AI shills Talking Points vs Legal Reality
Folks are online arguing they're "AI artists" and "AI writers" - while using work stolen from creators. Let's look at their excuses.
These main headings are the 6 top specious arguments I see most often from apologists for the AI industry, and since this is a long article I’ve linked the sections below so you can jump directly:
I want to call these out. But I’m not just shaking my fist at the clouds, I’m arguing for strong regulation — which AI companies are fighting. See this post highlighting the need for AI guardrails for more on that.
AI Training is like what People Do
My opinion: “Training” as used for hill-climbing type incremental setting of neural networks weight values is a Term of Art created by computer scientists.
In other words “training” an AI is a metaphor not a reality, it’s meant to evoke an idea. It’s not literally “training” as we humans understand it. All the detail about hill climbing and error terms is complex so we use this shorthand word “training” although the AI process is nothing like what humans actually do.
AI model weights being adjusted over time during model building is a form of lossy encoding. It’s akin to how we store frames of video into an MPEG container file, and tolerate some small errors as long as human perception cannot see them.
It’s an important concept and I wrote about it in this Substack article.
AAP Lawyers say AI is Encoding
In a recent court case against Meta, known for social media platform Facebook, an Amicus Curiae brief was filed by The Association of American Publishers in Court in California 11 April 2025 which says in part:
II. Meta’s Mass Appropriation and Exploitation of Expressive Works to Train Llama Was Not a Fair Use
A. LLM Training Is Not a Transformative Use of Copyrighted Works
1. Training Is Not “Learning” About Works, But the Encoding of Expressive Content
Meta’s claim of fair use is largely predicated on two false narratives.
Seeking to establish that its exploitation is “transformative” under the first fair use factor of section 107 of the Copyright Act, Meta misleadingly portrays the LLM training process—in which works are systematically reproduced and ingested into Llama word by word—as merely recording “statistical information” about the works rather than capturing the content of the works. The LLM algorithmically maps and stores authors’ original expression so it can be used to generate output—indeed, that is the very point of the training exercise.
and
Contrary to Meta’s claims, there is nothing transformative about the systematic copying and encoding of textual works, word by word, into an LLM.
Word by Word
AI models like GPT 4.5 or LLAMA 4 are like a storage facility in that they contain data. That data bounds what it can produce. AI doesn’t have the ability to create ex nihilo1.
Right now AI image generators are interpolating between pieces of stored images, in the way analogous to the old face morphing technologies:
Humans cannot remember every detail of a photograph or painting; or complete tracts of text from thousands of books. AI can do that. That stored information is where AI gets its outputs from.
We know instinctively that it’s wrong, but there are people still defending AI by making this “training” claim. LLM’s are in fact at the storage end of the spectrum, not the intelligent end. They contain a lot of information but no smarts.
The nail in the coffin for the “it’s training” argument is that you can literally get AI models to regurgitate tracts of text out in response to crafted prompts. AI companies view this as an exploit and try to stop it, but it’s been repeatedly shown by researchers that they do memorise and regurgitate text of creative works.
This shows that the Amicus Curiae argument above regards copying is correct.
People Cannot Tell the Difference
This is a stupid argument for virtually anything you might apply it to.
Counterfeit $100 notes
Fake Rembrandt pictures
As creatives we sell our work based on our authenticity. Fake AI slop that pretends to be real, can fool some folks, and that is not on people who in good faith believe that stuff to be real.
Humans who are not creatives are not expected to know the real thing after a quick cursory inspection and they should not have to. This is a completely dishonest and corrupt idea.
Anyone saying “No-one can tell” doesn’t understand the concept of provenance. You can make a fake Rolex, and it doesn’t matter how good it is at telling the time, it’s not a real Rolex.
If I see an image created with paint and a paint brush I can read the artists statement, maybe at a gallery, and marvel at how the artist made my eye believe there was a field of flowers with a few spots of paint.
The point is not the pixels. It’s the provenance.
That is worth more than any AI fake image of a field of flowers. The point is not the pixels. It’s the provenance.
