How do students perceive universities offering them access to AI? It’s a crucial question given increasing numbers of students providing access to ChatGPT, Claude and Copilot yet it’s one we know relatively little about. This Oxford SU report, in spite of the terrible response rate, offers some important insights about the issues raised. For a university to offer enterprise AI is a signal to students but it is an ambiguous signal which often sits uneasily with messages they are receiving locally from teaching teams:
Many students think purchasing ChatGPT Edu suggested a less hostile response to GenAI use in education, but most still think policy and institutional attitudes are too vague and inconsistently applied for them to feel confident that they will not be accused of GenAI use. They particularly object to AI disclosure forms and a focus entirely on in-person exams, which treat everyone as guilty.
Students are using LLMs to make themselves more efficient. In our focus groups, most students described using ChatGPT to prepare revision materials more quickly, to work out what readings were most worthwhile (which they compared to reading an abstract, introduction or conclusion), or checking their work, especially programming work.
However, many of the same students described feeling they had to use LLMs in order to keep up with their course loads and to avoid falling behind other students who were using LLMs. One student described a collective anxiety among non-ChatGPT users that, after the University bought ChatGPT Edu licences, ‘we’d sort of be cheated disadvantaged by the ones of us that were using ChatGPT’, expanding that they need ChatGPT Edu to keep on work. Another expressed a concern that ‘f you don’t use ChatGPT you will just fall behind because everyone else will be like flying and soaring through things.
They also felt that, even after the purchase of ChatGPT Edu licences, policy had not been clear, creating a sense that even while GenAI was officially permitted, one could never be sure if you were using it in a way that was institutionally accepted, or accepted by the particular members of staff you interacted with. This confusion was mixed with a sense that since students now all had access to ChatGPT Edu, one needed to use GenAI to avoid falling behind other students. This further muddied students’ sense of what the university expected of them and what they thought was best for their learning.
There can be few more powerful signals of normalisation than a university making a model freely available for all students. However if you normalise use in the absence of consistent criteria about using it appropriately/inappropriately then, I would suggest, the existing problem is made worse. It means use increases, anxiety increases but in a fragmented pattern which makes it even more difficult to establish a normative framework at a later stage. Another way of putting this is that if you accelerate the diffusion of AI through an organisation that has low readiness and literacy for making purposive use of it, you don’t accelerate AI integration in the sense of purposefully incorporated the technology into existing practices in a way that genuinely enhances them. You actually make it harder to do this because it drives anxiety about norms and fragmented practice. If you try to speed run diffusion you are extremely likely to damage integration.
