AILSA CHANG, HOST:
Substack recently unveiled a new feature. Users of the newsletter platform can now scan any post of a hundred words or more for an estimate of what percentage of that piece was AI generated. This new feature has polarized the Substack community. ALL THINGS CONSIDERED producer Zephyr Weinreich talked with Substack creators about the change.
ZEPHYR WEINREICH, BYLINE: When Sam Kriss found out about Substack's new AI-detection tool, he celebrated.
SAM KRISS: I was straightforwardly in favor of it. Obviously, the technology isn't perfect, but I think it's good enough.
WEINREICH: Kriss is a freelance writer based in the U.K. His popular Substack, Numb at the Lodge, has covered culture and politics for four years. He says this move is a good thing for readers' experience of the site and for Substack.
KRISS: The only way that any platform is going to be able to survive into the medium future is if it can guarantee that the things people are reading were actually written by another person.
WEINREICH: Others doubt that this is a step in the right direction. Sam Illingworth is a Substacker (ph) and professor based in Scotland. His work focuses on thinking critically about when it's most useful to use AI and when it's not. He fears that this new update could enable what he calls a witch hunt.
SAM ILLINGWORTH: It creates an environment of suspicion, and, you know, the line always uses, you're guilty until proven human.
WEINREICH: Illingworth says the risk of false positives should be taken very seriously. Because Substack allows users to subscribe and pay their favorite writers directly, those who are inaccurately flagged as AI users could stand to lose a portion of their income. The change is a partnership with the AI-detection software service, Pangram. Illingworth says the way Pangram detects AI - by looking for phrases that are predictable and structured - makes it far more likely for certain marginalized groups to be mislabeled.
ILLINGWORTH: Anyone who's learned any other language as a second language, you learn it to be much more structured than somebody who is a native speaker. So this means that AI-detection software is biased against non-native English speakers. It's also the case for people who identify as being neurodiverse.
WEINREICH: Illingworth agrees that pure AI content is bad for Substack, but he thinks a percentage-based system doesn't provide the whole picture of how a person might be using AI.
ILLINGWORTH: Imagine I had an entire idea myself, and then I ask AI to riff with me on that idea. What percentage of that is AI?
WEINREICH: Supporters of the new feature argue that people often use AI to bypass the most important parts of the creative process.
SARAH FAY: Editing is voice. That's when we're doing style. That's when we're looking at word choice and syntax and how we're putting our sentences together.
WEINREICH: Sarah Fay teaches creative writing at Northwestern University and is an active user of Substack herself. She says she was excited about the new feature, so the backlash surprised her.
FAY: I just didn't realize so many people would feel caught out. There's just nothing wrong with using AI, but I do believe in transparency.
WEINREICH: Substack CEO Chris Best tells NPR he doesn't think using AI in the writing process is inherently bad, but he worries about too much AI slop.
CHRIS BEST: Even creators who are using these tools extensively have the same fear that if all of the platforms they're on get flooded with low-effort AI slop, it'll devalue their work.
WEINREICH: For those worried about the changes, Best says Substack isn't taking a knee-jerk approach to AI.
BEST: I think we're going to find that there are cases where people are totally fine with reading things that are heavily AI-assisted, but that - there will be new norms and new expectations. In general, creating some transparency and some agency on behalf of the reader is actually good for everybody who's doing high-quality work.
WEINREICH: And whether they're happy with this update or not, Substackers agree that conversations like these are crucial as AI use grows for readers and for anyone who wants to publish their writing.
For NPR News, I'm Zephyr Weinreich.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.