Use of generative artificial intelligence in the PhD thesis at the MN faculty

Different forms of Artificial Intelligence

Artificial Intelligence (AI) is a broad concept with many applications. This page focuses on the use of generative AI in a PhD thesis at the MN Faculty at the University of Oslo. Below, you will find a rough overview of different forms of AI.

• Artificial Intelligence (AI):
  • Expert Systems: Rule-based systems, e.g., automated chatbots or email responses.
  • Traditional AI Tools: Programs that do not use machine learning, e.g., Stockfish (chess) or symbolic math tools.
  • Machine Learning (ML): Learning systems based on data, including neural networks (deep learning).
    • Modeling: AI based on ML to simulate complex phenomena, e.g., in biology, medicine, or climate.
      Classification: Sorting data, e.g., cells, stars, or molecules.
    • Generative AI: Tools based on ML that create new content
      • Image Generation: Dall-E, MidJourney.
      • Text Generation: ChatGPT, GPT UiO.
      • Other Media Generation: Udio (music), veed.io (video).

Use in a PhD thesis

The use of generative AI in a PhD thesis is permitted at the MN Faculty but must be carefully evaluated in consultation with your supervisor. If you are considering using AI in your thesis, you must review each of the eight points below thoroughly together.

Note that this applies to the use of generative AI in your thesis (the introduction or monograph) and any unpublished articles you choose to include. For the use of generative AI in published research articles, you must adhere to the rules set by the journals where you seek publication.

1. Data storage

When using generative AI, you typically send data to a cloud service where it is stored. UiO has strict rules about the types of data that can be sent and where. A full overview is available on USIT's website about data and information classification, but in brief, data is classified according to the level of protection required, ranging from green data (minimal protection needed) through yellow, red, and black. If you intend to analyze yellow, red, or black data, you must use a service with a data storage agreement authorized by UiO. More information about which services UiO has agreements with can be found on USIT's website on IT services.

If you plan to use AI in your thesis, ensure that no data is sent to a service unauthorized to handle it. UiO’s own GPT service, GPT UiO, can process green and yellow data. For advanced users, there are also open models that can be downloaded and run locally on your own machine, allowing you to analyze all types of data as long as they are stored in compliance with UiO's regulations.

2. Declaration

If you choose to use generative AI, you must declare the extent of its use at the beginning of your thesis. A template for such a declaration can be found in the Competence Centre for Teaching and Learning in Science (Norwegian acronym KURT)’s suggested declaration format.

Note that this page provides templates for how such a declaration might look and includes examples of how to declare different types of use. You yourself, in consultation with your supervisor, must assess what to declare in your thesis to ensure academic transparency and integrity.

3. Generation of images and multimedia

Generating images and multimedia for the thesis must be done with great caution. There is still uncertainty regarding copyright and plagiarism rules for art created using generative AI. It can also be challenging to ensure that a generated image is not owned by someone else. It is discouraged to use generated images that could be mistaken for real photographs.

You are responsible for all multimedia published in your thesis. This means you must ensure that no plagiarism is involved. If there is suspicion of plagiarism in the thesis, pointing out that the images were generated by AI will not be a valid defense. More details on this responsibility are covered in point 8 regarding accountability for the thesis.

4. Consider whether the use is appropriate with respect to how the thesis will be evaluated

Part of writing a thesis is demonstrating your ability to produce a long, cohesive, precise, and engaging scientific work. Consider the following points with your supervisor regarding how AI might impact the evaluation of your thesis:

Poor use of generative AI for text generation often results in inconsistent tone and quality in the writing. This is a negative indicator and can lead to poor evaluations.

Your personal voice is essential and should be evident in the thesis. Careless use of generative AI can result in a low-quality thesis.

The MN Faculty encompasses many fields and specializations, and evaluation criteria vary significantly. Is the language and uniqueness of your phrasing an important evaluation criterion in your field? If so, you should exercise caution in using generative AI.

5. Assess whether the use is appropriate from a learning perspective

Another important goal of writing a thesis is to learn how to work scientifically. Future workplaces might not accept extensive use of generative AI, so becoming overly reliant on it is not advisable. Consider the following points from a learning perspective:

• Are you short-circuiting your own learning process through your use of generative AI? For instance, there is significant value in reading others' articles and summarizing their work yourself.
• Writing is a crucial part of the learning process. Much of your understanding is developed through the act of writing. Therefore, you should always strive to write on your own.

6. Assess whether the use is appropriate based on how your research community views AI

Just as different parts of the MN Faculty at UiO have varying perspectives on the use of generative AI, so do research communities. It can be valuable to reflect on and discuss with your supervisor how the use of generative AI is perceived within your research community.

7. Assess whether the use is appropriate from an environmental perspective

The use of generative AI typically involves high energy consumption, which can be problematic in today's context from a climate perspective. Extensive and uncritical use can result in a significant environmental footprint. This should be part of your assessment when deciding on its use.

8. Assess whether text generation is appropriate

Using AI to generate text can be problematic for several reasons:

• Generated text may include factual statements that are incorrect. Language models like ChatGPT produce new text that may not be sourced or accurate.
• Language models are trained on human-written text, which may reflect biases present in the source material. This can lead to bias in the text generated by the model.
• Similar to multimedia generation, text generation can result in unintentional plagiarism, as the generated text might closely resemble existing material.

The most important principle for using generative AI is the same as in academia in general: as the author of your work, you must be able to fully defend 100% of its content, arguments, and references. This applies to everything in your thesis, from cover to cover, including both text and images.