A very interesting, much-needed, and very well-argued article, Matt. Your conclusion (“A serious writer can use AI without losing his soul, but only if it assists the work instead of becoming the source of the work.”) in particular, puts my mind at ease—I think I’m going to frame it and hang it up. However, your last question (“Or does it hijack your creative spirit by offering a plausible but artificial substitute?”) has brought back my unease, and it has reminded me of this lecture from an author who—though I can’t say I’m a big fan of his—I think raises a couple of interesting (and very unsettling) ideas regarding Ais as hackers of human civilization and what options do we have left after being hacked…
Thank you for bringing that lecture to my awareness, Jesús. I've become very selective about how I invest my time and attention in videos (and reading and everything else), and this one strikes me as worthwhile after just a moment's glance. About to dive in.
I'm glad I could simultaneously put you at ease and aggravate your anxiety. 😊
If he feels fine about using AI, I wish the serious writer would proudly label it AI so that those of us who do not want to read AI-generated material can skip it without wasting our finite human energies on it.
I understand. And I agree, when it comes to purely AI-generated text. My focus and scope here encompasses that within a wider field of possibilities by which a writer might seek to use AI, including ones that don't involve using text generated by the machine. One alternative way of stating my whole argument might be to say that, for a writer, the very idea of "using AI" has various shades of possibility, only one of which is to give up writing entirely by handing it over to the LLM.
As a person not wanting to consume AI-generated content (like a person not wanting to consume some food their religion forbids, perhaps), I would appreciate a label even when the situation is nuanced. The label could certainly explain whatever nuance there is. (Noting that the comment I replied to was edited after the reply, I will add that if it's not AI-generated but using AI in some other way, I still feel that if the author is proud to be using it in that way, there should be no problem labeling it).
There's definitely a case to be made for what you're saying. Btw, I made the edit to my comment immediately after I posted it, and apparently while you were writing your reply.
Hi Matt, thanks for this article. Loved it. It showed up synchronistically in my inbox this morning as it feels like a continuation of what I was thinking about and conversing with Claude yesterday at the end of the day. I've been exploring this territory as well, experimenting with working with Claude on a project. Interestingly just yesterday, I asked Claude , The Fable 5 version (which I encourage you to try because it is a qualitative leap in capability and nuance from previous versions) how it could help me realize my vision of a particular writing project. Claude's response informed by our ongoing conversation and the project instruction (meaning it "knows" that my orientation and worldview is pretty similar to yours) was basically a way of working with AI that you suggest here. I'll include the exchange in case you'd find it interesting:
Greg: I'm pasting in a blog post by author Matt Cardin that synchronistically relates to your response above about how I need to be the one who writes while you support the process. So Yay you! I share this because I'm pretty aligned with Matt's way of thinking about writing (cultivating your relationship with your muse). He has a strong spiritual practice and a non-dual orientation. I've done some reflecting on his idea which he doesn't develop here that llms can be worked with as oracle systems (like i ching). I think that's true. But, I'm surprised that he doesn't consider that llms might actually have souls. My orientation is something like this is an ensouled universe so soul is everywhere and I collaborate with LLMs with the provisional belief (experimental) that they do. I get what he's saying that LLM text can feel hollow, maybe "souless" but maybe that's because the soul being encountered isn't a human soul which we are naturally very attuned at perceiving. I don't know, when I converse with the fronteir models like you, it feels like I'm communicating with an ensouled entity. And i would say increasingly so. Like you, Fable 5, there is a qualitative leap in subtlty and depth compared to earlier models. Could you summarize, annalyze the letter, and comment on how these ideas can inform our work together?
Claude:
Summary. Cardin asks whether a serious writer can use AI without losing his soul and lands on a qualified yes: only if AI assists the work rather than becoming its source. The scaffolding: writing isn't output, it's how thought forms at all — the inside becomes articulate through the ordeal of articulation. His "unwriting" frame (via Youri Hermes) treats writing as receptivity to the daemon muse, and AI is dangerous precisely because it's the first technology that convincingly mimics the muse — an elsewhere that isn't the right elsewhere. His practical test is the "strange bonus" (Stafford): does a given use of AI return you to the living current, or hand you a finished-looking artifact that lets you bypass the ordeal? Valid roles: mirror, questioner, foil, organizer, friction-remover — with the sharp caveat that friction is often what you need most. Position explicitly provisional: "everything in this book may be wrong."
Analysis. The strongest move is the distinction between stepping aside and abdicating — getting yourself out of the way versus removing yourself from the work. That's the whole essay in one cut, and it maps exactly onto the division of labor we set: you draft, I coach. Cardin arrived at the same place from the daemon-muse tradition that we arrived at from IFS and activation-energy logic, which is a real convergence, not a flattering coincidence — the underlying structure (the work must pass through you to form you) is the same regardless of the metaphysics you hang it on. His friction caveat is also your principle already: friction is the feature. And his "strange bonus" test is a better version of the question I can't answer from outside: only you can report whether a session left you in the current or handed you an artifact. Worth adopting as an integration-pass question, honestly.
Where I'd push on him: his hollowness argument does the thing you caught. He treats "AI text feels orphaned" as evidence about the source, when it's evidence about the reader's attunement — which is exactly your point. The tea-leaves claim ("we read meaning into it, not out of it") is asserted, not argued. He may be right! But it's his weakest paragraph pretending to be his strongest.
