This document lays out guidelines for the use of AI in the Corpling lab. The guidelines were decided on following a lab survey on preferred policies and member concerns in November 2026 and may be changed from time to time in the lab as a result of open discussion.
Throughout these guidelines, we distinguish ethical concerns (ways in which we believe AI should or shouldn’t be used regardless of its benefits or risks for us personally) from practical ones (how we should or shouldn’t use AI for our own good).
In the former category, we are primarily concerned with fairness, sustainability and harms to society. This includes cases where AI makes mistakes (hallucinations, misinformation) but also ones where AI works as intended, but we feel that it shouldn’t be used. In the latter category, we are concerned about the potential that AI has to harm our learning and education, and again, this can apply even in cases where we think AI is doing the job we are asking it to do (but potentially shouldn’t ask).
Beyond a general awareness of the costs of using AI, including monetary, environmental and societal impacts, we identified two overarching principles that inform the points below:
- Disclosure: Using AI with integrity means being completely open about its use. If you don’t feel comfortable disclosing that you are using AI to do something, you probably shouldn’t be using it to do that.
- Level of expertise: Only use AI at the level of your expertise, even for permitted applications. This means using AI primarily to do things you already know how to do yourself. While we can consult AI to learn how to do new things, leaving them completely to an agent is both dangerous (lack of oversight) and harmful to learning.
Below we list specific use-cases which we have come to a decision about in the lab.
- To write any part of a peer review
- For brainstorming where the substantive ideas for a project are the product of AI (plagiarizing bots); but see below on allowed brainstorming uses
- Copy-pasting text written by an LLM into a paper (excluding uses permitted below)
- Photorealistic image generation – we view the risks of deep fakes seriously and take the position that ‘photo’ means ‘real’. Please use stock images if you need random pictures of objects, place or people. We do not propagate realistic images of non-existent people.
- Free-form analysis of data (but see below on coding a well-defined analysis) – allowing AI models to explore data freely to generate hypotheses allows systems to steer rather than support our research process, and in some cases can invalidate the meaning of p-values (how do we know how many possible correlations a system checked? Are we adjusting p-values for multiple testing?)
- Indirectly writing your paper by submitting an incomplete draft for review with the expectation that the model will propose the missing parts (see below on asking a model to review a READY draft)
- Use LLMs to formulate rebuttals to peer review - you should know how to answer about problems in your paper, and there may also be data privacy issues in uploading reviews to AI models
- Make slides or a poster out of your paper – these are skills and creative processes we feel are valuable; but see the allowed uses below while making your slides/poster, which still apply
- For Brainstorming – when looking for possible answers to questions of the type ‘how to best do something I decided on’ (e.g. generate pro/con comparisons of available tools, or even weight options)
- Spellchecking, grammar checking - only ever ask models to point out errors so YOU can rewrite the text, we do not let models write text and paste it in (see above)
- Literature search (NOT writing a lit. review section of a paper); AI can suggest papers the same way you could Google for papers or use Google Scholar etc., but ALWAYS actually verify the papers exist, look at them, read what you cite, look for papers the system may have missed, etc.
- General coding assistance - use of AI agents for coding is generally OK, but take care not to a. inadvertently plagiarize non-open source code; and b. work at your level of expertise (see above). The latter point is essential for learning.
- Write code to analyze data according to a human generated idea (e.g. “write a script to rank the items in a table with these columns by chi squared residuals”). See above about NOT using AI for free-form data analysis.
- Generate diagrams from clear instructions (e.g. to facilitate encoding diagrams in Latex using tikz etc. – we view this as is similar to coding assistance)
- Write a constructively critical review of a READY paper draft before submission to identify weaknesses. See above about prohibited misuse of this case to complete a partial project.
Keep in mind that we need to disclose our code for analyses as per venue guidelines (e.g. ARR), and we can’t really say “our process was to ask Claude” – refer to the principle above: if you’re not comfortable disclosing it, you should be using AI for that.