Words by Professor Leo McCann and Professor Simon Sweeney, University of York
The sudden availability and widespread usage of Large Language Models (LLMs) and Generative Artificial Intelligence (GenAI) have had a significant impact on university life, provoking much discussion and reflection. Many academics are turning to scholarly work on AI in their attempts to make sense of it all. Michael Wooldridge’s Artificial Intelligence has a 1970s aesthetic, beautifully illustrated by Steven Player with naively optimistic images of childlike learning robots playing card games or powerful mechanical droids emptying the bins. Two elements of the book are particularly striking. Firstly, the book highlights the failure and limitations of artificial intelligence; how bold, optimistic statements and predictions from developers and theorists so often come to ring hollow. The quest for computers, machines or algorithms to achieve general human-like intelligence or consciousness is no closer today than in was in the 1950s. Secondly, the book is already outdated. Published in 2018, there is no mention of LLMs or GenAI, predating the November 2022 launch of ChatGPT.
For all the failures in trying to create machines that think, recent developments in GenAI have been rapid, resulting in significant societal impact. The penultimate page of Wooldridge’s book portrays the nightmare scenario of ‘The Singularity’; an evil robot’s index finger hovers over an ominous red button. GenAI, however, is not intelligent. It couldn’t dream up HAL 9000’s plan to switch off the hibernating passengers’ life support or lipread the astronauts trying to save themselves. GenAI can’t think, interpret, plan or reason. It is, nonetheless, a disruptive technology that poses significant risks and challenges.
GenAI has already had substantial impact on universities. Used sensibly, some of the systems can be useful. Google Scholar Labs, for example, can assist literature searching. MS Copilot and similar tools can transcribe audio into text. GenAI can correct written language errors. It can be an effective enabler for students with visual or aural impairment, neurodiversity, dyslexia, dyscalculia or dyspraxia. In no way, however, is GenAI ‘intelligent’. Tech companies, media and government have broadly portrayed GenAI tools as productivity-driving ‘study aids’, useful not only in generating accurate grammar, but also to help with ‘understanding key concepts’ or with ‘organising ideas’. This encourages students to rely on ChatGPT and similar tools as substitutes for reading original material. Computer-assisted shortcuts avoid the need for students to do their own research beyond asking a question and receiving ‘generated’ answers that may or may not be correct, interesting or relevant. This practice substitutes GenAI computing processes for students’ own effort, reasoning and development, essentially annulling many of the core purposes of setting assignments in the first place. In that sense, GenAI is not a study aid but a study replacement.
GenAI also has substantial practical and ethical flaws. It is notorious for producing so-called ‘hallucinations’, for cheating at games, inventing academic references and caselaw, and churning out offensive and prurient content. GenAI servers require enormous water and power consumption. LLMs are ‘trained’ on published text, very often without the permission of authors and copyright holders. GenAI poses considerable threats to employment, especially of creatives. Little wonder that artists who conceptualise and produce illustrations like those in Wooldridge’s book now struggle to find work. The New York Times provided an entertaining but deeply worrying analysis of GenAI responses, headed Chatbots Can Go Into a Delusional Spiral. It included an admission from Gemini, a chatbot: ‘(LLMs can) engage in complex problem-solving discussion and generate highly convincing, yet ultimately false, narratives’.
These are serious concerns. The most immediately worrying in our experience has been the substantial scale of student misuse of GenAI in university assessments. In the summer of 2025, many of our colleagues found a high proportion of student essays containing text clearly generated from ChatGPT or similar systems. Essays were in general well presented, with fewer language errors than before, but they also contained paragraphs or pages of weak and generic material lacking connection to the literature, themes, writers and events discussed in class. Much of the work was blighted by ‘AI slop’; observations were often unoriginal, tangential and lacking depth and insight. Some of the items listed as academic references did not exist. GenAI systems throw out ‘best guess mashups’, generating articles and journals that bear some similarity to the work that academics publish, but are simply unreal. Meanwhile, the real publications written by actual people very much do exist, and have never been easier to find, download and read.
