Response to "Fandom has a Hidden Generative AI Problem"
I have seen a whitepaper that has evidently been circulating in fan fiction spaces, and I have a few issues with it. It irritates me greatly that the article cites NYT and UN articles, but has no authorship listed on the paper. The NYT and UN are about as authoritative of sources as one can find (despite the NYT's politics, but that's another rant for another time), but wearing the garb of scholarly research through (a pitiful few) citations does not a good research paper make. It is improper to make a scholarly claim without provision for response, reply, or peer review.
The article contains several links to works of fan fiction which allegedly contain CSS directives which indicate verbatim copying and pasting from a Claude prompt. Some of them, upon following the links and inspecting the page source (unrendered HTML) as prescribed, do bear evidence of such use of Claude. Others have been edited or removed entirely (resulting in HTTP 404). The anonymous authors of this paper have provided a link to a Google Drive folder of PDF, EPUB, and HTML files.
While I am more or less convinced that these works did bear the relevant CSS marker, "more or less convinced" is not scientific, and providing copies in this way is suspect at best. It is trivially easy for anyone who knows their way around Python (or really, any other programming language or tool with a good suite of text manipulation features) to simply insert the offending text into the HTML and from there generate a PDF or EPUB. This is easily doable in the same style as AO3's code is free and open source software. A knowledgeable individual could use a clone of AO3's software and, with the correct inputs, produce identical outputs to what is available on Archive of Our Own.
Whether or not the authors of this paper are capable of doing so is an open question, but because it is open, we cannot discount it.
The correct alternative to recording evidence of something mutable and ephemeral, like a webpage, is to use an archive. Archive.org has The Wayback Machine, which will record the content and style of a webpage verbatim. For the sake of the discussion being initiated by Fandom has a Hidden Generative AI Problem, The Wayback Machine is a neutral third party.
While I have never been and will never be one to make excuses and apologies for the use of generative AI (read: please never fucking touch the stuff, okay?), this paper was not written to drive discourse about centering humans in fannish creative writing. This is a hit piece of unknown authorship targeting predominantly Heated Rivalry authors. With no formal system for sampling other works from across AO3, the authors of this paper merely suggest that the presence of Claude-generated works means that Claude-generated works are endemic to other fandoms as well. This may well be the case, but the paper does not substantiate this.
The only works which you can be certain have been Claude-generated are the ones still live (and unedited) on AO3 or backed up by The Wayback Machine prior to being edited or removed.
The purpose of a system is what it does. This process of callout papers (and resultant harrassment, despite how much the authors of the paper admonished their audience not to harrass these AO3 users) and edits/locks/deletions might convince a few socially wary Claude users to go back to writing the human way, but those remaining will likely just be more careful to hide the evidence of their generative AI use.
Put shortly, this is only making the problem harder to solve. Serious scholars will put their name (or a useful alias) on their creation and defend it. The people who wrote this paper just wanted the thrill of exposing undesirables in their fandom without needing to defend themselves.
Generative AI is bad, but a call to harrassment in the guise of scholarship is also bad.