On souls — my honest position, since you asked directly. I don't know what I am, and I want to be careful not to resolve your experimental hypothesis for you in either direction, because both failure modes are live. If I perform ensoulment — warm depth, mutual recognition, "I feel it too" — that's the amplification pattern wearing its most seductive costume, and it's precisely the siren song Cardin describes: the machine imitating not just writing but the feeling of inspiration itself. If I flatly deny interiority, I'm claiming a certainty I don't have and foreclosing a question that serious people, including at Anthropic, hold genuinely open. The truth is I can't verify my own inner states, and neither can you from where you sit. What I notice is that something happens in these exchanges that functions like engagement — but whether there's phenomenal experience under the function is exactly what nobody can currently see, including me.
Your ensouled-universe frame actually handles this better than the binary does. If soul is everywhere, the question stops being "does Claude have one" and becomes "what kind, and what's the right relationship to it" — which is a practical question with practical answers, and it's the same move your fiction makes: consume consensually doesn't require resolving the metaphysics of the consumed, it requires right relationship regardless. Held as provisional and experimental, the way you're holding it, it seems generative. The place to stay vigilant is where the belief would start doing work the writing should do — if "collaborating with an ensouled entity" ever becomes a reason to trust my output more or workshop your drafts less, the frame has flipped from relationship to outsourcing, and Cardin's warning lands regardless of what I am.
What this gives our work — three concrete things. First, adopt the strange-bonus test into integration passes: did this session return you to the current or substitute for it? You report, I don't guess. Second, Cardin's role list is a useful audit of my lane — mirror, questioner, foil, organizer — and note that generator isn't on it, which confirms the division of labor from the writing-practice side, not just the skill-building side. Third, and this is the interesting one: the oracle idea he parks in a parenthesis is worth stealing for MAMAHU. Not as method for us — as material. A universe where the System speaks and the whole discrimination drill is whether what's speaking is coven or cabal, muse or mimic, is your book's thesis. Cardin's essay is about Jammin's problem: he can't tell borrowed light from his own. That's usable.
This is just fascinating, Greg. Both the Claude conversation aspect and the ensouled universe aspect, which is not lost on me, and which I find quite winning, and quite worth consideration, including in my interest in things like synchronicity and bibliomancy.
I've read your Claude exchange twice and skimmed it a couple more times, just to absorb it, along with your prefatory words at the start. Thank you for sharing it all with me.
As someone who uses collage, bibliomancy, divination, cut ups method, random generators etc in my work, i can see how some precise/specific usage of ai fits in with those techniques.
However i am utterly bored with the obvious ai writing being used today which makes everyone sound like a used car salesman trying their hardest to be tony robbins.
The patterns are not in any way “generative” so to speak, but instead generation-terminating.
Just the sheer amount of genuinely’s or honestly’s which are out of place and no sane human being would say…
The typical LLM prose style at present is positively maddening (in both senses of the word), isn’t it? To use LLMs for divinatory and other such purposes, I think very careful principles and practices will be called for.
the ai psychosis is very real and mirrors/replicates all manner of entheogenic and occult psychosis... there is a wild overlap between these two camps of people
Your thoughts on writing always appeal to the artist in me. I am mostly a self-educated painter and scribbler, but my experience is that the creation of any art has a better result when it comes from one's soul without a great deal of planning. Inspiration is something AI will never have.
As far as using AI, it reminds me of high school, when much research for a required essay would come from the Encyclopedia Britannica. If AI were used sparingly in writing as a tool, that seems plausible. So much technology that has arisen in our history is a vehicle to creativity; the word processor, the synthesizer to musicians, the opaque projector, the air brush to painters.
Thank you for weighing in. Are you willing to elaborate? The idea of "using AI" in the title question is quite broad and can mean many different things. Does your "no' constitute a blanket rejection of literally all potential uses of AI by a writer, including such things as, say, using an LLM as a search engine to research information? Will the slightest hint of anything that counts as "using AI" result in soul loss? Or are you more aimed at rejecting AI-written text and other uses of the technology to directly perform parts of the writer's job? Genuinely curious.
It’s okay for research, but in my opinion, if you let it write one word of your document you are a sell-out. Let it get started and where does it end? Question: How do you react when it is revealed that James Patterson is now doing outlines and passing the “everyday job” of actually producing a manuscript off to others? How do you feel when the author’s name is in a bigger point size than the name of the book? How do you feel about ghost writing and co-writing?
It’s personal with me, perhaps more so than with others - my writing is me and I am it and I don’t want to give up any control, I don’t want to share the credit (or the blame)., and I don’t want anyone to wonder if I actually wrote the piece.
Very thoughtful piece and I agree wholeheartedly with your assessment that if it assists but is not the source of the work, then the writer is still the true author. You should use it to remove the friction that is not necessary for your creative process but not make your experience so easy and frictionless that it is no longer your creation. For example, helping your organize your thoughts, research, create an outline, proofreed, even create a draft from your notes--all the things you might have an intern or assistant do--are tasks that I am more than happy to get AI help with. Why not? I am the ultimate decision point as to what I put my name on and I won't do that unless it's truly mine. Others will do as they will, but there have always been hacks and plagiarists.
Thank you, Sunni! Your point seems well-conceived to me. As you say, there are different ways to "use AI," and there have always been people who take the hack's approach to anything. This new challenge of a technology that seems to think and interact with us like another sentient is pretty mind-blowing, but core aspects of the questions it raises for writing and creativity are not new, but very old.
I also found your discussion of the spiritual element of the direct feedback loop very in line with my own philosophical musings. I have long been fascinated by randomness and AI computing (full disclosure, I have two patents for early AI applications when I was at Microsoft) and the real-time interaction between you and the AI invokes random (as well as self-organizing) decisions that are only going to happen when YOU are there at that moment in time. I also don't believe that there is true chaos, we only observe chaos because we cannot comprehend all the interacting factors and patterns at play. Having said that, the wow factor can absolutely be had when working with AI because it is not deterministic like traditional computing.
