Making AI work for students: reflections on institutional practice

AI is widely used by students, including to explain difficult concepts, summarise lengthy documents, generate practice questions and support revision. There may be particular pedagogical benefits for students who are neurodivergent or have a disability, or whose first language is not English. Staff adoption of AI into teaching varies significantly. Some academics described embedding AI in their pedagogical approaches, including in the design and delivery of teaching, learning activities and materials, while others use it very little or not at all.

Whether AI improves or weakens learning depends on how it is used. Used critically and carefully, it can help students identify gaps in their understanding and engage with learning in ways that cater to their needs and circumstances. There is headroom for institutions to further exploit these benefits. However, using AI to replace reading, research, problem-solving or other challenging tasks – often called ‘cognitive offloading’ – can circumvent the productive struggle needed to develop knowledge and skills.

Effective institutional practice in this area is likely to include considering whether and how the integration of AI tools across different courses can support teaching and learning; assuring the quality of AI-supported teaching materials and feedback; and communicating clearly to students where and how AI has been used in the process of developing them.

AI poses significant risks to the validity and effectiveness of assessments. Where assessments are unsupervised, it is frequently possible to complete them to a high standard using AI tools without demonstrating the knowledge and skills they are intended to test. Evidence also suggests that AI detection tools do not consistently and reliably identify AI use. This creates risks to the credibility of awards and the reputation of the higher education sector. It also reduces incentives for students to engage in the process of learning.

Approaches to assessments vary widely. While many institutions have introduced AI policies, students are frequently unclear about whether and how they are permitted to use AI in assessed work, and it is usually not possible to conclusively establish whether a student has adhered to their institution’s AI policy. Some institutions have redesigned assessments to reduce their susceptibility to AI use, including through increased use of closed-book exams, presentations and other assessments that take place in controlled environments.

Institutions should develop accessible and robust approaches to assessment, which effectively assure student achievement. Secure approaches designed to prevent students from using AI will often be the most effective as they provide a high degree of confidence that students are demonstrating their own knowledge and skills. The use of innovative assessment approaches that reflect the ‘real world’ application of students’ knowledge and responsibly integrate AI will often be positive, but these require thoughtful design to ensure they do not compromise the validity and reliability of assessments as a measure of students’ knowledge and skills.

AI literacy refers to the knowledge, skills and capabilities students need to be effective, responsible users of AI while they study and in their future professional and personal lives. Many students see the ability to use AI well, both in general and in the context of their discipline, as important to their future employability.

The provision of AI literacy training for students is uneven. While some institutions have developed ambitious programmes intended to build students’ AI capabilities, many students reported receiving limited training or none at all. As the ability to use AI effectively becomes increasingly important to students’ success during and after higher education, institutional consideration of the required general and subject-specific capabilities of students also becomes increasingly important Students need opportunities to develop these capabilities, taking account of differential starting points and unequal access to AI tools.

Staff AI literacy is an essential enabler of effective institutional responses to AI, and the differing implications and applications of AI across disciplines means the most valuable AI literacy training will often be subject-specific. At present, AI literacy training for staff is frequently optional and often has low take-up.

At many institutions, approaches to AI have been driven by the experimentation of individual staff rather than strategically by senior leaders. Some institutions have established governance groups or transformation programmes, while others have appointed AI leads or ‘champions’ to help support the development of staff AI literacy and steer departmental approaches to AI. However, relatively few institutions that contributed to this research have adopted strategic approaches to AI and embedded these throughout their institutions.

Various examples of how individual staff are using AI to improve efficiency were shared. These include creating content, inputting data, automating routine tasks and deploying student-facing chatbots. However, most staff participants felt AI had not yet delivered significant efficiency gains. Barriers to the systematic integration of AI include lack of confidence in AI outputs, meaning extensive human oversight is required; limited staff capability; immature data infrastructures; and slow institutional approval processes.

Institutions have an opportunity to use AI strategically to improve services and free up staff to spend more time on high-value activities. Realising benefits at scale is likely to require a coordinated approach driven by senior leaders while ensuring individual staff and teams can continue to innovate in their own practice.

Participants called for sector, regulatory and government bodies to adopt a collaborative approach to AI, working with institutions to support improvements in their approaches. Many highlighted the value of sharing effective practice in addressing common challenges. Participants generally suggested that new conditions of registration from the OfS on AI could be counterproductive and would risk constraining innovation. Instead, some participants called for greater regulatory clarity on complying with existing conditions of registration in the context of AI.