In August 2022, Jason M. Allen won first place at the Colorado State Fair Fine Arts Competition. The category for his submission was “digital arts/digitally manipulated photography”. Normally this award wouldn’t make national news headlines, but the prize-winning image was controversial in its creation. Allen used the AI image generator Midjourney as the base to create his piece, leading to a divisive debate: are AI images truly art, owned by the artist? Are these AI generators another artistic tool, or are they just the product of randomized computer code?
The strong backlash against Allen’s piece, titled “Théâtre D’opéra Spatial”, mainly centered on the argument that submitting art created by an AI generator did not make Allen an artist. While artificial intelligence research has been around for years, the use of AI to create images has recently exploded in popularity, most notably with DALL-E, a neural network created by the research company OpenAI. The novelty of entering a string of text to create a fun image to share with friends captured the attention of many on social media. With so many users submitting unique prompts every day, these AI generators have been able to level up the capabilities of their algorithms.
As the quality increased, so did the viewpoint that these generators could be taken seriously as an artistic medium. However, the question of who really creates / owns the intellectual property of the generated artwork has led many to question if it falls under the category of modern / digital art, or if it is simply “cheating”. One social media user on Twitter compared Allen’s submission into the competition to letting a robot compete in the Olympics.
Who Really Owns the Images?
Many believe that the process of AI Image generation disqualifies it as an artistic medium for original works, arguing that they require existing images to create the “new” images. For this argument, the complications lie in the ethics and legality of using AI as a medium.
The question of who the copyright owner of a computer-generated image is was recently brought up in a US court case, when the CEO and President of the company Imagination Engines, Dr. Stephen Thaler, filed a suit against the United States Copyright Office (Thaler v. Perlmutter) because they refused to register a copyright for artwork he used an AI generator to make. Thaler listed the AI generator as the “author” of the work. The application for copyright registration on the artwork was denied by the office because it did not have “human authorship”
Under the Copyright Act of 1976, “original works of authorship” are protected. Thaler argued that human authorship should not be necessary, citing how the Copyright Act has applied to corporations. He also argued that AI generated art satisfies the legal standards for originality, based on how it can be indistinguishable from a work made by a human. As of the writing of this article, the final decision in this case has not been made.
So, How do AI Image Generators Work?
Both Midjourney and Open AI’s DALL-E use text-based “transformer language models”. DALL-E specifically inputs data through tokens. Tokens are described as “any symbol from a discrete vocabulary”, much like how a letter is a “token” from the English alphabet. DALL-E is a straightforward decoder-only transformer that models each text and image token (received by the generator in a singular stream of data) autoregressively, meaning it predicts future values by evaluating previous values. DALL-E Mini is the popular, open source version of DALL-E.
Midjourney, which was the AI generator used by Jason Allen to create his contest entry, allows users to enter prompts via a Discord channel. While it only allows for a limited amount of entries for free, it also stands out among other generators by allowing you to choose variations of the images created by your input. It also allows the option to “upscale” the images, creating a larger and more detailed version from the original.
Are AI Images Generators Just the Next Phase of Art?
Following the strong reaction to his win at the Colorado State Fair, Allen defended the legitimacy of his work by explaining how he didn’t just press a button and win a competition. He says he spent hours tweaking the phrasing for the text input until he got the exact results he was looking for, and then used Photoshop and another AI program to clean up, enhance, and improve the resolution of the images before having them printed on canvas. Competition judges stand by their decision, describing it as a “beautiful piece”.
Does Allen’s win symbolize a new era of AI generated images as a legitimate digital art medium? Would it be different if AI images were their own category for submission? While the future of artistic legitimacy and owner copyright is still in question, it is clear that now that AI image generation is here, it is here to stay.