Five Australian universities let AI mark student work

An unlikely sentinel

Armin Alimardani has blue hair, a smartwatch on his wrist, and calls himself "a tech enthusiast." He teaches law and technology at Western Sydney University, has been experimenting with AI for years, and has worked as a consultant for OpenAI. The Guardian, which published his warning on Sunday, calls him an unlikely sentinel.

He was speaking from Palermo, where he was attending a series of university workshops on the technology, when he explained what worries him. His colleagues have lately been asking whether AI could lighten the marking load. He finds the conversation "extremely concerning."

His exact words: "We've got to be really careful where we are stepping here. Because I think there's always a risk, at the moment, for higher education to collapse." Coming from a reflexive critic, the line would slide right past you. Coming from someone who builds AI assessment tools himself, it sticks.

The workload comes first

Before laying out the problem, it's worth laying out the reason for it. In Australian universities as everywhere else, marking eats up an enormous share of academics' time, and Alimardani says it himself: plenty of staff are overworked to "breaking point." That's not background color, it's the engine of the whole story.

Some of those markers are sessional academics for whom marking is their main source of income, in an already precarious line of work. When a technology promises to cut a task nobody can finish down to a fraction of the time, it doesn't get forced on anyone. It arrives as relief.

Alimardani goes further: AI could even improve marking quality, acting as "a second pair of eyes." He's working to build models that can catch inconsistencies in human assessment. His objection isn't to the tool. It's to what happens once you roll it out at scale.

What the policies actually say

Five institutions currently let their staff use AI to mark assignments and write feedback: Western Sydney University, the University of Newcastle in New South Wales, Deakin and RMIT in Victoria, and the University of Adelaide. The Guardian describes it as a qualified green light, and the conditions aren't the same everywhere.

Deakin is the most explicit: staff can use AI to "improve efficiency and support assessment and administrative activities," but not to "assign grades." Western Sydney claims "a human-centred approach," in which "the awarding of marks, grades and feedback must also always be the responsibility of academic staff." Newcastle, for its part, gives students the right to opt out of AI involvement.

For RMIT and Adelaide, the Guardian only notes "similar" allowances to Western Sydney's, without spelling out the detail. And three of the country's highest-ranked universities, the University of New South Wales, Melbourne and Sydney, draw the line at AI in marking, even as they enthusiastically welcome it elsewhere in the classroom.

So the short version that's about to circulate needs correcting. Where the detail is public, no policy hands the grade to the machine: at both Deakin and Western Sydney, a human stays at the end of the chain. That's the safeguard.

The verification drift

That's when Alimardani turns his objection into an image. Take a hundred academics handed an AI marking tool, told to verify every assessment. "Let's say 80 do exactly as advised, there are going to be 20 who do not."

He calls it "verification drift." The mechanism is disarmingly ordinary: after a few AI-generated assessments come back looking fine, the human in charge of oversight, often overworked themselves, stops being quite so vigilant. And the machine's hallucinations pass for facts.

It's the same reflex as a bank teller who, after the hundredth matching bundle of notes, stops actually looking. The safeguard isn't removed by any single decision. It wears down through repetition, without anyone signing off on it.

Alimardani points to scandals of exactly this kind that have already played out in courtrooms, where AI-drafted filings have reached judges with invented case citations. In university assessment, he calls them almost inevitable, with two knock-on effects: institutional reputations, and a sense of unfairness among students.

Readers of yesterday's story will recognize this

We published a piece yesterday about people who stop saying "I don't know" the moment an AI is sitting next to them. What's happening in Australia is the institutional version of that: an individual who stopped doubting becomes an organization that stops checking, and writes it into policy.

The comparison also clarifies the opposite choice. When the French-language Wikipedia community banned AI-written text in September, it wasn't because editors judged the writing to be bad. It was because they knew they couldn't reliably detect it: a rule you can't enforce protects nothing. Australian universities picked the other option, allow it and rely on oversight.

The critics' phrase, and what it doesn't measure

Some critics have coined a phrase for this: the "slop-cycle." Students submit AI-generated assignments, which academics then mark using AI. The same camp worries about seeing the work of campus academics replaced by algorithms in datacentres.

The phrase is catchy, and it should be credited for what it is: a line from unnamed critics, not a measured finding. No real usage figures appear anywhere in the reporting. Nobody knows how many academics are actually using AI to mark, or on what share of assignments.

What is documented is the contradiction staff themselves are raising. "We are telling students all the time that there are integrity issues with their use of AI in assessment," one academic said, speaking anonymously. "What message does it send if we then say: but we are going to use it to mark you?"

What it costs, from the student's seat

Alimardani's strongest argument isn't a moral one, it's an economic one. "Students are already financially struggling," he says. "If they are not receiving proper feedback, if they feel like their degree is not as meaningful for employers and if they think that having a university degree doesn't help them that much with guaranteeing a job, why would they come to university, really?"

Alex Cameron, a creative and professional writing student at Queensland University of Technology and an editor at the campus magazine, doesn't hesitate. His three years have been "awesome," he says, surrounded by interesting people, "the whole point of universities." He wouldn't pay $1000, he says, for a degree where an AI does the marking.

"At that point, we might as well just have AI everything," he says. "Just get our degrees through AI and have AI write our assignments." And how would he feel about it? "I would be irate. I would be absolutely bloody incensed. We like to do human writing, we like to have humans look at it, and we like to get feedback from humans."