The other day, I was playing with Stable Diffusion and found myself thinking hard about the ethics of AI-generated images. Indeed, I found myself in an ethical quandary. Or maybe quandaries.
More specifically, I was playing with putting famous haiku poems into the “Generate Image” box and seeing what kinds of images the Stable Diffusion generator would concoct.
It was pretty uninspiring stuff until I started adding the names of specific illustrators in front of the haiku. Things got more interesting artistically but, from my perspective, murkier ethically. And, it made me wonder if society has yet formulated way to approach the ethics of AI-generated images today.
The Old Pond Meets the New AIs
The first famous haiku I used was “The Old Pond” by Matsuo Bashō. Here’s how it goes in the translation I found:
An old silent pond
A frog jumps into the pond—
Splash! Silence again.
At first, I got a bunch of photo-like but highly weird and often grotesque images of frogs. You’ve got to play with Stable Diffusion a while to see what I mean, but here are a a few examples:
Okay, so far, so bad. A failed experiment. But that’s when I had the bright idea of adding certain illustrators’ names to the search so the generator would be able to focus on specific portions of the reticulum to find higher quality images. For reasons that will become apparent, I’m not going to mention their names. But here are some of the images I found interesting:
Better, right? I mean, each one appeals to different tastes, but they aren’t demented and inappropriate. There was considerable trial and error, and I was a bit proud of what I eventually kept as the better ones.
“Lighting One Candle” Meets the AI Prometheus
The next haiku I decided to use was “Lighting One Candle” by Yosa Buson. Here’s how that one goes:
The light of a candle
Is transferred to another candle—
This time I got some fairly shmaltzy images that you might find in the more pious sections of the local greeting card aisle. That’s not a dig at religion, by the way, but that aesthetic has never appealed to me. It seems too trite and predictable for something as grand as God. Anyway, the two images of candles below are examples of what I mean:
I like the two trees, though. I think it’s an inspired interpretation of the poem, one that I didn’t expect. It raised my opinion of what’s currently possible for these AIs. It’d make for a fine greeting card in the right section of the store.
But, still not finding much worth preserving, I went back to putting illustrators’ names in with the haiku. I thought the following images were worth keeping.
In each of these cases, I used an illustrator’s name. Some of these illustrators are deceased but some are still creating art. And this is where the ethical concerns arise.
Where Are the New Legal Lines in Generative AI?
I don’t think the legalities relating to generative AI have been completely worked out yet. Still, it looks like does appear that artists are going to have a tough time battling the against huge tech firms with deep pockets, even in nations like Japan with strong copyright laws. Here’s one quote from the article “AI-generated Art Sparks Furious Backlash from Japan’s Anime Community”:
[W]ith art generated by AI, legal issues only arise if the output is exactly the same, or very close to, the images on which the model is trained. “If the images generated are identical … then publishing [those images] may infringe on copyright,” Taichi Kakinuma, an AI-focused partner at the law firm Storia and a member of the economy ministry’s committee on contract guidelines for AI and data, told Rest of World….But successful legal cases against AI firms are unlikely, said Kazuyasu Shiraishi, a partner at the Tokyo-headquartered law firm TMI Associates, to Rest of World. In 2018, the National Diet, Japan’s legislative body, amended the national copyright law to allow machine-learning models to scrape copyrighted data from the internet without permission, which offers up a liability shield for services like NovelAI.
How About Generative AI’s Ethical Lines?
Even if the AI generators have relatively solid legal lines defining how they can work, the ethical lines are harder to draw. With the images I generated, I didn’t pay too much attention to whether the illustrators were living or dead. I was, after all, just “playing around.”
But once I had the images, I came to think that asking the generative AI to ape someone’s artistic style is pretty sleazy if that artist is still alive and earning their livelihood through their art. That’s why I don’t want to mention any names in this post. It might encourage others to add the names of those artists into image generators. (Of course, if you’re truly knowledgeable about illustrators, you’ll figure it out anyway, but in that case, you don’t need any help from a knucklehead like me.)
It’s one thing to ask an AI to use a Picasso-esque style for an image. Picasso died back in 1973. His family may get annoyed, but I very much doubt that any of his works will become less valuable due to some (still) crummy imitations.
But it’s a different story with living artists. If a publisher wants the style of a certain artist for a book cover, for example, then the publisher should damn well hire the artist, not ask a free AI to crank out a cheap and inferior imitation. Even if the copyright system ultimately can’t protect those artists legally, we can at least apply social pressure to the AI generator companies as customers.
I think AI generator firms should have policies that allow artists to opt out of having their works used to “train” the algorithms. That is, they can request to be put on the equivalent of a “don’t imitate” list. I don’t even know if that’s doable in the long run, but it might be one step in the direction of establishing proper ethics of AI-generated images.
The Soft Colonialism of Probability and Prediction?
In the article “AI Art Is Soft Propaganda for the Global North,” Marco Donnarumma takes aim at the ethics of generative AI on two primary fronts.
First is the exploitation of cultural capital. These models exploit enormous datasets of images scraped from the web without authors’ consent, and many of those images are original artworks by both dead and living artists….The second concern is the propagation of the idea that creativity can be isolated from embodiment, relations, and socio-cultural contexts so as to be statistically modeled. In fact, far from being “creative,” AI-generated images are probabilistic approximations of features of existing artworks….AI art is, in my view, soft propaganda for the ideology of prediction.
To an extent, his first concern about cultural capital is related to my previous discussion about artists’ legal and moral rights, a topic that will remain salient as these technologies evolve.
His second concern is more abstract and, I think, debatable. Probabilistic and predictive algorithms may have begun in the “Global North,” but probability is leveraged in software wherever it is developed these days. It’s like calling semiconductors part of the “West” even as a nation like Taiwan innovates the tech and dominates the space.
Some of his argument rests on the idea that generative AI is not “creative,” but that term depends entirely on how we define it. Wikipedia, for example, states, “Creativity is a phenomenon whereby something new and valuable is formed.”
Are the images created by these technologies new and valuable? Well, let’s start by asking whether they represent something new. By one definition, they absolutely do, which is why they are not infringing on copyright. On the other hand, for now they are unlikely to create truly new artistic expressions in the larger sense, as the Impressionists did in the 19th century.
As for “valuable,” well, take a look at the millions if not billions of dollars investors are throwing their way. (But, sure, there are other ways to define value as well.)
My Own Rules for Now
As I use and write about these technologies, I’ll continue to leverage the names deceased artists. But for now I’ll refrain from using images based on the styles of those stilling living. Maybe that’s too simplistic and binary. Or maybe it’s just stupid of me not to take advantage of current artistic styles and innovations. After all, artists borrow approaches from one another all the time. That’s how art advances.
I don’t know how it’s all going to work out, but it’s certainly going to require more thought from all of us. There will never be a single viewpoint, but in time let’s hope we form some semblance of consensus about what are principled and unprincipled usages of these technologies.
Featured image is from Stable Diffusion. I think I used a phrase like "medieval saint looking at a cellphone." Presto.