Artists using AI in their work are regularly accused of theft and if we’re going to have any kind of reasonable conversation on the subject that needs to be addressed. On a recent 60 Minutes episode artist Molly Crabapple said generative AI constitutes “the greatest art heist in history.” In the broadcasted interview she expanded on that saying “When we talk about art heists, typically, we’re talking about one painting being taken from a museum, two, three… They stole billions and billions of images.” Crabapple isn’t mincing words here, this is a sharp accusation and it deserves to be taken seriously.
In order to talk about if something is theft we need to understand what theft means. It’s a legal term that hinges on “taking the property of another person with the intent of depriving the rightful owner of the possession.” Deprivation is key to theft, something has to be taken away. Generally we’re talking about physical objects, but there’s certainly historical precedent for thinking about theft in relation to intangibles as well. Most famously is Crazy Horse who refused to allow his photograph to be taken, holding the belief that the resulting image would steal part of his soul and thus shorten his life. His documented response to such requests was “Would you imprison my shadow too?” How much this belief was held by others or can be found in other cultures is somewhat debatable, but there’s no question he firmly believed that a photograph would steal something from him, leaving him with less as a result. It would be theft. Thomas Jefferson took the other side of that argument, in an 1813 letter to Isaac McPherson he challenged the notion that ideas could be stolen. In that he coined the now famous analogy of how lighting a candle from another candle doesn’t deprive the first candle of the flame. “He who receives an idea from me, receives instruction himself without lessening mine; as he who lights his taper at mine, receives light without darkening me.” Lawrence Lessig expanded on that when he wrote “The physics of piracy of the intangible are different from the physics of piracy of the tangible” in his 2004 book Free Culture when noting that copying an MP3 file does not remove the file from the source.
This tension around theft of the intangible isn’t a new problem, at some point if the accusation is made there needs to be a rigorous definition applied to it in order to find a solution. The most famous and best trial of the concept was the Recording Industry Association of America (RIAA) file sharing cases in the early 2000’s. The RIAA argued that every download would have otherwise been a sale and so the actual loss was tangible. Even if you accept the “lost-sale logic” at face value, the moment you start doing the math the premise collapses entirely. At the time the maximum claim, per infringement, was $100k. Importantly, this figure had been previously determined assuming commercial violations, applying it to individuals was new legal territory. The recording industry didn’t bother to try and justify it anyway, they just adopted that number in their suit, which ended up being 1,071x the actual cost per track. Jammie Thomas, a single mother of two, was found liable for $222,000 in damages for sharing 24 songs on Kazaa. A shocking number, but peanuts compared to the most egregious example, the case against LimeWire. There the RIAA claimed damages of $72 trillion, which was more than three times the GDP of the entire planet at the time. This absurdity was noted by the judge in the case who scolded the plaintiffs for “suggesting an award that is more money than the entire music recording industry has made since Edison’s invention of the phonograph in 1877.” In hindsight one might hope that these cases ended up being laughed out of court, but they mostly proceeded leaning into infringement rather than theft, resulting in significantly reduced damages. Turns out without loss, there was no argument for theft.
(LimeWire file sharing interface, 2008)
If making a copy of something is infringement, but not theft, then how do we discuss that accusation if instead of copying we’re talking about imitation? Influence, inspiration, and even blatant biting happen all the time in art, in literature, in every creative field you can imagine. Kirby Ferguson argues in his Everything Is A Remix series that this is by design, that this is how creativity works. We see something and that inspires us to make something. This is core to why artists visit museums and galleries, why musicians listen to records, why authors read books; chasing inspiration and hoping to find an idea worth building on. Anyone can read Cormac McCarthy and recognize that he has read a lot of Hemingway. That’s not some dirty secret desperately covered up, McCarthy himself acknowledges this influence openly. You can find that reference in any number of interviews and critical reviews, none of which progress to accusations of theft. And McCarthy isn’t unique, ask any artist how they got started and they will tell you a story about being inspired by someone else’s work. We commonly group artists together who have overlapping inspirations and don’t consider them thieves. Consider the impressionists, cubists, modernists, anime, etc… We judge those artists based on their individual works, not on their influences.
That doesn’t mean all inspiration leads to powerful new creative works, some artists just straight up copy other artists’ styles. You don’t need to look any further than Etsy to see countless examples of no-name artists selling work that looks strikingly similar to their very well known contemporaries, and the market recognizes the difference. The value of an original Banksy is orders of magnitude larger than any one of the millions of Bansky knock offs. And perhaps more importantly, Banksy’s work isn’t devalued because other artists copy him. And even there, artists who make nothing but derivative work heavily biting the styles of others may be criticised for being lazy or unoriginal, but they wouldn’t be accused of theft. In its most basic and least creative usage (a point we’ll spend more time on in a future essay), that’s very similar to what AI does. It makes something new that looks very much like something it’s seen before. It’s not making an exact copy of it like we discussed with file sharing, it creates work inspired by other work it’s seen. Fair Use laws exist for this very reason, because it’s understood and accepted that creative works often build on inspiration and that’s the standard AI companies training these models continue to point to. When a human does that it’s imitation. The moment an AI generates the image, regardless of how involved a human prompting it was, the accusations of theft start flying. The principle and standards being applied to AI training are never applied to human learning. This asymmetry is important to sit with, and raises the question if theft is the thing we’re actually afraid of?
