Last year, I wrote about the short anime series Time of Eve, a science fiction story dealing with the relationship between humans and a set of androids with newly found sentience. In the course of that deep dive into the anime, I went on a tangent into AI “art” generation tools, asking whether artificial intelligence can truly create art as opposed to simply whipping up something derivative based entirely on user prompts and inputs and drawing from a pool of human-created work.
I’m happy to say, reading that post back that after just a year, that it feels badly outdated, because we now have tools available to create far more impressive images, word collections, and sound files. The image generator Midjourney is so impressive that it generated a piece that won first prize in a digital art contest at the Colorado State Fair — a choice by the board of judges that pissed a lot of people off and got a ton of publicity for the winner, one Jason Allen, a tabletop RPG creator.
And I’m right there with the angry torch and pitchfork mob this time. The idea that an AI-generated image can win an award in the category of “digitally manipulated photography” might seem pretty logical — after all, that sounds like digital manipulation, doesn’t it? But digital tools made for artists like Photoshop and GIMP still require complete human control and input, whereas Midjourney requires that the user enter text prompts. If you’ve used DALL-E, you know how this works: type in whatever it is you’re looking for and you’ll get some depiction of said thing, assuming the AI can work out quite what it is you want. These tools don’t create art, as far as I can tell: they generate images that may or may not resemble art depending on how broadly you define the term.
In that Time of Eve analysis last year, I suggested that this material isn’t art, in part because there’s no intent to express anything behind the generation of each specific work. Some people may define art differently, but I can’t consider something art without at least this intent to express. I don’t like abstract expressionism; the drip paintings of Jackson Pollock and the color field paintings of Mark Rothko leave me completely cold, but I would never accuse them of not being art for just that reason. The same is even true of two guys I completely dumped on a while back, Damien Hirst and Jeff Koons — I find their work totally soulless and empty, but I can’t say they’re not trying to express something through it, even if that something is just “I like tricking rich people into thinking this piece has value” (in which case I actually respect that, but that’s a different matter.) And no matter how I might feel about their work, all four undoubtedly used techniques and put thought into those techniques in the creation of their work.
I actually like this “Théâtre D’opéra Spatial” image in itself more than a lot of what the above four guys have created. If you had just showed this to me and told me a human had painted it, I’d believe you, and it seems the judges felt the same way. It’s aesthetically pleasing.
It’s also not art. It’s a mix of elements from pieces, images many of which were created with human thought and intent behind them, but put together in a way without that intent to express, by a machine fulfilling the requirements of a series of prompts.
Shockingly enough, US law isn’t lagging as far behind on this high-tech matter as it usually would be. One of the most relevant cases to this issue has to do not with AI but with animals. You may have heard of the “monkey selfie” case Naruto v. Slater, billed as an Indonesian macaque named Naruto (?) suing a wildlife photographer, David Slater, over the ownership of photos a group of macaques took when Slater placed his camera on the forest floor and let them approach it.
A few of these photos turned out well, and when they were published on Wikipedia without Slater’s permission, he came down on the Wikimedia Foundation arguing that he held copyright to the “monkey selfies”, with Wikimedia arguing in its defense that no one held copyright because the creator of the photos was a non-human and hence that they were in the public domain. The US Copyright Office found in favor of Wikimedia, putting the case to rest and letting Slater at least compile these and other photos into a book of his otherwise copyrighted work.
And now in comes People for the Ethical Treatment of Animals, inserting themselves into a controversy that nobody asked their opinion about but that they insisted on contributing to. PETA disputed both Slater’s position that he held copyright and Wikimedia’s that nobody did, instead asserting that Naruto himself held copyright since he snapped the photos. PETA sued as next friend of Naruto, a method of bringing legal action on behalf of a minor or of someone not mentally competent to act in their own interest. (They also very generously volunteered to act as administrator of the proceeds resulting from Naruto’s copyright.)
