Article
You Let Me Believe You Wrote It: Undisclosed AI Use Is Lying
Photo by Kristina Flour on Unsplash
“Tesla never made a machine like the one I asked for.”
“I never said he had.”
“Well, you let me believe that he had!”
When you publish prose, you imply more than you say.
Unless you say otherwise, readers generally assume:
- … that the ideas are your own. They expect you to give credit for other people’s ideas – unless they’re so well-known they’ve become part of the cultural background, like general relativity or calculus.
- … that the words are your own. They expect you to use quotation marks otherwise – unless a quote is so well-known that any reasonable person can spot it right away, like “to be or not to be”. Readers also expect quotes to be accurate. Misquoting is a form of dishonesty, because it puts words in people’s mouths.1
- … that you hand-wrote it; more precisely, that your effort is roughly what a reasonable person would spend writing the piece themself. If someone or something else wrote the text, readers generally want to know.
Telling people otherwise is part of a writer’s integrity. If you violate any of these expectations, you can’t just say, ‘but I never said I’d meet them.’ They’re implied. Just because they’re largely unstated expectations set by the cultural background doesn’t mean you can violate them without lying.
Lying is, roughly, letting others (or yourself) believe a falsehood, given some context. It’s failing to take reasonable steps to find out or communicate the truth. So if people come to your blog or social media wanting to read your words, and you let them falsely believe they’re reading your words when really you’re just a proxy for your AI, that’s lying.
But many people have dishonest ideas about honesty, so their lies are entrenched. They roughly think they’re being honest as long as they’re not rubbing their hands together, laughing evilly, and lying on purpose. But that’s not good enough. Garden-variety dishonesty is, sadly, very common. Liars often evade the issue of honesty first, in the privacy of their own minds, and then lie to others later, as a result. In bad cases, they have a dishonest psycho-epistemology:2 they’ve automatized dishonest patterns of thought. Once automatized, those patterns are subconscious and so harder to notice. That doesn’t let habitual liars off the hook, though: it’s still on them to introspect and correct the error. But why expect habitual liars to be honest with themselves about how much they lie?
The Fountainhead by Ayn Rand is a great read on honesty. So is ‘Cargo Cult Science’ by Richard Feynman.3 He describes scientific honesty as “kind of leaning over backwards”, and he contrasts it with dishonest practices in advertising. One type of frying oil, Wesson oil, was said not to soak through food. So, given the context of competition, the ad implied that competing oils do. But that was misleading:
The fact that should be added to that advertising statement is that no oils soak through food, if operated at a certain temperature. If operated at another temperature, they all will—including Wesson oil. So it’s the implication which has been conveyed, not the fact, which is true, and the difference is what we have to deal with.
Feynman is right to draw a connection to advertising. I like that social-media platforms, for example, require people to mark posts as advertisements. I recently bought a dental appliance because a dentist I follow on Instagram recommended it. She posted a video praising it. But the video wasn’t marked as an ad. I happened to watch it again after buying the appliance and noticed the production quality was oddly high. When I commented on the video to ask about that, she quietly added a hashtag to mark the video as an ad after the fact. Still, I felt a bit tricked, even though the product does what the video said. By implication, she was recommending the product only because she personally believed in it. And she had let people falsely believe that.
So it is with writing: when you publish prose, you say more than the words themselves. You imply that you wrote it; that you put in the time and effort to create that piece. The difference is what readers have to deal with. Maybe someone recommends your writing to others. But if you’re actually a lousy writer without AI, the recommendation is wrong, and they wouldn’t know it.
@davies has argued, in reference to Karl Popper, that the origin of our knowledge isn’t important. If you publish something true, what does it matter whether you wrote it yourself? Just because AI wrote it doesn’t make it false. I agree, but that isn’t the issue. The issue is, again, what you dishonestly imply. Also, as I recall, Popper wrote that in response to the idea that truth is manifest, which invites totalitarianism because it attributes ignorance to bad faith. So he suggested replacing the question about the sources of our knowledge with another: “How can we hope to detect and eliminate error?”4 That’s a key insight about knowledge, but it has nothing to do with issues of authorship and honesty.
