Research with Integrity – the pitfalls and potential of generative AI

As universities wrestle with how to address the impact of generative AI on many levels, what does it mean on an individual level for a postdoc or PhD student in a lab interested in using generative AI to support their work?

If this is you, perhaps one simple question to ask yourself is: do you want to lean into generative AI or keep your distance from it?

Benefits and limits

To lean in, you’ll want to invest your time in understanding how best to use these new tools – thinking deeply about the implications in areas such as copyright and plagiarism, confidentiality and data security, ethical practises and environmental sustainability – and figure out how the concerns and opportunities in those areas interact with your research.

Or, perhaps, now is not the time for you. And this isn’t simply a scepticism or distrust of the technology but more a weighing of the cost/benefits at this particular time, for you, in your work.

There’s no doubt that some people are benefiting from using generative AI. I’ve personally heard from people using it to help with coding and bioinformatics, as a starting point for learning about new subject areas and for cutting down large amounts of text to make it more digestible and readable. There are also people who are embedding generative AI directly into their research processes, for instance to help with analysing large data sets.

If you do see an opportunity or potential for generative AI to benefit your research –  then I would suggest it’s hard to do this well without going into the deeper ethical and practical  questions. If you are leaning in, it’ll be worthwhile – maybe even essential – to really understand the mechanistic basis of Generative AI and Large Language Models (LLMs), their strengths but also their limitations.

If in two years’ time your thesis or paper will contain research methods that have used generative AI today, then – as your work will need to follow the commitments in the Concordat to Support Research Integrity and the codes of practice of research organisations in the UK that flow from this – it’ll be your responsibility to ensure that you keep generative AI under careful consideration, use it according to the highest standards in your field, and seek ethical approval where necessary. That describes the sort of package of evidence you will likely need when questioned by an examiner in your viva, or when signing declarations with a publisher.

However, that is a considerable amount of work to do well. And while generative AI is a fascinating topic, if it’s not essential to your work then maybe you are better investing that time, energy and thought into something else.

If not now, when?

I’m not saying generative AI will never be useful to you, but the question is: do you need to use this now?

There may well be a point where generative AI tools reach market saturation and most of us use them daily in the way we use spreadsheets and email. But we are not at that point. If an organisation like Apple is not able to make their generative AI-based news app work properly (requiring a very public recall); if people reviewing CVs and cover letters are reporting on social media that they can very much tell which ones are written using AI, and if people in criminal law cases are still referencing fake cases based on hallucinations from generative AI, I think it’s clear that this technology has not yet reached a final, stable form.

I am labouring this point because one of the biggest issues I believe researchers are facing with respect to generative AI is FOMO – the fear of missing out. The hype level in the generative AI news cycle is extremely high. Most of it is driven by the companies who have invested enormous sums in making these products. It’s in their interests to contribute to a sense of urgency in getting people on board with using generative AI.

Other contributing – but possibly linked factors – include government-level calls for incorporating Generative AI, along with examples of researchers themselves sharing how they are using these tools. But these are mainly those who are ‘leaning in’ to generative AI, spending time learning the software, adapting it to their own needs, using it in creative ways, to receive that benefit. But this comes at a cost in terms of time, energy and learning.

Therefore, I suggest that if you are in the position of having tried to use generative AI but still see a long road ahead to beneficial proficiency, that stepping back from using it at this time may be a completely rational approach.

Questioning your place on the AI spectrum

I’m conscious that I’m presenting this as a dichotomy – that, clearly, is an oversimplification. So, if you fall somewhere in the middle or you’re not sure where you sit, let me ask you five broad questions to help you find a way through. Most universities are developing or have developed guidance for their researchers, including training and frameworks for responsible use of generative AI, so I would always advise researchers to be aware of and align with their local guidance. But hopefully these questions are complementary and can help you think through the process.

Hopefully these sorts of questions can help you do a relatively quick triage to determine whether you are going to ‘lean in’ to generative AI or keep your distance for the time being. If you are going to ‘lean in’, then do so in a wholehearted way, seeking out the information, training and support needed to use it responsibly. And I am looking forward to seeing all the fantastic use cases that come from thoughtful, dedicated researchers adopting this new technology.

But if the concerns and questions here are big enough, I think it’s a perfectly sensible choice to actively choose not to use generative AI at this time and spend your energy on some other important part of your research.