AI Guidance for PGRs

You should become an expert in your field by engaging with and thinking critically about research.

As a PGR, you are becoming an expert in your field. The development of your expertise takes time and effort. Your research and your writing are exercises in critical analysis, synthesis, and knowledge development. One of the most crucial skills you develop as a PGR is problem solving.

The end product of your doctoral research is more than your written thesis: it is your in-depth engagement with your area of research. While GenAI can support you in certain areas, such as automating processes, it cannot replicate that learning for you – it is vital that you have ownership over the knowledge you’re developing. To do this, you have to give yourself time to think, research, and write your way through your doctoral research.

Your written work must therefore be your own; you must be the one to produce it.

Please read Queen Mary’s Academic Misconduct Policy, available on the Policy Zone page, for full details.

When you do choose to use GenAI, you want to make sure that you have well-thought out reasons for doing so and that you keep careful track of the ways in which you have used it.

With your supervisors, you should decide if you need to use GenAI.

GenAI is very environmentally costly because it requires a lot of energy to run, particularly compared to traditional AI functions, such as browser search engines, literature database search engines, etc. For example, an average hyperscale data centre uses 2.1million litres of water per day (Barker, 2024).

The amount of energy required depends on the task; it takes less energy, for example, to produce GenAI text than it does an image. However, AI developed for a specific task, such as searching research databases, uses less energy than a generalist AI programme, which is often performing multiple functions at once (Heikkilä, 2023). As such, GenAI might not be the right tool for the task you have in mind.

With your supervisors, decide which system(s) you will use, why you will use them, and how you will use them.

The GenAI landscape is constantly evolving and there are a number of tools you can use. When selecting which tool to use, make sure you understand what are you agreeing to as a user. Read the Terms & Conditions – don’t simply click ‘accept’ without knowing what you’re committing to. It is possible that by inputting your work into a GenAI means that you are agreeing for your work to become part of the public domain and/or to be used in future training sets.

Understand your options for which GenAI tool to use: check if there are any for which your institution, school/institute, or funder hold a license and which will ensure that your work is not made public or used in training sets.

You should keep detailed records of your inputs and outputs.

Some GenAI programmes keep a log of your interactions – don’t rely solely on these! Remember that this is your work, so you need to keep track of it in a way that you know is reliable.

Work with your supervisors to determine how to best maintain that information.

As general guidelines, you want to record:

• Why you used GenAI.
• How you used it.
• When you used it.
• Which programme(s) you used.
• What you input.
• The outputs.
• How you used the outputs

You should keep detailed records of your inputs and outputs.

The Library has guidance on how to reference GenAI on their Referencing guides and resources page.

You shouldn't use GenAI as a means of avoiding the in-depth thinking and analysis you need to undertake to complete your doctorate.

Relying on GenAI to do the critical and conceptual work for you can limit your ability to develop expertise in your field.

This means that you, as a researcher, are less present in your research. Your authorial voice is unique and develops through the practices of writing. If you don’t write, or if you rely on GenAI to ensure that your work sounds professional and reads well, you are losing your own authorial voice in the process.

GenAI is weighted towards information that aligns to the middle ground, meaning that it ignores outliers or information that doesn’t quite ‘fit’. These outliers are often what drives change and so relying on GenAI to shorten necessary steps in writing means you might miss key pieces of information that can generate novel discussion.

This also means you’ll be less familiar with the information in your written work, as you are not working through it yourself and not thinking about it in as much depth. GenAI is prone to introducing errors, and if you are not familiar with your content or with your subject matter, it can make it much more difficult to spot these errors.

Ultimately, this can also make it much more difficult to defend your thesis in your viva because you will be less familiar with the content and will have engaged much less critically with it.

You shouldn't use GenAI without having established a clear plan and rationale for its use with your doctoral supervisors.

If you use GenAI without a clear plan and rationale, you risk using it unethically, even if you are doing so unintentionally. Your supervisors have oversight of your research project and are here to guide and support you in the decision-making processes for your research. While it is up to you to make final decisions about the research, your supervisors should be agreement with your decisions. If you use GenAI without informing them and working with them on the rationale, you risk being in conflict with them over this decision or they may not be able to help you defend your decision if they are not aware you’re using GenAI.

You shouldn't rely on the accuracy of GenAI outputs – always check to see if the information is true and accurate.

The information GenAI outputs is not perfect and is prone to hallucinations and errors.

Hallucinations are made up information. Not all of the information GenAI provides is actually true and taking it at face value means you could be introducing misinformation into your work.

Errors are information that is simply wrong. This may result from errors in the training database that are then replicated if you include them in your work.

When you use information output by GenAI, you always need to check it carefully and make sure it’s both true and accurate. The more you have engaged with the research in your field, the easier this will be to do.

The training sets used for GenAI are biased towards Western perspectives. It is tempting to think that the information produced by GenAI is neutral; however, it is only as good as its training datasets. These training datasets are selected by the individuals who input them, who will have biases, whether conscious or unconscious. This is known as value lock-in, when the information crystalises around the values of those contributing to the training sets.

The information the GenAI has access to is also unlikely to be completely up-to-date. Training datasets are only current up to the point at which the dataset was generated and used to train the software. This could make your own work out-of-date and you risk missing important new results or information.

You shouldn't input your work into GenAI without reading the terms and conditions. Understand what you’re agreeing to because your work could become part of the public domain and/or future training sets, meaning you could lose control of it.

Before you input any of your work into a GenAI programme, you need to understand how it can be used by the programme and whether you are transferring any ownership of your work.

This means reading the Terms & Conditions to make sure you know how your work will be stored, used, and distributed.

Without knowing this, you risk your work becoming part of the public domain. This has many potential implications for your work, including:

• Whether you can then publish it in an academic journal. Individual journals have their own AI and publication policies, and journals typically will not consider work for publication that they deem has been published elsewhere.
• Violating institutional ethical policies. If you input sensitive or potentially sensitive data (e.g., participant details), that information could become part of the public domain and therefore no longer confidential.
• Intellectual property (IP) and commercialisation implications. It’s important to be aware of who owns the IP for your work. IP is retained differently for PGRs depending on your funder and any industry agreements with the university – what is crucial is that you, as a PGR, are unlikely to own the entirely of your IP for your doctoral research. Putting your work into GenAI can cause complications with IP ownership.
• Security risks. If your work is sensitive, classified, or embargoed, putting it into GenAI could breach any security requirements imposed on your work.
• Loss of ownership. Your work could become part of future training sets for that GenAI programme and you authorial voice and/or your work may show up in the work of others.

You shouldn't disregard Research Integrity and Academic Integrity considerations – your work must be your own.

It’s important to remember that a doctoral thesis is an assessed piece of work being submitted for a degree. In order to meet research integrity and academic integrity standards, your work must be your own. In your doctoral viva, you will be required to demonstrate that ownership, and that in-depth knowledge can only come from putting the work into your doctorate yourself.

Using GenAI increases the potential for fabrication, falsification, and plagiarism – even unintentionally. Your responsibility as a researcher when using GenAI is to ensure that you are honest, rigorous, careful and respectful, transparent and open, and accountable in your research.

Please read Queen Mary’s Academic Misconduct Policy, available on the Policy Zone page, for full details.