The AI Writing Apocalypse is Here
The writing is on the wall, but you can't tell if a human or computer wrote it
A non-movie post for today.
When last I wrote about ChatGPT, in that long-ago dark age known as this past April, the viral spike in ChatGPT had hit and the program “changed the world.”
Now AI writing in general, not just ChatGPT, is more alive and mainstream, with the hype dying down. What else has changed in the last several months?
Everything.
It has changed so much that I am now putting the disclaimer here on Substack and on Twitter that “no AI tools are used to write these Substacks, ever.”
And you have no way of knowing if I’m being honest.
The Teaching of Writing is in an Apocalypse Now
Let me address all teachers out there, any teacher in any school at any level.
You have no way of knowing for sure if a student heavily relied on AI or not.
The key phrase is “for sure.” Epistemology, applied to the Internet, is now a trainwreck. There is no way to tell the real from the artificial, fake, or computer-generated. At least not with good fakes.
Your hope is in two things. One is in AI watermarks working. The other is in your own eyeballs. If you witness a student writing anything, save that and use it as a baseline for anything they turn in later. You will have to compare the two, carefully.
As for me, I now don’t know what is computer-generated and what is not.
That means, as a teacher, I cannot give take-home writing assignments anymore, unless I’m fine with 20-100% of the writing being done by AI.
Two problems are happening: 1) decent writers are being flagged as AI-writing by AI detectors, when in fact they wrote something themselves; 2) writers of any caliber, including your less educated, low test-score types, are capable of producing perfect professional prose, even without knowing it.
AI detection software is a total failure. It’s a castle wall with a hundred holes in it. There are no strong defenses against the offensive onslaught of AI-writing programs, which still grow exponentially in ability.
So let’s repeat: no homework assignment is guaranteed to be created solely by a human student anymore. You either proctor an essay in person, or you risk the hazards of trying to tell AI from non-AI writing, which you aren’t going to be able to do.
At this point, AI-writing programs are so good at well-crafted prose that even students who are incapable of putting sentences together can use them to create perfect punctuation, perfect structure, and perfect wording.
Moreover, they can use programs to try to “wash” their AI prose of the sheen that AI detectors would notice. The result is that my school’s AI detector, Turnitin, is a failure. It’s so cautious about false positives that even if a student gets an 87% AI score, it will warn you about the likelihood of false positives. What good is a detector that can detect nothing at all?
I have seen the wildest things this year. Students go from ordinary undergraduate writing to turning in writing that is in the top 0.5% of prose stylists I’ve ever seen from a student. How? I rarely know. I tell these students that they have become the greatest undergraduate stylists of the thousand or more students I’ve taught, on par with the abilities of those who go to upper-echelon grad schools or who become professional writers.
The problem is obvious. The AI tools can craft awesome writing, just in terms of grammatical exaction and higher-order vocabulary. But they aren’t helping anybody with reading comprehension or critical thinking. Poor students can be excellent writers, via cheap AI, but not any better readers or thinkers.
The consequences for any writing class are these: you either shift to a “how to think” class, and/or you teach them how to use AI-writing more proficiently than they already can on their own.
All language teachers, and anybody relying on writing for assessments, will have to change. Some of them will have to change what they are doing entirely.
The Consequences for Substack
You might’ve heard about the Sports Illustrated scandal. In short, SI was using computer-generated text and creating fake authors for AI-written articles. When they were called out on it, they deleted the evidence.
This is called, in the article I linked to, “unscrupulous.” Which is true, I think, because it’s a fraud — appearing to be human, and pretending to be human, when it’s not.
Still, the day is here when a journalist, a Substack writer, or somebody writing for a living can just mine AI-writing programs and create their own content. And nobody would know, again, because the detectors are useless at this point.
If I were unscrupulous, I would try to start a Substack empire. I could churn out dozens, maybe hundreds, of newsletters quickly. See which ones hit and stick, make more of those, and off we go.
Obviously, the threat is that I would get caught and called out. But I would not have put a disclaimer in my newsletter, “No AI writing ever,” so caveat emptor.
That is why I’m doing the disclaimer with my newsletter — “no AI writing ever.”
Now I don’t mind computers aiding us as the tools they are, and God knows I’ve needed Microsoft Word’s Spellcheck and red-squiggles features to correct many keyboard errors. I have eternal trouble spelling “conscious” and “conscience” correctly. I misspelled “caveat” above, and my Grammarly extension caught the error. So you could say I do use a form of AI. But that’s not what I’ve been talking about in this piece, of course.
Anyway, go ahead and use the advanced AI programs as you wish. It will, perhaps, write novels and scripts for you eventually. At the moment, it can write a decent, readable essay. If you are enterprising, you generate the essay, you tweak it yourself, and voila!, a newsletter.
We aren’t going back to the guarantee that writing is by humans. What the total effect of that is on journalism, education, and the like remains to be seen. I of course think it’s apocalyptic — totally new and totally different. Maybe you all are fine with reading 60% content from a computer? I doubt that’s what people want now on Substack. That might change.
But, in my opinion, the situation is now immediate and career-altering for those of us teaching writing.
And I’m trying to distinguish myself, as a writer, by saying “no AI ever.” That might be as vain and short-lasting as those late ‘70s albums that labelled themselves with “no synthesizers ever.” Queen had those, and then they of course caved to the synths.
We’ll see if I cave to the synths as well.
P.S. “Apocalypse” means massive change. It’s not always a disaster or total destruction. It is, though, a near-complete change in vision and practice, because of material and/or political changes.
P.P.S. I am aware that “AI” is largely or solely computers that utilize advanced pattern recognition, and not AGI. Advanced pattern-recognition is what I mean here. “AI” is just easy shorthand.