Cool New Tool Or Flat-Out Theft? Artists Split Over AI Revolution
Like photography, new forms of artistic expression arise in every age. Now, new artificial intelligence is making it possible to create incredible images in an instant — but it opens up an ethical and philosophical debate.
ROME — In 2017, when Google launched Deep Dream, an artificial intelligence image generator seemingly obsessed with psychedelic dogs, the internet's immediate reaction was curiosity and amusement.
But some people also wondered what would happen as the technology improved. Even from those naïve drawings of dogs, it was clear that it wouldn't take long for more capable neural networks to be developed, powerful enough to draw whatever a user asked.
That time is now. Last summer, as beta versions of AI programs Dall-E, Stable Diffusion and Midjourney (for now, the most effective of the networks), social media was filled with strange drawings. The style was still a bit grainy or dreamy, but immediately recognizable as something novel and significant. This was how the world became familiar with text-to-image software, which can create images based on written or dictated instructions.
At first, the reaction as these programs improved was playful. Illustrators got to know the new tool, as well as writers, including a comic strip written by an author and drawn by Midjourney.
But how do we know that Midjourney and other programs like it are just tools? Francesco D’Isa, artist and philosopher is editor of L’Indiscreto, the first magazine to publish images and stories made with Midjourney. D'Isa is convinced the technology can be useful for art.
Even before debates and controversies erupted around Midjourney and other AI image generators, D’Isa wrote that the programs were just tools, comparable to photography. The first cameras were met with amazement (if not dismay) at the rapidity with which the new medium could produce images. Even then, people feared the new technology would spell the end of painting and illustration. People complained about how easy cameras were to use — something where all "you only had to press a button” could never be art. We know how that turned out.
Still, there are a few obvious differences between photography and programs like Midjourney.
First, they generate images through text commands, which makes it possible to create simple and complex images. As a prompt for the program, “Imagine a girl” works just as well as “Imagine an oil painting of a band of masked girls, in the spring, with fractal Platonic solids, in the 1890s, reminiscent of Hockney, metaphysical, geometric, fractals, with a touch of Ingres, symbolism, academic painting, old masters, De Chirico, trigonometry.” Try it yourself: Midjourney and most other AI tools have user-friendly trial versions.
Power of a prompt
Updates are continuous, and the quality of results is astounding, in a wide variety of genres from classic fantasy illustration to futurist acrylic painting. In many cases, the result is indistinguishable from a human work. Moreover, it all happens in a matter of seconds. With an hour or so, a sketch artist with half an idea in their head can generate literally hundreds of high-definition images.
The one thing Midjourney continues to have difficulties with is hands. In the image archives, or datasets, used to train the program, whole hands are rarely visible because they are often grasping something. As a result, the software doesn't "get" how to draw them, often leading to bizarre results.
It's now certain that we are facing a revolution comparable in scope to the advent of photography or digital graphics, argues programmer and writer Gregorio Magini. How we should relate to this revolution, however, is up for debate. We still don't fully understand all of its implications.
“The oil painting of a masked girl posse, in the spring, with fractal platonic solids, in 1890, reminiscent of Hockney, metaphysics, geometric, fractal, with a touch of Ingres, symbolism, academia painting, old masters, De Chirico, trigonometry.”
Vanni Santoni via Midjourney
Theft of authorship
Critic Demetrio Paparoni told Galápagos magazine that there are no real risks to artists, because the artist will always remain the one who inputs commands to generate images — even if this allows a person who can't even hold a pencil to generate fifty Diego Vélazquez-esque paintings in minutes.
Illustrator and cartoonist LRNZ, born Lorenzo Ceccotti, disagrees. LRNZ sees the advent of these image generators as a threat to all graphic design professions. His doubts about the new technologies are not only artistic or technical, but also ethical: companies that make the software, and charge to use them, have trained their products on billions of images online, without asking permission of any of the image creators.
Midjourney now “knows” countless artists and can therefore draw in their style. This dataset no longer exists in a literal sense: after training, the AI works on its own, without searching for images in some database, as if it were “imagining” images that resemble a given artist. Can we therefore afford to legitimize this “theft” (as LRNZ calls it) of images? Many artists argue that we can’t. On Art Station, a large portal of mostly amateur illustrators, creators have been in an uproar about the use of artificial intelligence.
Risk of deepfakes
Then there are other problems inherent to the technology itself.
One is indistinguishability. A Midjourney image is already in many cases indistinguishable from those made by a human being. This can be the source of scams (an illustrator selling images made in seconds with Midjourney), and there have already been cases of art contests won with AI-generated works. This problem encapsulates others — like the possibility of easily creating deepfakes (videos of real people modified by AI) — and may only be solvable through a sort of paradox: soon, only AI will be able to detect work made by AI, just like the androids in Blade Runner.
