This piece was so on-point it deserved a real, live, human-generated comment. I've been living with this question for some time--what does disclosure look like for writing, or any creative work, in the AI era? I'm in the process of setting some rules around this, with the intention of preserving uniquely human cognition while managing the realities of a world trained on synthesizing content at the speed of light. This certainly adds to my thinking!
So pleased to read this, Jory. If you ever feel like sharing your own rules, please do so in comments here or on the accompanying "Our Rules" piece, linked at the bottom. We who so value human writing would all benefit, I think, from hashing this out together...
Tom, your invitation to share rules prompted me to surface a framework I've been developing — the Human Agency Architecture — which I apply to writing decisions the same way I apply it to workforce ones. It maps writing acts across five categories, from fully delegated (AI operates without review) to human retained (provenance is non-negotiable, regardless of AI capability). The governing principle: accountability follows judgment. An AI-assisted piece is only legitimately yours if you brought something the AI couldn't — organizational context, relational knowledge, a specific formed perspective. Light editing doesn't transfer accountability; it erases it. I'm still working out the writing-specific application, but your "5 Rules" gave me useful guardrails to pressure-test it against. Will share more as it develops.
The fifth rule lands because it asks what authorship means when the named person still matters to the reader. I use AI as a thinking partner, but the moral choice, voice, and responsibility for the work must remain human. That tension is central to the techno-thrillers I write.
Thank you for this Tom, very enjoyable read. One question on 3. The Off switch: This relies on the following to not happen, correct?
"But AI developers are pursuing ever-better writing tools, which could raise the opposite problem: increasingly mistaking AI writing for human. In which case, we’ll need further antibodies.
And do I understand you correctly that here you mean a future without all writing, not all human writing?
"A future without writing—as opposed to today’s decline of deep reading—is possible only if we find a technology more proficient at transmitting ideas and experiences from brain to brain. Brain-computer interfaces are not near any such text-ending magic."
If I understand both points correctly I would be very interested if you have given some thought to a future in which AI-writing is ubiquitous and undetectable.
tl;dr - KIDDING! Tom, I read this from start to finish (in 2 sessions, I'll admit). It's the best essay I have read on AI and writing. Thank you for your time, your effort, and your craft.
Here's an AI-age angle you did not cover, but which is needling me these days: Audiobooks. My audiobook subscription forces me to listen to at least 12 books per year. Meanwhile, The New Yorker, The Economist, The Atlantic and nearly every other publication I subscribe to offers the 'listen to this article' option.
The problem is sleep. Audio books and articles — whether read by increasingly realistic AI bots, actors, or (yay!) the authors themselves — put. me. straight. to. sleep.
Don't get me wrong, reading puts me to sleep, too. But at least I have that lurking fear that I'm going to drop the book (or other reading device) right onto the bridge of my nose and then bang onto the hardwood floor, jolting me out of that weird dream that the book has colored in some creepy sepia. I'll remember more and find my place easily the next night.
Audio books are such a fast-acting soporific that I never remember much. If I forget to set the timer, the book will play to the end, and I lose my place and skew my stats. My sister tells me I should listen while out on walks, but I like to hear the birds, the wind in the trees, and that 3-ton hybrid SUV creeping up on me.
Maybe direct-to-brain books will be better.
All the best to you and your readers,
Jim
PS - I, too, am really curious about the handwritten letter. Secret admirer?
I think audiobooks gained popularity (partly) because people are so overwhelmed with choice, and anxious about keeping up, so they try to feed text passively into their heads at all times. It reminds me of the fallacy popular when I was a kid that you could learn a language by playing a record of it while you slept!
To be clear, I have loved some audiobooks--when living in Paris, I went on long walks listening to early modernist classics, such as Ulysses and To the Lighthouse, and finally *got* them from hearing them read brilliantly! But in many cases, I suspect that audiobooks convey the false sense that one is actually absorbing the text, when it's just flowing past, leaving the faintest residue. It reminds me of something in our recent interview with Carl Hendrick (https://aipolicyperspectives.substack.com/p/the-science-of-learning), where he made the point that low-effort learning isn't really learning--that effort *is* the learning.
