Skip to content

AI Policy

This is to clarify my subjective rules on use of (generative AI) for my courses and research work. tl;dr: scientific hypotheses and interpretation as well as presentation structure (as text or figure) needs to be original work by you as human. It can be improved or being iterated over with the help of AI.

Feel free to ask clarifying questions anytime.

Currently, my mental model is: "If I delegate a task which I can NOT do myself on the computer without AI AND if AI was a person, would it be good scientific practice to have them do this part of my work without attribution?" If the answer is no, you can't use AI for this. If the answer is "Maybe, I'm not sure.", this probably means you cannot use AI for this. Scientific integrity also requires that you can defend any aspect of your work: this means that you need to have understood everything that you claim as your own. To sharpen this statement, please find a list of possible questions and their answer based on analogy.

Teaching

  • Can I use AI to understand the task better? Yes, sure: it is akin to asking your fellow student. Just make sure to instruct the AI to not spoil the solution as it is implicitly assumed when discussing the task with humans.

  • Can I use AI to solve my assignment? No, not at all - it does not help the learning process anyway. I will undertake no attempt at testing this, though. The exercise sessions are tailored around interactive discussions where the prepared solution does not help.

  • Can I use AI to ask questions to the lecturer/tutor? Yes, since you could also have asked your colleague to sharpen the question. Usually, the rough question is much more elucidating about where exactly the problem lies, so better ask the question as it is in your head.

  • Can I use AI to repeat the explanation from the lecture? Yes, since you could also asked your fellow students. Better use a book (as it follows a well-established teaching path) or ask in the lecture so I can explain again. Chances are that if you are stuck there, others are as well. AI may give inadequate answers if they are incomplete or misleading.

  • Do I have to disclose AI usage? Typically no, since tasks are to be done without AI anyways and all other usages discussed before are permissible without disclosure. If you want feedback on your AI usage and how to be most effective, feel free to ask questions and discuss.

  • Can I use AI for presentations? Only for slide generation or beautification, not for structure or content curation. Using a pretty template has always been permissible, and we learn visual design through imitation. Having your friend produce the content of your presentation never was permissible in the first place.

  • Can I use AI in exams? No, all my exams are closed book exams without tools.

Research

  • Can I use AI to generate code I use for research? Yes, as long as the design of WHAT and HOW to do things is done by you, since this is the critical aspect of solving an issue. "Coloring in" as in implementing a plan you devised is not necessary. Analogue: you don't evaluate an integral by hand, you have the computer evalutate it, but you choose which integral and whether to do this symbolically in a computer algebra system or numerically.

  • Can I have AI suggest causes for bugs in my code? Yes, as long as you judge wheter this indeed is a bug. AI struggles with unseen concepts, i.e. in research almost everything is unseen. AI will see frequently see code as errorneous solely because it does something differently. Analogue: static code tests or linters are standard practice and regular tools in development.

  • Can I have AI suggest research questions? No, absolutely not. Hypothesis derivation and the testing thereof is a key intellectual contribution to scientific work. AI can do part of this, but it is lacking both context and direction. Analogue: You cannot ask your friend how to solve a problem and then claim you figured it out yourself.

  • Can I use AI for ideation ("brainstorming")? Yes, for e.g. fuzzy literature search ("Which scientific topics use method XY?", "What is the method called that does ...?", "Where could XY be helpful?"). Not ok for research questions: "How to best build a model to predict ...?"), but ok for research ("How did people build a model to predict...?") Analogue: this is just accelerating searching original sources.

  • Can I use AI for improving my communication? Yes, finding typos, unexplained symbols, ambiguous phrases, spotting inconsistencies between texts written by co-authors or your past self, all fine. Analogue: asking somebody whether this text is legible and clear is common practice; style manuals exist already.

  • Can I use AI to write my thesis / manuscript text? No, not at all. If the bullet points you put in the prompt would suffice, all papers would just be bullet point lists. Communication is not piling factoids, it is organising data, shaping a picture and imprinting it on someone else's brain. This adds information and communicates your personal views on the matter: this is why writing a paper is hard, because you need to phrase this concisely. This work is the sole responsibiltiy of the researcher. Analogue: ghostwriters are not accepted. Rephrasing is ok, rewriting is not (analogue: ghostwriter). Rephrasing is: "The point of the following is to show that the addition of two and two yields 4." -> "We now show 2+2=4.". Rewriting: "There are twos and there are fours and they are related by some mathematical operation." -> "Two plus two is four." (See how rewriting adds information and is more precise than the original - whenever something is added, it's rewriting.)

  • Can I use AI to typeset my math? Yes, taking a photo of notes and asking "latex me this" is fine, as long as you are the author of the math. Knowing LaTeX surely is helpful but not necessary to write a good paper. Analogue: OCR is permitted.

  • Can I use AI to improve my scientific argument? Yes, as far as it is critizing where gaps are only closed implicitly. "You show that from A follows B, but the text does not show that A is true for all real numbers.": Probably you know that A is true for all real numbers and just forgot about writing it there because its clear to you. Analogue: reading the text several times very carefully would spot this incompleteness as well. You CANNOT turn this around: "I have that from A follows B, tell me how to show A is true." would have scientific work be done by AI. Analogue: Asking a friend how to extend your math proof without attribution is not acceptable.

  • Can I use AI to improve my understanding of literature? Yes. "This paper shows that from A follows B, and states that A is true for all real numbers, but I don't see how they get there." is ok, because you later would cite that paper and use that they showed "A is true for all real numbers." and not claim that you figured it out. Analogue: you ask a colleague who already read that paper for help.

  • Can I use AI to generate figures? Yes, provided the content of the figure is your idea. Analogue: "Place this box top left, write XY in there, make an arrow to the next box, ..." is fine and common, e.g. with graphviz; on a technical level, that is what you do with a vector graphics editor, just without language. "Take this code and make a flowchart of the function calls." is fine. Analogue: callgraphs are standard tools. "Take this code and make a high-level flowchart of how it solves the problem." is acceptable. Analogue: this is re-hashing the code logic flow you already designed.

  • Can I use AI to interpret results? No, not at all. Analogue: Ghostwriter, see above.

  • Can I use AI to summarize papers? Yes, beware that AI might be wrong. Analogue: Asking a colleague to explain a paper to you.

  • AI did something wrong, is it my fault? Yes. Analogue: Regular code that has a hand-crafted bug.

  • Do I need to prove that I did not use AI? You can't prove a negative. Keeping logs is not crucial.

For even more details and possible legal aspects (e.g. regarding PhD declarations), please see here: https://www.uni-kassel.de/uni/aktuelles-aus-der-universitaet/infothek/themenfeld-kuenstliche-intelligenz-ki/empfehlungen-zum-umgang-mit-ki/kuenstliche-intelligenz-in-der-forschung.html