The legal academy recently worked itself into a lather over AI-generated law review submissions. According to one law professor writing on Twitter, the AI detection platform Pangram had flagged a law review submission as having substantially AI-generated text. This set off a round of sanctimonious back-patting about how disgraceful it is that a legal scholar would employ AI to pad out the world’s most boring account of the postmodern implications of ERISA or whatever the hell law reviews talk about these days. As though the legal academy hasn’t built itself upon judging scholars on their capacity to turn a 5-page thought into an 80-page article.
I jumped into this social media frenzy to point out that this was not a good look for legal scholarship. “The implication of this moral panic is that law can’t evaluate a quality argument on the merits, without leaning on the sweat equity of keyboard toiling.” In response, the professor wrote, “No.” Which was, objectively, very funny. Replying to the claim that professors care more about human authorship of every word than producing a well-reasoned argument with an undeniably human, yet entirely vapid, conclusory post seemed like solid self-aware comedy. I waited for the inevitable, detailed follow-up.
Instead, he seems to have blocked me.
Lawprofblawg already wrote up the risks of journals arbitrarily screening submissions through a questionable AI tool more than capable of delivering a career-damaging false positive. Those points are all valid, but let’s come at it another way: why should we care if Pangram is right?
Is it really that hard to just read? Unedited AI-generated content is, mostly, painful. But the thing about Pangram and the internal watermarks that Anthropic and Google place into their output is that it exists to flag AI writing even after someone took the time to scrub away the most obvious slop. And at that point, who cares? Either the article makes a coherent, insightful argument or it doesn’t.
Instead, law professors are exhibiting the apotheosis of AI slop brain: we can’t stand AI writing so much that we need *another* AI tool to tell us if we should hate this writing that we otherwise couldn’t identify as AI. It’s like The Crucible, except all the devout God-fearing villagers decide to marry Satan themselves so they can know who all the witches are. Reading and weighing an argument takes time, but running the file through AI to see if it contains any AI takes mere seconds! Gatekeepers are so worried about authors hitting the easy button that they need an easy button to outsource evaluation.
Part of the panic over AI submissions might well be the honest terror that no one actually remembers what amounts to a good argument. When article selection descends into rewarding an author’s established clout or helping some former right-wing SCOTUS clerks publish specious, armchair history so their old bosses have something to drop in a footnote while reverse engineering a historical precedent for “sure, presidents cancel elections all the time,” it’s understandable to subconsciously fear that ChatGPT might slip one past that substantively empty net.
The fear broke free of the legal academy sandbox over the weekend. The Argument, a publication custom-built for Pangramphiles as the writing is stylistically comfortable and the substance is consistently vacuous, argued that Pangram critics are mostly people getting caught. The article frames itself around attacking a Wired article, which is described as “bizarre,” presumably because it’s serious tech journalism as opposed to a Substack grievance. The author claims the Wired piece “amplifies a wide range of false arguments from Pangram skeptics: that AI detectors are racist (they’re not) or ableist (they’re not) or don’t work (this one does).” Each of these reductionist conclusions misses the point. She rejects the racism argument with a hand-wave that Black authors have only gotten detected because they used AI — which is circular — but also fails to grapple with the most common articulation of the racism argument, which is that it’s not so much a technology deficiency as the fact that AI detectors are catnip for racists.
Writers of color get run through detectors by bad faith actors who would never question white authors and when a score falls short, Black writers lose publishing deals and white authors get excuses. The author at least cites something for the ableism argument — a Pangram blog post saying it’s not ableist — that also wildly misses the point. Even if it doesn’t false positive neurodivergent writing, the product is designed to gatekeep neurodivergent writers who might want to use the model to communicate ideas to a broader audience. The “this one does” claim is backed by the footnote:
Many non-Pangram detectors don’t work, but many possible airplane designs also don’t work, and no one considers this a general argument against aviation; the thing that matters is whether any detection works
See what I mean about that publication being both reliably human-written and an intellectual wasteland?
In any event, that article set off some excited dunking on social media, as users grabbed clunky lines from Pangram critics to point at and laugh. One post I saw quoted in the pile-on came from William Crichton declaring “Suspicion isn’t a side effect here, it’s the product being sold.”
But beyond the “it’s-not-X-it’s-Y” construction, one of the most irritatingly prolific AI-isms out there, he’s making an excellent point. The market for this product rests on a crisis of trust. Whether it’s in other writers, in the readers secretly worried that they wouldn’t be able to identify human work anymore, or in the racists who want something to confirm for them that a minority author couldn’t possibly have done this on their own. Whatever the motive, Pangram sells suspicion to people who would rather signal than read, and business is good. Apparently, they have enough market interest that there’s a whole browser extension scoring what you read in real time. Start paying $20 (or more) a month and you never have to risk burning your virgin eyes on a sentence of potentially machine-produced writing.
Pangram will burn tokens on your behalf, so you can enjoy your moral position of not contributing to the AI economy!
Professor Panos Ipeirotis of NYU’s Stern School of Business recently started running an experiment where he makes a Tweet-sized, straightforward argument and runs it through an LLM until Pangram flags it as 100% AI. Then he watches as people point at the scarlet letter rather than address the actual argument. “It’s a fucking tweet,” he writes. “They are literally using a machine to avoid reading it.”
If you cannot read a tweet, understand the thought it contains, and engage with the argument before outsourcing your judgment to Pangram, then the slop is inside your skull.
Go read Trithemius and In Praise of Scribes. Five hundred years later, some people still think the intellectual value of a text lies in the fingers that typed it. Fucking hell.
Someone in my cursed “For You” timeline on social media posited that anyone who isn’t a good writer doesn’t have good ideas, a position that is deeply messed up. Have you met a mathematician? A lot of those folks can’t construct a paragraph. Non-native speakers work in a second language that makes it difficult to consistently convey thoughts that the median native speaker will understand. Neurodivergent people have thoughts that don’t always route cleanly into neurotypical prose. That piece in The Argument suggested that the people mad about detection are the bad writers, but it actually flows the other way: the people most worked up over AI-assisted writing are the mediocre writers frightened that someone will read a machine’s writing and say, “this… isn’t WORSE than what that guy did.”
And, to be clear, AI writing is usually pretty meh. It’s generally not terrible these days, assuming the model is advanced enough. But it’s a prediction engine reaching for the most likely next word. By design, it’s trying to write middling output. That doesn’t mean the user isn’t trying to convey something of value.
There’s an article of faith being shared amongst the boneheaded that “writing is thinking,” which people translate as the “if these fingers did not produce it, no thought occurred” logic Professor Ipeirotis flagged. But that’s not how writing works generally, and it’s not how lawyers work at all. We farm work out to juniors and mark it up. “Writing” is thinking, but “writing” isn’t the superficial act of putting the first words on paper. Writing is in the reviewing, editing, rewriting, reframing, interrogating, and iterating — including, yes, iterating with the machine if that’s your bag — until the thing says what the author actually means.
If a law review author is truly cutting all that out of the process, a quick read will expose that. If the author put real insights through a word calculator to flesh out the text, then there’s something there for the law review editors to fix. If there’s a real, valid argument buried in a distracting mess, the editors should join the collaborative process and repair that mess. Don’t rely on a bot to give you a number, read the thing and use human judgment to figure out if anything can be salvaged.
Can’t we all just suck it up and try reading? You know, so we can stay human.
Earlier: Law Reviews And AI Detection, A Letter To Law Review Editors
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news.
The post Pangram Loving Law Professors Have Reached Peak AI Slop Brain appeared first on Above the Law.