How AI is Killing Human Artists—And Why “Theft” is the Only Word for It.

The emergence of Generative AI is often framed as a miracle of modern engineering, yet beneath the sleek interfaces lies a ledger of unpaid debts. As the proliferation of AI Art begins to systematically displace Human Artists, it becomes clear that this billion-dollar industry was built on a foundation of non-consented, uncompensated labor. We are told this is a “democratization” of creativity, but for those whose life’s work was scraped to train the very machines now undercutting their livelihoods, the more accurate term is theft. This isn’t just a shift in tools; it is a calculated heist of human experience.

In 2022, Stability AI was valued at approximately one billion dollars. The company’s flagship product, Stable Diffusion, had been trained on a dataset called LAION-5B — five billion, eight hundred million images scraped from the internet without the knowledge, consent, or compensation of the human beings who made them. Among those images: the concept art, illustration work, photography, and painting of working artists who had spent years, sometimes decades, assembling a portfolio and building a livelihood. The billion-dollar valuation was constructed, brick by brick, from unpaid labour. The artists whose creativity formed the very foundation of that valuation received a zero-dollar cheque and, as a consolation prize, the unique honour of now competing in the open market against a machine trained on their own stolen work.

This is what the tech industry means by democratisation.

The Billion-Dollar Valuation of Unpaid Labor

The argument that AI art is killing human artists is not, at this point, a hypothesis. It is a documented fact that the people responsible for it are managing very carefully — through the language they use, the press coverage they cultivate, and the speed with which they deploy legal resources when anyone uses the word that most precisely describes what happened. The word is theft. They prefer innovation.

Let us be clear about what was actually built and how. The large generative AI image models — Stable Diffusion, Midjourney, DALL-E, Adobe Firefly, and their various descendants — were trained on images that were taken without asking. Midjourney’s founder David Holz, to his modest credit, at least admitted this plainly in a Bloomberg interview, confirming the company had used scraped data and had not sought consent from creators. He was not subsequently shunned by the industry. He was subsequently celebrated by it. In January 2023, Getty Images filed suit against Stability AI for the unauthorised use of more than twelve million photographs from its archive, the suit noting — with the photographic precision that legal documents occasionally achieve — that some AI-generated images even reproduced corrupted versions of the Getty watermark. They had scraped the images so completely they had scraped the proof of their own scraping.

One might, I suppose, argue that this is simply how the internet works — that data has always flowed freely, that the concept of digital property is philosophically murky, that the old frameworks don’t apply to new technologies. One might argue that. People with a financial interest in that argument do argue it, loudly, in conferences and opinion pages and congressional testimony, wearing the bright, sincere expression of men who have spent considerable money on the opinion they are now offering for free.

But the artists know what happened to them.

Decoding the “Luddite” Myth and the ArtStation Protest

The ArtStation protest of December 2022 — in which illustrators and concept artists flooded the platform with “No AI Art” images until the front page was a wall of refusal — was not, as some technology journalists described it, a Luddite panic. It is worth pausing on that word, Luddite, because it has been doing significant ideological work since the nineteenth century. The actual Luddites — the framework knitters and croppers of Nottinghamshire and Yorkshire who broke machines between 1811 and 1816 — were not, as the myth insists, ignorant peasants who feared technology. They were skilled tradespeople defending a livelihood and a craft that was being deliberately automated out of existence by mill owners who understood exactly what they were doing. They were right about what was happening. They lost anyway. And every time a tech evangelist calls a worried artist a Luddite, he is — consciously or not — repeating a two-hundred-year-old propaganda campaign on behalf of the people who won.

The artists on ArtStation were not afraid of the future. They were watching the present. They were watching concept art forums fill up with clients asking for “Midjourney-style images” — which is to say, images that looked like their work, generated by a system trained on their work, at a price point that made their work economically pointless to commission. They were watching the job boards thin out. They were watching rates drop. They were watching clients who had previously paid three thousand dollars for an illustration propose two hundred dollars — because, the client explained with the cheerful obliviousness that characterizes this particular moment in the history of creative labour, they could just use AI if the price wasn’t right.

This is how AI is killing human artists: not in some dramatic, cinematic way, but in the grinding, incremental way that economic devastation actually works. A commission gone here. A contract not renewed there. A client who used to hire eight illustrators for a game project now hiring two and using AI for the rest. The violence is quiet and it is distributed, which makes it very easy for the people who benefit from it to describe it as a natural market correction and move on.

The Economic Reality: From $3,000 Illustrations to “Prompt Engineering”

The defence that is always offered — that new technologies always displace old jobs but create new ones — deserves the brief, fluorescent examination it has earned. It is true that technological disruption has historically created new categories of work alongside the old ones it destroyed. It is also true that this fact has been used, with remarkable consistency over two centuries, to tell the people being displaced right now that they should stop complaining because something will turn up eventually. It is the economic equivalent of telling a drowning person that statistically, most people who fall into water survive. Accurate, broadly speaking. Useless to the person who is drowning.

The new jobs that AI is supposedly creating for artists are, in the main: prompt engineer (which requires no artistic training and which will itself be automated within five years, a fact the people offering it as consolation know perfectly well), AI image curator (a job that pays a fraction of what illustration paid and that will also be automated), and AI trainer — which is to say, the job of providing human feedback to improve the system that is replacing you, at wages that are, in documented cases across the outsourcing economy, below subsistence. The pipeline of creative work that sustained a generation of commercial illustrators, concept artists, book cover designers, and editorial illustrators is not being redirected. It is being closed.