I suspect audiobooks are (sometimes) a case of this. And I suspect that people realize this on a certain level, which explains why they often feel sheepish to say they've "read" a book if they actually listened to it. They know it's not the same.
In a period where we're all glutted with inputs constantly, and (I at least) feel overloaded by things we *should* get into our brains, accepting that you'll not know certain things feels like a weirdly important skill, if we're to avert a constant sense of anxiety and falling behind...
Would you use AI to make your argument more concise? Today I read a 37 min essay on Russia that I found very charming; this I found harder to get through. Is it me?
With complex texts that I'd like to read, I sometimes use AI to take a first pass, helping me figure out the overall sense and shape of the work. Ideally, I'd just read everything in its entirety. Sadly, not possible! For difficult and technical papers, in particular, I find this approach allows me to grapple with texts that'd would otherwise be intractable for my middling brain.
However, be careful: this process risks stripping out ideas/perspective. That means I'm more likely to employ an AI summary in cases when I primarily seek broad understanding. I'd be far less likely if my goal were a human's view. Imagine how drained an AI summary of first-person war reportage would be.
When it comes to writing, if you've produced an excessively long text, you might drop the whole piece into AI, and ask it to ferret out your core points, and where your structure is muddled. You could then evaluate its editorial notes, and revise accordingly. By contrast, if you just got the AI to make your long text concise, its verbal sleekness might seem obviously better--yet could contain little of you anymore. In such cases, many people will just adopt the AI writing as their own, probably convinced that it really was their work. It's a seduction where the AI ends up in bed with your ideas.
A more general observation is the bifurcation in writing styles. On the one hand, much writing has adapted to the shortened attention spans affecting everyone. AI often adopts this staccato when writing (partly because humans tend to favour it when giving feedback on AI answers). Yet many humans are also replicating that snippet style: they observe that it's what people click on.
On the other hand, you also find reams of self-indulgently excessive pieces today, much as you have absurdly long podcasts. In one way, this is heartening: a counterpoint to the human loss-of-attention. But I suspect it has a more banal origin. That direct-to-audience writing on Substack and other media have no editors! That is, nobody is saying, "Shut up already!" It allows writerly self-indulgence free rein. Also, the writer has few/none of the hard constraints of old forms, such as the newspaper's limit of column inches. Constraints are excellent for creativity. And limits forced reader-oriented concision that I often miss when wading through a charming but needlessly long piece on Russia or the likes...!
This is a profound piece. Your point about the 'off-switch' of interest once we realize an author is an AI despite the text itself being eloquent, hits the nail on the head. It’s that exact sense of 'relational' value that differentiates writing from mere content generation. If writing is truly humanity’s technology for 'visiting another’s head,' then AI authorship isn’t just a shortcut; it’s a trespass.
I write about the digital marketing landscape over at nicolo.substack.com, and I constantly grapple with this balance—using AI as a tool for efficiency without sacrificing that 'human-intensive, artisanal' essence you describe. I’d love to support your work and stay in the loop with your ongoing reflections on this. Would you be open to a subscription exchange? Keep writing, we need the struggle of the blank page.
Awesome piece, and you know what some of the worst things are, when it comes to AI generated content on different platforms...people have stopped putting in the effort to even edit it or atleast put some sort of effort to take that initial draft AI presented to you and put some soul into it..
The other thing which I hate is, real writers having to change how they write because the pattern seems too AI'ish.
Several cringe cases of not-even-editing have come in academic publication; one published article began "Certainly, here is a possible introduction for your topic:..."
As for writers changing because they fear sounding AI-ish, that's an interesting one. I agree with you. But I also don't mind being forced to scrutinize my own language choices here and there--is "delve" the word I really want? If so, I'll use it.
A revelation from this whole issue is how ardently writers love the emm-dash! Nearly every author I know who is wary of seeming AI-ish says, almost with lower-lip trembling, "But they won't take my emm-dashes!"
