Thank you. I’m now thinking about AI writing in a whole new way: it’s just a fancy sentence organizer and grammar checker. The ideas, thought and decision making behind it are yours or mine — just as they have always been when, eg, a ghostwriter compiled and organized John Kennedy’s ideas in Profiles in Courage, or Malcolm X’s Autobiography.
Yes, and more. The question is who thought it, and then wrestled it into form and being, gave it shape and made it live.
Thought is the conception.
The midwife is the personality or character of the person who generated the thought, and that personality is the product of a life lived.
The midwife cannot be a machine if the product is to be a genuine human artifact. An artistic creation is the stamp of a personality, an existential utterance.
One of the benefits of this technology is efficiency.
I wonder if this achievement serves the goals of capital more that thoughtful knowledge production.
Don’t we already have an over production problem?
Maybe not ?
The increase in AI production will exponentially increase the imbalance between written production and readers.
The decrease in the cost of publishing increased the production of books.
The increase of internet sites that host written production reduced the costs of publication even more. Production of written pieces increased by greater participation online.
Increased access to publishing would seem to be a good thing .
Is there a point at which this widespread access and production could produce an intrinsic downside?
The efficiencies Ai are likely to vastly increase output and reduce the possibility of discriminating between AI and human production.
Is this a good thing?
I think that Mike is saying that discrimination is irrelevant and that Ai used correctly as a tool only makes for better transmission of the writers ideas.This view poses questions about authorships importance as well as traditional copyright protections
May authorship is not so important. Is it time to abandon copyright laws? Of course that is unlikely as it benefits large corporations the most.
It is worth asking whether the probability of the increase in the imbalance of written production will eventually cause downside effects?One possibility is that judgement and review will be decreased in relation to production.Another could be a reaction to this imbalance will be the production of robot readers. They will become a necessity to digest all of the production. Already Science journals are overwhelmed.
If this environment increases competition, the use of AI for whole compositions and books will increase. If we reach a point at which the production imbalance with readers necessitates the use of robot readers it raises the likelihood of complete AI production.We may arrive at a point where AI supplies the only capable writers and readers at speed/scale.
There could be good reasons this does not happen. Failure of the companies,or government regulation perhaps.
However if even a significant part of this develops, it will create an ever increasing dependence on AI for written knowledge production,digestion, storage and retrieval.
Will this reduce human capabilities? It does seem that many memory tasks have been replaced with easy access to records.And schools in many cases have decried the distractions phones and in some cases found AI to be more obstacle than help.
Keep in mind that the Ai dependence even if not as complete as the scenario above contemplates,
that all of Ai information has these attributes:
It has been shaped by the AI companies for both salability and political alignment.
Is this a good thing? I guess it depends. Probably not.
It has been shaped to eliminate access to dangerous information and some types of pornography.
Is this good thing? Depends on who judging I think.
So participation in the AI loop will likely strengthen the largest corporations and will enable a group of ever increasing downside effects.Some of which may be existential.It will increase our dependence on AI for knowledge metabolism. They AI systems are at present unreliable, giving inaccurate information, creating some out of whole cloth, do not obey their creators and those human creators don’t know completely how Ai works.
The existential downsides may be remote. But even the adoption at present increases or dependence as well the likelihood of overproduction as well as the reproduction of AIs inaccuracies.
Lastly, keep in mind that this system is all connected runs on energy production that while having many redundancies has many choke points. If it fails what does that mean for all of the stored knowledge? Even if significant parts fail , are we seriously threatened?
I am not sure.
Think of all the leisure time we’ve been promised!
There is one serious contradiction here. You rightly claim that LLMs are the product of massive amounts of stolen data. And that massive data reflects the recorded thoughts of an unknown large percent of human history.The LLM is completely dependent on that data. Without it,it does not exist. That must mean that the frontier labs existence relies on the greatest intellectual heist in history. A theft of astronomical proportions.
In the world of crime a theft transmits in part to the receiver of what has been stolen. The most culpable is the person who knowingly receives the stolen property. This would mean that, if not in a literal sense)that all the users of LLMs are committing a crime downstream from the unprecedented crime committed by frontier labs. Of course when a crime is so widely and easily committed we tend to absolve everyone.
