What if the Anti-AI Writing Police Make a Mistake? (cnn.com) 46
Writing contests face "a literary version of the ruined world" in Blade Runner, writes CNN. "Anxiety and suspicion about robot infiltrators are everywhere — and so are people claiming to hunt the robots down."
Multiple companies now offer products that promise to spot the difference, with often contradictory, biased or erratic results. The makers of Pangram, an AI detection platform that has been cited as authoritative in many of the recent controversies, claim it can distinguish between human and synthetic writing with near-perfect accuracy... The company's own estimates claim that false positives — in which the tool wrongly reports that a piece of human writing was created — are 1 in 10,000 and shrinking. Peer-reviewed studies from the University of Chicago and the University of Brussels support that claim, though one 2025 study from the University of Maryland found that Pangram erroneously labeled 2% of documents as AI.
A widely referenced 2023 Stanford study that found AI writing detectors produced higher rates of false positives on writing from non-native English speakers was conducted before Pangram was created. Pangram ran an internal study using a database of non-native English texts and self-reported that false positive rate was around 8 in 10,000... Given the thousands upon thousands of pieces of writing published each day, a 1 in 10,000 chance of a false accusation — let alone an 8 in 10,000 chance — means that even the most trusted AI detector is going to be wrong, and wrong about a lot of people.
Pangram isn't able to tell users the truth about someone's writing, but it is able to introduce something much more complicated: doubt. Unless an accused writer confesses to using an LLM, there is no bulletproof way to prove an accusation, short of being given full access to someone's computer history and finding the chatbot prompts... To some people and institutions on the literary side, using automated AI writing detectors to fight the influx of AI writing is just trading one form of technological domination for another.
Pangram "has entered the public consciousness at a time of distrust that goes beyond writing," CNN notes, when people want to block "at least some of the daily overflow of misinformation and misrepresentation" recently studied by John Warner, the author of More Than Words: How to Think About Writing in the Age of AI .
But he's one of the people who believes AI-detecting tools can create a new dependence on another kind of machine. "Like all other technology of automation, that is going to de-skill people from discerning these things for themselves."
A widely referenced 2023 Stanford study that found AI writing detectors produced higher rates of false positives on writing from non-native English speakers was conducted before Pangram was created. Pangram ran an internal study using a database of non-native English texts and self-reported that false positive rate was around 8 in 10,000... Given the thousands upon thousands of pieces of writing published each day, a 1 in 10,000 chance of a false accusation — let alone an 8 in 10,000 chance — means that even the most trusted AI detector is going to be wrong, and wrong about a lot of people.
Pangram isn't able to tell users the truth about someone's writing, but it is able to introduce something much more complicated: doubt. Unless an accused writer confesses to using an LLM, there is no bulletproof way to prove an accusation, short of being given full access to someone's computer history and finding the chatbot prompts... To some people and institutions on the literary side, using automated AI writing detectors to fight the influx of AI writing is just trading one form of technological domination for another.
Pangram "has entered the public consciousness at a time of distrust that goes beyond writing," CNN notes, when people want to block "at least some of the daily overflow of misinformation and misrepresentation" recently studied by John Warner, the author of More Than Words: How to Think About Writing in the Age of AI .
But he's one of the people who believes AI-detecting tools can create a new dependence on another kind of machine. "Like all other technology of automation, that is going to de-skill people from discerning these things for themselves."
Not gonna happen (Score:2, Redundant)
Unlike AI, humans don't make mistakes. They can figure out, with absolute certainty, if AI was used in someone's writing.
No one has to worry about this scenario. /s
Re: (Score:2)
Truly no-one has to worry about this scenario. Anyone who interprets a rejection of AI slop writing as an AI classification problem misses the point.
The point is: humans have many ways to separate the in-crowd from the out-crowd, and it's not well defined what criteria are used.That's what nerds in High School always misunderstand. They think it's about breaking that one specific unfair rule. It's not and never will be.
People who fake their own writing using AI slop can be found out, either after the fi
Re: (Score:1)
If you fuck up you're out. So don't fuck up in trivial ways that are easy to spot.
