Adam Serwer has an excellent piece on Texas’ censorship on its college campuses. The whole thing is worth a read, but this part seems relevant (boldface mine):

Texas Tech’s approach to censorship is notable for being both onerous and clumsy, in part because it relies so heavily on artificial intelligence…

The process still involves layers of review from administrators and university leaders, but the faculty’s role has been replaced in part by the censorship bot, which scans course materials considered potentially controversial…

For many years, Sara Spurgeon has taught a literature class focused on social justice and the environment. She has often assigned Ta-Nehisi Coates’s Pulitzer Prize–winning memoir Between the World and Me, but this time, the book was flagged by the bot as promoting “the belief that meritocracy is a construct of oppression designed to exploit Black bodies,” and it therefore “directly violates the system’s Prohibition of Prejudiced Advocacy.” (The word meritocracy does not appear in the book.) Creighton has insisted that “studying Black history, discussing the realities of racism, and treating every student with respect are not prohibited.” Spurgeon told me that works by science fiction and fantasy authors were also flagged.

In other professors’ syllabi, the bot flagged the Pulitzer Prize–winning play Angels in America by Tony Kushner, essays by the Mexican American MacArthur genius grant recipient Sandra Cisneros, and materials related to a class about the Holocaust (that professor was told to drop material regarding Nazi persecution of LGBTQ people).

Putting aside the ideological slant of the new rules, the bot, like many large language models, seems to “hallucinate,” or make mistakes. The AAUP lawsuit lists texts banned for their “sex-based ecofeminist framework” and “gendered ecofeminist frameworks.” These phrases might sound persuasively like academese, but the academics behind the lawsuit had never heard of them; these concepts “do not exist,” the complaint says. Matthew Pehl, a history professor at Tech, told me that he knew several people who’d received reports from the system that were “plainly written by an AI that had hallucinated content or confused content from one class to another.”…

“The fact that the system AI doesn’t work, that it makes inaccurate pronouncements—that’s a feature, not a bug. They don’t care if it’s accurate or not. They just want to be able to say, Look, we’re making sure that your children are not being exposed to these horrible ideas from these radical professors,” Barenberg said. “A key part of how authoritarian systems often work is that, through the vagueness and sending up signals and vibes, they’re essentially saying, You know what not to do, but being deliberately vague about it.”

This can’t all be laid at the feet of AI: other Texas institutions of higher education are also being censored, but with an artisanal censorship process*. But like many real-world implementations of AI (not all!), it is taking a bad thing, in this case state censorship, and making it worse, both in terms of quantity and accuracy.

BuT It’s a mYsTeRy wHy sO MaNy pEoPlE DoN’T LiKe aI!

*Including the HBCU Prairie View A&M, founded in the Reconstruction Era as a segregated college, and which might not even be able to discuss how it literally came to be.