Journals and AI

Many (math) journals require the authors to acknowledge the use of AI (LLMs or otherwise) in the preparation of the paper.

Let’s agree that this is a requirement from the side of the journal that the authors advertise the tools they used. If the use of AI was unsuccessful, the authors are not required to report it. If it was successful, they have to report it (i.e., advertise it).

It is interesting that this didn’t happen for so long with LaTeX, Linux (or even Windows or MacOS), or Tex editors and their spell checkers, or computational systems like Sagemath (or even Mathematica, Maple and Matlab), or the hardware, without which none of this would be possible anyway.

As long as there is no free and open source alternative, it is corporate advertisement to mention use of AI in journal articles. Researchers act like influencers, adding in their content some reference to a tool or service they used. Influencers are either paid or given perks, when advertising a product by showing to others how useful it was when they used it. Will authors get a payment or at least a free month of usage from the so and so corporation whose product they advertised in their research paper, as obliged by the journal’s policies?

Note that acknowledging the use of computer (hardware or software) is not a new idea. Doron Zeilberger co-authored a number of papers with Shalosh B. Ekhad, his computer.

As Petkovic explains: “Since the computer was instrumental in his work, Zeilberger credited it as if it were a co-author. Sometimes he even signed papers under the pseudonym alone. It was not just a joke: it was also a gesture, a way to emphasize the growing role of computers in mathematical research.”

Funny thing, Zeilberger had problems with journals. They would not accept papers where the co-author is not human, presumably because the authors have to take responsibility for a number of things, for example about the content. The authors assume full responsibility about the content and they cannot claim that the spell-checker corrected their main theorem being for prime instead of primary ideals for example. Same responsibility holds if the authors used a computational tool, say Mathematica, that returns a false result and they publish it. Wolfram is not responsible to retract the paper, it is the authors responsibility, in the same way that it is “their result” if it is correct. Same responsibility holds if authors used AI or any other tool.

I agree with journals not allowing Prof. Ekhad as a co-author. I don’t see, except for the corporate backing, why journals went 180 degrees on that and require authors to acknowledge AI. I expect soon Claude and friends will be top candidates for academic positions having contributed to so many publications.

I just wonder if Anthropic will take responsibility for the (unavoidable, just as in non AI assisted,) mistakes in the published papers that acknowledge Claude. Clearly, nothing personal with Anthropic and Claude, except that I personally use that one.