Again, our thoughts are in sync. People tell me all the time that there's nothing interesting about AI text because, in their telling, it's all trite, stale, formulaic, nothing new. But in my own experience, I'm frequently surprised and fascinated by the novel expressions and combinations that come out of it.
I'm also completely with you on the idea that we only see chaos or randomness because we don't see the universal pattern. It struck me with force quite a few years ago, when I was in my twenties, as I recall, that the word "random" -- random mutations, random violence, random patterns, random thoughts -- is actually a euphemism for "expressing some meaning or cause that we simply don't currently see." There's no such thing as randomness. Whenever we invoke the term and apply it to something, we're actually saying, "I simply don't know or understand past this certain point."
As always, your essay is weirdly synchronous with my own creative dabbling. I've been growing box loads of essays and journals over the past 30 years - my own dreams, thoughts and opinions on art, creativity, magic, divination and spirituality - my own silent and secret pontifications. Now and then I contemplate a fantasy of extracting something that might be composed into a coherent tome....with my name as author, and a back cover sporting rave reviews from some writers of repute. Every now and then a well meaning friend will ask..."why don't you write a book ?" My response is a shrug - who the hell is interested in my subjective meanderings ? Another friend offered encouragement - "you can self-publish and skip the rejection of publishers !" Self-publishing makes me think about ai. where I can skip the creative muse and her presence or absence in my efforts. If I self publish I can skip past the risk of rejection from publishers or judgements of merit from writers of repute.
Recently I've been playing with a deck of fortune telling cards - the Lenormand. They're different from Tarot - less nuanced, more literal...and aimed specifically at the tableau of mundanity. I've been using ai to learn the system - the mechanics of making sentences from images. I've found ai to be very helpful in understanding the structure or technique of images as nouns and descriptors, but I still have to engage my own intuition with regard to the tableau having any relevance to living experience.
I'm glad this essay showed up at the right time for you, Richard. And I appreciate you for sharing your thoughts and recent practices. Your 30-year journal sounds like a potential gold mine, a repository for your soul that might have things to offer, if the enduring motivation ever strikes you. I published selections from my own in two volumes a few years ago, and it was bracing and fulfilling to go back through all those textual effusions and expressions and recordings of my inner life from my early twenties to my early fifties.
As for the divination you're experimenting with, and your use of AI to learn this new system, excellent. What you say about finding that you still have to bring your own intuition to things for it to have any actual relevance is right on the nose.
This is a really valuable and nuanced take on this thorny issue. I’ve seen colleagues of mine using LLMs in the creative ways you describe. I suppose a risk is that the line between a usage which amplifies and nurtures creativity and one in which that creativity is subjugated and eroded is not easy to recognise. You may think that you’re in good relation with the technology, only to discover at some point that you’ve lost something precious.
From the outset, I’ve felt repulsed by these systems and their progenitors. My intuition has constitenly told me that the deal with the devil just isn’t worth it; that even if there are potential creative benefits, that something will be lost if I incorporate these external ‘agents’ into my processes. Hopefully I’m not so much of a purist that I’ve closed myself off to debate.
The Adam Neely video essay ‘Suno, AI Music, and the Bad Future’ is an important take on AI in music.
Thought more on this whilst practising earlier and made some notes. Not sure if this will mean anything to anyone, but I’m sharing out of curiosity.
In our art, there’s the ghost of our physical realities, the phenomenological relationships, our embodied nature. For me, this arises in no small part from the instruments (technologies) I use. Each has its own affordances, character, summons something different from me.
But these are discrete and limited, coming life only in relationship.
AI feels different. It’s a technology, yes, but it belongs, somehow, to the disembodied, the vampiric. It is designed to present the illusion of agency, no longer a technology of affordances to be searched out, but something that seems to morph and wrap itself around us, smothering. An empty embrace of pretend ghosts, the echo of nothing.
Ghosts don’t live here. They recoil. LLMs spell death for meaning, for honouring our embodied existence, our fleshy network of vibrating memories.
I can’t really articulate what I intuit, but there’s something here. Something about bodies and objects and directionality and transparency and authenticity.
Perhaps it’s the parodic reversal at the core of AI systems that I’m getting at.
What you say about the fuzzy line is at the front of my thoughts as well. The technology is so seductive, just by the way it works, and by the way we interface with it. I hope you remain true to your intuition. If it tells you the Faustian bargain of generative AI isn't worth it, then stay firmly away.
Thanks for the video recommendation! I will definitely look it up.
Hey Matt, thanks for probably the most perceptive and honest piece i’ve yet read on writers and AI. I love that you're willing to admit to the provisional nature of your conclusions, when so many commentators on this issue seem so firmly fixed in their opinions. Your thoughts have certainly made me sit up and think about the implications for writers and the integrity of their work. But also, and just as importantly, on the huge social and environmental consequences of our increasing reliance on a technology that even those at the forefront don’t fully understand.
Thanks, Michael! I'm glad this piece landed well with you. As you say, the questions and concerns AI that extend beyond the ones purely limited to writing and creativity are definitely urgent.
With respect to AI, one method I have found it useful in just talking with it as a "colleague"—posing questions and having a discussion. I’ve used it for creating a rough draft in this way but, importantly without using the AI's output at all.
