Principles for Using GenAI in Research and Innovation

Overview of principles

Your use of generative AI must align with the University’s expectations for responsible research and academic integrity. The focus here is on the use of GenAI in research and innovation, and excludes research on GenAI itself. The counterpart of this policy for education is available from Elevate.

Generative AI holds the promise of revolutionising how research is conducted, from productivity aids in writing and presenting results, through to the generation of results themselves. Gen AI tools are trained on vast amounts of diverse types of data, allowing them to create new text, images, computer code and audio. While there are continuing concerns around the sustainability and ethics of GenAI, it is incumbent on us to consider how we might integrate and realise the significant opportunities this unleashes, while ensuring the integrity of our research endeavours in what is a very rapidly evolving landscape. However, the application of such tools is nuanced, with variations even by discipline, and the potential risks posed by GenAI will depend on the context of its application.

This page sets out some overarching principles to guide the evaluation and use of GenAI tools. Researchers must use Generative AI effectively whilst maintaining rigour and integrity in the research process. This is about being aware of the difference between appropriate use of GenAI tools and using them to provide intentionally or unintentionally misleading information. This might concern authorship or information accuracy, for example.

Alongside the standard tenets of research integrity (as described in the Good Research and Innovation Practices (GRIP) policy (PDF, 254 KB)), the overarching principles that should guide any particular use of GenAI are the following:

• Appropriateness – Is this an appropriate use of GenAI? You must be able to explain and justify the use of GenAI in your research. Does the GenAI tool(s) complement human expertise, critical thinking and judgement in the research process, rather than replacing it? Have you considered the sustainability implications of using GenAI? Are you using it only when it is offering significant value and potentially contributing to meaningful and sustainable outcomes?
• Attribution – You must transparently acknowledge the use of GenAI at any point in the research or writing process. For example, ‘this manuscript was prepared with the assistance of Generative AI <cite the tool>’. Where there is verbatim reproduction of GenAI output in a section, this should be explicitly indicated. For example, ‘the Abstract and Introduction were produced by Generative AI <cite the tool>’. You should include the name and version of the tool, the date it was used, and whether the output has been validated. To ensure transparency, you should record the prompt texts used.  You must not use GenAI to create synthetic data, without attribution, for presentation in a research study.
• Accuracy – You must check the accuracy of any GenAI output used in research. These technologies are known for ‘hallucinating’ and ‘confabulating’ in text, imagery, etc. and for making programming errors. You should pay particular attention to references generated, syntheses of prior research, and attributions of fact, as these may not be reliable, and may be biased.
• Adherence to data protection principles (including access to data and confidentiality) – The use of GenAI comes with the risk of leakage of and loss of control over sensitive and proprietary information to a third party. Some GenAI tools take users’ data for the purposes of further general training of the model. In the case of proprietary research data, sensitive personal data, etc. you must avoid all exposure to public GenAI tools (management of and responsibilities regarding sensitive data are also covered in GRIP). The selected tool should have the appropriate level of safeguarding of users’ data, not using it for training at all, or only for training of a version of the tool that is private to the user (the Terms and Conditions for these tools will detail how information inputted into them will be used). Using a third party’s data on GenAI without their permission would equally be inappropriate (eg using industrial partner data in GenAI). Other sensitivities, like export controls, might also need to be considered here. Further guidance on which tools have what levels of data protection is available later in this document (see also the Information Commissioner’s Office Guidance on AI and Data Protection).
• Accountability – You must ensure that the use of GenAI, whether in the results of research or work presented for assessment, complies with ethical standards and the tenets of research integrity in line with the Good Research and Innovation Practice (GRIP) policy, as well as the policies of relevant funders and publishers.  You remain accountable for your research when using GenAI and must, given the fast-moving nature of GenAI, keep up to date with developments of the tools you are using.