Very interesting article! Brings a few thoughts to mind:
It reminds me of debates I had in my religious days about the authorship of the Bible. So much of the religious tradition rested on the fact that it had divine authorship, and yet many wished to argue that the text itself was divine regardless of where it came from, because we decided to treat it as such. But given that Torah Study in particular was seen as a way to become more god-like in nature, and you take on the mind of the author you read by coming into relationship, I wonder what sort of being you take on when you enter the 'mind' of an authorless LLM, and what that will do to readers who take it seriously. It's hard to imagine something religion-like not springing up.
Also, the idea of custom novels feels a bit off-putting too. There is something that feels wrong about being able to change the ending of a book you do not like, for example, or have a book written just for you. On the one hand, that power would be incredible to use, but also I think reading is supposed to be an encounter with limits. The text stands on its own, and changes you, and forces you to wrestle with it, with something that stands outside of yourself. Being able to control or change it on command disrupts our entire sense of a shared reality, in addition to endorsing a sort of narcissism that prevents true formation from occurring. Almost like an algorithmic social media filter bubble on steroids. I think the accelerated collapse of an agreed upon shared reality (which is already happening) is also a huge problem when you lose the authority of authorship.
In any case, it feels nice that I can engage with you (a human being who I could enter into relationship with) and not a faceless pile of bits here!
Sam, some interesting thoughts here. Religious texts are an interesting case because they have an authority that exceeds that of any other text. In fact, Barthes thinks that the concept of authorship as such partakes in some small way of the supreme authority of the author of scripture: the author's authority is just a smaller and more limited version of the ultimate Author of the ultimate Text.
And yes, books should be a challenging encounter with a perspective different from your own, not an affirmation of the self. The opportunities to accelerate echo-chambers and filter-bubbles really are disturbing.
Very tangential, but it makes a lot of sense that you're a Johnny. Often makes me regret not going there when I meet Johnnies and see how much fun they are in conversation.
Honestly, the Great Books stuff gets all the attention--and that's an important part of what the College does, obviously--but I can't overstate how formative it is to be thrown into conversation over and over again with a small group of other students who just read what you've read. There's a good reason Johnnies tend to be fun people to talk to!
Patrick, this is a wonderfully thoughtful essay. Thank you! Your treatment of Barthes is nuanced and does real work in a short space. There’s a lot here to think about besides, and I’m glad to have taken the time to slow down and engage with your writing today.
Thank you for reading, Gregory! These longer pieces make demands of the reader that are out of step with the fast-paced scroll of Substack Notes, so I'm grateful whenever someone is willing to slow down and spend some time with it.
What if the LLM is not the author but the text. Until LLMs can experience time and alteration, they are a medium, not an author, and their author is all the authors they have assimilated (plus, as you mentioned, the prompters, including the corporate owner)
Yes, that's what makes an LLM a different sort of 'author' from any that came before--they are an aggregation of data coordinated to produce a certain kind of output rather than a single, unified identity.
Patrick, I really appreciated this walk-through some of the postmodern ideas and their relationship to LLMs. I came up in my graduate studies in French so these were good reminders for me. More recently I’ve been reading Jean Louis Chretien and one of his very lovely images came to mind that might hit a happy medium between over emphasis on the reader creating meaning and the author creating meaning. It goes like this:
“everyone who writes knows that he is merely the one who ferries words from silence to silence from the silence in which he meditates, and from whence his as yet unuttered speech arises, to the visible silence of the page which welcomes it and preserves it, without the blank whiteness disappearing.” -Ark of Speech
Your core speculation is the one that troubles me the most. If the AI produces satisficing or better texts, what's the toll that we pay for reading it?
Theories of authorship are interesting, but pretty damn abstract. And they can be complicated by rather plebian means. For example, when I read the translation of "The Three Body Probem' by Cixin Liu, am I receiving a reliable depiction of the authorial intent? Insert a historical religious work in a dead or minority language and ask the same question. Then do the same with Shakespeare, even as a native English speaker. Without a basis for lexical and sociological translation, distortions of the authorial intent seem inevitable.