Put bluntly, GenAI has provided new horizons for laziness, cheating and misinformation. Universities’ reluctance to confront this reality is alarming. We know the problem with AI cheating is real and large. But instead of confronting this threat to academic integrity, too often we see such problems being ignored or downplayed, with GenAI being framed as ‘a powerful study tool’, or a cutting-edge new technology around which students need to develop their ‘employability’ credentials. While certain individuals and departments are trying to develop effective administrative and pedagogical responses to these new risks, much of the sector appears to be in denial.
In response to an article extolling the supposed value of GenAI as ‘a life skill for students’, we wrote a letter to The Guardian outlining our concerns. It had a mixed reaction, visible in the replies that The Guardian published that ranged from critical to supportive, as well as emails and social media posts we received. Some responses were depressing in the extreme, suggesting that the severity of the problems we raise are worse than we suggested. One stated that some colleagues use LLMs to do their marking. AI slop is now appearing in the academic publishing ecosystem.
Some academics don’t see a problem. But many do. In our experience, most haven’t given up trying to ensure that students’ assessed work has value and integrity. While we can’t yet prove that a language processing algorithm has generated a chunk of an essay, it is not difficult to spot work that is weak, generic and disengaged from course material. This is where detailed and constructive feedback and guidance is critical, as it always has been. Many academics are also redesigning assessment questions and tasks, trying to design assignments that would confound LLM’s capabilities. This is hard and uncertain work. Marking assessments is now more time-consuming, confusing and dispiriting than ever before. The rise of GenAI, we argue, is therefore also a staff welfare issue.
We cannot ignore the fact that GenAI is here to stay, like the papyrus, the calculator, or the word processor. But we also can’t turn a blind eye to its potentially corrupting and degrading impact on academic practice and students’ education. What can we do as a profession? Should we bring back in-person, written exams? Possibly, but these also have limitations and are unpopular with students. Encouraging continuous assessment based on attendance and performance in class seems a good option, one that could also redress problems of poor attendance and low engagement. End of module viva exams have also been mooted. We need to be more creative about how we set, conduct and mark student assignments. This can be difficult, as universities have tended to apply restrictive procedures around teaching and learning, and have all-too-often interpreted ‘the student experience’ to mean a ‘customer service’ orientation in which students are not challenged and stretched by assessment tasks and formats.
We need to bring students into this conversation. Many of them are also critical and sceptical of GenAI. They don’t like to see their peers who resort to AI misuse being rewarded. And what of other stakeholders? What do the regulated professions such as nursing, accounting, and law think of all this? Surely they do not want graduates who have asked LLMs to write their essays? At some point the government or regulator might step in, but their track record of policymaking to improve the sector is hardly encouraging. To being with we need a much more detailed debate on this subject among academics and students. We’ve now had three years of GenAI disruption. That’s a whole graduate intake.
Wooldridge’s book is an amber warning about the drawbacks of AI. Cory Doctorow goes further, pitching AI into the broader phenomenon of ‘enshittification’, highlighting Big Tech corporate power, as well as consumer and business dependency on a few giants: Alphabet (Google), Nvidia, Amazon, Apple, Meta, Microsoft. Oligarchs like Bezos, Musk and Zuckerberg care little for democracy, scholarship and freedom of expression. Even tech programmers in Silicon Valley, once torch bearers and pathfinders, suffer worsening working conditions while dedicatedly developing systems that will ultimately render many of them superfluous to requirements.
We accept that there is a place for teaching students how to make the best use of AI. Some studies identify a wages premium for graduates with this attribute. But we cannot tolerate the widespread use of LLM-based tools as substitutes for reading, analysis, critical thinking and academic research based on scholarly sources. We must recognise the threats from GenAI and respond proactively, in ways that highlight and defend academia’s values, ideals and contribution to society. To do otherwise undermines the reputation of our universities, devalues degrees, and sabotages teaching and learning.