Generative + Agentic Artificial Intelligence Use

A thesis is a piece of substantial intellectual work that demonstrates the student has mastered a field of study and, in the case of a doctoral thesis, advanced knowledge in that field. The Faculty of Graduate Studies strongly discourages students from using agentic and generative artificial intelligence (Gen AI) tools to create textual elements of their theses, including for summarizing or synthesizing the work of other scholars.

Agentic and GenAI tools should be employed only with the prior and explicit approval of the student’s thesis supervisor. Students who want to use agentic and GenAI tools for thesis research and/or writing support should get prior approval in writing from their academic supervisor describing the specific uses that are approved. Evidence that a student had approval to employ specific AI tools can help protect a student from accusations of violating academic integrity. 

Here are some things to consider before you employ GenAI tools for your thesis research and writing.

Want more information?

Learn about GenAI and thesis research in our Canvas Course

Use GenAI tools in ways that enhance your learning and development as an expert in your field

Let’s face it, writing a thesis or dissertation is hard. But the struggle is the thing that makes you a scholar and enables you to be proud of your accomplishments. Be judicious in your use of GenAI tools and think carefully about which specific tasks you will delegate to GenAI. What are the benefits and trade offs? Can you accomplish the same tasks with a conventional method? Will you be able to verify the accuracy of the output? If you decide to use GenAI to support your data collection, analysis, and or writing, do so in ways that don’t shortcut your opportunity to develop your analytical capacities and subject knowledge.

Use GenAI tools in ways that respect the intellectual property of others

Chatbots and other GenAI tools make use of Large Language Models (LLMs) that are trained to recognize the patterns in language, music, and visual art. An overwhelming portion of this training data are works protected by copyright used without seeking the permission of the creators. One of the characteristics of GenAI is that its inferences (responses to user queries) are supposedly unique. In fact, chatbots and other GenAI tools sometimes repeat verbatim phrases or images from their training data or the internet. More worrying for academic work is that citations generated by chatbots are often missing or wrong. For example, a chatbot may provide responses that, if you had read them in a book, you would almost certainly credit the author with a citation. Will you be able to associate ideas generated by GenAI with their authors to avoid inadvertently plagiarizing other scholars? Will you be able to explain and defend the ideas in your thesis?

Use GenAI tools in ways the protect the privacy of your research participants and maintain the security of your data, inventions, and discoveries

Make sure that you know how to prevent AI tools from collecting your data to use to train future generations of AI tools. Data collection is the default setting for many chatbots as well as for other AI tools such as transcription, translation, and grammar checking software. Use only tools that allow you to opt out of data collection or, even better, that you can run on a local computer or server. The version of Copilot provided with your SFU subscription to Microsoft 365 runs on SFU servers and does not share your data with Microsoft, but the same is not true of many other AI tools you might use. Even supposedly anonymized data can be combined with information already in public and commercial databases in ways that put people at risk. You might inadvertently expose proprietary data or affect any patents you could apply for. Make sure any research participants and/or collaborators understand the risks to their privacy and data security that may result from your use of AI tools. Finally, if you have human research participants, be sure to include all planned uses of GenAI in your ethics application and consent documents. If new uses arise as you plan data analyses, be sure to have these approved and to include these uses a revised consent form.

Use GenAI tools within the norms for your discipline and with the written approval of your supervisor(s)

There is no one, single acceptable way to use GenAI tools in academic research and writing. Norms for using GenAI tools in academic research are in flux and vary across different academic disciplines. Even within the same discipline, different scholars will arrive at different conclusion about which uses are appropriate and which research tasks may be supported with GenAI tools. The key is transparency; are you willing to acknowledge that you used an AI tool to help you analyze your data or to draft several paragraphs of your thesis?

While you are responsible and accountable for the content of your thesis, taking care to document your AI use and having your Supervisor’s advance permission are critical to protecting you in the event that an examiner or other person raises questions about your adherence to SFU’s academic and research integrity policies.

How to acknowledge AI use in your thesis

AI tools used for any aspect of thesis work involving data (i.e. methodological decisions, data collection, data analysis, translation or transcription, or visualization) should be fully described within the thesis as you would for any research method.

AI tools used for any aspect of writing or editing, including summarizing or synthesizing previous research, must be identified and clearly described in the Acknowledgements pages of your thesis. You may find it helpful to employ an AI statement generator. GAIDeT Declaration Generator is one example.

Use GenAI tools in ways that accord with your values

Generative AI tools come with ethical questions about labour practices and environmental sustainability. Extraordinary amount of water, electricity and critical minerals are required to create the hardware required for GenAI. Water and energy are also consumed when the models are trained, and especially, when users seek answers to their queries. Equally concerning is the exploitation of low-wage workers in the global south who are tasked with labelling and annotation tasks that allow GenAI to appear to work like magic

Use GenAI tools in ways that preserve your authentic voice

LLMs have a tendency to produce homogenized and flattened prose. It may be grammatically correct, but it is often boring and lifeless. Rather than offering new ways of thinking, chatbots and AI drawing tools tend to reproduce the racial, gendered, ageist, classist, and other biases and hegemonic ideas that exist in their training data.

Faculty of Graduate Studies