These issues aren’t new – AI has just sparked more discussions about them. Take Einstein again: imagine he stole special relativity. (Some people claim his first wife co-authored it without getting credit. Having read Walter Isaacson’s biography of Einstein, I don’t think that’s true, but assume for the sake of argument that it is.) That would change nothing about whether special relativity is true as an explanation of physical reality. But Einstein would still have been a liar. It’s not that he would have let people falsely believe special relativity was true when he should have known it wasn’t, or vice versa. It’s that he would have let people falsely believe he alone came up with the idea. You can lie about one thing and not the other.
For @moritz-wallawitsch, what matters is the creativity that goes into prompting the AI. But even the most elaborate prompt still takes far less effort than writing the article yourself – if it didn’t, writers wouldn’t bother with AI in the first place. And readers don’t get to see the prompt anyway, so how could they possibly judge it?
Granted, writing good prompts does take some effort, and going back and forth with AI to correct errors and improve the results takes even more. For example, I made the Veritula Chronicles with the help of AI. At the time of writing, there are five chronicles, which took about two days of full-time work for me to make. I worked closely with the AI, read everything, made improvements along the way, etc. The idea to make the chronicles was mine: I wanted to honor some of history’s greatest thinkers. But I’d never claim I wrote the chronicles myself. Because I didn’t. So they each say, on every page, that AI wrote them.
People have oddly relaxed requirements around honesty – until they realize they’re being lied to, then they’re suddenly much stricter. But effort matters. Honesty about effort matters.
To see why, imagine buying a beautiful photorealistic painting for a large sum. Much of the value comes from the great skill that must have gone into creating it. Making a photorealistic painting is hard! The piece is that much more special and rarer for it; the painter has earned his pride and your praise. But now imagine finding out later that the ‘painting’ is nothing more than an actual photo; that the ‘painter’ had counted on you not noticing so he could get your money and admiration. Would you not be at least a little disappointed? Would you not feel tricked?
Some people say they don’t mind this kind of thing since it changes nothing about the artwork visually. But that’s just like the special-relativity example again: there’s a difference between content and implication. At the risk of giving yet another example: if you found out after your wedding that your wife had AI write her wedding vows,5 wouldn’t that make them at least a little less special?
The issue of disclosure doesn’t just apply to writing articles. If someone uses AI in a debate without telling you, they have a leg up. They can reply faster, with far more text. That’s an uphill battle you don’t even know you’re fighting. It’s clearly not fair. (Worse, someone recently tried to cheat me out of $3,000 by using AI in a Veritula bounty, without telling me and against the posted rules. His ‘excuse’ was basically that he hadn’t laughed evilly and lied on purpose.) But when people are upfront about their use of AI, you can adjust: you can use AI yourself to level the playing field, or you can choose not to continue the debate.
Here’s what I recommend for disclosure. Be honest and upfront about how much you use AI for your writings. If you copy/pasted AI output about an idea you came up with, say so, and make that distinction. If you hand-wrote text about an idea your AI came up with, say so. Any mix of the two, again, say so. For example, my article about a mathematical proof Claude had found was written by Claude and only lightly edited by me, so I gave the corresponding credit. Make the disclaimer prominent enough so a reasonable person will spot it early – ideally at the beginning of your article. Veritula makes it easy to both mark and identify articles as AI-assisted. That’s why I’m marking this article as AI-assisted: Claude came up with the title, helped me correct errors, and suggested some of the wording.