I loved this article. So many good points. Will be pondering it. I’m curious how you might rethink some of these points if we were to apply them to the calculator? A long time ago people’s whole profession was to sit in a room with other people to make mathematical calculations. Their job title was Calculator. Now, we have a tiny machine in our phones that used to take a whole room of people. Everyone in school (as far as I know) is still learning basic math. But we would never spend days making a complex calculation. It’s too much time and no one is complaining this is a bad thing. Could some of this thought be applied to AI writing?
On the calculator, that technology is sometimes cited in discussions of "cognitive offloading"--the question of how much handover is good. Another case in point is the memory on devices today, where we've outsourced tons that we used to recall. As a kid, I knew dozens of phone numbers. Today, I know one. Literally, one. (Plus those from my childhood that remain in memory!)
I think there are a few differences between the calculator and AI writing. Among these are that the calculator swiftly processes something external to us, numbers. If we lose any way to do, or understand those processes, then we have lost much. That is why maths continues despite calculators.
By contrast, writing isn't about processing information that is identical for any calculator. It's about moving what's inside your unique mind to the outside world. Even the writer doesn't know all that's inside; we have inclinations, preferences. But it's largely a mental mist that the constraint of language forces into meaning. This applies to speaking aloud, to artistic expression, and to writing, I think. If we're not doing the communicating then everyone is locked inside themselves.
It’s best to bring the idea and logic of the piece, before using any AI LLMs. Then use the AI Model as a brainstorming partner or tool to nail the final product.
I agree. I also wonder how we tease apart our efforts from its efforts. In some domains, this may not matter; it'd simply be better to have an intellectual ally. But in certain cases, it'd undermine one's own thinking. Working effortfully leaves a mark in a good way.
Much will become clear, I suspect, as we continue to use these tools, and as they grow better...
I do worry that, over time, AI writing becomes ubiquitous, whether because the taboo against generating text dissipates or because the detection systems cannot keep up. The former is the far greater danger because there's no coming back. And that's precisely why I think detection right now is urgent: we cannot maintain social norms if there's no way to identify when someone violates them.
Tom, I appreciated this post. A larger and important point you touch on is where you write: "News publications ban it, educators resist it, and academic journals flounder. Yet social norms could be shifting subtly toward acceptance. Increasingly, you hear the view that authorship is immaterial; it’s about making something worth reading."
This may be true broadly speaking, but it overlooks certain cases where relying on AI will make it less likely to result in 'making something worth reading'.
To take only one of many examples, I don't think leaning on AI heavily to write a court brief in a factually complex case will get you there. And despite how many prizes go to short stories that AI helped write, AI will likely never generate prose with the range and power of someone like John Updike or David Foster Wallace. (The Monet test does nothing to persuade me otherwise.)
I found the essay strongest when it treats writing not merely as output, but as a formative act: a way of thinking and becoming responsible for one’s own meaning.
That is why I was surprised by the confidence in detection technologies and their institutional enforcement. If the central concern is that writing is tied to authorship and consciousness, there is a strange irony in relying on machines to decide whether a human consciousness was present in the text.
Perhaps the deeper issue is not simply whether AI wrote the words, but what kind of human judgment and responsibility the writing process required.
The irony: rules for AI writing assume a stable, centralized thing we're all writing about, but if AI itself fragments into thousands of sovereign models optimized for different domains and values, the "rules" become context-dependent too. That's not a weakness; it's actually closer to how language evolved. What do you think happens to writing guidance when intelligence itself becomes distributed?
Now I want to know: Who sent you the actual handwritten letter and what did it say? :)
This piece was so on-point it deserved a real, live, human-generated comment. I've been living with this question for some time--what does disclosure look like for writing, or any creative work, in the AI era? I'm in the process of setting some rules around this, with the intention of preserving uniquely human cognition while managing the realities of a world trained on synthesizing content at the speed of light. This certainly adds to my thinking!