The Ninth Circuit on appeal found that PETA hadn’t properly established next friend status with Naruto, but far more importantly for our purposes here, it also found that PETA had failed to show that Naruto had standing under the Copyright Act, standing being someone’s right to sue in the first place.1 Part of the reasoning behind the decision not to extend copyright protection in such a case is that non-human animals don’t have the ability to express an “original intellectual conception.”2
The caselaw surrounding AI-generated images and other media is still nearly nonexistent, but early this year, the US Copyright Office confirmed that it will not recognize copyright in such media because it lacks sufficient human authorship. This decision arose from an ongoing fight in which one Dr. Stephen Thaler, co-creator of another AI image generation system called the Creativity Machine, claimed copyright over an image titled “A Recent Entrance to Paradise”. Dr. Thaler has filed suit disputing the Copyright Office’s rejection of his application to the DC district court, where the case is now pending.3
As Thaler’s attorney has argued (linked in the Smithsonian article above):
A.I. is able to make functionally creative output in the absence of a traditional human author and protecting A.I.-generated works with copyright is vital to promoting the production of socially valuable content. Providing this protection is required under current legal frameworks.
Setting aside Mr. Abbott’s use of the extremely vague term “socially valuable” to describe AI-generated content (it might have some social value anyway, sure, but that may even work against his argument considering the social value of the public domain concept) I believe his argument is total nonsense.
So let’s take it apart. AI makes “functionally creative output in the absence of a traditional human author.” But there is a human author involved. Many human authors, in fact, since the AI wouldn’t be able to generate a damn thing without its pool of existing human-created work to draw upon. That argument also ignores the human input necessary to generating said output, though I’d argue as the Copyright Office does that creating prompts, even many of them for the purposes of fine-tuning, doesn’t count as “human authorship” and doesn’t invest the prompt-writer with copyright ownership.
I suppose that’s why Abbott inserted the “traditional” to qualify his use of “human author” here, but it hardly matters, because again, this is an unforgivably vague term. What’s a non-traditional human author? The prompt-writer? The thousands or more of artists whose work was used (with permission or without, I’m not sure — that’s another issue entirely) to train these AI generators?
And while we’re picking apart his words, how about the use of the term “functionally creative”? Either something is creative or it isn’t. Using “functionally”, as far as I’m concerned, is basically an admission that both Abbott and his client know they can’t exactly argue AI-generated works are “created” in the same way as human-made works are, with actual thought and consideration instead of mechanical processes. In fact, I’d argue Naruto the macaque had a better claim to copyright over the photo he took than Dr. Thaler has to “A Recent Entrance to Paradise” through the Creativity Machine for the simple reason that a non-human animal can at least act independently and make its own choices without human prompting, whereas an AI can’t.4
Not yet, at least. But I’ll leave that for the second part of this post when I move away from the legal aspect of the AI “art” generation question and into the philosophical and moral ones. That’s something I’m not actually qualified to talk about, but I will anyway, because this really is far more than just a legal issue and I take an interest in both of its sides.
In the meantime, if you have any interest at all in these questions, be sure to follow the progress of Thaler v. Perlmutter, linked above at Court Listener. It may turn out to be a landmark case, though I hope not for the wrong reasons. Really, the suit should probably be thrown out on summary judgment.
Before I end this post and start working on the next, I’ll make a depressing prediction (as you expect from me!) I believe that as large, influential corporations see the value of AI-generated content, they and their special interest groups will push for changes to the Copyright Act to get rid of any ambiguity around the human authorship requirement commonly read into it in favor of AI-generated work that they don’t have to pay real flesh-and-blood visual artists for (and eventually also writers and musicians.) At the very least, I think they’ll be attracted by Stephen Thaler’s “my AI system is working for me under the work for hire doctrine” argument. I don’t see how this can be avoided unless these corps can somehow be shamed into not taking this route.
