AI use is now universal in higher education. Why are universities failing to adapt?

Artificial intelligence (AI) use among university students is now near-universal in many countries. Yet, as of a year ago, fewer than one in five universities in a 90-country UNESCO survey had a formal AI policy. Since ChatGPT arrived in late 2022, the threat to academic integrity has loomed large. Unis must now ensure students earn degrees by showing what they can do, not what AI can do. This requires policies that drive structural change to assessment and curriculum, not just rules about when AI can be used. But it’s not so straightforward. The challenge for universities Universities, in particular, have struggled to adapt to generative AI more than primary and high schools. This is because university classes are often large and contact time between students and teachers is limited. Together, this makes it hard for teachers to get to know students personally and to ensure they learn what is required for their degrees. UNESCO has published an AI competency framework, but it focuses primarily on schools and stops short of providing specific guidance for higher education. University AI policies are often vague A study published in July examined the public AI policies and guidelines of 39 Australian universities that teach information and communications technology. This is a subject area where AI can readily complete coding tasks. The analysis found wide variation among the universities’ AI policies. Often, AI rules were buried in institutions’ misconduct policies rather than assessment design procedures. Out of the 39 unis, only one had published specific guidance around AI. The universities’ general guidance on AI use was often vague, too. One simply required AI use to be “ethical, pedagogically sound, transparent and purposeful” but did not provide any further detail. The researchers argued that AI was disrupting assessments faster than universities could build the structures to govern it. Only seven of the 39 universities had documented a body, such as a task force, to coordinate their response. Many have also left it to individual unit coordinators to spell out what AI use is allowed. This means the rules can differ between units. The risks also go well beyond cheating. Global discussion among leading researchers and policymakers this year has shifted to a deeper issue: using AI to avoid learning altogether. This is when students’ offload their work to AI entirely and, as a result, learn nothing themselves. It is difficult to gauge how much of this offloading is already occurring. Universities’ current policies aren’t sufficient in addressing student offloading to AI. Students cannot be forced to learn through a policy. Teachers have also been using AI tools to detect whether students have included AI-generated text in their submissions. But universities are now moving away from this as the technology is not always accurate. Experts have warned that relying solely on AI detectors can lead to false accusations of student academic misconduct. Assessment reform Instead of focusing on more systemic responses, universities have mainly responded to AI by changing how assessments work. This has usually meant: a greater emphasis on in-person exams and oral presentations where AI cannot be easily accessed permitting the use of AI in some assessments, such as take-home essays, where students have cited their use of the technology. But while redesigned assessments can show whether learning has occurred, it is still not enough. It also matters what students are assessed on. What are other countries doing? In other parts of the world, such as the European Union, attention is turning to explicitly developing students’ capabilities to adapt to constant change. These include teaching students to: make sound judgements about how much they know about a subject critically evaluate information, both with and without AI make decisions about their learning processes and goals independently of AI or an educator. However, these capabilities are hard to develop and mostly don’t transfer between subjects. Instead, they must be taught explicitly to students, in the context of their subject area or discipline. One way of doing this would be to incorporate specific modules into courses that teach this. So, what’s the right approach? In 2023, a number of academics, including myself, wrote evidence-based assessment principles for the Australian higher education regulator. The principles were formed to guide unis on how to adapt assessments to account for students’ AI use. One principle focuses on assuring students are actually learning content and not offloading work to AI, which institutions have attempted to address through assessment redesign. These attempts at redesign are having an effect. The second principle is about preparing students for a world where AI is ubiquitous. Students need to learn how to effectively use AI for when they enter the workforce. This principle has proven much harder for unis to adapt to, as skills tied to today’s AI tools quickly become outdated as the technology rapidly advances. This means simple “AI literacy” is not enough for students alone. While there is no one-size-fits-all approach, unis would also benefit from collaborating with each other to tackle the problem. However, institutions tend to compete rather than share resources. Only one of the 39 in the study I previously cited mentioned learning from other institutions. A national strategy would help In 2024, the higher education regulator required every uni to submit AI action plans outlining what they are doing to adapt to AI. The regulator has also since published guides outlining strategies. But a regulator can only set expectations. It cannot fund the long-term work of enabling changes to policy and practice across the country. That was the job of the Office for Learning and Teaching, which funded teaching innovation and helped universities share what worked. It closed in 2016, and a promised replacement never materialised. A national approach would mean an enabling body – independent of any single university – sharing evidence and guidance across the sector. This might look similar to Ireland’s Higher Education Authority, which has published a national framework on AI in teaching and learning. However, there are challenges in setting this up in Australia, given how protective universities are about their autonomy. There would be questions around who funds such a body. And it would be difficult for a national authority to keep pace with technology that changes faster than policy. Still, unis must engage deeply and collaboratively with this debate about how to adapt to AI. Otherwise they risk becoming redundant in the age of intelligent machines.