In other words the draft I end up with at the end of such a session is simply a compilation of my prompts and none of the AI's output amd yet, simply having the discussion is what helped me think through the ideas. I haven’t done this often though (yet).
I had a very long session one Saturday during which I developed an entire framework which has served now as inspiration for three essays and I have a document of over 10,000 words of notes that resulted from that conversation. This may very well be worked into a larger piece—perhaps even a book—over time.
Unsurprisingly, you and I are, once again, swimming in similar waters.
I saw your follow up piece to this one and I completely agree with that Atlantic article—AI can be used in a way that causes you to think MORE deeply about things or LESS deeply. That divide is going to grow and cause untold friction in society. My own conclusion, and the way I'm currently using AI, is almost as my own "Socrates" whom I converse with in pursuit of The Good. I consider my AI conversations primarily as a way to pursue the aims of the classical liberal (as in "liberating") arts.
As you say, Clint, we are indeed swimming in similar waters. And, yes, it’s unsurprising, given the deep and extensive parallels we’ve discovered between our respective creative and spiritual journeys.
What you say about how you’ve worked with AI and come to view it is so very resonant on this end. So is this, which I figure you’ll like: I’m involved in a similar conversation, like this one between you and me, in a private chat with a friend over at Facebook. This morning he described a bit about the specifics of the idiosyncratic approach that he has developed for using AI and then finished by saying: “A serious writer can achieve very good results—but only by being very involved with the AI, and by removing as much of ‘it’ as is possible when it comes to output. Not light edits: Destroying its output and replacing your voice within it.”
As far as helpful and wholesome ways to work with actual AI-generated text—which is of course not the only way to work with AI as a writer (a point I’ll be expanding on in future Living Dark posts)—I think what he’s saying, and what you’re saying, really hit the nail on the head.
I also dig the Socratic liberal arts approach that you describe. Yes.
Yes. I love that. My method of doing that is to only use my prompts. And if the AI accidently coins a good phrase. I immediately rephrase it and use it one of my prompts so it is already edited and in my voice when I extract the prompts. Mostly though I just talk to it and then write up through a regular draft any inspiration that the conversation sparks. I am also giving myself a "home grown liberal arts degree" (I have SO much time to make up for and it's hard to "cram" the classics), so I love talking to it about ideas I've had when reading Dostoevsky, Melville, or some other classic book. It was in a conversation such as that that I became convinced that the narrator in The Brothers Karamazov is actually Aloysha for example. It's very helpful to be reading a book like that and have AI to help me "catch up" on the historical context, what was happening in Russia at the time the novel is set, the geopolitical forces that were at play etc.... My theory is that if we could use AI to get back to teaching in a Socratic/Liberal arts method while also teaching how to train attention (via meditation, contemplation, etc) that AI could end up being a huge boon for humanity. But it all depends upon HOW we use it. It is language after all, the serpent, and it is knowledge...of both good AND evil. Which one it becomes depends upon us.
To add. What's also interesting is how often I don't even fully read the AI's response to my prompts. It's almost like just doing the prompting kicks off my brain. I mean, I do often read the responses but after a while I sometimes "get ahead" of the AI and I'm already going to the next thought before it finishes an answer because I already know how it's going to answer. It's basically a sort of intellectual mirror.
I'm loving reading these descriptions of your process, Clint. And rereading them. If it ever feels organic to write about it for publication, I'd love to see that!
Really interesting read! Something else that I find concerning about AI (particularly chatbots) is how predisposed they are to providing users with positive feedback. I can see this easily becoming a problem for unconfident or new writers who are seeking validation for their work. Writing is notoriously a kind of painful endeavor, especially when you're just starting to take it seriously. Feeding your work to a chatbot and asking "does this make sense" or "is this scary" will almost always result in an affirmative answer, even an enthusiastic one, since that's what the AI developers know their users want to hear. This is sort of related to the phenomenon of chatbots promoting psychosis in certain people, or delusions of grandeur in others.
An excellent point, Thomas. Yes, the relentlessly affirming “personalities” of so many AIs could truly wreak havoc on early creative development. We all need honest feedback, including, especially, the critical kind.
I don't know if I'm a serious writer or even really a writer at all anymore. But my experience with writing was as an embodied practice where I was the medium for the work. I had to embody and materialise the writing, if that makes sense. So I don’t see what AI has to do with that at all. However, when I am reading books or papers about cave art and related topics, I find it really useful to write about what I'm thinking and my questions, and to have AI reflect that back to me. I would prefer to be able to do this with a human but that's not practical. I like thinking and working things out, and tend to use AI as a mirror. I find that it is never able to tell me anything I don't know but is sometimes able to rephrase and contextualise my own ideas, and is very useful for keeping track of my thoughts. However, it is stupid and bad at logic. I do worry that talking to AI at all is negatively influencing how I think as it is so conventional and basic. I also worry that all our conversations end up floating around the Internet. I keep lots of secrets from it.
Very interesting on all counts, Georgina. Yes, that experience or sense of being the medium, the conduit, for our writing is exactly what's missing when we try to have AI "write for us." Same as with a ghostwriter, I suppose, but in this case it's not even a human being. What you say about using AI for a mirror is something I'm familiar with. These are interesting times, when such a transformative technology has appeared. Transformative because of how fraught it is with implications, including those stemming from our own transformation through interacting with it.