Well, regular people like us have just about no power over that. It still matters, but as with so many other things in this world, it feels to me like watching a trainwreck in slow motion — you can’t stop it from happening but just have to look on in horror as other people cry and laugh at it all.
But since it’s all useless, an effort may as well be made, because what do we have to lose? More on that whenever I can get to it, and maybe even without an endnote section nearly as long as the post proper. Until then.
1 I should note that the standing of animals under US law in general is a more complex issue. However, the Copyright Act specifically excludes non-human standing, and as a practical matter it’s hard to imagine what a macaque would do with its copyright ownership if the court had found in its favor. Naruto certainly didn’t know or give a damn about any of this nonsense, and the court rightly expressed doubts about PETA’s ultimate motives in their involvement.
2 Arcane legal hair-splitting time now, because it can’t be avoided. The US Copyright Office states in section 310.5 of its guide to copyrightable authorship that it “will not consider the author’s inspiration for the work, creative intent, or intended meaning” in determining the originality of authorship. However, it also states in section 306, The Human Authorship Requirement, that “copyright law only protects ‘the fruits of intellectual labor’ that ‘are founded in the creative powers of the mind.’ (Trade–Mark Cases, 100 U.S. 82, 94 (1879).)” (Citation included in case you really want to dive deep into the exciting history of US copyright law.)
I read this to mean that while the Copyright Office doesn’t care about the artist’s specific intent and won’t bother making courts try to be mind-readers in that sense, it also demands that there be some kind of intent to create, which is proved simply by the creation of the work. Whether any of the smartest/most potentially self-aware non-human animals are capable of “original intellectual conception” (higher apes? Dolphins? Maybe crows?) is an interesting scientific question, but the legal one has been answered, at least for now.
Finally, I should note to be fair that the Copyright Act doesn’t explicitly require human authorship — this is just the understanding of the Act’s plain language by the Copyright Office and courts up until today. There’s good reason to believe that won’t remain the case too much longer, as I note near the end of the non-endnote section of this post.
3 Thaler is trying to pull a pretty absurd trick here. While acknowledging that the Creativity Machine can’t legally hold copyright in its product, Thaler himself claims the copyright under one of two possible legal theories: first, as the owner of the AI and therefore the AI’s product according to the accession and first possession doctrines of property law, just as a farmer owns the calf birthed by a cow he owns, or as the owner of a 3D printer owns whatever it produces. The trouble for Thaler here is that neither these nor any of his other comparisons (see p 13 of his complaint on) have anything to do with copyright investment and ownership.
In the alternative (meaning if the court were to properly reject the above argument) Thaler argues that he holds copyright in the Creativity Machine-generated piece under the work for hire doctrine. This is a well-established doctrine that grants copyright ownership to employers who hire artists and writers to create works.
Thaler is using this doctrine in a novel way here, to put it politely. To put it impolitely, I think his interpretation of work for hire is a load of shit. But we’ll see what the court thinks assuming his suit gets that far.
And just one more dig at Dr. Thaler and/or his attorney because I can’t help it: read paragraph 31 of the complaint. This argument of “well we could have tricked the Copyright Office by not telling them it was AI-created” is so beside the point it’s almost hilarious. But it’s also a scary point to make considering that the best AI-generated visual “art” does closely resemble human-created work.
4 I believe this point highlights one of the contradictions from the pro-AI copyright camp in this argument: on one hand, some defending Jason Allen and “his” piece claim a program or system like Midjourney or Stable Diffusion is merely a tool like a paintbrush or Photoshop, but then Dr. Thaler in his complaint claims that these systems possess creativity (see the “functionally creative” comment from his lawyer above.)
I see these as more tools than not, though they’re clearly not simply like a paintbrush or a program manipulated directly by a human artist as some are claiming (and even disingenuously claiming, maybe.) But I think the “AI generated content is art” crowd will have to pick an argument and stick with it. You can even see this self-contradiction in Thaler’s complaint: his first argument implicitly takes this “my system is just a tool” approach.