However, there are other troubling things about your analysis. Using LLMs is not the same as using ghost writer in the normative sense. First a ghost writer agrees to be anonymous and gets paid for it. Second it is almost always detected if not immediately then posthumously. There were always doubts about PT #109. And speechwriting is a particular job that is compensated and the public is aware that it is part of the entire political machine. The speechwriters are usually known and the collaboration is expected.
The use of the Autobiography Malcolm X is a strange choice. Alex Haley’s writing was known to be integral to the success of that book. Furthermore their relation was printed cover of the book-as told to Haley by Malcom X. So the book was not an example of ghost writing but more a way of reporting that became popular in the period: In Cold Blood by Truman Capote fit the the type. Today we recognize the film versions as documentaries or docudramas. All have clear authors and use various points of view.
I fervently oppose LLM detectors as well as LLM watermarks and do so especially due to the consequences of their imperfections- as you point out.
It is the process of shaping LLMs as well as imperfections or mistakes produced by LLMs that I also think are worth considering. If indiscriminate writers use them to help them write in domains they in which they are not already accomplished, they risk of transmitting false information goes up .The argument against this is that it happens anyway, without LLMs. Sloppy, ambitious people are to blame not the tool. However, it is precisely the time saving, the ease that opens the door to an increase in transmissions that are false. LLMs provide instant access to knowledge that a writer is has need of but is not familiar with.This is especially true in polemical writing.
It is true that improved arrangements of written knowledge have tended to increase accessibility.
The invention of the printed book was one of the most important dispersal technologies. The spread of libraries of increasing size greatly extending access.
Eventually came the internet and accessibility seemed miraculous. Of course you could copy and paste documents and word processing eased the job of moving sections of your writing and corrected the spelling and eventually suggested words. All of these technological developments did not significantly change our views of authorship. It was easy to surf new domains however it still required some reading to make sense of this unfamiliar information. It was encyclopedic at whole new speed.
The trend in knowledge transmission follows most other technological development. It satisfies desires, increases production and efficiency. Ai as tool does all of them. It does seem true that increased dispersion of the written word has increased our understanding of our world. It created new domains of practice, increased inventions and so on.
Of course we are increasingly aware of the trade offs, despoiling the environment, use of massively destructive weapons and global climate change. Some see that the use of AI will empower the feedback loop between the downside effects and perceived positive developments.
Since what underlies all of this development at increasing speed is capitalism . Without the accumulation of capital in the corporations, it would be unlikely to unfold in this way. And it seems now incontestable that vast amounts of capital accumulated in the hands of private elites degrades our democratic institutions.
This brings us to some more existential questions.
What are we willing to trade for AI?
Further degradation of our environment!
A narrowing of access as well as decreasing the power of our constitutional legitimized enfranchisement ?
The increase of AI weaponry?
The increase of internal AI surveillance?
The unknowable prospect of AI produced superhuman intelligence?
It is true that some these threats were on the horizon and almost no one,even the most critical,could resist use of the internet. That buy in produced the most powerful corporations in history.
LLM use we can only rationally conclude, feeds the entire system -not just more data-but the buy in financially feeds their corrupt exclusive forms of development and marketing and in turn accelerates the side effects listed above.
Stolen is a strong word. Training a neural network on publicly-available data seems like fair use to me—kids will be doing this in their basement in 5 years, no Anthropic or OpenAI required. However, reproducing copyrighted content by virtue of replicating that content for commercial or otherwise, unauthorized use is infringement and subject to litigation. That's the only position that seems reasonable.
Someone arguing that ANY output from an LLM is plagiarized—which seems to be the vibe I get from some of the online discourse—because it's trained on copyrighted content seems like a overdetermined stance.
If you ask a model to write a random sentence and it writes back: "The quick brown fox jumped over the lazy dog." — is that plagiarism happening? I don't think so.
The real problem one might have with the notion that a trained neural network on copyrighted content makes all derivative output from that training set "stealing" it does need to be said that nearly everything everyone ever says or writes is something they heard elsewhere. Novel thoughts are a rare breed.
Not entirely sure if this is pertinent, but here goes:
As a professional (quite good) graphic designer, I occasionally dream the solution to a visual problem. It doesn’t happen that often and it doesn’t always work, but fuck, when it does, it’s incredible! Dreams are also as indisputable as they are unimaginable. No machine/android sheep will ever dream. How could they?
Thank you. I’m now thinking about AI writing in a whole new way: it’s just a fancy sentence organizer and grammar checker. The ideas, thought and decision making behind it are yours or mine — just as they have always been when, eg, a ghostwriter compiled and organized John Kennedy’s ideas in Profiles in Courage, or Malcolm X’s Autobiography.