What is the Harm
The model is already trained right? The art is already stolen. What is the harm in someone having some fun making a doll picture of themselves?
Using AI to generate images:
boils the planet
enriches oligarchs
traps developing world workers in exploitation
creates a cycle of continual destruction of IP
This YouTube documentary — already reference above — features Zhang Jingna’s photography work and how she was hounded by AI incels for pushing back when her work was stolen to be used in countless “Asian girl in the style of Jingna” prompts, directly stealing from her Intellectual Property.
In books the same problem occurs when pirates take books and then unethical AI companies cynically use pirated books in creating new AI models. AI companies are only able to do this because people keep using the data that the companies supply.
If you don’t use fraudulently obtained data from unethical AI companies then their count of MAU - monthly active users goes down, and that is everything to a startup.
When I challenge people online about AI usage sometimes I get a nice result where they genuinely did not realise the problems with what they were saying and doing. Sometimes we agree to disagree. But there are also some who are flat-out really horrible people.
If you’re a defender of AI look around at who you’re on the same side with. The Mikkelson twins for example, and scammers who know they’re ripping people off and revel in that.
You cannot stop it
A non-argument. It’s basically “stop arguing, you can’t change anything” - so not addressing any factual argument at all.
We can also dismiss this one by simply examining the form of this “argument”.
Tidal waves, Gonorrhoea, Domestic violence… you can’t stop it.
No, but if it’s bad, and you can do something to lessen the impact of that bad thing then ethically it behoves you to do that.
My position (see the top of this article) is that AI must have regulation. This is something the AI companies are fighting by trying to cast themselves as innovative geniuses pushing technology forward and anyone in their path as luddites who don’t understand the technology.
In fact under this completely dishonest and fatuous argument I’d also place:
You don’t understand AI
You can’t stop progress
You just need to learn AI
You’ll be replaced by someone else using AI
…and all the rest.
Go here to see my credentials in particular. I am no expert on AI but I have found in general I have a better understanding of it than some of the folks I encounter.
AI is not a new technology. The transformer architecture, and the attention algorithm are innovative; but the real “innovation” was large scale theft of data, creating models with billions and billions of parameters.
This “you can’t stop it” thing is just a way to insult and attack a person, without actually addressing what they’re saying.
Edgelords are on social media dissing artists, saying they should just quit and get a job at McDonalds. In a video by Mohammed Agbadi about “AI artists fixing real artists work” he shows this guy in the screenshot above telling artists to quit and be a plumber. Mohammed’s whole video is instructive if you want to know what non-artists in some cases think of art. Basically to some AI is a big “f*$k you for having skills” weapon against artists.
Its just a tool
Often the people online saying this contradict themselves in the next post or even next sentence.
The long form (probably AI generated) “School Debate Club” style essay entitled “The Final Argument: AI-Generate Images Are Art” says it will oppose the claim AI can’t make art because it lacks consciousness.
The author says:
AI is not devoid of human input.
…and then — without a trace of irony — he goes from that to:
The AI is a tool, like a paintbrush, a camera, or a chisel.
So which is it? A) human/artistic/conscious/whatever or B) a tool like a chisel?
To be honest that whole word salad is free of any actual meaning.
Or are they trying to say it’s the human input that makes the difference? If that’s true then the proposition AI-generated images are art that the whole essay is trying to prove is out the door, because it’s human input not AI that causes the art.
Obviously Not Just a Tool
Honestly trying to discern any actual train of thought or cogent argument from these “ai is a tool” people is hopeless. And we should not bother because AI is not a just a tool: it’s got to be one of the least “tool like” things on the planet.
Generative AI LLM’s are Chat GPT 4.5 are giant models that have billions of parameters requiring whole data centres to run. Huge teams of engineers work on deploying, updating and managing these, and they are costing trillions of dollars in investment. And they rely on massive troves of stolen data. Millions of books for example.
AI data centres cost billions of dollars to run. Each time you run a query the real cost is subsidised by investor money, and requires a gallon of cooling water for a postage stamp of imagery.
