week placement and made her no longer eligible for the government’s HELP-HECS loan scheme. The ACU claims that students were made aware not to discuss exam questions as they can be reused and that by doing so, she may have given other students an unfair advantage. The student denies knowledge of this and is suing to have the university declare that she didn’t cheat. It’s a tricky one - students not reading rules is pretty commonplace and isn’t a strong defence for a breach, but the policy and its application overall do seem on the harsh side. This is the same university that was in the news last year for using AI detectors to accuse 6000 students of academic misconduct.

Generally I’ve grown a little weary/wary of many of the art-creation aspects of GenAI but I can’t deny that the capacity of Google’s Genie 3 open world generator to build navigable 3D spaces is impressive. Ethan Mollick shares some of his experiments with paintings by de Chirico, Munch, and Turner in these Bluesky ‘skeets’.

Reply

or to participate.