Now, to move into what is hopefully fresh ground: what if the author is a type of intelligence that is alien to you? Sure, an AI is one example--but what if they are truly alien, e.g., from another solar system. Can we rely on our ability to form a mental link to the author? Can we intuit authorial intent? Or are we left with the mere text and a (hopefully) neutral reading of it?
I don't know the answer to any of this. Which is fair support of your position that writing is in a different baliwick from, say, shoemaking. I drew an analogy between AI and shoemaking earlier today, via John Galsworthy's short story "Quality", and now I wonder if it was apt.
Thanks for kicking this article back into the fray. That's the only reason I saw it.
I like the piece! I do wonder, though, why you emphasize the importance of "who is writing" and "who is speaking" and then pivot to oral traditions in which narratives did not seem to belong to a single author but rather were circulated widely and transformed in the process without anyone necessarily knowing the original author. I don't know much about oral traditions (mostly I just vaguely remember reading Eric Havelock's Preface to Plato in grad school), so this is intended as a clarifying question rather than a critique. It seems like oral storytelling is a bit closer to LLM output in a way since it seems to emerge from a mass of anonymous human storytellers rather than a single author. Oral narratives feel almost like internet memes which pass through "communities" and transform along the way without being attached to a particular author. On the other hand, I do understand that oral storytelling is embedded in community and rich human experience which is definitely different from LLMs. I guess I wanted to hear your further thoughts on this, because it feels like I've seen people kind of avoid oral storytelling when discussing the tension between LLMs and authorship, and I was intrigued to see you engage with it directly.
That’s an excellent point. When I write these essays, I often find that I’m doing some pretty careful threading in the first 70% or so, before I open things up in the concluding section where I glance off a number of other possible directions which I don’t have time to adequately pursue before the end.
In an oral context, the obvious answer to the question “who is speaking?“ is the storyteller, but you’re right to point out that, in a deeper sense, the story doesn’t really have its origin with that particular person. Modern LLMs almost seem like an odd return to this sort of collaborative process in which a collective narrative legacy is passed down over time.
I was thinking I’d like to return to this topic again some time soon, and that observation will be on my mind as I keep thinking about this. Thank you!
That seems like a good approach to an essay like this. Substack allows us to be a little more speculative than certain other forms of writing which is what I like about it. We can test out ideas we're still feeling our way through, and we can offer invitations for thinking to the reader that wouldn't be possible if we had to perform certainty all the way though. You've done a good job presenting a well-grounded argument throughout while also honoring the origins of the essay at the end by including an exploration of unsettled ideas to figure out what you think and invite the reader to do the same.
Really enjoyed this. Like how near the end you touched on things that deal with the purposes for speech. One thing I have always wondered about with the "text itself" crowd is if it's possible for an account to be given as to why write at all.
Related to this is a question I have about LLMs productions of speeches. Isn't there a distinction between 'producing a speech' and 'speaking' that is worth articulating. Perhaps LLMs or machines can do the former but I have my doubts about the latter.
You're right that, even though all speech and writing has the power to affect the world in some way, not all speech or writing is produced as the result of the same sort of intentionality. We still have to ask: are these words an expression of agency, or merely of automatic processes? That difference matters a lot.
Thank you for the insightful essay, Patrick! Took me right back to my comp lit studies decades ago :) What about the embodied aspect of the human author, though? I feel your inquiry would perhaps benefit from bringing in some feminist perspectives to text production via lived experiences, etc…I’m not expert enough to give direct recommendations to such thinkers though, sorry.
Thank you, Aki! I didn't manage to get into embodiment here in any explicit way, but you're right to point in that direction. It's where I'm going in the final section here, with the discussion of the author's worldliness. Embodiment, I think, belongs to the worldly and historically-situated character of the author, and if I was to expand the final section, I think that would be a natural place to go.
Thinking back on this discourse now it seems as if what it really is amounts to something like a kind of prophetic cultural prediction of LLMs. But I’ve also found the whole discourse on the death of the author so much thinner than I used to.
Yes, same. I've seen a number of different versions of this argument--that LLMs represent a sort of vindication of the poststructuralists of the last century, who were basically right but just too early. For me, it was the implausibility of this way of thinking about AI-generated text that calls into question the theory which it rests on.
Some really great food for thought here, thank you. I like your writing style, not just your arguments.