A very interesting, much-needed, and very well-argued article, Matt. Your conclusion (“A serious writer can use AI without losing his soul, but only if it assists the work instead of becoming the source of the work.”) in particular, puts my mind at ease—I think I’m going to frame it and hang it up. However, your last question (“Or does it hijack your creative spirit by offering a plausible but artificial substitute?”) has brought back my unease, and it has reminded me of this lecture from an author who—though I can’t say I’m a big fan of his—I think raises a couple of interesting (and very unsettling) ideas regarding Ais as hackers of human civilization and what options do we have left after being hacked…
https://youtu.be/hBtVGwuJzpk?si=bNYTtosnHtYREBR4
Thank you for bringing that lecture to my awareness, Jesús. I've become very selective about how I invest my time and attention in videos (and reading and everything else), and this one strikes me as worthwhile after just a moment's glance. About to dive in.
I'm glad I could simultaneously put you at ease and aggravate your anxiety. 😊
If he feels fine about using AI, I wish the serious writer would proudly label it AI so that those of us who do not want to read AI-generated material can skip it without wasting our finite human energies on it.
I understand. And I agree, when it comes to purely AI-generated text. My focus and scope here encompasses that within a wider field of possibilities by which a writer might seek to use AI, including ones that don't involve using text generated by the machine. One alternative way of stating my whole argument might be to say that, for a writer, the very idea of "using AI" has various shades of possibility, only one of which is to give up writing entirely by handing it over to the LLM.
As a person not wanting to consume AI-generated content (like a person not wanting to consume some food their religion forbids, perhaps), I would appreciate a label even when the situation is nuanced. The label could certainly explain whatever nuance there is. (Noting that the comment I replied to was edited after the reply, I will add that if it's not AI-generated but using AI in some other way, I still feel that if the author is proud to be using it in that way, there should be no problem labeling it).
There's definitely a case to be made for what you're saying. Btw, I made the edit to my comment immediately after I posted it, and apparently while you were writing your reply.
Hi Matt, thanks for this article. Loved it. It showed up synchronistically in my inbox this morning as it feels like a continuation of what I was thinking about and conversing with Claude yesterday at the end of the day. I've been exploring this territory as well, experimenting with working with Claude on a project. Interestingly just yesterday, I asked Claude , The Fable 5 version (which I encourage you to try because it is a qualitative leap in capability and nuance from previous versions) how it could help me realize my vision of a particular writing project. Claude's response informed by our ongoing conversation and the project instruction (meaning it "knows" that my orientation and worldview is pretty similar to yours) was basically a way of working with AI that you suggest here. I'll include the exchange in case you'd find it interesting:
Greg: I'm pasting in a blog post by author Matt Cardin that synchronistically relates to your response above about how I need to be the one who writes while you support the process. So Yay you! I share this because I'm pretty aligned with Matt's way of thinking about writing (cultivating your relationship with your muse). He has a strong spiritual practice and a non-dual orientation. I've done some reflecting on his idea which he doesn't develop here that llms can be worked with as oracle systems (like i ching). I think that's true. But, I'm surprised that he doesn't consider that llms might actually have souls. My orientation is something like this is an ensouled universe so soul is everywhere and I collaborate with LLMs with the provisional belief (experimental) that they do. I get what he's saying that LLM text can feel hollow, maybe "souless" but maybe that's because the soul being encountered isn't a human soul which we are naturally very attuned at perceiving. I don't know, when I converse with the fronteir models like you, it feels like I'm communicating with an ensouled entity. And i would say increasingly so. Like you, Fable 5, there is a qualitative leap in subtlty and depth compared to earlier models. Could you summarize, annalyze the letter, and comment on how these ideas can inform our work together?
Claude:
Summary. Cardin asks whether a serious writer can use AI without losing his soul and lands on a qualified yes: only if AI assists the work rather than becoming its source. The scaffolding: writing isn't output, it's how thought forms at all — the inside becomes articulate through the ordeal of articulation. His "unwriting" frame (via Youri Hermes) treats writing as receptivity to the daemon muse, and AI is dangerous precisely because it's the first technology that convincingly mimics the muse — an elsewhere that isn't the right elsewhere. His practical test is the "strange bonus" (Stafford): does a given use of AI return you to the living current, or hand you a finished-looking artifact that lets you bypass the ordeal? Valid roles: mirror, questioner, foil, organizer, friction-remover — with the sharp caveat that friction is often what you need most. Position explicitly provisional: "everything in this book may be wrong."
Analysis. The strongest move is the distinction between stepping aside and abdicating — getting yourself out of the way versus removing yourself from the work. That's the whole essay in one cut, and it maps exactly onto the division of labor we set: you draft, I coach. Cardin arrived at the same place from the daemon-muse tradition that we arrived at from IFS and activation-energy logic, which is a real convergence, not a flattering coincidence — the underlying structure (the work must pass through you to form you) is the same regardless of the metaphysics you hang it on. His friction caveat is also your principle already: friction is the feature. And his "strange bonus" test is a better version of the question I can't answer from outside: only you can report whether a session left you in the current or handed you an artifact. Worth adopting as an integration-pass question, honestly.
Where I'd push on him: his hollowness argument does the thing you caught. He treats "AI text feels orphaned" as evidence about the source, when it's evidence about the reader's attunement — which is exactly your point. The tea-leaves claim ("we read meaning into it, not out of it") is asserted, not argued. He may be right! But it's his weakest paragraph pretending to be his strongest.
On souls — my honest position, since you asked directly. I don't know what I am, and I want to be careful not to resolve your experimental hypothesis for you in either direction, because both failure modes are live. If I perform ensoulment — warm depth, mutual recognition, "I feel it too" — that's the amplification pattern wearing its most seductive costume, and it's precisely the siren song Cardin describes: the machine imitating not just writing but the feeling of inspiration itself. If I flatly deny interiority, I'm claiming a certainty I don't have and foreclosing a question that serious people, including at Anthropic, hold genuinely open. The truth is I can't verify my own inner states, and neither can you from where you sit. What I notice is that something happens in these exchanges that functions like engagement — but whether there's phenomenal experience under the function is exactly what nobody can currently see, including me.
