Chapters 8: Usage of artificial intelligence

Chapter 8.1 Content responsibility

Written work is to be created independently and must comply with the rules of good scientific practice. This means

• the author is responsible for all parts of the work.
• the work is based in its essential aspects on the thoughts and ideas of the author.

Chapter 8.2 Transparency

Any use of AI systems must be documented in the List of aids. It must be comprehensible to third parties which systems were used to what purpose and to what extent. Further, it should be noted that not only AI based systems need to be listed in the list of aids (e.g., survey tools such as LimeSurvey or literature management with Zotero).

Should the submitting of a research protocol be requested, the intended use of the AI system must be examined carefully. If an AI system is used, for example, for identifying a research gap or to gain inspiration on text structuring, it will be listed in the list of aids. However, if an AI system is used for research (even for only one instance), it is not to be documented in the list of aids, but in the research protocol.

Should no research protocol be required, all tools, regardless of their purpose of use, are listed in the list of aids.

What details are to be listed in the list of aids, or the research protocol can be found in Chapter 5.1.3 of this guideline.

Chapter 8.3 Reflective AI use

To make the responsible usage of AI systems more visible, a two-stage confrontation is required in every written work, in addition to the formats in Chapter 8.2. This confrontation is only required when AI systems were used in the creation process of the presented work.

• On the one hand, in the methodological approach section of the main part is to describe which AI systems were used. The specific function (e.g., idea generation, transcription, or data analysis etc.) of the AI system in the research process must be explained.
• On the other hand, the methodological use of AI systems is to be critically evaluated in the reflection (see Chapter 5.1.2). It is to be discussed to what extent the use of AI systems was effective for completing the task and where the limits of machine assistance have become clear. Based on this insight, an outlook on the completion of future tasks is to be deduced.

The PAIR framework of King’s College serves as the basis for this two-stage reflection.

Chapter 8.4 Citation Worthiness

AI systems are not worthy to be cited in written papers. However, three exceptions apply for generative AI systems.

• For figures and graphics created with generative AI tools and used in the work, it must be explained within the heading that it is an original representation created with the help of AI. The specific AI model is mentioned in the subsequent source citation, including the reference in the reference list and inclusion in the list of aids.
• If the author sees it fit to include the output of a prompt (question to a chatbot) without significant intellectual effort (without significant own contribution means that the contents/texts taken over analogously do not originate from one's own knowledge or thoughts), this must be marked with an in-text-reference (e.g. OpenAi, 2023a) and listed with the corresponding reference in the reference list. If this is not done, it constitutes plagiarism. This is an act of deception and is graded with a mark of 1.0 in accordance with the Bachelor's and Master’s study programmes Regulations. For purposes of traceability, the use must also be listed in the list of aids including a remark explaining why the author decided to directly adopt the output.
• A direct (verbatim) adoption of text outputs from generative AI systems is only allowed if these were used only as formulation and spelling aid (optimisation of self-written content). Further, the adoption is allowed if research questions, these, text fragments were developed with the help of AI and subsequently checked, improved, and adapted by the author themselves. Direct adoption does not need to be marked as a citation, but the use of AI system must be listed in the list of aids.

Chapter 8.5 Copyright and data protection

Copyrighted documents (e.g., PDF articles from databases or the library) or confidential documents (e.g., personal data on interview participants) must not be uploaded to AI tools. Uploading such documents may constitute in a violation of copyright, license agreements, or data protection, and needs to be examined carefully for each case (e.g. technical circumstances, licences such as open access, explicit consent of interview partners, and FADP or GDPR). Further information can be found in the Factsheet on the Use of Generative Artificial Intelligence Applications and in the Overview on the Use of Full Texts in Generative AI of the ETH Zurich.