One of my initial reactions was that there is one thing AI will never be able to do, and that is to write in my private diary. Yes, there will probably come a day when the levels of electronic surveillance we are routinely subjected to will be such that an AI might be able to produce a passable simulacrum ('Dear Diary, today I went to the supermarket [phone tracking, facial recognition at the checkout], then I read Substack [tracking on the web], then I sat in the garden [satellites, phone location]'). But the AI won't be able to describe what I thought or what I felt in my secret heart about these activities.
But then I read your other piece on 'the AI/god of the gaps' and how the argument against using AI to do one's writing or any other heretofore seemingly quintessentially human activity shouldn't rely on what AI does or does not have capacity to do. And I agree with that. But I was reminded of something I read (probably in Zuboff's Surveillance Capitalism) about how the tech companies know it is difficult to monetise our interiority, because they can't yet literally read our minds, but what they do instead is try to control our innermost thoughts and feelings via the algorithm, by increasing the time we spend marinating in the discourse, and so on. If they do this successfully enough then they really can write my private diary for me because they know how I will think and react, having 'implanted' those thoughts and reactions themselves. At that point its conceivable that the machine will know me better than I know myself. But so what? I will still write my diary because the act of writing it helps me understand the world and myself.
Anyway, these are just a few of the tangents I went on, inspired by your very interesting essay.
Oh this is very good, and would easily support much more, so I encourage you to expand this project.
To that end, a couple of thoughts. (i) Even when there is no "author" speaking, the reader hears a voice. Think of other collaborative endeavors, e.g., movies, or HR. Somebody is speaking. (ii) This is not just the reader's imagination. Requarth in Persuasion just had a nice piece on the alignment, imputation of personality, that users of Grok or Anthropic experience, which seems to have political consequences. (Your murky corporate authorship.) (iii) this piece is framed by, but also somewhat limited by, the figure of "the artist." But if we think of other LLM outputs, e.g. code writing or legal documents or even scientific research, the figure of the authority/author rememerges as the system owner/administrator, the legal party, or the scientist not incidentally seeking personal/professional credit for her work.
There's a lot more, and I should write my own damn essay, but let me close by inverting Barthes: to understand a string of marks to signify, as a text, presupposes an author, attempting to represent. The author is need not be a person, certainly not a single live artist (the Romantic figure), but it is an epistemological necessity. Anyway, very strong outing! Hope to see more!
Really appreciate the kind words, David. I'm thinking I might try to stay with this theme a little longer and explore it a bit more. I have a few ideas about how I might do that.
You raise a very interesting point about the imputation of personality--the fact that, even though there's nothing but data under the hood, people still feel that their chatbot, say, has a distinct personality, and they imagine it to be a 'who' in the sense I mean here. I'll have to think about the implications of this somewhat more.
Fantastic piece of writing, Patrick - on this theme, one question to add is whether an LLM can create a "body of work" in the same way as a human. People grow and change and read more and revisit their old ideas, so situating a piece of writing in a particular moment in time and/or comparing it to the author's other works is another piece of the puzzle that doesn't map onto LLMs in the same way. I guess you could look at training data from different model years but I think you get my point. Nice shout out for SJC - I always went back and forth on the prohibition of outside info, but in the end fell on the side of really adhering to the text. Once that falls, it's very easy to veer off track.
Thanks, Stephen! The question of the 'body of work' is a really interesting one because it highlights the way that the author has a particular perspective on things which evolves over time. It suggests that the work changes as the author changes. It only really makes sense, I think, if you assume that authorship has an inherently subjective and situated quality, something that an unworldliness of data doesn't really support. I had wanted to explore this question a little bit here, but ultimately couldn't make it fit.
Not a fully formulated take, but I've been wondering lately if the death of the author is really an epistemological take that's been treated as an ontological one. Obviously, it's impossible to ever know an author's intentions definitively--the best you can hope for is trusting them. But the fact that this intention can't be definitively known doesn't seem to necessitate that the intention is irrelevant to its meaning. The world's full of unseen phenomena that hold everything together.
Yes, that's interesting. I'm not sure I would call it an epistemological take, though, because unknowability isn't really the issue for Barthes. The reason he wants to deny the author isn't because the author's intentions can't be known, but because those intentions, whether they can be known or not, are extraneous to the text itself.
