AI Generated Imagery

There’s a long, complicated debate online right now about Artificial Intelligence (AI) and generated imagery. Is it art or not? Can you copyright it? Is it theft? Etc… And it’s pretty simple, actually. AI images are not art, you can’t copyright them, and they’re, essentially, theft. Let’s dive in.


art [ ahrt ]


Art is, objectively, a genre of work built on personal and creative expression. As society and civilisation has progressed, the mediums have changed and grown, but the value of art has remained the same. It is, as a whole, invaluable to society because of its unique ability to recount, express, and share the experience of humanity through time. This is what separates art from objective history, experience.

To create art is a personal experience and process; it takes time, patience, skill, and talent to produce remarkable work, but anyone of any skill-level can, in fact, create art.

Artificial Intelligence (AI)

artificial intelligence [ ahr-tuh-fish-uhl in-tel-i-juhns ]

noun Computers.

    • the capacity of a computer, robot, or other programmed mechanical device to perform operations and tasks analogous to learning and decision making in humans, as speech recognition or question answering.
    • a computer, robot, or other programmed mechanical device having this humanlike capacity: teaching human values to artificial intelligences.
  1. the branch of computer science involved with the design of computers or other programmed mechanical devices having the capacity to imitate human intelligence and thought.
    Abbreviations: AI, A.I., artificial intelligence

We’re a long way from the sentient AI you see in the movies. There’s no imminent threat of Skynet or VIKI looming on the horizon, at least not in commercial research labs. At the moment, AI is limited to complex decision-tree algorithms… that’s the simplest way I can euphemistically explain it, honestly.

How does AI image generation actually work?

AI algorithms and software are “trained” on vast datasets of pre-existing work, whether that’s books and news articles for text generators or collections of artwork for image generators. The software records patterns that it recognises in the “training data” in association with metadata and keyword tags entered by the trainers/AI engineers, and then it reproduces those patterns based on a prompt the user enters. That’s it. That’s how it works. The level of detail with which the software reproduces elements from its training dataset depends entirely on the quality of the algorithms written into the software.

The problem…

The problem with AI image generators is that most, if not all, of them have been “trained” by scraping visual artwork from the internet without permission from the artists whose work the software is, essentially, copying. The issue is consent! So much of these datasets are based off of copyrighted work without a license such that users can specify an individual artist in their prompt to emulate that specific artist’s work. And it doesn’t stop there! There are plenty of examples of AI images that feature real-life artists’ signatures because the work in the dataset is so lopsided—by which I mean that they’ve been “trained” solely from gallery and portfolio sites, again, without permission—and the algorithms are that bad at plagiarism.

But even if the algorithms get better so that they don’t reproduce signatures, the cat is out of the proverbial bag… That they can and have belies the fact that AI image generators are stealing from artists to recreate work outside of copyright and undercut artists whose jobs it is to create art for clients. Take the money directly from our pockets and the bread from our hands, it’s the same thing. AI imagery hurts real, live people. That the process obfuscates the theft with computer code makes no difference, AI users are stealing from artists every day but we’re stuck having Twitter arguments about the premise of theft. It’s ridiculous!

AI Images Are Not Art

The US Patent and Trademark Office has said that AI generated work cannot be copyrighted because work “must be created by humans to gain official copyright protection.” This objective reality, that the work is not created by a human, is the whole game. Prompt scripting or not, an AI, at this point, cannot create, they can only emulate and copy existing work. Without the datasets the algorithms are trained on, they would not be able to produce anything at all.

AI images are stealing artwork by scraping the internet in the exact same fashion that facial-recognition software scrapes social media sites to train and improve its algorithms. It’s unethical in both cases. As the NBC News story discusses, it’s nearly impossible to successfully opt-out of these datasets and equally difficult to have your photos/images/artwork removed from the training data once it’s been scraped. 

