AI – Artist or Paintbrush?
Taking first place in digital fine arts at the 2022 Colorado State Fair was a piece named “Théâtre D’opéra Spatial.” The artist, Jason Allen, garnered a lot of attention when, after winning, he revealed how the piece came to be: He used an artificial intelligence (or AI) program to generate the image, plus another AI program to increase the quality.
Much of Allen’s post-reveal attention questioned whether he was truly an “artist” – and if his win signaled an oncoming automation of the arts. AI programs have proliferated in the realm of visual media and become a consistent point of contention.
While Allen’s state fair win is not the first time AI programs have caused us to question what it means to create, it does provide a glimpse into how we will have to resolve that question in the future.
How AI and Copyright Fit Together
The U.S. Copyright Act provides protection for works of art to their creators. Protection also means monetization, and it is through this protection that artists can seek compensation for the use of their work.
Whether Allen is truly an “artist” is more a philosophical question – but our interest lies in determining if Allen’s work would qualify for Copyright protection. Because, for better or for worse, whether these works can be protected and monetized (or not) matters to would-be users of AI.
The Copyright Act protects “original works of authorship,” which is broken down into two requirements:
First, the work must be original: meaning it can neither be a copy nor something so basic that it lacks originality (like sheet music depicting a major scale). Pieces made using AI like Allen’s don’t seem to have any worries on meeting the originality requirement under the Copyright Act.
Second, the work must be created by an author, which is understood to mean a natural person. The “author” requirement prevents copyright protection from being used for things like the human genome sequence, because that sequence was not created by a natural person. But the “author” requirement can impact traditional works when atypical methods are used by the artist.
Alternative Interpretations of “Natural Person”
Imagine you take a photograph of a monkey while you are at the zoo. That photograph is an original work, and you, as the photographer, are the author of that work.
Now imagine you set the camera on a tripod with a timer. Are you the author of the resulting photograph even though you did not deliberately press the button on the camera? Most agree that you would still be the author of the resulting photograph.
Final experiment: What if you placed your camera in the enclosure, and the monkey ended up picking up your camera and taking a selfie? Who is the author: you or the monkey?
That real world example occurred when a wildlife photographer staged an area to entice a monkey to use his camera, resulting in a monkey selfie. The photographer claimed the copyright, but others used the photograph freely, asserting that it was in the public domain (because the monkey was the author). The question “Monkey or human owner of the copyright?” remains unanswered, because litigation ultimately settled.
As the distance between the artist and the work increases, through processes or elements outside of the artist’s control, the question of authorship becomes more relevant. At its core, this question seeks to identify the source of the intellectual labor in creating the work.
What is the source of the intellectual labor for works created using AI?
The Steps From Artist to Work
Turning now to “Théâtre D’opéra Spatial,” Allen’s reported technique consisted of three steps.
First, Allen used Midjourney, an AI program that generates images based on a user’s written prompt, to generate the base image. The images that Midjourney can create based on even minimal user prompt language are astounding.
Allen noted that he had been experimenting with a “special prompt” (that he will make public later), and he used that prompt to generate hundreds of images from which he ultimately chose his entry. This process is where the authorship question could be a hurdle for copyright protection.
Is Allen’s written prompt the primary source of intellectual labor in creating the work? Or is the AI making significant artistic choices in generating the resulting image based on his prompt? If Allen’s work began and ended at this stage, there would remain significant questions as to authorship – and that is something other users of AI programs should keep in mind. Luckily for him, Allen didn’t stop there.
In his second step, Allen used Photoshop to modify and alter the generated image. Digital artists and their works created through visual arts software have long been appropriate subjects for copyright protection. The user input typically required for software such as Photoshop is viewed akin to any other artistic tool. It does not replace the artist as the source of intellectual labor.
Once Allen finished the modifications, the third and final step was increasing the quality and clarity of the image, a process known as “upscaling,” using the Gigapixel AI program. The upscaling AI uses deep learning to predict and generate what a higher resolution or higher fidelity version of the image would be and fills in the gaps to generate the upscaled version of the image.
Does This Equal Authorship?
Allen’s three step process likely encompasses enough control and attributed intellectual labor to establish his authorship of “Théâtre D’opéra Spatial.” If anything, his win shows how quickly the tools available to artists have progressed in just the past few years.
It also exhibits how AI programs have created a new dynamic to the questions of authorship and creation. We may soon see programs that tip the scale away from the artist as the primary source of intellectual labor.
Then the question becomes whether we refuse copyright protection to those works as having no “author” – or do we change the definition of “author” to include non-natural persons? Either way, it promises to be interesting.
Images from https://www.midjourney.com/showcase/ follow along with the user input prompt that generated the image.