We are adapting our policy in response to the emerging AI-scape, with a curiosity bound by our commitment to PREreview’s core values.
At PREreview, we are deeply committed to celebrating our shared humanity. We are also deeply committed to our core values, including accessibility, diversity, equity, inclusion, justice, and openness, and to learning to do better when we fail.
We are navigating the emerging AI-scape with a curiosity bound by these commitments.
Recently, we’ve found it necessary to revisit and expand our AI policy (currently living in our Code of Conduct) in response to different kinds of community member behavior while authoring reviews assisted by AI. Sometimes we find a review that seems to have been authored with AI, even though it’s not declared; other times we notice that reviewers are publishing multiple reviews very quickly while declaring their use of AI appropriately. Given the variety of these emergent cases, we want to maintain an AI policy that makes it clear when someone has used AI without being punitive. We expect our policy to evolve even more over time, and we’re committed to sharing these changes—and our thinking about them—as openly and as often as possible with you.
First, we believe and expect that all reviews should be authored by people.
When this is not the case, anyone publishing a review on PREreview.org must declare the use of AI if they use it to help generate new ideas or overarching organizational structures for a review, but reviewers should never feed an entire research object to AI and ask it to produce a review for them to subsequently publish on PREreview.org.
It’s important that preprint authors, journal editors, and other community members reading our reviews know whether or not a review was written by a person or a machine so that they can apply their own measures of authenticity and trust to those reviews.
To make it especially clear when someone has published a review declaring the use of AI, we will take the following steps to differentiate it from a wholly human-composed review:
- The review will carry a new “AI-supported review” label near the top of the published review on PREreview.org. This is meant to alert readers to the use of AI ahead of their consideration of the review.
- The review will be licensed CC0, or under the public domain, since we cannot accurately attribute something AI “writes” using uncredited and uncompensated training materials taken without consent from other humans.
- The review will not be added to its creator’s ORCID record, as we cannot accurately credit them as the author of the review once AI is used to generate its content.
- The review will carry the usual affirmation of the use of AI under the main body of its text, where this kind of declaration and the declaration of any competing interests already appear.
Second, we understand that bias is an inherent part of the current scholarly publication ecosystem. The kinds of bias that show up in this space include, but are certainly not limited to, geographical, linguistic, ethnic, and racial bias.
Therefore, we do not currently require reviewers on PREreview.org to declare the use of AI tools for copyediting or grammar checks, as we do not want such declarations to bias preprint authors, journal editors, or other community members against reviewers publishing in English when that is not their preferred language. While this policy may change, reviewers on PREreview do not need to declare the use of AI for checking grammar, spelling, and style in a sentence or polishing, for example, the transitions between complicated clauses or sentences.
In cases where we at PREreview believe generative AI has been used without it being declared (for example, when an author or group of authors publishes multiple, lengthy reviews with the same structure within a short time frame), we will investigate further as with any CoC violation and act on our findings accordingly.
We also believe that you should never feed an entire preprint or dataset to an AI model and ask it to do your review for you, regardless of the prompts you use. This creates the risk of adding that text to the AI model’s training corpus without the human author's consent or any recognition for them. It also generates review text that may have been similarly stolen from human authors without giving them credit or compensation.
Ultimately, we value the people engaged with open peer review the most. If an author wants an AI review, they can prompt one for themselves. Our AI declaration is meant to make sure people understand who has authored a review, not to enable the non-stop publication or spamming of AI reviews.
To summarize:
If you use AI in a PREreview, you must declare its use whenever you:
- Ask an AI model to review a preprint, dataset, another kind of modular research output, or any part thereof for you in any way whatsoever.
- Ask an AI model to structure an entire review for you in any way whatsoever.
- Use an AI model to review a preprint, dataset, another kind of modular research output, or any part thereof for you if you lack domain knowledge or other forms of expertise and experience in its research area.
- Use AI to rapidly review numerous preprints, datasets, other modular research outputs, or any parts thereof, whether they are requested or not.
- Use AI to clean up or translate language across all of your reviews. Please feel warmly welcome to publish your feedback in a language you feel comfortable composing in, instead of “outsourcing” your review production to AI.
You do not need to declare the use of AI if you:
- Did not use AI to come up with new ideas or to compose, structure, polish, or translate the entirety of your review.
- Use an AI service to clean up or polish a few sentences of complex information occasionally in your review, while preserving human agency to critically decide over the obtained outputs.
Please remember that while using AI does not necessarily violate our Code of Conduct, failing to declare its use is a violation. If PREreview suspects AI has been used in a review that doesn’t declare its use, staff will initiate a Code of Conduct investigation and take action to address any violation they find likely to have occurred.
While we don’t want to contribute to the institutional and systemic biases already present in scholarly research and communication, we think it’s essential to preserve and recognize the human connection, expertise, mutual learning, and skills that go into all kinds of communication, including academic, research, or scholarly communication. We think it’s important that we, as readers, understand when a review is human-authored or machine-authored, regardless of the quality of the review. Furthermore, we want to celebrate, reward, and show gratitude for people and the labor they contribute to bettering research for the benefit of all.
It’s also important to acknowledge that PREreview uses some AI tooling too. For example, we use some AI-assisted translations with human approvals for our localization and translation work, and we use AI to summarize and share details about published reviews and requests for review on our community Slack. We’re also using AI while developing our upcoming matchmaking features to help reviewers find relevant research to review.
AI can support us in moving more quickly on these kinds of tasks than our small team could alone. We focus on using AI in clearly-bounded and reasoned ways that connect people with people and people with content that is most relevant to them. Ultimately, PREreview’s use of AI will always prioritize human agency, community stewardship, and collective benefit from knowledge evaluation and dissemination.
Finally, remember: our AI policy is evolving and may continue to change - even frequently. We will continue to navigate those changes alongside you with as much accessibility, clarity, inclusivity, openness, reason, and transparency as we can.
If you have a question about using AI for any reason in a review, please let us know at help@prereview.org so we can work through any doubts you have before you publish your review.
Stay connected
You can follow us on Mastodon, BlueSky, LinkedIn, and YouTube. You can also keep up with all the latest news by subscribing to our newsletter and joining our Slack community.