You can connect to it over the internet from your phone or laptop, but ChatGPT, LLAMA or Claude - these things are monsters. It can do the homework of thousands of kids, while summarising a thousand articles all while spewing out dozens of derivative works from stolen copyrighted book or pictures; all in the same minute.
It’s like saying an aircraft carrier is a tool.
It's about as opposite from being "just a tool" as you can get.
AI is Not Like Photography
Shills say: When photography was first invented, it was seen as not an art form. Here’s an example of it in the wild:
The long form (probably AI generated) “School Debate Club” style essay entitled “The Final Argument: AI-Generated Images Are Art” that essentially argues the same thing. The form of the argument is:
People say AI images are not art.
But AI is like photography.
Photography was rejected as art
AI will therefore be accepted as art
Here are the problems with it:
Begging the Question
Today everyone in their right mind accepts that photography is an art form. Mapplethorpe’s portraits are in the Tate gallery so it’s common knowledge that it’s possible to create art with a camera.
So by saying in 2. above that AI is like photography they are trying to smuggle in the conclusion — that AI images are art — into the premise. There’s no proof of 2. AI is like photography offered by the people who argue this.
They also don’t prove item 3 - photography was rejected by the art community - beyond some quoting Baudelaire or some others, but there’s never any demonstration that widespread grassroots art communities, galleries and society in general rejected photography.
Analogical Argument
Using analogies in discourse is illustrative, in general. You have to prove all the points in the argument still. At best it can suggest an outcome, providing some support, but it’s inconclusive. And it’s especially so when you never prove the likeness in the first place.
Art Definitions are a Red Herring
But all of this is subterfuge designed to get us drawn into an argument about what is art and what is not, all to divert attention from the arguments most glaring problem.
Say for a moment that the output from AI is art: that does not entail that any person who may’ve initiated it is an artist. If they could prove the output was art, they still haven’t proven that the person who initiated it is an artist. Put a book mark in this: I’m coming back to it.
Say the output from AI is art. That does not entail a person who initiated it is an artist.
The AI images are art guy says - in his big long list of definitions of art:
AI-generated images, presented as art, fulfil this criterion by their very exhibition.
No quarrel with that. I think artists could use AI if they want. They can use a banana so why not AI?
Right now there are various artists who are presenting AI images in galleries and submitting them to art competitions:
Berlin based photographer Boris Eldagsen created an image using AI
It won a photography competition, but Eldagsen did not accept the prize
He’s shown in a YouTube documentary describing himself as a Promptographer - he gets some interesting images from AI
Sydney photographer Jamie Sissons heads an Art Studio which created an image
Under a fake name the image won a local competition
Jason Allen entered AI pieces into a Colorado art fair in 2022 under the “digital arts/digitally-manipulated photography” section and won a $300 prize
The first two are both cases where someone self-described as a photographer got an image from AI and won a photography competition with it. These people were artists before they took an AI image to a gallery or art competition.
In these the AI involved was not the thing that made the images into art. Boris Eldagsen person - an artist - used AI to make something - then presented it as art, giving a story as to why he made it. His artist statement. So something else — the artist and the provenance — is determining the images “art-ness”, not the AI.
So something else is determining the images “art-ness”, not the AI.
Colorado State Fair’s fine art competition allows any “artistic practice that uses digital technology as part of the creative or presentation process” but this was at a time when no-one knew what Midjourney was. Allen used Photoshop to add a head to one figure whose head was missing. The judges may have assumed that Jason Allen had painted the whole work in Photoshop using a stylus, in which case the provenance would have reflected that.
when another user asks if he explained what the software does, Allen replies, “Should I have explained what Midjourney was? If so, why?” eliciting face-palm emoji reactions from others in the chat.
Here also the AI made something which was presented by someone as art, but the judges were not aware it was AI. This was a case of something looking like art, being presented as art, and judged as art, but later doubt being placed on that judgement once the provenance of the piece was known.
Here the judge kept their decision in place, but there was doubt thrown onto the provenance of the piece once its provenance was known.
