Evidence Synthesis Methods Interest Group

Hi everyone,

I had a faculty member reach out with a question about uploading abstracts to an AI model like ChatGPT to help with screening abstracts for a scoping review. They are looking to use python and a spreadsheet to communicate with the AI model. They tried it with a local model that wasn't connected to the internet. It worked, but they'd like to use a more powerful version to make it quicker and easier.

They are wondering if it would violate copyright to upload abstracts to a tool like ChatGPT. There are a number of articles out there that discuss using ChatGPT for article screening. So, it looks like others are doing it. Does anyone know if there are copyright issues with this? I need to dig into the research more, but I thought I'd see if anyone has heard this question before.

Thanks,

Kathleen

------------------------------
Kathleen Flynn
Biology, Chemistry, Physics, Mathematics & Statistics
and Atmospheric and Environmental Sciences Librarian
University at Albany, Science Library
518-437-3943 | kflynn@albany.edu
------------------------------

There is no settled law that I know of, but there is a lawsuit from the New York Times against the ingestion of copyrighted content. 

As an addendum, I'm not sure why you'd want to use something like ChatGPT for this, rather than one of the systems trained to do this, like Elicit, Scite, or the Scopus AI if you subscribe (or I'm sure there are more.) Those systems are set up to do article screening and have better relationships with the publishers.

------------------------------
Rebecca Hedreen
Head of Research & Instruction, Life Sciences & Distance Learning Librarian
Buley Lib-Periodicals
Southern Conn State University
She/Her/Hers,They/theirs
------------------------------

Hi Kathleen,

We are currently doing this in a few of our projects as well. Since the researchers are using GPT via the API, the data they give it will not be used to train their model. Beyond that, whether there are other licensing restrictions that prohibit use of titles/abstracts with large language models (even closed ones), I don't know. However, in my view, this should be an allowable use since it is not being used to develop a commercial product that would compete with the publishers or impact their ability to sell their content. In some ways, I have a hard time seeing how this is much different from any other use of titles/abstracts for research purposes, as long as OpenAI isn't ingesting the data and training their model on it. 

That's my two cents but then again, I'm certainly no copyright lawyer!  And like you pointed out, many people are doing this now and publishing about it, and I'm not sure the publishers will be able to shut this down at this point. I think the NYT case is a little different since that's about OpenAI using their content to train the model.

Sarah

------------------------------
Sarah Young
Liaison Librarian
Carnegie Mellon University
------------------------------