I think he would say that the imputation of authorship onto a text reflects a certain set of assumptions about what a text IS--assumptions which, prior to the appearance of LLMs, happened to always be true: namely, that a piece of writing is the work of a particular person. His point, as I understand it, is that we can simply choose to stop making those assumptions, and start treating authorial intention as wholly irrelevant to the interpretive project. In the '60s, this was little more than an intellectual exercise, because the truth of the matter was that texts still DID have authors, whether or not you wanted to acknowledge their intentions. The argument that Gunkel (and many others whom I haven't mentioned) is making is that 'the death of the author' is no longer an intellectual game, but has become literally true in the case of AI-generated text.
But to the degree that Barthes is making an ontological argument (namely, that the text operates within an altogether different realm than the real, historically-rooted world which produced it), I'm essentially making the opposite one: not only do human-authored texts not plausibly manage to escape completely from the real world, but AI-authored texts are no more successful in doing so. The truth is that both of these continue to be rooted in real-world reality in ways that 'the death of the author' thesis seems to want to deny, even if those realities are different whether we're talking about a human-authored text or an AI-generated one.
Thanks Thomas! - I mentioned in our meeting a full two weeks ago that this was basically finished, but it's taken until now to finally get it posted. Glad you enjoyed it!
What did photography do to painters? What did the digital camera do to professional photographers? What are AI generated images doing to photographers in general?
With AI writing anyone can now be an “author” because they can “prompt” their own story as much or as little as they wish, and then share their “story” in whatever is the equivalent of instagram.
The upside is that now everyone can use stories as tool for creating and exploring their world and creating new ones. In other words, this is the democratization of people being able to create stories for their own psychological spiritual development. For example, if I’m dealing with a significant decision or psychological problem, I can generate a story for each possible path (perhaps with myself as the protagonist) and then see how it plays out as a way of helping decide or heal.
Very interesting article! Brings a few thoughts to mind:
It reminds me of debates I had in my religious days about the authorship of the Bible. So much of the religious tradition rested on the fact that it had divine authorship, and yet many wished to argue that the text itself was divine regardless of where it came from, because we decided to treat it as such. But given that Torah Study in particular was seen as a way to become more god-like in nature, and you take on the mind of the author you read by coming into relationship, I wonder what sort of being you take on when you enter the 'mind' of an authorless LLM, and what that will do to readers who take it seriously. It's hard to imagine something religion-like not springing up.
Also, the idea of custom novels feels a bit off-putting too. There is something that feels wrong about being able to change the ending of a book you do not like, for example, or have a book written just for you. On the one hand, that power would be incredible to use, but also I think reading is supposed to be an encounter with limits. The text stands on its own, and changes you, and forces you to wrestle with it, with something that stands outside of yourself. Being able to control or change it on command disrupts our entire sense of a shared reality, in addition to endorsing a sort of narcissism that prevents true formation from occurring. Almost like an algorithmic social media filter bubble on steroids. I think the accelerated collapse of an agreed upon shared reality (which is already happening) is also a huge problem when you lose the authority of authorship.
In any case, it feels nice that I can engage with you (a human being who I could enter into relationship with) and not a faceless pile of bits here!
Sam, some interesting thoughts here. Religious texts are an interesting case because they have an authority that exceeds that of any other text. In fact, Barthes thinks that the concept of authorship as such partakes in some small way of the supreme authority of the author of scripture: the author's authority is just a smaller and more limited version of the ultimate Author of the ultimate Text.
And yes, books should be a challenging encounter with a perspective different from your own, not an affirmation of the self. The opportunities to accelerate echo-chambers and filter-bubbles really are disturbing.
Very tangential, but it makes a lot of sense that you're a Johnny. Often makes me regret not going there when I meet Johnnies and see how much fun they are in conversation.
Honestly, the Great Books stuff gets all the attention--and that's an important part of what the College does, obviously--but I can't overstate how formative it is to be thrown into conversation over and over again with a small group of other students who just read what you've read. There's a good reason Johnnies tend to be fun people to talk to!
Patrick, this is a wonderfully thoughtful essay. Thank you! Your treatment of Barthes is nuanced and does real work in a short space. There’s a lot here to think about besides, and I’m glad to have taken the time to slow down and engage with your writing today.