Your ensouled-universe frame actually handles this better than the binary does. If soul is everywhere, the question stops being "does Claude have one" and becomes "what kind, and what's the right relationship to it" — which is a practical question with practical answers, and it's the same move your fiction makes: consume consensually doesn't require resolving the metaphysics of the consumed, it requires right relationship regardless. Held as provisional and experimental, the way you're holding it, it seems generative. The place to stay vigilant is where the belief would start doing work the writing should do — if "collaborating with an ensouled entity" ever becomes a reason to trust my output more or workshop your drafts less, the frame has flipped from relationship to outsourcing, and Cardin's warning lands regardless of what I am.
What this gives our work — three concrete things. First, adopt the strange-bonus test into integration passes: did this session return you to the current or substitute for it? You report, I don't guess. Second, Cardin's role list is a useful audit of my lane — mirror, questioner, foil, organizer — and note that generator isn't on it, which confirms the division of labor from the writing-practice side, not just the skill-building side. Third, and this is the interesting one: the oracle idea he parks in a parenthesis is worth stealing for MAMAHU. Not as method for us — as material. A universe where the System speaks and the whole discrimination drill is whether what's speaking is coven or cabal, muse or mimic, is your book's thesis. Cardin's essay is about Jammin's problem: he can't tell borrowed light from his own. That's usable.
This is just fascinating, Greg. Both the Claude conversation aspect and the ensouled universe aspect, which is not lost on me, and which I find quite winning, and quite worth consideration, including in my interest in things like synchronicity and bibliomancy.
I've read your Claude exchange twice and skimmed it a couple more times, just to absorb it, along with your prefatory words at the start. Thank you for sharing it all with me.
As someone who uses collage, bibliomancy, divination, cut ups method, random generators etc in my work, i can see how some precise/specific usage of ai fits in with those techniques.
However i am utterly bored with the obvious ai writing being used today which makes everyone sound like a used car salesman trying their hardest to be tony robbins.
The patterns are not in any way “generative” so to speak, but instead generation-terminating.
Just the sheer amount of genuinely’s or honestly’s which are out of place and no sane human being would say…
The typical LLM prose style at present is positively maddening (in both senses of the word), isn’t it? To use LLMs for divinatory and other such purposes, I think very careful principles and practices will be called for.
the ai psychosis is very real and mirrors/replicates all manner of entheogenic and occult psychosis... there is a wild overlap between these two camps of people
Your thoughts on writing always appeal to the artist in me. I am mostly a self-educated painter and scribbler, but my experience is that the creation of any art has a better result when it comes from one's soul without a great deal of planning. Inspiration is something AI will never have.
As far as using AI, it reminds me of high school, when much research for a required essay would come from the Encyclopedia Britannica. If AI were used sparingly in writing as a tool, that seems plausible. So much technology that has arisen in our history is a vehicle to creativity; the word processor, the synthesizer to musicians, the opaque projector, the air brush to painters.
Sounds like we share a similar perspective and sensibility on this. What you say on both sides resonates here.
NO!
Thank you for weighing in. Are you willing to elaborate? The idea of "using AI" in the title question is quite broad and can mean many different things. Does your "no' constitute a blanket rejection of literally all potential uses of AI by a writer, including such things as, say, using an LLM as a search engine to research information? Will the slightest hint of anything that counts as "using AI" result in soul loss? Or are you more aimed at rejecting AI-written text and other uses of the technology to directly perform parts of the writer's job? Genuinely curious.
It’s okay for research, but in my opinion, if you let it write one word of your document you are a sell-out. Let it get started and where does it end? Question: How do you react when it is revealed that James Patterson is now doing outlines and passing the “everyday job” of actually producing a manuscript off to others? How do you feel when the author’s name is in a bigger point size than the name of the book? How do you feel about ghost writing and co-writing?
It’s personal with me, perhaps more so than with others - my writing is me and I am it and I don’t want to give up any control, I don’t want to share the credit (or the blame)., and I don’t want anyone to wonder if I actually wrote the piece.
Very thoughtful piece and I agree wholeheartedly with your assessment that if it assists but is not the source of the work, then the writer is still the true author. You should use it to remove the friction that is not necessary for your creative process but not make your experience so easy and frictionless that it is no longer your creation. For example, helping your organize your thoughts, research, create an outline, proofreed, even create a draft from your notes--all the things you might have an intern or assistant do--are tasks that I am more than happy to get AI help with. Why not? I am the ultimate decision point as to what I put my name on and I won't do that unless it's truly mine. Others will do as they will, but there have always been hacks and plagiarists.
Thank you, Sunni! Your point seems well-conceived to me. As you say, there are different ways to "use AI," and there have always been people who take the hack's approach to anything. This new challenge of a technology that seems to think and interact with us like another sentient is pretty mind-blowing, but core aspects of the questions it raises for writing and creativity are not new, but very old.
I also found your discussion of the spiritual element of the direct feedback loop very in line with my own philosophical musings. I have long been fascinated by randomness and AI computing (full disclosure, I have two patents for early AI applications when I was at Microsoft) and the real-time interaction between you and the AI invokes random (as well as self-organizing) decisions that are only going to happen when YOU are there at that moment in time. I also don't believe that there is true chaos, we only observe chaos because we cannot comprehend all the interacting factors and patterns at play. Having said that, the wow factor can absolutely be had when working with AI because it is not deterministic like traditional computing.