I truly love your writing style. It makes me want to master the art of boorish poignancy as means of self-care. Well, attempt anyway.
Yes, and more. The question is who thought it, and then wrestled it into form and being, gave it shape and made it live.
Thought is the conception.
The midwife is the personality or character of the person who generated the thought, and that personality is the product of a life lived.
The midwife cannot be a machine if the product is to be a genuine human artifact. An artistic creation is the stamp of a personality, an existential utterance.
Continuation:
One of the benefits of this technology is efficiency.
I wonder if this achievement serves the goals of capital more that thoughtful knowledge production.
Don’t we already have an over production problem?
Maybe not ?
The increase in AI production will exponentially increase the imbalance between written production and readers.
The decrease in the cost of publishing increased the production of books.
The increase of internet sites that host written production reduced the costs of publication even more. Production of written pieces increased by greater participation online.
Increased access to publishing would seem to be a good thing .
Is there a point at which this widespread access and production could produce an intrinsic downside?
The efficiencies Ai are likely to vastly increase output and reduce the possibility of discriminating between AI and human production.
Is this a good thing?
I think that Mike is saying that discrimination is irrelevant and that Ai used correctly as a tool only makes for better transmission of the writers ideas.This view poses questions about authorships importance as well as traditional copyright protections
May authorship is not so important. Is it time to abandon copyright laws? Of course that is unlikely as it benefits large corporations the most.
It is worth asking whether the probability of the increase in the imbalance of written production will eventually cause downside effects?One possibility is that judgement and review will be decreased in relation to production.Another could be a reaction to this imbalance will be the production of robot readers. They will become a necessity to digest all of the production. Already Science journals are overwhelmed.
If this environment increases competition, the use of AI for whole compositions and books will increase. If we reach a point at which the production imbalance with readers necessitates the use of robot readers it raises the likelihood of complete AI production.We may arrive at a point where AI supplies the only capable writers and readers at speed/scale.
There could be good reasons this does not happen. Failure of the companies,or government regulation perhaps.
However if even a significant part of this develops, it will create an ever increasing dependence on AI for written knowledge production,digestion, storage and retrieval.
Will this reduce human capabilities? It does seem that many memory tasks have been replaced with easy access to records.And schools in many cases have decried the distractions phones and in some cases found AI to be more obstacle than help.
Keep in mind that the Ai dependence even if not as complete as the scenario above contemplates,
that all of Ai information has these attributes:
It has been shaped by the AI companies for both salability and political alignment.
Is this a good thing? I guess it depends. Probably not.
It has been shaped to eliminate access to dangerous information and some types of pornography.
Is this good thing? Depends on who judging I think.
So participation in the AI loop will likely strengthen the largest corporations and will enable a group of ever increasing downside effects.Some of which may be existential.It will increase our dependence on AI for knowledge metabolism. They AI systems are at present unreliable, giving inaccurate information, creating some out of whole cloth, do not obey their creators and those human creators don’t know completely how Ai works.
The existential downsides may be remote. But even the adoption at present increases or dependence as well the likelihood of overproduction as well as the reproduction of AIs inaccuracies.
Lastly, keep in mind that this system is all connected runs on energy production that while having many redundancies has many choke points. If it fails what does that mean for all of the stored knowledge? Even if significant parts fail , are we seriously threatened?
I am not sure.
Think of all the leisure time we’ve been promised!
There is one serious contradiction here. You rightly claim that LLMs are the product of massive amounts of stolen data. And that massive data reflects the recorded thoughts of an unknown large percent of human history.The LLM is completely dependent on that data. Without it,it does not exist. That must mean that the frontier labs existence relies on the greatest intellectual heist in history. A theft of astronomical proportions.
In the world of crime a theft transmits in part to the receiver of what has been stolen. The most culpable is the person who knowingly receives the stolen property. This would mean that, if not in a literal sense)that all the users of LLMs are committing a crime downstream from the unprecedented crime committed by frontier labs. Of course when a crime is so widely and easily committed we tend to absolve everyone.
However, there are other troubling things about your analysis. Using LLMs is not the same as using ghost writer in the normative sense. First a ghost writer agrees to be anonymous and gets paid for it. Second it is almost always detected if not immediately then posthumously. There were always doubts about PT #109. And speechwriting is a particular job that is compensated and the public is aware that it is part of the entire political machine. The speechwriters are usually known and the collaboration is expected.