Fallacy of the Excluded Middle
The excluded middle fallacy occurs in an argument when a conclusion is drawn that ignores other possible facts. Here the claim is if an AI produced an artwork, that means the person who initiated the AI production is an artist. However there are other possible facts that could explain an artwork coming out of an AI.
The word entail in formal logic means necessarily implies. It’s written with an arrow like this →
For example:
B initiated an AI artwork → B is an artist
of the form
A → C
When we know the antecedent (before the arrow) is true, then we know the consequent (after the arrow) must also be true. The whole thing together is true if the antecedent always necessitates the consequent.
But if there are in fact other possible cases that are ignored so that sometimes A occurs but C does not, then the whole statement of A → C is wrong. Those “other cases” are excluded middles.
AI Images are not Art in Some Cases
Let’s have a quick look at another logical fallacy - one easier to tackle. The false syllogism, a form of sneaky argument.
This is the mistake of thinking that A → C means C → A. For example Cats have four legs; does not mean anything that has four legs. My friend Bowser the four legged dog shows that is wrong.
To make this clearer, we know that sometimes people make AI images for other reasons than art. So its not true that all AI images are automatically art. Here’s an analysis I did of an AI slop image made for the purpose of selling someone’s services on Facebook. The AI slop image was for sales so not intended to be taken to a gallery or entered in an art competition.
So: proving some output from AI was art doesn’t get you anywhere because there are plenty of examples of outputs from AI that their authors freely admit are not art.
Check my other article under the heading “AI fakes are used to Rob People” for more examples of AI images that are not art.
Also if AI was an art form would not prove that the person pursuing it had talent.
You’d want to see them produce art with that art form first. Picking up your cousins Stradivarius violin every day for a month while they’re away on holiday, doesn’t make you a concert violinist even if you get some kind of noise out of it.
So this whole idea of “what is the definition of art” and “who gets to decide” — it’s all misdirection.
Stepping into an Artists Shoes
Let’s return to that book mark:
Say the output from AI is art. That does not entail a person who initiated it is an artist.
Smac McCreanor performs crushing objects at the National Gallery of Victoria or NGV. Don’t think it’s art? Go argue with the National Gallery.
If you went to see her piece “Hydraulic Press Girl” at the NGV but someone else was there in her outfit, you’d be upset that you’d missed the real thing. You didn’t see smac’s art. This is different from an authorised gallery curator installing an art piece, it’s someone arriving at the penultimate step not knowing smac’s art and stepping in to her shoes.
They’re kind of a fake smac. The performance art failed to meet its provenance.
Here is our excluded middle. And it doesn’t care what you define art as.
Let’s imagine we agree on some definition of what art is. That definition — call it d — is known and we have a person A who is about to produce a piece of output that complies in every meaningful way with d.
We don’t have to know what d is, just that A knows it intimately and has a system set up to produce an output2. A readies the system, then leaves the room for a minute, and a bystander B walks in and presses the button, and takes the output. B does not know d. They could not build the system, but they can see enough to turn the key.
A person does not become the artist by virtue of extracting the output
Here B did not become the artist by virtue of extracting the output3. They were just cos-playing as artists.
It’s a Searle’s Chinese Room type argument where B is in the Chinese Room and is not the artist.
Photography something, something… AI image makers are Artists?
No it doesn’t follow.
I have noticed that being an artist makes images into art, regardless of if they came out of AI.
Here’s an OK definition that I like for what “being an artist” means for me:
The conscious use of the imagination in the production of things intended to be contemplated or appreciated as beautiful, as in the arrangement of forms and colour.But as I mention above it doesn’t matter how you define “art” or “artist”.
My thought experiment above is just an illustration, so it’s not conclusive. What is conclusive is that people saying “AI images are art” are failing to meet their burden of proof.
Actually — sidebar — it’s not totally clear what is meant by AI images are art. Does it mean:
“AI images win art competitions so AI images are all art” or
“I’m an artist if I make any AI image”
Both of these are not true, see the section AI images not art in some cases.