There's also the question of rights. Every time Midjourney generates an image, it does so only because it has been able to use a dataset of billions of images. Each image from Midjourney then belongs in some way to each artist in the dataset, to whom copyright should be acknowledged, even if their individual contribution to the dataset was minimal. For now, there is no solution for the many authors who have asked to be excluded from the dataset. Once Midjourney has learned, it cannot unlearn.
One can at best hope for an exclusion in the next version.
The author question
It is also difficult to understand exactly how much an AI-generated work is created by the person who types the source text and presses the enter key. Is this person an artist? They do not draw or paint: they type a string of text. Are they a programmer?
It seems more like something between a DJ, who mixes other people’s tracks, and a curator or decorator who sets up an exhibition or furnishes a room, choosing from what they have available.
Certainly, “prompting” is a very easy art form. A smattering of artistic knowledge and sufficient good taste are enough to quickly generate thousands of aesthetically satisfying images. It is likely that, in the near future, the prompter — with their writing, evaluating and selecting activities — will slowly be considered an artist. This is what happened with photography as the numbers of photographers grew and the medium was refined — the result of a process that has gradually made it possible to distinguish amateurs from professionals and artists.
“Prompting” is as difficult as searching for something on Google Images: type in words and images emerge. But it is understood that in art, technique has lost its value, replaced by the idea — which is now everything, or almost everything. So can we deny that, in arranging images in a narrative order, selecting and reassembling them in a certain way or even just giving them a title, one is not in some way making art? Hardly.
People are already testing the bounds of automation, like programmer and writer Massimiliano Geraci who has devised a prompt to use ChatGPT — the chatbot launched by OpenAi, the manufacturer of Dall-E — to take rough ideas and produce textual prompts optimized for Midjourney.
“The oil painting of a showdown between noble ladies and masked men, spring breakers, platonic solids, in 1890, metaphysic, geometric, fractal, in the style of Ingres and Bouguereau, symbolist, academia painting, Old Masters, weird, highlights.”
Vanni Santoni via Midjourney
Beyond all this, the fact remains that Midjourney, Dall-E, Stable Diffusion and the other AI, while opening up new technological and artistic scenarios in our world, are being enriched by huge datasets composed of images taken without permission (although it could be argued that even an artist visiting galleries takes without asking just by looking at art). This has become intolerable for an increasing number of artists, to the point that some have decided to try to take on the AI-makers through political and legal means.
In Italy, the artist LRNZ leads this movement, along with artists Sio, Francesco Artibani, Paola Barbato, Elena Casagrande, Manuele Fior, Ariel Vittori and others. They plan to launch a major crowdfunding effort to “support the legal costs necessary to have the way these companies collect their data regulated at the European community level.” They have already raised more than €30,000 of their €70,000 goal. With the help of a specialized law firm, they aim to create binding European regulations for companies that produce image-making software.
Artists v. Midjourney
Others are already trying to do this.
On Jan. 13, artists Sarah Andersen, Kelly McKernan, and Karla Ortiz sued Stability AI (the maker of Stable Diffusion) and Midjourney, claiming that the companies infringed on the rights of millions of artists to train their software. Getty Images has also joined the lawsuit.
Will their claims succeed? As Terminator, not Blade Runner, teaches us, the victory of humans over machines is never assured.
But would it really be a victory? According to Francesco D’Isa, there is a risk of unintended consequences effects: “If the claims of Getty Images and the illustrators are successful, they will not eliminate these technologies from the market as some hope, but will create monopolies,” he explains. “The big tech companies that can afford to buy immense stocks of images will be able to develop artificial intelligence; the others will not.”
It doesn't seem to be a coincidence that well-equipped parties are on both sides — and that among the three major text-to-image software companies, the only one not to have been sued is OpenAi, which is backed by Amazon Web Service and Peter Thiel, co-founder of PayPal. Not a small amount of economic and political power.
The images in this article were created by Vanni Santoni with Midjourney. The captions are the original prompts entered into the artificial intelligence program to create the images.
- Why Artificial Intelligence Is Simply Impossible ›
- What Happens When A Ukrainian Asks ChatGPT About Crimea ›
- Artificial Intelligence And The Limits Of 'The Imitation Game' ›
- Ukraine Denies Pipe Sabotage, Georgia Protests, Holi Kickoff - Worldcrunch ›
- My AI Image Experiment In Dream Analysis - Worldcrunch ›