If you get credibly accused of.... So don't get credibly accused of....
That's the reality.
Now, if writing in a style that is "bot like" constitutes "f'ing up" then I'll concede your point. But someday soon, bots will be able to write in ways that, in a pre-LLM era, we would call "excellent writing." Even absent LLMs, computers already out-do humans at some intellectual tasks, like playing chess.
Re: (Score:3)
Actually, and IIUC, they can do that now with proper prompting. They just make mistakes about what the universe is like. World models should be fixing that within the next year.
It's already "you get accused of", not "there is clear proof of". I've known some really bad writers that made worse world model mistakes than are typical of a wlll-prompted recent AI. (Actually, I've read some good authors who also make that kind of mistake, the difference is that do it on purpose and it fits well into the story
Re: (Score:2)
That's right, my point is exactly that. Humans live in the human world where herd mentality is the law and your job as a child is to master enough social patterns so that you can thrive during your lifetime. That game is a lot more complex than trying to pass off an LLM text as a human text.
On your second paragraph, I would merely indicate to you that the high water mark is not the same as the low watermark, or the median or the mean. Pointing to a few instances of "excellent writing" is statistical noise
Re: (Score:2)
AI disclosure: yes, I used an AI [Copilot] as a drafting assistant. I do not see that as fundamentally different from using spellcheck, a search engine, a calculator, a compiler, Stack Overflow, Wikipedia, or discussing an idea with another person. The argument is mine. The responsibility for it is mine. Judge the reasoning, not the tool.
You're conflating social acceptance with correctness.
"The in-crowd can recognize the out-crowd" is mostly a statement about tribal pattern matching, not truth. History is f
Re: Not gonna happen (Score:2)
So you are saying that without an AI (in this case CoPilot) you are no longer capable of writing coherent and useful _posts_to_a_blog_?
No, AI ist not at all like a spellchecker or a calculator. It takes away the ability to think for yourself and act for yourself. Which in the rund takes away the ability to make decisions in your own. You replace the 'wwjd' armbands with 'wwaid' mental cuffs.
In your case, meaning from the way you describe your workprocess, it seems that what you write is the output of an ai
Re: (Score:2)
No, AI ist not at all like a spellchecker or a calculator. It takes away the ability to think for yourself and act for yourself. Which in the rund takes away the ability to make decisions in your own. You replace the 'wwjd' armbands with 'wwaid' mental cuffs.
Then so does ever human editor out there. If someone has written the words, has gone over them a few times to rework them, then goes to an LLM and says, "Look this over and suggest corrections", that is no different than going to a human editor. A huma
Re: Not gonna happen (Score:2)
And why on earth would I go to an llm or a human editor for /posting on slashdot/ ?
Re: (Score:1)
And why on earth would I go to an llm or a human editor for /posting on slashdot/ ?
You or the person you are replying to may not need a human editor, but here are people whose grammar or spelling is so bad they could really benefit from an editor, human or otherwise.
There are also people whose first language is not English, or who are from a part of the world where English is spoken or written in a different style or dialect than Slashdot and they want their posts to not stick out like a sore thumb.
Re: (Score:1)
AI disclosure: The underlying opinions are mine. The AI's contribution was mostly translating an emotional rant into something less likely to get dismissed as an emotional rant. If that's cheating, then so is spell-check.
---
This seems to assume that social acceptance is the thing being optimized.
Why should it be?
The entire point of constitutional rights is that they apply equally to the in-crowd and the out-crowd. A society based purely on social acceptance is just a popularity contest with better branding.
Re: Not gonna happen (Score:2)
So again, you are saying that you are incapable oft writing coherent thoughts in a blog without an AI?
And in this very special case: you get triggered so heavy emotionaly by people who want real discussions with humans or (university context) input from real students rather than an AI that you can no longer write on your own?
You are a very good slave.
Re: (Score:3)
J'accuse martin-boundry !
The above post has tells of AI use (proper spelling and grammar -even paragraphs!) /s
You are hereby banned forevermore from the interwebs for suspicion of tech-heresy.