(Screenshot example of Banksy knock off on Etsy)
There is a very important distinction between how humans and AIs learn though, and that’s scale. Systematic, industrial volume ingestion of work is very different from organic human learning and I won’t pretend otherwise. That’s not an after the fact realization either, that’s literally the value proposition. An LLM based AI has a memory of more imagery than a human could experience in an entire lifetime, which means a human using that AI as a collaborative creative tool has a reference library billions of times larger than a human relying on only their own memory. This absolutely changes the scope of what is being discussed, but the question we’re left with is does that scale change the ethical category of the work or just the degree of it? It’s easy to understand that stealing a record and making a copy of it are different things, and making 10,000 copies of it doesn’t change that very real difference – but it’s not as easy to explain the ethical difference between using 10 paintings as inspiration vs using 10 million. There are as many differing opinions as there are paintings, and I don’t have a concrete answer. This is all brand new territory that all of us will have to figure out as we go.
While some things remain unanswered, that doesn’t mean we can generalize across the board. If everything gets called theft and everyone involved thieves, then the debate stays at that categorical level and specific companies, policies and actions avoid scrutiny. And they need to be scrutinized, it’s important to be able to point out the problematic choices and decisions so that we know who to hold accountable for any harm they cause. It’s equally important to avoid conflating corporate actions with the medium itself, or with artists engaging with the medium. Lumping those together allows legitimate concerns to be written off as simply being “anti-progress.” This oversimplification provides cover for bad actors, and delegitimizes serious concerns. Corporate bad behavior is separable from the technology, and criticism should be specific. It would be silly to blame or write off the entire genre of painting because some art supply manufacturers use stolen materials to make paint brushes. It would similarly be misdirected to harass painters for the actions of those companies. A better approach is to cleanly separate the companies from the technology from the people using that technology, so that criticisms of the companies land squarely and users of the technology can help build momentum around modifying those actions. In my silly analogy, painters are more likely to join you in challenging the supply chain of their brushes, and help change industry standards, if you aren’t blaming them for the actions of the companies.
The first step in solving a problem is identifying what the problem is. Going back to the 60 Minutes episode mentioned at the beginning, Crabapple lists the grievances “No artist has been asked for their consent. No artist has received compensation… we don’t even see credit.” The good news, if you like the idea of problems being solved, is that this complaint is demonstrably overstated. The best example of artists being asked for consent, being compensated and receiving credit all in one go is the Shutterstock Contributors Fund. Hundreds of thousands of artists opted-in to allow their work to be used as training data and were both compensated for it and granted ongoing royalties for use. And that’s not the only example, there are a number of other models already being used where artists can sign up to allow their work to be used in training models and receive compensation for it, and they are quite popular. Even in cases where there wasn’t initially consent, such as in the Universal Music Group vs Udio case, we’re seeing settlements which include opt-in structures that give copyright owners and creators the ability to sign off on specific usages. Another counter point from the same 60 Minutes episode is the artist Refik Anadol who since 2020 has trained his models exclusively on what he calls “ethically sourced data” that they license directly from the source in a transparent and permission-first way. However even if Crabapple’s assertion is overstated, there absolutely are artists who haven’t consented, been compensated or given any credit. The good news, again, is that solutions are already in place. This isn’t an unsolvable problem inherent with the technology, it’s a critique of bad corporate behaviour which reiterates the importance of, as mentioned in part two of this series, making sure we’re worried about the right things.
(Still image from Refik Anadol’s “Living Paintings Immersive Editions” RAS)
While “theft” remains the most common accusation, it’s also the one with the least supporting evidence. But “not theft” doesn’t mean there are no other concerns, or that those other concerns aren’t valid. What words we use to name the problem determines what solution we work towards. Name the wrong problem and we end up with the wrong solution, leaving the real problem in place and unaddressed. Our criticisms need to be more precise. I’ve stated from the beginning of this series that there are real concerns and real harm that needs to be discussed, from economic to structural, but we need to be accurate in our complaints, target the appropriate actors, and identify the specific policy problems.
[Header Image: Letter from Thomas Jefferson to Isaac McPherson, Page 6, August 13, 1813]
This essay is part three of a series, part one can be found here.