Watching people doing things may become an essential, not a luxury

Even if you were to optimistically regard as exaggerated outliers the tales of untrained, or non-expert, or seemingly unfeasibly rapid markers whose feedback is thin, confusing or contradicts the marking rubric, the sector has hardly enjoyed a period of rude reputational health over the academic integrity of its awards this summer.

But as newspaper FOIs sweep the sector looking for evidence of students being accused of, or caught for, using generative AI to cheat in their assignments, that academic integrity issue facing UK universities may well be about to get much much worse.

I made a thing for you to look at

The basic problem boils down as follows. A significant and dominant method for assessing the extent to which a student has met their learning outcomes involves said student producing a digital asset – an essay, project or whatnot – that is then uploaded, asynchronous marked (and those marks then moderated) against a set of criteria.

This is already a process whose fairness consistently receives poor scores from students on the National Student Survey – where qualitative research tells us that students think work is marked inconsistently, don’t trust moderation and worry about the way in which their personal circumstances sometimes are or are not factored in in the process of both production and grading.

This summer at least, the anecdotal evidence is that at least the essay mills all seem to be fading from view (and students’ WhatsApp groups) as ways to cheat that system.

But once generative AI makes it almost impossible to judge whether that digital asset that has been supplied is the sole and original work of the submitter, the sector’s standard summer line – that students’ work must be their own – becomes a thumb up against a flood.

There have been plenty of hopium and copium pills to take – academic staff using the free version of Chat-GPT to prove to themselves that its use is easy to spot underestimate how sophisticated the tools have been getting since launch, and ignore how all but the most desperate or daft students are now using the myriad tools.

The reality is that many if not most of the innate academic skills that appear in a marking rubric can now be (and are being) satisfactorily automated, and many of the wider and emergent transferable skills that we might build into more creative forms of assessment can be automated too.

Don’t need roads

There are important and earnest debates about all of this that concern the future of education, the purpose of assessment, the nature of passing foundational knowledge on to new generations and the way in which the tsunami of change will impact the labour market and society at large.

Some are excited by generative AI, keen to embrace it and see the way it might free them and their students to focus on creativity and the application of knowledge. There are scary visions of the future too – often concerned with the way in which large language models and their image based equivalents will eat themselves on their own gathering of AI produced material, or the way in which they will cause humans to lose essential skills.

Some are having interesting conversations about the basic structure of the degree, and the way in which assessment might promote kinds of behaviours or how it might promote the application of rather than acquisition of knowledge.

But on the basis that higher education often defaults to tweaking last year a bit or dreaming dreams so far away as to be impractical, I figure that finding a way to look at next academic year will matter more.

Let’s imagine that an August of the press working out that nobody’s been caught for using Generative AI in their assignments all summer combines with some embedded tweets of students sharing the hamfisted feedback they’ve had on their dissertation.

These are journalists who, just like students the sector educates, will marvel at how their copy of Microsoft Word will suggest not just better spelling and grammar, but what to write next based on what they’ve previously written. And they’ll think back to their degree when doing so.

At some point hostile press coverage means ministers will wake up to the issue, and then like clockwork the regulators – especially the Office for Students – will wake up too. The pressure will be on to demonstrate that assessment policies and the practice used to detect cheating are robust.

An arms race involving spending more and more on AI to detect cheating when students are spending more and more on AI to produce cheating will be a seriously defeating zero-sum game, where the only winners will be swimming in sector cash in their silicon valley pools.

And while some muse on the future of mankind, that gives the rest of the sector four real options down at module level:

• Glide gently back to in-person, paper and pencil exams – which we now know to be hugely problematic both logistically and from an access perspective;
• Double down on online exams and timed assessments – where expensive proctoring continues to cause ethical, EDI and efficacy concerns and students get ever more confused about what counts as cheating;
• Have more assessments that involve watching people doing things – despite the fact that doing so is demonstrably harder to scale and chock full of concerns about both adjustments for Disabled students and the way in which personal characteristics can cloud the judgements of those doing the watching;
• Like when you’re using those scanners in Asda, call in one on X students who’ve submitted a digital asset to do a mini-viva in order to establish whether they really did understand what it was that the asset suggested they did.

None of these options are easy to take, and to succeed all involve processes that require a serious and prolonged commitment to engaging with those that teach and students whose programmes will be impacted in order to be accepted and to succeed.

And yes, finding capacity to engage in creativity beyond “we had a good discussion at Senate/Academic Board” is hard when students have no time, industrial strife still surrounds the sector and the tech itself is moving so quickly.

But mark my words – whether they’ve been produced by Chat-GPT or not.

Anyone imagining that there will be awards left by this time next year that substantially depend on the submission of written work to be asynchronously marked later is almost certainly kidding themselves. Accepting that now will be better than being bullied into bodging it in February.