Aethlinga, hear my rhyme,
a tale of new technology most sublime.
For AI art has come to be
a wondrous new way to see.
With algorithms and datasets galore,
artistic visions can now explore.
Colors and shapes of every kind,
all generated with a special kind.
No longer do we need brush and paint,
for AI art can show what we can’t taint.
No longer do we need a muse
for AI art can create and amuse.
Aethlinga, hear my plea,
for AI art is here to stay, you see.
A wondrous new way to behold
A new form of art to behold.
You’ve just read a poem written by one of OpenAI’s text generators, called the GPT-3 text-davinci-003 model. This is their latest model; it was released on 28 November 2022. The input I gave it was: “write a poem about AI art in old English.” However, as you can see, the poem is mostly written in modern-day English. This generator can write practically anything you ask for — poems, songs, answers to your questions, blog posts, articles etc. — but it doesn’t always give the desired output.
Artificial intelligence art (AI art) is any artwork created using artificial intelligence (the simulation of human intelligence processes by machines). When we say “AI art”, we usually think about images generated by artificial intelligence, but AI can also write poetry, compose music, write lyrics, and it’s widely used in the making of movies — from scriptwriting to video editing.
Now, would you say I wrote the poem above? I gave the machine a prompt and it gave me back a complete piece of writing based on that. You’d probably say that the AI wrote the poem. But what is “the AI”? Even if we give it a name — Dall-E 2, Stable Diffusion, Midjourney etc. —, can the AI be treated as a person, and the artwork be attributed to it? So, is “it” the owner of the poem? Does it own copyright to it? And, if the poem were an image, and it were sold, who would be the seller? Am I the seller for having given the prompt to the AI? Do the creators/owners of the AI get to sell it? We certainly cannot humanise “the AI” to a level where it can own artworks.
The first significant AI art generator was developed by Harold Cohen in the 1960s; its name was AARON, and it created simple black and white drawings with the help of an algorithm. The current most popular AI art generators (which were released in the past couple of years) — like Dall-E 2, Midjourney, Stable Diffusion, Nightcafe, Deep Dream Generator (by Google) etc. — all create imagery in a similar way. You don’t need to learn programming anymore to use these software, because they are based on a very user-friendly text-to-image method. According to OpenAI’s website, DALL-E 2 is trained on billions of captioned images from the internet, but some of these images are removed and reweighed to change what the model learns (for example, violent and/or sexual images are removed from the dataset). Apart from relying on large datasets, they all use a technique called stable diffusion, which in short means that they all manipulate these datasets somehow, ideally to remove biases. For instance, if there are more Labradors than Chihuahuas in the “dog” subgroup of the dataset on which the AI is trained, then a part of the Labrador images is removed. On a larger and more important scale, AI could turn out to be a racist, a feminist, a Buddhist, or any other “type” based on the images it’s trained on.
In 2018, the New York-based auction house, Christie’s became the first one to offer a work of art created by an algorithm. It was sold for $432,500, while its estimate value was between $7000-$10000. It’s a portrait of Edmond Belamy by Obvious, a collective of AI researchers and artists. The artwork is signed with the algebraic formula of its creating algorithm (bottom right of the painting). This painting generated a plethora of controversy.
In June 2022, Cosmopolitan came out with the world’s first issue having an AI-generated magazine cover. This raises questions about whether AI is taking away graphic designers’ job. Many argue that artwork such as illustrations for articles, books or album covers may soon face competition from AI. Others say that it should be viewed as a positive side effect, because this process democratises art. In October 2022 Shutterstock, a popular electronic image library, partnered with OpenAI — choosing to befriend the enemy, rather than to fight it.
A couple of months ago Jason Allen’s artwork won first prize at the Colorado State Fair’s annual art competition in the digital art category. The artwork submitted by him was a picture created with Midjourney. Its title is “Théâtre D’Opéra Spatial” and he submitted it under the name “Jason M. Allen via Midjourney”. His success started a large wave of controversy and debate around the topic, about whether he is an artist or not, and if his art could be considered real art. Many were attacking him, claiming that he was a contributor to the death of artistry. Others defended him, saying that photoshop was no different and that he still needed to be imaginative with the text prompt. Mr. Allen encouraged artists to overcome their objections and went on to say: “Art is dead, dude.”
At this point, we must ask: is AI creative? What even is creativity? This trait has always been referred to as something inherently human. We are struck by inspiration (the nature of which we often have no idea about) and we create things, but we’re never sure about where this ray of inspiration comes from. Moreover, we haven’t even reached a consensus on the definition of human creativity as it is. Thus, it’s hard to define it for computers. One thing is certain: AI can’t express human emotions (yet). It can’t mimic human stories behind art, it lacks depth. When you sit down to paint a picture, or just create something from scratch, whatever the outcome is, working on it by yourself gives you a sense of satisfaction. AI making art kills the procedure.
As for creativity, we can define it in a way that makes it possible for machines to achieve and master this skill. Did humans ever create anything new? Or did they just take the already existing things and transform them? Margaret Boden in her book “Natural Man”, published in 1987, says that the new thoughts that originate in the mind are not completely new, because they have their seeds in representations that already are in the mind. To put it differently, the germ of our culture, all our knowledge and our experience, is behind each creative idea. The greater the knowledge and the experience, the greater the possibility of finding a unique relation that leads to a creative idea. According to this definition, we can base our creativity on a huge but finite jigsaw puzzle, where we have trillions and trillions of puzzle pieces, and we can choose how many of them to use to create an image. We can connect the pieces however we want to, but it depends on their context whether the outcome makes sense or not.
Apart from the social ones, several legal problems concerning copyright arise when dealing with AI art. First, the training material of these AI art generators is concerning. The huge datasets used include artworks protected by copyright, created by artists who are still alive. In the prompt users give, they can include the name of an artist, which makes the generator create the image in that artist’s style. After a while, the work of these artists won’t be distinguishable from the images created by random people in their style.
Another issue is the copyright of the image created. By reading through these AI art generators’ frequently asked questions, one can get an idea about just how messy the legal background of artificial intelligence is. Some of them say that the created image belongs in public domain, others attribute it to the person who gave the prompt, but there are also a couple that want to abolish the idea of copyright as it is, because they think of it as an outdated system that constrains the coming AI era.
In 2011, the wildlife photographer, David Slater, dropped his camera and a macaque took a selfie with it, which Slater then published and claimed copyright ownership for. PETA (People for the Ethical Treatment of Animals) sued him, and the story went on until 2018, when the “Monkey Selfie” case was finally closed with the statement that an animal cannot own copyright. If not even our “close relatives” can own copyright, then how could a machine possibly do it? Under the law of most European countries, machines aren’t recognized as legal entities. It varies between countries, but if the artwork wasn’t created by the collaboration of a human and a machine, but just a machine, then the ownership usually goes to the creator of the machine (or the algorithm). However, there has been a case in India where they recognized copyright shared between the artist and the AI generator. Ankit Sahni gave his AI-based app a picture he took, plus Van Gogh’s Starry Night, and gave the name “Suryast” to the outcome.1
AI art will continue to develop in the following years. It’s the new art-creating technology of the 21st century. The camera was once regarded as the debasement of human artistry, and the same goes for digital design. AI may be different in terms of increased autonomy of machines, which are trained before usage. But there’s no point in resisting change, rather, learning to benefit from it can be fundamental.