Thank you for reading, Gregory! These longer pieces make demands of the reader that are out of step with the fast-paced scroll of Substack Notes, so I'm grateful whenever someone is willing to slow down and spend some time with it.
What if the LLM is not the author but the text. Until LLMs can experience time and alteration, they are a medium, not an author, and their author is all the authors they have assimilated (plus, as you mentioned, the prompters, including the corporate owner)
Yes, that's what makes an LLM a different sort of 'author' from any that came before--they are an aggregation of data coordinated to produce a certain kind of output rather than a single, unified identity.
They may be the text, not the author
Patrick, I really appreciated this walk-through some of the postmodern ideas and their relationship to LLMs. I came up in my graduate studies in French so these were good reminders for me. More recently I’ve been reading Jean Louis Chretien and one of his very lovely images came to mind that might hit a happy medium between over emphasis on the reader creating meaning and the author creating meaning. It goes like this:
“everyone who writes knows that he is merely the one who ferries words from silence to silence from the silence in which he meditates, and from whence his as yet unuttered speech arises, to the visible silence of the page which welcomes it and preserves it, without the blank whiteness disappearing.” -Ark of Speech
That is a truly excellent passage - thank you for sharing! I'm going to have to try to get a hands on a copy of that book!
Your core speculation is the one that troubles me the most. If the AI produces satisficing or better texts, what's the toll that we pay for reading it?
Theories of authorship are interesting, but pretty damn abstract. And they can be complicated by rather plebian means. For example, when I read the translation of "The Three Body Probem' by Cixin Liu, am I receiving a reliable depiction of the authorial intent? Insert a historical religious work in a dead or minority language and ask the same question. Then do the same with Shakespeare, even as a native English speaker. Without a basis for lexical and sociological translation, distortions of the authorial intent seem inevitable.
Now, to move into what is hopefully fresh ground: what if the author is a type of intelligence that is alien to you? Sure, an AI is one example--but what if they are truly alien, e.g., from another solar system. Can we rely on our ability to form a mental link to the author? Can we intuit authorial intent? Or are we left with the mere text and a (hopefully) neutral reading of it?
I don't know the answer to any of this. Which is fair support of your position that writing is in a different baliwick from, say, shoemaking. I drew an analogy between AI and shoemaking earlier today, via John Galsworthy's short story "Quality", and now I wonder if it was apt.
Thanks for kicking this article back into the fray. That's the only reason I saw it.
Well done and cheers!
Glad to hear I've stirred up these questions in your mind! I'm thinking I might need to return to some of them in a future post.
A phenomenal essay!
I like the piece! I do wonder, though, why you emphasize the importance of "who is writing" and "who is speaking" and then pivot to oral traditions in which narratives did not seem to belong to a single author but rather were circulated widely and transformed in the process without anyone necessarily knowing the original author. I don't know much about oral traditions (mostly I just vaguely remember reading Eric Havelock's Preface to Plato in grad school), so this is intended as a clarifying question rather than a critique. It seems like oral storytelling is a bit closer to LLM output in a way since it seems to emerge from a mass of anonymous human storytellers rather than a single author. Oral narratives feel almost like internet memes which pass through "communities" and transform along the way without being attached to a particular author. On the other hand, I do understand that oral storytelling is embedded in community and rich human experience which is definitely different from LLMs. I guess I wanted to hear your further thoughts on this, because it feels like I've seen people kind of avoid oral storytelling when discussing the tension between LLMs and authorship, and I was intrigued to see you engage with it directly.
That’s an excellent point. When I write these essays, I often find that I’m doing some pretty careful threading in the first 70% or so, before I open things up in the concluding section where I glance off a number of other possible directions which I don’t have time to adequately pursue before the end.
In an oral context, the obvious answer to the question “who is speaking?“ is the storyteller, but you’re right to point out that, in a deeper sense, the story doesn’t really have its origin with that particular person. Modern LLMs almost seem like an odd return to this sort of collaborative process in which a collective narrative legacy is passed down over time.
I was thinking I’d like to return to this topic again some time soon, and that observation will be on my mind as I keep thinking about this. Thank you!