When do you not need to disclose your use of AI? Only when it’s strictly limited. For example, if you use AI only to research content, but then put in the work to understand the subject matter yourself and express everything in your own words, never copy/pasting from the AI output, then I think an otherwise reasonable person would agree that no disclosure is needed. That’s because you basically used AI in the same way you would a dictionary. But be careful: once you copy/paste from a dictionary, you do need to use quotation marks and cite the source. I think the same is true for AI. But once the ‘effort’ by the AI is tiny compared to your own, you don’t have to mention it. The same is true if you merely use AI to fix spelling and grammar. Still, I’d disclose it once you copy/paste words from it you didn’t come up with yourself, at the latest.
I keep referring to the standard of reasonableness because there’s no hard-and-fast rule for this kind of thing. First and foremost, you have to be honest with yourself about how much you used AI. If you can be honest with yourself about that, being honest with others usually follows. As Feynman said:
The first principle is that you must not fool yourself—and you are the easiest person to fool. So you have to be very careful about that. After you’ve not fooled yourself, it’s easy not to fool other scientists. You just have to be honest in a conventional way after that.
Not all creative fields treat this issue the same. The culture around programming, for example, is changing. So many programmers use AI now that not using it would be worth mentioning. Once the use of AI is implied and the cultural background expects it, disclosure isn’t needed anymore. The phrase ‘I had Claude make…’ is slowly turning into ‘I made…’, but readers generally know what it means, so it’s not lying anymore. It’s like an accountant saying ‘I calculated’, when in reality he typed something into a calculator. We all know (and hope) he didn’t do the calculations manually.
Maybe writing will go through a similar transition. Nobody knows. In the meantime, use Pangram to find out, with reasonable accuracy, whether some text is human-written.
Use my tool Quote Checker to quote accurately. ↑
Psycho-epistemology is a concept from Objectivism. It’s “the study of man's cognitive processes from the aspect of the interaction between the conscious mind and the automatic functions of the subconscious.” https://aynrandlexicon.com/lexicon/psycho-epistemology.html ↑
Feynman, Richard P. "Surely You're Joking, Mr. Feynman!": Adventures of a Curious Character (pp. 338-346). Kindle Edition. Feynman holds scientists to a higher standard than others:
I am not trying to tell you what to do about cheating on your wife, or fooling your girlfriend, or something like that, when you’re not trying to be a scientist, but just trying to be an ordinary human being. We’ll leave those problems up to you and your rabbi. I’m talking about a specific, extra type of integrity that is not lying, but bending over backwards to show how you’re maybe wrong, that you ought to have when acting as a scientist. And this is our responsibility as scientists, certainly to other scientists, and I think to laymen.
In other words, whereas your average Joe should take reasonable steps to find out and communicate the truth, a scientist should take exhaustive steps. ↑
Karl Popper, Conjectures and Refutations, ‘On the Sources of Knowledge and of Ignorance’ ↑
I heard this example somewhere. I don’t remember where, but it’s not mine. ↑
Can you refute this article?
2 criticisms are pending. Address them, or add your own.
This page shows only the replies leading to #6163. See them all
A reader’s preference doesn’t create a duty to disclose: We generally share finished work without volunteering which tools we used to make it. Readers who care about the process can ask or choose to read writers who disclose it. If an author explicitly claims to have drafted the prose themselves, or the setting reasonably implies it, they owe readers an honest account. To call silence about AI use a lie, you need to establish that it leaves readers with a false impression they had good reason to form. The fact that some readers would choose differently if they knew is insufficient.
More articles
web-haptics: Haptic Feedback Finally Comes to iOS Safari
lochie just published a package bringing haptic feedback to the web on iOS: web-haptics. I spent four years at Apple working mostly on UIs, so I pay close attention to design and UX. In terms of presentation, use case, and attention to…
5 Minute Creativity
TL;DR: When making a decision or working to solve a problem, spend 5 minutes (using a timer) just coming up with ideas. Managing your attention like this can supercharge your creativity. A few days ago I was helping a friend flesh out an…
In Defense of Ambition
Until about a week ago, when I debated this with my friend Tom Hyde, it seemed incredibly obvious to me that wanting to be rich is rational but I could not explain why. Fundamentally I agree with Popper and Deutsch that all life is problem…