And “opt-out” programmes are not a solution to this theft. Opt-out is the default. These datasets must be made opt-in, otherwise the systematic, programmatic theft continues. Opt-out programmes are almost always abusive and parasitic to users, hidden deep in settings menus, and presuppose consent without informing users in a sufficiently conspicuous and easy-to-understand way. And how would it be enforced on a private portfolio site away from places like DeviantArt, Flickr, and ArtStation? Is it a metadata or EXIF tag? In that case, the image would have to be scraped first and then processed… and that’s not likely to be ethical either.

Making the Argument

Should you find yourself in an online “debate” about AI generated imagery, here are some key things to remember:

  1. Don’t call it art, because it isn’t art.
    • Refer to it as “AI images” or “AI generated imagery,” not “AI art.”
  2. They’re “AI Users,” not “AI Artists.”
    • For the same reasons you shouldn’t refer to AI images as art, you shouldn’t refer to AI users as artists. They’re just not. They’re software users, nothing more.
  3. Training datasets for AI image generators don’t have permission or consent for the overwhelming majority of the work they’re trained on.
    • This is important because copyrighted work has to be licensed for any use outside of private display, and even then the artwork has to be purchased.
    • When making this argument, beware of the retorts about Michelangelo or Leonardo DaVinci, et al. Those masters are long dead and any claim to copyright or ownership of their work has long expired. It is not relevant to artwork scraped from the internet.
  4. Arguments that all art is derivative are disingenuous.
    • Point of fact, not all art is derivative and the idea that all artwork comes from copying or reusing pre-existing ideas, imagery, or elements belies a fundamental lack of understanding of art history. If, for example, all art is derivative, what inspired The Lion Man, and what inspired that, and what inspired that, ad infinitum… What specific, visual imagery is DaVinci referencing in the Sistine Chapel? When Caravaggio painted Saint Francis of Assisi in Ecstasy in 1595, what pre-existing work was he stealing from? When David Carson set an article in Ray Gun in Dingbats in 1994, what was he referencing? And who did Stafan Sagmeister steal his 1999 AIGA poster from? Which artists were the painters of the Lascaux caves in the Vézère Valley of France pulling from 10k+ years ago? I could do this all day…
  5. Arguments that AI images constitute an art movement are tragically flawed.
    • First, this argument presupposes that AI images are art. They are not.
    • Second, in order for something to be considered a “movement” in art, there have to be multiple artistic sources working in the same or similar media, in the same time, sometimes in the same place (hello, Impressionism), with similar techniques and principles or themes. AI images steal and copy real artwork from across time and geographic space without regard to original medium, the original artists’ principles or ideas contributing to the original work, and without any unifying characteristics.
  6. AI images are not analogous to photography.
    • Photography aesthetics and composition are complicated principles that merge traditional artistic practices with technical mathematics and mechanics, to argue that photography is to traditional art as AI images are to digital art is to misunderstand the process of artistic photography at every level.
    • Turn-of-the-century arguments against photography as art were misguided and didn’t deal with the same points of contention as the argument against AI images as art because photography doesn’t copy and isn’t generative based on existing work.
  7. Writing a “prompt” is not the same as, nor is it similar to, creating art.
    • A prompt may be helpful as a starting point for an artist, or even as a set of objectives for a finished, commissioned piece to achieve. But it’s not a work of art. And regardless of how many “hours of tweaking” the prompt it takes to get the right output, the AI user still isn’t doing the work of creating the image through the application of their skills and talents to the tools of artistic craft.
  8. Beware straw men.
    • Techbros and AI Users are passionate about their art theft and will construct elaborate straw men arguments with little understanding of art or art history in order to confuse or otherwise “trigger” artists and other people they’re arguing with. Don’t fall for this trap. Stick to the facts: AI images are not art, AI users are not artists.
  9. Don’t get angry.
    • It’s an infuriating topic and the cavalier, unsophisticated audacity of the pro-AI crowd in these conversations can easily get under your skin, but you have to remain calm and clinical about this if you engage in the “debate.” They are, essentially, trolls and will pick up on an emotional response and use it as a sign of victory. Don’t feed them.