I actually think the author means something more like:
“I made AI images and I say they are art so I am an artist now”
In that case the sections Art Definitions are a Red Herring and Stepping into an Artists Shoes show that even if the output from AI is art, that does not entail that the person who initiated it is an artist. Your definition of art doesn’t matter.
This is because there is an excluded middle, another explanation. The true provenance, the origin of the image, is “from the AI’s data”. Pieces of art were in the AI and they are smooshed together by the morphing algorithm.
Vibe Coding for Art
When I first heard the term “vibe coding” I understood what it meant as I had used AI systems that my previous employers provided to write code with. If you “just let the system have its head” it will produce tons of code. A bit like letting auto-complete on your phone just go and write a whole message.
The AI does not have a conscious idea of what code it is writing, its following its context — for example the function body the cursor is inside of, the other code around it — and it auto-completes in what seems a coherent way. Until you look and realise the code makes no sense, and you have to fix it. I found it frustrating and turned it off, as I’m faster without it.
I think in the case of the two photographic artists above they might work with the AI system to produce a consciously intended image.
I’m not sure about Allen. What he might have done is “vibe coded” it: given a prompt and tried many times to get one with a head on it — and eventually decided he like what he had gotten out of the AI well enough so he put the head on it using Photoshop. He was a game designer which is an arty sort of profession, so I’m sure he knows his way around Photoshop, and is good at briefing artists for his games.
I think what happens in some cases is that people use a prompt and vibe around a bit, then convince themselves that what came out was their imagination.
What is People’s Motivation When Creating AI Images?
Eldagsen says he and his mother saw old pictures including ones with his father, and wanted to capture that feeling. He says that sometimes “the AI comes up with good ideas” — those are in fact from the data set the AI model was built with.
But I think AI poisons any artists creative vision. I think the data set can lead folks to any place, but only ones its been fed from other peoples images. And as well as artistic images there are bad images.
I think they feel free to indulge their worse impulses
I think people using AI to create images are very poor judges of the merits of the image — witness the scrambled office desk, cigar and microphone-melted-penstand with typewriter above. I think they feel free to indulge their worse impulses at times.
Witness this:
We know that AI was fed pornography due to the work in Madhumita Murgia’s book Code Dependent where she talks about the developing world workers and others who are subjected to horrible images when labelling training data. We know that AI systems were training on images of children, and on child abuse imagery. AI companies claim this has stopped, but they’ve also gotten rid of safety protocols and staff responsible for stopping it.
We know people are using AI to create pornography including of underage targets for example in the GenNomis database:
The publicly exposed database was not password-protected or encrypted. It contained 93,485 images and.Json files with a total size of 47.8 GB. The name of the database and its internal files indicated they belonged to South Korean AI company GenNomis by AI-NOMIS. In a limited sample of the exposed records, I saw numerous pornographic images, including what appeared to be disturbing AI-generated portrayals of very young people.
Conclusion
My position with AI images and art is that AI doesn’t make you an artist. I think it’s possible for a few AI using folks to make art with it. For any given person claiming to be an artist, using AI, I am ready to be convinced given evidence. I would want to see provenance and an artist statement.
AI training is the same as people learning
FALSE
No-one cannot tell the difference
FALSE
What’s the harm
LOTS
You cannot stop it
YOU CAN REGULATE
Its just a tool
ITS DATA CENTRE’S FULL OF STOLEN IMAGES, NOT A TOOL
AI is like photography
PROVE IT
What I would like is for readers to support regulation of AI.
Thanks for reading.
Some AI can use Retrieval Augmented Generation or RAG to fetch information outside of the encoded model, feed it through the inference and get it out as part of the generation. This is how it deals with current events using news sources for example. I don’t think AI image generation is using this directly to create images, but it might use it to make sense of a prompt. I don’t know for sure.
A row of dominoes, ropes and pulleys, a player piano where they have punched holes in the paper tape themselves; whatever, it’s some turnkey system.
We agreed on d - the definition of art - it doesn’t matter what it is, so we know it would have been art had A done the final act required to realise the requirements of d.