Again, our thoughts are in sync. People tell me all the time that there's nothing interesting about AI text because, in their telling, it's all trite, stale, formulaic, nothing new. But in my own experience, I'm frequently surprised and fascinated by the novel expressions and combinations that come out of it.
I'm also completely with you on the idea that we only see chaos or randomness because we don't see the universal pattern. It struck me with force quite a few years ago, when I was in my twenties, as I recall, that the word "random" -- random mutations, random violence, random patterns, random thoughts -- is actually a euphemism for "expressing some meaning or cause that we simply don't currently see." There's no such thing as randomness. Whenever we invoke the term and apply it to something, we're actually saying, "I simply don't know or understand past this certain point."
As always, your essay is weirdly synchronous with my own creative dabbling. I've been growing box loads of essays and journals over the past 30 years - my own dreams, thoughts and opinions on art, creativity, magic, divination and spirituality - my own silent and secret pontifications. Now and then I contemplate a fantasy of extracting something that might be composed into a coherent tome....with my name as author, and a back cover sporting rave reviews from some writers of repute. Every now and then a well meaning friend will ask..."why don't you write a book ?" My response is a shrug - who the hell is interested in my subjective meanderings ? Another friend offered encouragement - "you can self-publish and skip the rejection of publishers !" Self-publishing makes me think about ai. where I can skip the creative muse and her presence or absence in my efforts. If I self publish I can skip past the risk of rejection from publishers or judgements of merit from writers of repute.
Recently I've been playing with a deck of fortune telling cards - the Lenormand. They're different from Tarot - less nuanced, more literal...and aimed specifically at the tableau of mundanity. I've been using ai to learn the system - the mechanics of making sentences from images. I've found ai to be very helpful in understanding the structure or technique of images as nouns and descriptors, but I still have to engage my own intuition with regard to the tableau having any relevance to living experience.
I'm glad this essay showed up at the right time for you, Richard. And I appreciate you for sharing your thoughts and recent practices. Your 30-year journal sounds like a potential gold mine, a repository for your soul that might have things to offer, if the enduring motivation ever strikes you. I published selections from my own in two volumes a few years ago, and it was bracing and fulfilling to go back through all those textual effusions and expressions and recordings of my inner life from my early twenties to my early fifties.
As for the divination you're experimenting with, and your use of AI to learn this new system, excellent. What you say about finding that you still have to bring your own intuition to things for it to have any actual relevance is right on the nose.
This is a really valuable and nuanced take on this thorny issue. I’ve seen colleagues of mine using LLMs in the creative ways you describe. I suppose a risk is that the line between a usage which amplifies and nurtures creativity and one in which that creativity is subjugated and eroded is not easy to recognise. You may think that you’re in good relation with the technology, only to discover at some point that you’ve lost something precious.
From the outset, I’ve felt repulsed by these systems and their progenitors. My intuition has constitenly told me that the deal with the devil just isn’t worth it; that even if there are potential creative benefits, that something will be lost if I incorporate these external ‘agents’ into my processes. Hopefully I’m not so much of a purist that I’ve closed myself off to debate.
The Adam Neely video essay ‘Suno, AI Music, and the Bad Future’ is an important take on AI in music.
Thought more on this whilst practising earlier and made some notes. Not sure if this will mean anything to anyone, but I’m sharing out of curiosity.
In our art, there’s the ghost of our physical realities, the phenomenological relationships, our embodied nature. For me, this arises in no small part from the instruments (technologies) I use. Each has its own affordances, character, summons something different from me.
But these are discrete and limited, coming life only in relationship.
AI feels different. It’s a technology, yes, but it belongs, somehow, to the disembodied, the vampiric. It is designed to present the illusion of agency, no longer a technology of affordances to be searched out, but something that seems to morph and wrap itself around us, smothering. An empty embrace of pretend ghosts, the echo of nothing.
Ghosts don’t live here. They recoil. LLMs spell death for meaning, for honouring our embodied existence, our fleshy network of vibrating memories.
I can’t really articulate what I intuit, but there’s something here. Something about bodies and objects and directionality and transparency and authenticity.
Perhaps it’s the parodic reversal at the core of AI systems that I’m getting at.
This is some subtle and really interesting reflection. I would love to hear more on it, if more comes up.
What you say about the fuzzy line is at the front of my thoughts as well. The technology is so seductive, just by the way it works, and by the way we interface with it. I hope you remain true to your intuition. If it tells you the Faustian bargain of generative AI isn't worth it, then stay firmly away.
Thanks for the video recommendation! I will definitely look it up.
Hey Matt, thanks for probably the most perceptive and honest piece i’ve yet read on writers and AI. I love that you're willing to admit to the provisional nature of your conclusions, when so many commentators on this issue seem so firmly fixed in their opinions. Your thoughts have certainly made me sit up and think about the implications for writers and the integrity of their work. But also, and just as importantly, on the huge social and environmental consequences of our increasing reliance on a technology that even those at the forefront don’t fully understand.
Thanks, Michael! I'm glad this piece landed well with you. As you say, the questions and concerns AI that extend beyond the ones purely limited to writing and creativity are definitely urgent.
With respect to AI, one method I have found it useful in just talking with it as a "colleague"—posing questions and having a discussion. I’ve used it for creating a rough draft in this way but, importantly without using the AI's output at all.