So pleased to read this, Jory. If you ever feel like sharing your own rules, please do so in comments here or on the accompanying "Our Rules" piece, linked at the bottom. We who so value human writing would all benefit, I think, from hashing this out together...
Tom, your invitation to share rules prompted me to surface a framework I've been developing — the Human Agency Architecture — which I apply to writing decisions the same way I apply it to workforce ones. It maps writing acts across five categories, from fully delegated (AI operates without review) to human retained (provenance is non-negotiable, regardless of AI capability). The governing principle: accountability follows judgment. An AI-assisted piece is only legitimately yours if you brought something the AI couldn't — organizational context, relational knowledge, a specific formed perspective. Light editing doesn't transfer accountability; it erases it. I'm still working out the writing-specific application, but your "5 Rules" gave me useful guardrails to pressure-test it against. Will share more as it develops.
The fifth rule lands because it asks what authorship means when the named person still matters to the reader. I use AI as a thinking partner, but the moral choice, voice, and responsibility for the work must remain human. That tension is central to the techno-thrillers I write.
Great piece. Rule 3, that outsourcing deletes your future wisdom, is the one that sticks.
Thank you for this Tom, very enjoyable read. One question on 3. The Off switch: This relies on the following to not happen, correct?
"But AI developers are pursuing ever-better writing tools, which could raise the opposite problem: increasingly mistaking AI writing for human. In which case, we’ll need further antibodies.
And do I understand you correctly that here you mean a future without all writing, not all human writing?
"A future without writing—as opposed to today’s decline of deep reading—is possible only if we find a technology more proficient at transmitting ideas and experiences from brain to brain. Brain-computer interfaces are not near any such text-ending magic."
If I understand both points correctly I would be very interested if you have given some thought to a future in which AI-writing is ubiquitous and undetectable.
tl;dr - KIDDING! Tom, I read this from start to finish (in 2 sessions, I'll admit). It's the best essay I have read on AI and writing. Thank you for your time, your effort, and your craft.
Here's an AI-age angle you did not cover, but which is needling me these days: Audiobooks. My audiobook subscription forces me to listen to at least 12 books per year. Meanwhile, The New Yorker, The Economist, The Atlantic and nearly every other publication I subscribe to offers the 'listen to this article' option.
The problem is sleep. Audio books and articles — whether read by increasingly realistic AI bots, actors, or (yay!) the authors themselves — put. me. straight. to. sleep.
Don't get me wrong, reading puts me to sleep, too. But at least I have that lurking fear that I'm going to drop the book (or other reading device) right onto the bridge of my nose and then bang onto the hardwood floor, jolting me out of that weird dream that the book has colored in some creepy sepia. I'll remember more and find my place easily the next night.
Audio books are such a fast-acting soporific that I never remember much. If I forget to set the timer, the book will play to the end, and I lose my place and skew my stats. My sister tells me I should listen while out on walks, but I like to hear the birds, the wind in the trees, and that 3-ton hybrid SUV creeping up on me.
Maybe direct-to-brain books will be better.
All the best to you and your readers,
Jim
PS - I, too, am really curious about the handwritten letter. Secret admirer?
Hey Jim--many thanks for the great comment.
I think audiobooks gained popularity (partly) because people are so overwhelmed with choice, and anxious about keeping up, so they try to feed text passively into their heads at all times. It reminds me of the fallacy popular when I was a kid that you could learn a language by playing a record of it while you slept!
To be clear, I have loved some audiobooks--when living in Paris, I went on long walks listening to early modernist classics, such as Ulysses and To the Lighthouse, and finally *got* them from hearing them read brilliantly! But in many cases, I suspect that audiobooks convey the false sense that one is actually absorbing the text, when it's just flowing past, leaving the faintest residue. It reminds me of something in our recent interview with Carl Hendrick (https://aipolicyperspectives.substack.com/p/the-science-of-learning), where he made the point that low-effort learning isn't really learning--that effort *is* the learning.
