Response to: ChatGPT in medical imaging higher education: Reply to Currie et al.

Dear Editor,
We thank the authors for the Letter to the Editor1 in relation to the article by Currie et al.2 In this rapidly evolving space we welcome all perspectives and discussion.

We are not convinced that the limitations of currently available generative artificial intelligence (AI) tools offer any advantage in expediting research and publication processes. Research and writing are a component of academic integrity and mastery of content, and this would be displaced by reliance on AI. Moreover, how time is saved to allow focus on research with the well documented limitations of generative AI is a mystery; it would create additional work for fact checking, reference checking etc. The value perhaps is in re-crafting work already researched and written, particularly for the novice author or those where English is a second language. Make no mistake, ChatGPT (OpenAI, San Francisco, USA) does not perform research. It is a predictive text generator and so the text it generates is predicted based on how that sequence of words has been previously used (in its training set). When ChatGPT produces content or references, it is not drawing these from computer memory, the internet or the resources it was trained on, it is predicting each next word from the sequence of words preceding it. This may well include language sequences out of context, inaccurate or outdated and produces plausible but fabricated references. As mentioned, with the exception of asking ChatGPT to reword your already created text, it should not be used for research or writing and researchers who do so either do not understand the nature of the generative AI or they do not understand the importance of research validity. This type of use is non-malicious but incidentally damages academic integrity and raises similar concerns as deliberate misuse.