Re: (Score:2)
Finally, someone sees the obvious (Score:2)
Unlike AI, humans don't make mistakes. They can figure out, with absolute certainty, if AI was used in someone's writing. No one has to worry about this scenario.
Honestly, thank you for pointing that out. It's a huge relief that humans can spot AI so reliably, this whole "AI detection" panic is just a non-issue. I don't know why everyone is stressing when the answer is this obvious. Or self-evident as the /s indicates.
Nothing major. (Score:2, Redundant)
Just life- and career-ruining accusations spread far and wide with "evidence". And when disproven, as with all false accusations, the apology will be half-assed if there's an apology at all.
Search results will smear the writer's name forever.
But hey, that's a small price to pay for protecting society from... *checks notes* Nothing.
Re: (Score:2)
Came here to say this.
"Strike them down, and God will know His own"
Re: (Score:2)
Just life- and career-ruining accusations spread far and wide with "evidence". And when disproven, as with all false accusations, the apology will be half-assed if there's an apology at all.
Search results will smear the writer's name forever.
But hey, that's a small price to pay for protecting society from... *checks notes* Nothing.
We're basically at the point where if you use Oxford commas, or write with any eloquence at all, you're flagged as AI. Because humans write slop, according to these checkers. If you don't misspell every third word, fuck you, you robot.
We're idiots in the aggregate, and we're scrambling desperately to make sure those of us who aren't idiots at the individual level are punished for it.
Ruined Worlds (Score:2)
Writing contests face "a literary version of the ruined world" [...]
Our history is littered with instances of technology ruining the previous world. The technology may actually be a net positive, but we can still acknowledge the negative side-effects.
We used to cook food from basic ingredients. It would take time and knowledge, but tasted wonderful. We now more commonly microwave frozen meals, which is quick, and requires next to no knowledge. But it has arguably ruined the culinary world. And yes, you can still find properly prepared meals, in the same way as you can still
Re: Ruined Worlds (Score:2)
And to stick to your frozen-meals analogy: When you consume frozen meals you consume stuff that someone else decided to put together. You are no longer in control. And the health benefits of frozen meals are devastating.
Writing contests (Score:2)
Do we really need competitive literature anyway?
Techically an impossible challange (Score:2)
That is, even if you wrote a 100% perfect human/AI distinguisher, there are times when it would be wrong because the AI produces words that were exactly produced by a human.
Re: (Score:2)
So if you're writing is so soulless and bad (Score:2, Insightful)
That said AI is fundamentally plagiarism on demand. This means that the same tools that have been developed for a long long time for detecting plagiarism will work just fine. The real problem is people are assholes and they will accuse people of using AI to write stuff and people will join in because people are assholes.
The solution is to start discouraging people from being assholes. But that is a big lift because behind every asshole is
Re: So if you're writing is so soulless and bad (Score:2)
I think what you are trying to solve is the symptom and not the core issue.
There is a fundamental culture problem, extending from the maximization of capitalism and the externalization of the side effects. Start ups have the job of being disruptors, that means smashing into cultures and ignoring all etiquette and moral decency. From this you get things like uber adding stores to their menu without permission and charging for it, stealing of copyright work for AI, and camera sunglasses for peeping.
We cr
Re: (Score:2)
It seems to disproportionately affect people on the Autism spectrum. I've been accused of it more than once, and seen people accused when to me their text reads perfectly normally.
These detectors seem to be a load of crap.
Obligatory (Score:1)
https://www.youtube.com/watch%... [youtube.com]
Re: Obligatory (Score:1)
https://m.youtube.com/watch?v=... [youtube.com]
Gemini can you respond? (Score:1)
AI disclosure: The underlying opinions are mine. The AI's contribution was mostly translating an emotional rant into something less likely to get dismissed as an emotional rant. If that's cheating, then so is spell-check.