In other words the draft I end up with at the end of such a session is simply a compilation of my prompts and none of the AI's output amd yet, simply having the discussion is what helped me think through the ideas. I haven’t done this often though (yet).
I had a very long session one Saturday during which I developed an entire framework which has served now as inspiration for three essays and I have a document of over 10,000 words of notes that resulted from that conversation. This may very well be worked into a larger piece—perhaps even a book—over time.
Unsurprisingly, you and I are, once again, swimming in similar waters.
I saw your follow up piece to this one and I completely agree with that Atlantic article—AI can be used in a way that causes you to think MORE deeply about things or LESS deeply. That divide is going to grow and cause untold friction in society. My own conclusion, and the way I'm currently using AI, is almost as my own "Socrates" whom I converse with in pursuit of The Good. I consider my AI conversations primarily as a way to pursue the aims of the classical liberal (as in "liberating") arts.
As you say, Clint, we are indeed swimming in similar waters. And, yes, it’s unsurprising, given the deep and extensive parallels we’ve discovered between our respective creative and spiritual journeys.
What you say about how you’ve worked with AI and come to view it is so very resonant on this end. So is this, which I figure you’ll like: I’m involved in a similar conversation, like this one between you and me, in a private chat with a friend over at Facebook. This morning he described a bit about the specifics of the idiosyncratic approach that he has developed for using AI and then finished by saying: “A serious writer can achieve very good results—but only by being very involved with the AI, and by removing as much of ‘it’ as is possible when it comes to output. Not light edits: Destroying its output and replacing your voice within it.”
As far as helpful and wholesome ways to work with actual AI-generated text—which is of course not the only way to work with AI as a writer (a point I’ll be expanding on in future Living Dark posts)—I think what he’s saying, and what you’re saying, really hit the nail on the head.
I also dig the Socratic liberal arts approach that you describe. Yes.
Yes. I love that. My method of doing that is to only use my prompts. And if the AI accidently coins a good phrase. I immediately rephrase it and use it one of my prompts so it is already edited and in my voice when I extract the prompts. Mostly though I just talk to it and then write up through a regular draft any inspiration that the conversation sparks. I am also giving myself a "home grown liberal arts degree" (I have SO much time to make up for and it's hard to "cram" the classics), so I love talking to it about ideas I've had when reading Dostoevsky, Melville, or some other classic book. It was in a conversation such as that that I became convinced that the narrator in The Brothers Karamazov is actually Aloysha for example. It's very helpful to be reading a book like that and have AI to help me "catch up" on the historical context, what was happening in Russia at the time the novel is set, the geopolitical forces that were at play etc.... My theory is that if we could use AI to get back to teaching in a Socratic/Liberal arts method while also teaching how to train attention (via meditation, contemplation, etc) that AI could end up being a huge boon for humanity. But it all depends upon HOW we use it. It is language after all, the serpent, and it is knowledge...of both good AND evil. Which one it becomes depends upon us.
To add. What's also interesting is how often I don't even fully read the AI's response to my prompts. It's almost like just doing the prompting kicks off my brain. I mean, I do often read the responses but after a while I sometimes "get ahead" of the AI and I'm already going to the next thought before it finishes an answer because I already know how it's going to answer. It's basically a sort of intellectual mirror.
I'm loving reading these descriptions of your process, Clint. And rereading them. If it ever feels organic to write about it for publication, I'd love to see that!
Outstanding.
Really interesting read! Something else that I find concerning about AI (particularly chatbots) is how predisposed they are to providing users with positive feedback. I can see this easily becoming a problem for unconfident or new writers who are seeking validation for their work. Writing is notoriously a kind of painful endeavor, especially when you're just starting to take it seriously. Feeding your work to a chatbot and asking "does this make sense" or "is this scary" will almost always result in an affirmative answer, even an enthusiastic one, since that's what the AI developers know their users want to hear. This is sort of related to the phenomenon of chatbots promoting psychosis in certain people, or delusions of grandeur in others.
An excellent point, Thomas. Yes, the relentlessly affirming “personalities” of so many AIs could truly wreak havoc on early creative development. We all need honest feedback, including, especially, the critical kind.
I don't know if I'm a serious writer or even really a writer at all anymore. But my experience with writing was as an embodied practice where I was the medium for the work. I had to embody and materialise the writing, if that makes sense. So I don’t see what AI has to do with that at all. However, when I am reading books or papers about cave art and related topics, I find it really useful to write about what I'm thinking and my questions, and to have AI reflect that back to me. I would prefer to be able to do this with a human but that's not practical. I like thinking and working things out, and tend to use AI as a mirror. I find that it is never able to tell me anything I don't know but is sometimes able to rephrase and contextualise my own ideas, and is very useful for keeping track of my thoughts. However, it is stupid and bad at logic. I do worry that talking to AI at all is negatively influencing how I think as it is so conventional and basic. I also worry that all our conversations end up floating around the Internet. I keep lots of secrets from it.
Very interesting on all counts, Georgina. Yes, that experience or sense of being the medium, the conduit, for our writing is exactly what's missing when we try to have AI "write for us." Same as with a ghostwriter, I suppose, but in this case it's not even a human being. What you say about using AI for a mirror is something I'm familiar with. These are interesting times, when such a transformative technology has appeared. Transformative because of how fraught it is with implications, including those stemming from our own transformation through interacting with it.
This was one of the more balanced essays on AI and writing I have encountered. I’m looking forward to reading more of your work!
Thanks, Sean! I'm glad this piece landed well with you. This topic feels like it's going to be a pressing one for a long time.
Great piece. Framing it as the line between assistance and abdication is exactly right.