In our January “how can we help you?” thread, a reader writes:

What should I do if I think my co-referee on an article used AI to write their report?

I’ve lately come across reports from my fellow manuscript reviewers (they’re visible to me) that stand out, and in exactly the same ways. Almost like a template. They are over-the-top glowing in their praise of the manuscript, unlike anything I’ve seen before (I’ve been refereeing for years); they spend numerous paragraphs – more than half – simply summarizing and at times quoting the MS (also something I’ve never seen before), and they all conclude with a suggestion that amounts to a minor word-change in an insignificant section or section title. It’s uncanny, when you see multiple versions of this. It’s almost as though they’re designed to (a) recommend publication in a way that forecloses the possibility of revisions coming back for further review, but (b) also in a way that showcases detailed (if uncritical) engagement with the manuscript.

I’m not alleging my co-referees gave the task over to AI completely. My suspicion is that they quickly got the sense the piece is on the level and, having satisfied themselves to that effect, didn’t want to invest more time. I say this in part because in all of the cases that stood out to me, the argument and scholarship had significant – but not immediately obvious – problems, but the style was impeccably expert. Enough to look exactly like good philosophy if you don’t have the time to read closely. Or so it seemed to me.

But I’m not sure what to do about this. Should I report my suspicion to the editors, and risk harming the reputation of professional colleagues on less than ironclad proof? Has anyone else noticed anything like this?

I haven’t noticed this as of yet, but most of the journals that I referee for don’t share referees’ reports with the other referees. In any case, if the OP suspects something, I don’t see anything amiss in them reporting their suspicions. If the editors want to look into it further, they can.

What do readers think? Have any of you encountered (or suspected) this?

Posted in ,

5 responses to “When you suspect referees used AI?”

  1. Anonymous

    This is quite awkward. On the one hand, I think the bar for calling someone out like this should be pretty high – you need strong evidence. But, on the other hand, we need to alert editors to misbehaviour (and the use AI for writing referee reports would count as such). As Marcus says, it is probably worth alerting the journal editor, but realize that one way journal publishers might respond is to no longer share access to others’ referee reports. (I review for a number of journals that allow you to see what the other referee wrote – I have found this feature generally reassuring … there is more agreement on details about a paper than one might think, given all the bitching about refereeing).

  2. Anonymous

    I believe this is probably happening all over the place. However, I almost always give multiple paragraph summaries of the paper while quoting the MS, so I’m not sure that’s a reliable indicator.

    I think it’s on the editors. Hopefully they have enough good sense to take that referee off their lists.

  3. Michel

    I think you should share your suspicions. As a referee, you’re better-placed to do this than the author is.

  4. Anonymous #3

    I also think that you should say something to the editor. I’d be livid if my paper was reviewed (even in part) by genAI, much more so than if it were desk rejected. While I recognize that not everyone feels the same, I think there are enough concerns about using genAI in refereeing (/other aspects of academic philosophy) that any such use would at minimum need to be made clear to authors & potential referees before submission.

    It is unfortunate that this might make us all more skeptical of our peers, like Anonymous #2 above, and we should keep that in mind. But the problem is the use (or unacknowledged use) of genAI, not the bringing up (potentially) problematic reviews.

    As an aside, I also like seeing other referees’ reports. In addition to the reassuring aspect mentioned by Anonymous #1 above, it can give insights into the thoughts of folks in overlapping subfields / with distinct focuses (assuming they are thoughts by folks, vs outputs by LLMs trained on stolen thoughts by folks…).

  5. Anonymous

    I would politely ask the editor to read it, in order to see if they agree with your assessment. Relying on ChatGPT to review papers is unacceptable, and it might destroy the peer review system.

Leave a Reply

Discover more from The Philosophers' Cocoon

Subscribe now to keep reading and get access to the full archive.

Continue reading