That seems like a good approach to an essay like this. Substack allows us to be a little more speculative than certain other forms of writing which is what I like about it. We can test out ideas we're still feeling our way through, and we can offer invitations for thinking to the reader that wouldn't be possible if we had to perform certainty all the way though. You've done a good job presenting a well-grounded argument throughout while also honoring the origins of the essay at the end by including an exploration of unsettled ideas to figure out what you think and invite the reader to do the same.
Patrick
Really enjoyed this. Like how near the end you touched on things that deal with the purposes for speech. One thing I have always wondered about with the "text itself" crowd is if it's possible for an account to be given as to why write at all.
Related to this is a question I have about LLMs productions of speeches. Isn't there a distinction between 'producing a speech' and 'speaking' that is worth articulating. Perhaps LLMs or machines can do the former but I have my doubts about the latter.
You're right that, even though all speech and writing has the power to affect the world in some way, not all speech or writing is produced as the result of the same sort of intentionality. We still have to ask: are these words an expression of agency, or merely of automatic processes? That difference matters a lot.
Thank you for the insightful essay, Patrick! Took me right back to my comp lit studies decades ago :) What about the embodied aspect of the human author, though? I feel your inquiry would perhaps benefit from bringing in some feminist perspectives to text production via lived experiences, etc…I’m not expert enough to give direct recommendations to such thinkers though, sorry.
Thank you, Aki! I didn't manage to get into embodiment here in any explicit way, but you're right to point in that direction. It's where I'm going in the final section here, with the discussion of the author's worldliness. Embodiment, I think, belongs to the worldly and historically-situated character of the author, and if I was to expand the final section, I think that would be a natural place to go.
Thinking back on this discourse now it seems as if what it really is amounts to something like a kind of prophetic cultural prediction of LLMs. But I’ve also found the whole discourse on the death of the author so much thinner than I used to.
Yes, same. I've seen a number of different versions of this argument--that LLMs represent a sort of vindication of the poststructuralists of the last century, who were basically right but just too early. For me, it was the implausibility of this way of thinking about AI-generated text that calls into question the theory which it rests on.
Some really great food for thought here, thank you. I like your writing style, not just your arguments.
One of my initial reactions was that there is one thing AI will never be able to do, and that is to write in my private diary. Yes, there will probably come a day when the levels of electronic surveillance we are routinely subjected to will be such that an AI might be able to produce a passable simulacrum ('Dear Diary, today I went to the supermarket [phone tracking, facial recognition at the checkout], then I read Substack [tracking on the web], then I sat in the garden [satellites, phone location]'). But the AI won't be able to describe what I thought or what I felt in my secret heart about these activities.
But then I read your other piece on 'the AI/god of the gaps' and how the argument against using AI to do one's writing or any other heretofore seemingly quintessentially human activity shouldn't rely on what AI does or does not have capacity to do. And I agree with that. But I was reminded of something I read (probably in Zuboff's Surveillance Capitalism) about how the tech companies know it is difficult to monetise our interiority, because they can't yet literally read our minds, but what they do instead is try to control our innermost thoughts and feelings via the algorithm, by increasing the time we spend marinating in the discourse, and so on. If they do this successfully enough then they really can write my private diary for me because they know how I will think and react, having 'implanted' those thoughts and reactions themselves. At that point its conceivable that the machine will know me better than I know myself. But so what? I will still write my diary because the act of writing it helps me understand the world and myself.
Anyway, these are just a few of the tangents I went on, inspired by your very interesting essay.
Thanks for reading, and I'm glad it was generative for your own thinking!
Oh this is very good, and would easily support much more, so I encourage you to expand this project.
To that end, a couple of thoughts. (i) Even when there is no "author" speaking, the reader hears a voice. Think of other collaborative endeavors, e.g., movies, or HR. Somebody is speaking. (ii) This is not just the reader's imagination. Requarth in Persuasion just had a nice piece on the alignment, imputation of personality, that users of Grok or Anthropic experience, which seems to have political consequences. (Your murky corporate authorship.) (iii) this piece is framed by, but also somewhat limited by, the figure of "the artist." But if we think of other LLM outputs, e.g. code writing or legal documents or even scientific research, the figure of the authority/author rememerges as the system owner/administrator, the legal party, or the scientist not incidentally seeking personal/professional credit for her work.