I suspect audiobooks are (sometimes) a case of this. And I suspect that people realize this on a certain level, which explains why they often feel sheepish to say they've "read" a book if they actually listened to it. They know it's not the same.
In a period where we're all glutted with inputs constantly, and (I at least) feel overloaded by things we *should* get into our brains, accepting that you'll not know certain things feels like a weirdly important skill, if we're to avert a constant sense of anxiety and falling behind...
All best,
Tom
Love this piece!
Would you use AI to make your argument more concise? Today I read a 37 min essay on Russia that I found very charming; this I found harder to get through. Is it me?
With complex texts that I'd like to read, I sometimes use AI to take a first pass, helping me figure out the overall sense and shape of the work. Ideally, I'd just read everything in its entirety. Sadly, not possible! For difficult and technical papers, in particular, I find this approach allows me to grapple with texts that'd would otherwise be intractable for my middling brain.
However, be careful: this process risks stripping out ideas/perspective. That means I'm more likely to employ an AI summary in cases when I primarily seek broad understanding. I'd be far less likely if my goal were a human's view. Imagine how drained an AI summary of first-person war reportage would be.
When it comes to writing, if you've produced an excessively long text, you might drop the whole piece into AI, and ask it to ferret out your core points, and where your structure is muddled. You could then evaluate its editorial notes, and revise accordingly. By contrast, if you just got the AI to make your long text concise, its verbal sleekness might seem obviously better--yet could contain little of you anymore. In such cases, many people will just adopt the AI writing as their own, probably convinced that it really was their work. It's a seduction where the AI ends up in bed with your ideas.
A more general observation is the bifurcation in writing styles. On the one hand, much writing has adapted to the shortened attention spans affecting everyone. AI often adopts this staccato when writing (partly because humans tend to favour it when giving feedback on AI answers). Yet many humans are also replicating that snippet style: they observe that it's what people click on.
On the other hand, you also find reams of self-indulgently excessive pieces today, much as you have absurdly long podcasts. In one way, this is heartening: a counterpoint to the human loss-of-attention. But I suspect it has a more banal origin. That direct-to-audience writing on Substack and other media have no editors! That is, nobody is saying, "Shut up already!" It allows writerly self-indulgence free rein. Also, the writer has few/none of the hard constraints of old forms, such as the newspaper's limit of column inches. Constraints are excellent for creativity. And limits forced reader-oriented concision that I often miss when wading through a charming but needlessly long piece on Russia or the likes...!
This is a profound piece. Your point about the 'off-switch' of interest once we realize an author is an AI despite the text itself being eloquent, hits the nail on the head. It’s that exact sense of 'relational' value that differentiates writing from mere content generation. If writing is truly humanity’s technology for 'visiting another’s head,' then AI authorship isn’t just a shortcut; it’s a trespass.
I write about the digital marketing landscape over at nicolo.substack.com, and I constantly grapple with this balance—using AI as a tool for efficiency without sacrificing that 'human-intensive, artisanal' essence you describe. I’d love to support your work and stay in the loop with your ongoing reflections on this. Would you be open to a subscription exchange? Keep writing, we need the struggle of the blank page.
Awesome piece, and you know what some of the worst things are, when it comes to AI generated content on different platforms...people have stopped putting in the effort to even edit it or atleast put some sort of effort to take that initial draft AI presented to you and put some soul into it..
The other thing which I hate is, real writers having to change how they write because the pattern seems too AI'ish.
Thank you, Mohib!
Several cringe cases of not-even-editing have come in academic publication; one published article began "Certainly, here is a possible introduction for your topic:..."
As for writers changing because they fear sounding AI-ish, that's an interesting one. I agree with you. But I also don't mind being forced to scrutinize my own language choices here and there--is "delve" the word I really want? If so, I'll use it.
A revelation from this whole issue is how ardently writers love the emm-dash! Nearly every author I know who is wary of seeming AI-ish says, almost with lower-lip trembling, "But they won't take my emm-dashes!"