Media Cowardice and the False Equivalence of the AI Debate

The media’s role in this deserves its own prosecution, because the coverage of generative AI as it applies to the arts has been — with honourable exceptions — a sustained act of epistemic cowardice dressed in the language of balance. The standard form of the mainstream technology article on this subject: quote an enthusiastic tech CEO, quote a nervous artist, conclude that “the debate continues” and that “both sides make valid points,” and file it before lunch. What this framing accomplishes, and what it is specifically designed to accomplish, is the creation of a false equivalence between the people who are being harmed and the people doing the harming — as though the question of whether it is acceptable to steal from working people in order to build a billion-dollar product is genuinely a matter on which reasonable observers can disagree.

They are not equivalent positions. The question of whether AI art is killing human artists has an answer. The answer is yes. The question of whether it matters has an answer. The answer is also yes. The only ambiguity in this situation is legal and technical — questions about what copyright law covers, what fair use permits, what training data is permissible — and those questions are being actively litigated in courts where the technology companies have incomparably more resources than the artists they scraped. This is what class war looks like when the winning side writes the terms of service.

Lessons from Napster: When “Democratization” Means Penury

There is a historical precedent that the industry’s cheerleaders prefer not to mention, because it rather undermines the inevitability narrative. When Napster and its successors dismantled the music industry’s revenue model in the early 2000s, the same argument was deployed: this is democratisation, this is the future, the old gatekeepers are finished and good riddance, the artists will find new ways to monetise. And some did. And the rest — the session musicians, the mid-level recording artists, the producers and arrangers and studio engineers who had built careers in a functioning industry — watched their livelihoods dissolve while a new class of platform owners captured the value that had previously, imperfectly, flowed to the people who made the music. Spotify‘s founders became billionaires. The median musician’s recorded music income fell. The democratisation was real; what was democratised was penury.

Abstract Forces vs. Human Decisions

The story of how AI is killing human artists is the same story. The technology changes. The structure of who benefits does not.

One concession, and I mean it as precisely as I give it: the tools themselves, in principle, are not the problem. A paintbrush is not a moral agent. A camera did not steal work from painters; it created a new form of art while painters found new territory. The problem is not that generative AI exists. The problem is the specific, documented, legally contested choices made by specific companies and specific executives about how to build it — choices that involved taking, without asking, the accumulated creative output of hundreds of thousands of working artists, in order to construct a product that directly competes with and undercuts those artists’ ability to earn a living. Those choices were made by people with names. Emad Mostaque built Stability AI on scraped data. Sam Altman‘s OpenAI trained DALL-E on images it did not license. David Holz’s Midjourney — whose annual revenue was reported to be in the hundreds of millions by 2023 — built its product on work it did not pay for and has not, to this writing, compensated.

These are not abstract forces. These are decisions. Decisions have authors. Authors have addresses.

The Legal Gray Zone: Extraction Without Human Authorship

In 2023, the US Copyright Office ruled that AI-generated images cannot receive copyright protection because they lack human authorship. The intended takeaway, in the mouths of tech advocates, is that this is merely a temporary legal ambiguity that will eventually be resolved. The actual takeaway — the one sitting quietly in the ruling, waiting to be noticed — is that the system producing these images required human authorship to train on, extracted that authorship from living artists without compensation, and now produces outputs that belong to no one, benefit the platform owners, and have the economic effect of devaluing the work of the humans who made the whole thing possible.

The artists made the food. They were not invited to the table. They are being asked to be grateful that at least someone is eating it.

Expropriation vs. Democratization: Choosing the Future of Art

The question of what to do about the fact that AI art is killing human artists is, in some respects, a legal and regulatory question — and those cases are moving through courts in the United States and elsewhere, slowly, against well-resourced defendants. It is also a question about what we think creative labour is worth and who we think it is for. If we have decided, as a society, that the accumulated creative output of human civilisation is a free resource to be industrialised by whoever gets there first with sufficient compute, we should at least have the honesty to say so explicitly. We should stop using the word democratisation when we mean expropriation. We should stop calling it the future when we mean the past, wearing new clothes.

The artists who flooded ArtStation in 2022 with those protest images understood something that the billion-dollar valuations prefer you not to examine too closely: that what was being destroyed was not just a market, but a practice — the particular, irreplaceable thing that happens when a human being with a specific history and a specific pair of eyes and a specific set of failures and obsessions sits down to make something that did not exist before. That thing cannot be scraped. It can only be valued or abandoned, supported or starved.

We are choosing. We are watching ourselves choose. And the people making money from the choosing are very much hoping we stay distracted long enough that the choice becomes irreversible.

It might already be.

What do you think? Is the “AI revolution” an inevitable evolution of tools, or are we witnessing the largest copyright heist in human history? Leave a comment below or share this with an artist who needs to hear this.

Further Reading:

• U.S. Copyright Office: Artificial Intelligence and Copyright Initiative — The official government resource for legal updates on AI authorship and ongoing copyright law reviews.
• UNESCO: Report on Artificial Intelligence and the Future of Culture — A global analysis of the systemic risks AI poses to cultural diversity and the livelihoods of creative professionals.
• World Economic Forum: How Generative AI is Reshaping the Creative Value Chain — An economic perspective on how AI platforms are restructuring the market and capturing value from human creators.
• The Getty Images v. Stability AI Landmark Ruling: A Legal Deep Dive — A technical breakdown of the most significant legal battle regarding unauthorized data scraping and intellectual property.