Guideline for the Use of AI in Reviews

10th International Conference on Applied Category Theory • Cambridge, Massachusetts, USA • TBD

Draft Feedback and discussion welcome.

We acknowledge that AI assistance can be a very helpful tool for reviewers. Given appropriate use, it can significantly improve the quality of reports. While we generally do not want to interfere with the reviewers' freedom to make their own choices of tools, we feel that it is important to spell out our expectations regarding the use of AI for reviews in particular.

  • Reviewers steer the conference by selecting what is deserving of the community's attention. At this time, exercising good taste is a task that requires human judgment and cannot be delegated to AI. Therefore we ask reviewers to refrain from using AI as a decision-making tool, and we urge them to be mindful about the biases that might creep in from its output. In particular, we encourage reviewers to engage with a submission themselves first before querying an AI.
  • Reviewers must handle submissions confidentially, which in particular means that an upload to an external cloud-based service is not acceptable. This does not prohibit the use of self-hosted AI tools. The use of external AI tools for conversations that do not compromise confidentiality is unproblematic.
  • PC members maintain full responsibility for their reviews, regardless of how they are produced.
  • While we adjust to the entrance of AI tools in our discipline, we encourage reviewers to disclose AI usage in their reviews. Benefits of disclosure include transparency and avoiding the uneasy feeling of wondering how and whether something might be AI generated. However, this decision is left to the reviewer's discretion.