The use of the Autobiography Malcolm X is a strange choice. Alex Haley’s writing was known to be integral to the success of that book. Furthermore their relation was printed cover of the book-as told to Haley by Malcom X. So the book was not an example of ghost writing but more a way of reporting that became popular in the period: In Cold Blood by Truman Capote fit the the type. Today we recognize the film versions as documentaries or docudramas. All have clear authors and use various points of view.
I fervently oppose LLM detectors as well as LLM watermarks and do so especially due to the consequences of their imperfections- as you point out.
It is the process of shaping LLMs as well as imperfections or mistakes produced by LLMs that I also think are worth considering. If indiscriminate writers use them to help them write in domains they in which they are not already accomplished, they risk of transmitting false information goes up .The argument against this is that it happens anyway, without LLMs. Sloppy, ambitious people are to blame not the tool. However, it is precisely the time saving, the ease that opens the door to an increase in transmissions that are false. LLMs provide instant access to knowledge that a writer is has need of but is not familiar with.This is especially true in polemical writing.
It is true that improved arrangements of written knowledge have tended to increase accessibility.
The invention of the printed book was one of the most important dispersal technologies. The spread of libraries of increasing size greatly extending access.
Eventually came the internet and accessibility seemed miraculous. Of course you could copy and paste documents and word processing eased the job of moving sections of your writing and corrected the spelling and eventually suggested words. All of these technological developments did not significantly change our views of authorship. It was easy to surf new domains however it still required some reading to make sense of this unfamiliar information. It was encyclopedic at whole new speed.
The trend in knowledge transmission follows most other technological development. It satisfies desires, increases production and efficiency. Ai as tool does all of them. It does seem true that increased dispersion of the written word has increased our understanding of our world. It created new domains of practice, increased inventions and so on.
Of course we are increasingly aware of the trade offs, despoiling the environment, use of massively destructive weapons and global climate change. Some see that the use of AI will empower the feedback loop between the downside effects and perceived positive developments.
Since what underlies all of this development at increasing speed is capitalism . Without the accumulation of capital in the corporations, it would be unlikely to unfold in this way. And it seems now incontestable that vast amounts of capital accumulated in the hands of private elites degrades our democratic institutions.
This brings us to some more existential questions.
What are we willing to trade for AI?
Further degradation of our environment!
A narrowing of access as well as decreasing the power of our constitutional legitimized enfranchisement ?
The increase of AI weaponry?
The increase of internal AI surveillance?
The unknowable prospect of AI produced superhuman intelligence?
It is true that some these threats were on the horizon and almost no one,even the most critical,could resist use of the internet. That buy in produced the most powerful corporations in history.
LLM use we can only rationally conclude, feeds the entire system -not just more data-but the buy in financially feeds their corrupt exclusive forms of development and marketing and in turn accelerates the side effects listed above.
Stolen is a strong word. Training a neural network on publicly-available data seems like fair use to me—kids will be doing this in their basement in 5 years, no Anthropic or OpenAI required. However, reproducing copyrighted content by virtue of replicating that content for commercial or otherwise, unauthorized use is infringement and subject to litigation. That's the only position that seems reasonable.
Someone arguing that ANY output from an LLM is plagiarized—which seems to be the vibe I get from some of the online discourse—because it's trained on copyrighted content seems like a overdetermined stance.
If you ask a model to write a random sentence and it writes back: "The quick brown fox jumped over the lazy dog." — is that plagiarism happening? I don't think so.
The real problem one might have with the notion that a trained neural network on copyrighted content makes all derivative output from that training set "stealing" it does need to be said that nearly everything everyone ever says or writes is something they heard elsewhere. Novel thoughts are a rare breed.
Not entirely sure if this is pertinent, but here goes:
As a professional (quite good) graphic designer, I occasionally dream the solution to a visual problem. It doesn’t happen that often and it doesn’t always work, but fuck, when it does, it’s incredible! Dreams are also as indisputable as they are unimaginable. No machine/android sheep will ever dream. How could they?
https://bsky.app/profile/orteil42.bsky.social/post/3mmz74x4cn225
https://anthonymoser.github.io/writing/ai/haterdom/2025/08/26/i-am-an-ai-hater.html
"I strongly feel that this is an insult to life itself."