STEM Update #11: Generative AI and Merit Review
Tuesday, January 9, 2024
Context: In my role as division director of IIS, I’m sending out a short message to the IIS mailing list on the Second Tuesday Every Month (STEM). Here’s the installment for January 2024.
Two days after my most recent STEM, NSF announced guidance on the use of generative artificial intelligence (GAI) technology in the NSF merit review process: https://new.nsf.gov/news/notice-to-the-research-community-on-ai . I wanted to share some information about that announcement.
The merit review process is NSF’s pride and joy and so this announcement is kind of a big deal. Or, to say it another way, it took us a year from when it became clear that ChatGPT was having a big impact before we felt ready to share our perspective, because we don’t take this kind of thing lightly. NSF is a very consultative place so everything is discussed in detail before anything visible happens. Regardless, I was happy to be a part of the process of putting it together and I am very happy it’s out there now.
Here’s what I think are the highlights:
Reviewers shouldn’t enter information from proposals into GAI tools. This rule stems from existing rules on confidentiality. Reviewers who enter proposal information into a cloud-hosted GAI system relinquish control over who can access it, and that’s not ok. So, at present, uploading a proposal into a chatbot to get help with reviewing is strictly forbidden. (Yes, we kind of finessed for now the deeper question about using GAI to aid in forming assessment of proposals. For now.)
Reviewers can apply GAI tools to publicly available information. For example, abstracts of funded proposals are fair game. We could imagine it being useful to generate summaries of publicly released information or quick reminders about scientific topics.
Proposers can use GAI in the development of the proposals, but they are expected to describe, in the project description, how it was used. This rule stems from existing rules around the inclusion of text in proposals not written by the proposer—including students and professional grant writers. (I hadn’t realized that this kind of disclosure existed. I know some institutions offer proposer writers help in composing their proposals. This rule applies to them as well. But I like the idea that getting help with grammar and flow is more broadly available now!)
Proposers should be careful with machine-written text, however. The use of GAI-generated text carries risks in terms of accuracy and originality. Existing rules on misconduct and plagiarism are still in force, so delegating to a GAI the task of generating an original idea could very well land you in hot water.
We plan to continue to monitor the landscape closely. As always, I welcome your feedback.
Until next time!
-Michael