There's a lot more, and I should write my own damn essay, but let me close by inverting Barthes: to understand a string of marks to signify, as a text, presupposes an author, attempting to represent. The author is need not be a person, certainly not a single live artist (the Romantic figure), but it is an epistemological necessity. Anyway, very strong outing! Hope to see more!
Really appreciate the kind words, David. I'm thinking I might try to stay with this theme a little longer and explore it a bit more. I have a few ideas about how I might do that.
You raise a very interesting point about the imputation of personality--the fact that, even though there's nothing but data under the hood, people still feel that their chatbot, say, has a distinct personality, and they imagine it to be a 'who' in the sense I mean here. I'll have to think about the implications of this somewhat more.
Fantastic piece of writing, Patrick - on this theme, one question to add is whether an LLM can create a "body of work" in the same way as a human. People grow and change and read more and revisit their old ideas, so situating a piece of writing in a particular moment in time and/or comparing it to the author's other works is another piece of the puzzle that doesn't map onto LLMs in the same way. I guess you could look at training data from different model years but I think you get my point. Nice shout out for SJC - I always went back and forth on the prohibition of outside info, but in the end fell on the side of really adhering to the text. Once that falls, it's very easy to veer off track.
Thanks, Stephen! The question of the 'body of work' is a really interesting one because it highlights the way that the author has a particular perspective on things which evolves over time. It suggests that the work changes as the author changes. It only really makes sense, I think, if you assume that authorship has an inherently subjective and situated quality, something that an unworldliness of data doesn't really support. I had wanted to explore this question a little bit here, but ultimately couldn't make it fit.
Not a fully formulated take, but I've been wondering lately if the death of the author is really an epistemological take that's been treated as an ontological one. Obviously, it's impossible to ever know an author's intentions definitively--the best you can hope for is trusting them. But the fact that this intention can't be definitively known doesn't seem to necessitate that the intention is irrelevant to its meaning. The world's full of unseen phenomena that hold everything together.
Yes, that's interesting. I'm not sure I would call it an epistemological take, though, because unknowability isn't really the issue for Barthes. The reason he wants to deny the author isn't because the author's intentions can't be known, but because those intentions, whether they can be known or not, are extraneous to the text itself.
I think he would say that the imputation of authorship onto a text reflects a certain set of assumptions about what a text IS--assumptions which, prior to the appearance of LLMs, happened to always be true: namely, that a piece of writing is the work of a particular person. His point, as I understand it, is that we can simply choose to stop making those assumptions, and start treating authorial intention as wholly irrelevant to the interpretive project. In the '60s, this was little more than an intellectual exercise, because the truth of the matter was that texts still DID have authors, whether or not you wanted to acknowledge their intentions. The argument that Gunkel (and many others whom I haven't mentioned) is making is that 'the death of the author' is no longer an intellectual game, but has become literally true in the case of AI-generated text.
But to the degree that Barthes is making an ontological argument (namely, that the text operates within an altogether different realm than the real, historically-rooted world which produced it), I'm essentially making the opposite one: not only do human-authored texts not plausibly manage to escape completely from the real world, but AI-authored texts are no more successful in doing so. The truth is that both of these continue to be rooted in real-world reality in ways that 'the death of the author' thesis seems to want to deny, even if those realities are different whether we're talking about a human-authored text or an AI-generated one.
Great essay. Really enjoyed it. Thank you!
Thanks Thomas! - I mentioned in our meeting a full two weeks ago that this was basically finished, but it's taken until now to finally get it posted. Glad you enjoyed it!
What did photography do to painters? What did the digital camera do to professional photographers? What are AI generated images doing to photographers in general?
With AI writing anyone can now be an “author” because they can “prompt” their own story as much or as little as they wish, and then share their “story” in whatever is the equivalent of instagram.
The upside is that now everyone can use stories as tool for creating and exploring their world and creating new ones. In other words, this is the democratization of people being able to create stories for their own psychological spiritual development. For example, if I’m dealing with a significant decision or psychological problem, I can generate a story for each possible path (perhaps with myself as the protagonist) and then see how it plays out as a way of helping decide or heal.