In a recent Emergency Medicine Australasia editorial, Professor Daniel Fatovich and Professor Geoff Hughes explored the use of AI in academic publishing.
We asked Prof Fatovich what responsible AI use might look like when writing and reviewing research grant applications.
A long-time supporter of EMF, Prof Fatovich is a senior emergency physician and clinical researcher at Royal Perth Hospital with more than 30 years’ experience in emergency medicine research. He is Head of the Centre for Clinical Research in Emergency Medicine at the Harry Perkins Institute of Medical Research and Professor of Emergency Medicine at the University of Western Australia.
AI can help researchers save time on routine, non-sensitive administrative tasks and basic editing. It can also assist with literature reviews, but you still need to check accuracy. As such, it can be considered a bit like a research assistant.
However, submitting sensitive or confidential details to AI, such as innovative ideas or unpublished results, could compromise your application and intellectual property. AI also hallucinates and creates false citations.
In my experience, AI uses sub-optimal language that superficially sounds good but can be meaningless. Human oversight is an essential requirement for the use of AI.
Should researchers disclose their use of AI in grant applications?AI is rapidly evolving and, in my opinion, moving faster than we can adapt to it.
Nevertheless, researchers need to disclose how they have used AI in their application, and funders should require this information.
A nice expression I’ve come across is that AI is not a truth machine. It’s also crucial to avoid inputting copyrighted or unpublished material.
What about AI use by grant reviewers?
Submitting an application to AI to assess it is strictly prohibited. Applications are confidential documents containing sensitive information, so they should never be submitted to AI.
However, if there’s a method in an application you don’t understand, it would be appropriate to Google it to improve your understanding.
In this rapidly changing area, the Swiss National Science Foundation has just released new guidelines which you can view here. This allows reviewers limited use of AI to assist with evaluations, “but it should not replace their own judgment’” and they must protect proposal confidentiality.
As AI makes it easier to produce polished applications, what should reviewers focus on?I’m not so sure that AI produces highly polished applications. It might help you to get started, but in my experience, the language used by AI is frequently distinctive – using generic terms that sound good but are ineffective and not precise. It doesn’t sound considered!
You can’t beat expert judgement by a human in assessing the track record of the team and how useful the project might be in terms of context, novelty and clinical relevance. AI also tends to want to please us, so it can give variable answers to the same question.
What’s your rule of thumb for using AI responsibly?Don’t believe everything you read; be sceptical! Maintain privacy and confidentiality, and always keep human oversight.
