The following scenarios have been designed to help clarify King's generative AI guidance for doctoral students, supervisors and examiners.

If you have suggestions for additional case studies, please contact doctoralstudies@kcl.ac.uk.

You are a doctoral student preparing your final thesis for submission.

Your supervisor has suggested that your discussion of research paradigms in your introduction needs to be clearer.

You are finding it difficult to articulate the difference between ontology and epistemology, and how they relate to your research project, so you ask ChatGTP to provide a concise definition of each term.

You’re pleased with its response, and don’t feel you can write clearer definitions yourself, so you paste them into your final thesis. When you submit your RD2 form, you declare that you have used AI software. 

• Pasting text directly from a generative AI tool into your thesis without any editing or revision exceeds what is permitted under the King's guidance. This cannot be considered your own words, and so it would be treated as academic misconduct, regardless of which declaration you had selected on the RD2 form.
• In pasting the output without editing it or cross-referencing with other sources, you are also assuming that the definitions are reliable when they may not be, potentially introducing errors into your text.
• It’s also likely that the underlying issue, which was your difficulty understanding key terms, has not been addressed. This could mean that when your examiner asks you questions about the philosophical underpinnings of your method during your oral examination, you struggle to answer the question.

As alternative approaches, you could:

• Find definitions of each term in a valid academic source and quote them directly, citing appropriately.
• Use generative AI to help clarify your understanding. You could ask it to define the concepts in terms suitable for a high school lesson, a first-year sociology undergraduate lecture, and a master’s seminar. Then set those definitions aside, and attempt to write clear definitions in your own words. Compare what you’ve written with your most recent thesis draft, and if your next text is clearer then integrate it into your next draft.

You are a supervisor to a doctoral student for whom English is a second language.

The student wishes to run some sections of their thesis through ChatGPT to ensure their argument is expressed as clearly as possible.

You are somewhat wary about the use of generative AI tools, as you are not very familiar with them, so you are unsure how best to advise the student. 

• The student is proposing to use generative AI only to ensure the quality and standard of the English used. Provided that they carefully review and edit any outputs before incorporating them into their thesis, and acknowledge the use of AI software on their RD2 form at submission, this would be permitted under King’s guidelines.
• However, before they do so, you should discuss with them whether there might be any confidentiality, copyright, data protection, and ethical considerations relating to them uploading parts of their thesis to an external, third-party service.
• To increase your knowledge and confidence about generative AI, you should also familiarise yourself with King’s guidance on generative AI for teaching, assessment and feedback.

You are a doctoral student preparing to submit your thesis.

When completing your RD2 form to accompany your submission, you see that you are required to declare whether or not you have used generative AI tools when writing your thesis.

This isn’t something you’ve really thought about before. You’ve certainly never used ChatGPT. But you did use Microsoft Word to write your thesis, and Word does often suggest correct spellings or different phrases while you are typing, which you sometimes accept and sometimes ignore. You also use Mendeley to organise your papers and references, which often suggests related papers that might be relevant to your project, some of which have proved very useful.

All of this leaves you feeling anxious and unsure which declaration to select on the form.

• You have not used any software in such a way that your authorship of your thesis would be undermined. You can select the first declaration on the RD2 form – that no part of this thesis has been generated by AI software. You can confidently select this declaration because you have not used generative AI tools to rewrite, rephrase, or paraphrase parts of this thesis, and using software for English grammar and spell checking is consistent with the first statement.
• You can also be reassured that both potential declarations on the RD2 form describe acceptable approaches to thesis authorship. If you were to select the second statement on the form through caution or uncertainty, you would not be penalised for doing so.
• It is a valid approach to make use of papers suggested by Mendeley – this would not compromise your authorship of your thesis, as you are not incorporating text generated by Mendeley, so you do not need to declare it on the RD2 form.
• You should not rely solely on the recommendations of any single platform when conducting a literature review, so before submitting your thesis you should reflect on how you have approached your literature search and assure yourself you have used good practice.

You are a doctoral student preparing to submit your thesis, and are making final revisions to your second chapter.

You first drafted this chapter early in your second year, and one section in particular contains a lot of factual information that you haven’t revisited since then. To reassure yourself that there are no silly inaccuracies in the text, you paste the section into Chat-GPT and ask it to correct any errors.

You’re relieved when it identifies two incorrect dates and corrects a few other small details, and you amend the chapter accordingly.

• This is not a permitted use of generative AI, which can be used to correct and clarify your English, but should not be used to correct factual inaccuracies.
• It is also not good practice more generally – by relying upon a generative AI tool to provide factual information without further cross-referencing or revision, it is possible you have introduced inaccuracies into your text rather than corrected them.
• Even if the generative AI tool has genuinely corrected factual errors, it will provide no insight into why you originally made those errors. By revisiting your source material or data and checking for yourself, you might realise that those errors arose out of a deeper misunderstanding that you can now clarify. This will leave you better prepared to discuss and defend your project during the oral examination.

You are an academic at another university and have agreed to examine a doctoral thesis submitted by a King’s student.

Reading the thesis in preparation for the student’s oral examination, you have some concerns about one chapter in particular. The students cites an extensive range of relevant literature, but in doing so they sometimes seem to miss key nuances of certain papers. In some cases, there are outright errors that suggest a lack of understanding.

Although you haven’t engaged with it closely, you’re aware that there is a growing debate about the impact of generative AI on research integrity. Perhaps the student has used Chat-GPT to generate their literature review, and this has caused the errors?

A colleague suggests uploading the section of the thesis into Chat-GPT to see if it can detect any AI-generated text.

• Examiners must not upload any part of a thesis they have been asked to examine into a generative AI tool, because doing so may violate copyright or intellectual property rights, or breach data privacy.
• Currently available tools also cannot be relied upon to accurately identify AI-generated text – as the King’s AI guidance states, ‘detection tools are at best unreliable and, at worst, may present false positives’.
• Clumsy use of a generative AI tool is one possible explanation for the issues identified in the case study. However, it is also very plausible that they could have arisen from more ‘traditional’ circumstances – a hurriedly completed final draft, or a literature review written early in the project and insufficiently revised.
• In the absence of any overwhelming and unambiguous evidence of academic misconduct involving generative AI tools, examiners are asked to assess doctoral theses on the basis of the existing King’s criteria and procedures.

In the case study above, potential issues have been identified relating to the criteria that a thesis should 'demonstrate a deep and synoptic understanding of the field of study through its critical assessment of the relevant literature'. These issues are best addressed by the examiner asking questions focusing on this aspect of the thesis during the oral examination, and giving the student the opportunity to demonstrate their knowledge and understanding.

The examiners would then use their judgement to determine whether this issue will influence the outcome of the oral examination (for example, they might request revisions to this section of the thesis).