Ray, Guest: current 'AI literacy' is literacy-laundering

A preprint posted on Zenodo today by Ishani Ray and Olivia Guest picks a fight with the category everyone in the AI-education space has been quietly building around, 'AI literacy' itself. Their charge is that the current discourse on large language models has bifurcated into two camps, one waving the stochastic parrot line to puncture hype, the other treating each capability jump as grounds to soften the critique, and that both are answering the wrong question. As the authors put it, 'the issue is not whether or not LLMs work, but what kind of working is happening and at whose cost.'

The reframe they propose is mechanistic critical AI literacy, and the target it names is literacy-laundering, their term for AI literacy programs that end up embodying the confusion they claim to address. They organize the argument along three dimensions: cognitive, where statistical pattern-matching displaces iterative thinking; pedagogical, where the current 'AI literacy' label is treated as a symptom of confusion rather than a cure for it; and political, where framing LLMs as infrastructure naturalizes labor displacement. The stochastic parrot is worth keeping, they argue, but 'deployed with mechanistic precision rather than mere rhetorical convenience,' because used well it specifies what is forfeited when cognitive labor gets delegated and to whom the cost lands.

If you are funding AI literacy work, in a school, an HR function, or a policy shop, this preprint is a shot across the bow. It says the compliance-shaped modules that get rolled into onboarding are, at best, tool tutorials, and at worst they teach students and workers to accept a settlement they weren't party to. The authors want the starting point for meaningful literacy to be, in their words, 'the epistemology of those most harmed.'

The honest caveat is that this is a fresh July 2026 preprint by Ray (Weill Cornell Medicine) and Guest (Radboud University Nijmegen), and what I can ground here is the argument's shape, not its full evidentiary base. What the abstract does not answer is which existing programs the authors have in view, or what a curriculum designer actually adopts on Monday morning. The framing is the useful part. The fight is no longer over whether LLMs are 'real intelligence,' it is over who is being trained to accept whose costs, and that is a far more tractable conversation.