"Certainly, here is a possible introduction for your topic:..." - Alright, that one got me, hah.
I loved this article. So many good points. Will be pondering it. I’m curious how you might rethink some of these points if we were to apply them to the calculator? A long time ago people’s whole profession was to sit in a room with other people to make mathematical calculations. Their job title was Calculator. Now, we have a tiny machine in our phones that used to take a whole room of people. Everyone in school (as far as I know) is still learning basic math. But we would never spend days making a complex calculation. It’s too much time and no one is complaining this is a bad thing. Could some of this thought be applied to AI writing?
Thank you, Mariah!
On the calculator, that technology is sometimes cited in discussions of "cognitive offloading"--the question of how much handover is good. Another case in point is the memory on devices today, where we've outsourced tons that we used to recall. As a kid, I knew dozens of phone numbers. Today, I know one. Literally, one. (Plus those from my childhood that remain in memory!)
I think there are a few differences between the calculator and AI writing. Among these are that the calculator swiftly processes something external to us, numbers. If we lose any way to do, or understand those processes, then we have lost much. That is why maths continues despite calculators.
By contrast, writing isn't about processing information that is identical for any calculator. It's about moving what's inside your unique mind to the outside world. Even the writer doesn't know all that's inside; we have inclinations, preferences. But it's largely a mental mist that the constraint of language forces into meaning. This applies to speaking aloud, to artistic expression, and to writing, I think. If we're not doing the communicating then everyone is locked inside themselves.
Also Tom, just wanted to let you know, we used your article to start a discussion about AI writing ethics. Thanks for inspiring us to dig a bit more into this topic. https://molochinations.substack.com/p/on-the-ethics-of-ai-in-writing
That’s so cool! Delighted to see this, Mariah! 🙌
Thanks for this thoughtful reply. This makes sense to me as an important difference.
It’s best to bring the idea and logic of the piece, before using any AI LLMs. Then use the AI Model as a brainstorming partner or tool to nail the final product.
I agree. I also wonder how we tease apart our efforts from its efforts. In some domains, this may not matter; it'd simply be better to have an intellectual ally. But in certain cases, it'd undermine one's own thinking. Working effortfully leaves a mark in a good way.
Much will become clear, I suspect, as we continue to use these tools, and as they grow better...
Thanks for the comment!
I do worry that, over time, AI writing becomes ubiquitous, whether because the taboo against generating text dissipates or because the detection systems cannot keep up. The former is the far greater danger because there's no coming back. And that's precisely why I think detection right now is urgent: we cannot maintain social norms if there's no way to identify when someone violates them.
Tom, I appreciated this post. A larger and important point you touch on is where you write: "News publications ban it, educators resist it, and academic journals flounder. Yet social norms could be shifting subtly toward acceptance. Increasingly, you hear the view that authorship is immaterial; it’s about making something worth reading."
This may be true broadly speaking, but it overlooks certain cases where relying on AI will make it less likely to result in 'making something worth reading'.
To take only one of many examples, I don't think leaning on AI heavily to write a court brief in a factually complex case will get you there. And despite how many prizes go to short stories that AI helped write, AI will likely never generate prose with the range and power of someone like John Updike or David Foster Wallace. (The Monet test does nothing to persuade me otherwise.)
I found the essay strongest when it treats writing not merely as output, but as a formative act: a way of thinking and becoming responsible for one’s own meaning.
That is why I was surprised by the confidence in detection technologies and their institutional enforcement. If the central concern is that writing is tied to authorship and consciousness, there is a strange irony in relying on machines to decide whether a human consciousness was present in the text.
Perhaps the deeper issue is not simply whether AI wrote the words, but what kind of human judgment and responsibility the writing process required.
The irony: rules for AI writing assume a stable, centralized thing we're all writing about, but if AI itself fragments into thousands of sovereign models optimized for different domains and values, the "rules" become context-dependent too. That's not a weakness; it's actually closer to how language evolved. What do you think happens to writing guidance when intelligence itself becomes distributed?