Choose timezone
Your profile timezone:
AI Discussion Forum
→
Europe/Stockholm
A5:1003 (AlbaNova Main Building)
A5:1003
AlbaNova Main Building
29-09-2026: Codes of conduct
- Note taker: Renske
- Discussion leader: Renske
- Topic introduction: N/A
Summary:
First of all: feel free to take inspiration from the material published here, and share what you have yourself!
We started by having Sofia and Anna share their current codes of conduct.
Sofia:
General setup: use is allowed, but we need to be ethical about it. Transparency and honesty is integral. AI use should not hamper the learning, e.g. only automate with AI once you've done it yourself. Use for commercial and personal projects is currently prohibited, though she's not entirely sure on the prohibition on personal projects. Sensitive data (e.g. CVs, meeting recordings) cannot be uploaded to external models under any circumstances.
Anna:
Emphasis on confidentiality of collaboration data & protection of personal data, which is also enforced by an addition to the Claude prompt for accounts within the organisation. Commercial projects are a no-go, but personal projects aren't as explicitly prohibited as in Sofia's code of conduct. Reasoning being that you're learning skills when doing personal projects so it can help you with your work in the end. However, personal projects shouldn't cause you to run low on tokens, i.e. they should not interfere with your work. No real guidelines on how to use compared to Sofia's code of conduct, Anna is first going to see how students use it and then they can sit together and come up with usage rules.
The question of what to do with students producing large amounts of output came up, since at least one person had experienced that bachelor/master students had a tendency to just produce output and then send it directly to the supervisor.
- Ownership and accountability are even more important now
- When starting with bachelor/master students, set expectations, have the conversation about what is a useful supervision discussion and about transparency. They can be very receptive to these discussions, and sometimes even glad that expectations are clearly set for them.
The next question was when is it appropriate to use AI in the learning journey anyway? Anna is of the opinion that PhD students should definitely learn how to use agentic AI, but master students should still do things themselves. Sofia has the "do it once" guideline, so only automate things that you yourself have already done so that you are better equipped to judge the AI's output.
There was an inciting incident that prompted Anna to make a code of conduct: someone used the team plan for a commercial project (that didn't benefit SU). There are some indications available to team leaders within Claude that can help diagnose these things (ranking of how many tokens people use, see what tools are used by different people), but you cannot see what people are actually doing. Sofia didn't have an inciting incident, but was motivated to set a quality standard, and also make this standard searchable by putting it on the group website. This should help people also when they move on to a next job, since the hiring team can then directly see from what kind of environment they came. And having a code of conduct helps people trust each other's output.
The final question we discussed what amount of AI usage is acceptable in writing (especially students writing reports, theses, articles). Anna had an instance where a Japanese student was using AI (a year ago) to write, and it was producing good-looking but bad-content text and the student wasn't checking the physics and so they had to iterate a lot. To first order, the answer is that learning to write scientifically is a skill and so it shouldn't be offloaded. (also because it teaches you how to think scientifically). However, with Sofia's "do it once" guideline in mind, once a student has shown that they can write high-quality text themselves, then maybe some amount of AI automation is okay? Another scenario that came up is that sometimes when writing an article, supervisors will just rewrite large pieces of text without actively including the students (because the priority is getting the article out). If this happens, what's the difference between rewriting as a supervisor or if AI rewrites it? Difficult question, but maybe the difference is the weight behind the rewrite: a student can trust (or is supposed to trust) the supervisor and can use the rewrite as a learning opportunity, whereas with AI we are teaching the students to always question its output.
It would be good to host another meeting with the wider OKC audience to get more team leaders thinking and formalising these things.
There are minutes attached to this event.
Show them.