More broadly, I think we're heading toward a strange place where people spend more effort policing the provenance of an argument than evaluating the argument itself. If the reasoning is wrong, it should be wrong because the logic is faulty. If it's right, it should be right because the logi
Threshold for labeling as AI generated? (Score:2)
What counts as AI generated? If I use a spell checker? How about a grammar checker? What if the grammar checker suggests replacements? Often those replacements are either corrections of nonstandard phrases or more standardized diction that introduces no new ideas.
What if an LLM is used to insert a phrase, a sentence, several sentences, a paragraph, several paragraphs? At what point does the LLM usage transform the composition into "AI"?
I often use GMail's suggestions to improve diction. I imagine the underl
Re: (Score:1)
It's pretty simple, you'd be an idiot to use an energy-inefficient algorithm like an LLM for spell-checking or grammar. Syntactic analysis is a long-solved problem that can be done efficiently using traditional algorithms that produce repeatable results without any plagiarism. The recommended replacements are syntactic corrections.
The point at which an LLM has been used is the point at which an LLM has been used. I hope that clears things up for you.
University's also. (Score:1)
Re: (Score:2)
I am exceptionally skeptical that your characterization has any basis in reality.
I suspect that your university changed its policy, and your proposed reason is merely what you heard people say about it. People who weren't involved in the decision-making and didn't consult whatever public information might have been available. Instead, they made a statement of hyperbole that fit their preferred narrative. And you credulously repeated that characterization as if it was a statement of historical fact.
Critical
False positives (Score:5, Interesting)
My kid is taking a class in high school that involves a lot of writing. Essays are submitted online using a tool that checks for AI and plagiarism. You can see the score of your draft (and make edits) before submitting.
On one of the first assignments, said kid complained that the tool was reporting an AI score over 60% even though the writing was original work. Adding misspellings and run-on sentences lowered the score.
I tested the tool for myself by writing a paragraph on a random topic. I didn't use AI tools. I didn't even refresh my memory of the topic. I just wrote a few sentences based on what I remembered from the course when I took it 30 years ago. The tool said my completely original writing was over 50% AI.
The teacher has said that he expects students to have AI scores below 20%. I advised my kid to make handwritten outlines and drafts, to retain any printouts that they had marked up or highlighted, and to carry these with them to school every day so that they could be provided immediately to the teacher if there was ever an accusation. Short of filming yourself working from start to finish, I don't see how you can prove you didn't use AI.
Lots of people are going to get burned by this.
Re: (Score:1)
It's no different than turnitin. If your business involves catching witches you will never run out of witches to catch. Even if it means manufacturing some.
Re: (Score:2)
Short of filming yourself working from start to finish
I suspect this will be the future, but not before things get much, much worse.
Re: (Score:2)
What is it worse? "You filmed yourself copying what AI wrote on your screen"
and if it is not a live stream people can claim you've cheated in other ways or even the whole video is generated.
Re: (Score:2)
I used to hate long form writing because it hurt my hands, due to arthritis. When I learned to type I was suddenly freed from the pain and I started to enjoy it.
So while hand-written notes are a nice idea, they won't work for everyone. I think recording the screen is going to be the only option for some kids.
I can see this working both ways (Score:2)
What if humans started training to become experts at writing like AI? We could end up with a whole industry of poorly paid, not-very-knowledgeable scammers in Indian call centres, cranking out utter fabrications that look like AI output but require a small fraction of the computing power.
Hooray for humanity, I guess?
If it was a mistake (Score:2)
Then new companies and products will spring up, that are more permissive, and the old guard that fights AI writing, will gradually fade.
It's kind of like the transition from traditional artist-rendered portraits, to photographic portraits. There was once major opposition, people criticized photographs as not being real art. But today, photography has become an art form in its own right, not replacing traditional art, but co-existing alongside it.
AI is likely to spawn its own forms and eventually be recogniz
Everyone's pointing fingers. (Score:2)
Only I didn't. I write 7 days a week. I have over 30 novels in print. I type fast and, after 35 years in the biz, I know how to write clean copy while pantsing the plot 1-2 chapters ahead.
I've even tweeted my progress and word counts most days, and my