Who owns copyright in AI-generated works?

If you’ve been keeping up with new trends in the art world, or have read literally ANY news lately, you may have heard of the new AI explosion – including “AI-generated art.”

AI-generated art has become increasingly popular and profitable for companies, and has been quite controversial with artists and other content creators, to the point of various lawsuits being filed on behalf of these creators.

But what does that mean, though? Who owns art generated by artificial intelligence?

In this blog post, we’ll talk about this issue and the latest lawsuits that have been filed regarding AI-generated art and similar legal issues.

Big thanks to Billy Bradley for his help with researching and writing this article!

What is copyright?

Copyright is one of the most valuable forms of protection the law affords to artists, game developers, writers, and other creators.

By protecting the artistic expression of ideas, copyright gives the author an exclusive right to:

• Reproduce;
• Create derivative works based on;
• Distribute copies of;
• Publicly perform; or
• Publicly display their creative works.

This applies to drawings, writing, characters, settings, music compositions, and other creative choices you made (but not the underlying “idea” of them). In other words, while the idea of medieval kingdoms are not protected under copyright law, the specific creative expression that makes up the Kingdom of Hyrule is.

Read more about this in my post on the idea/expression dichotomy here.

In that same vein, if something is customary to or by necessity included in a work because of its genre, it is barred from copyright protection under the scènes à faire doctrine. A saloon would not be a copyrightable element in a Wild West story since it is common to the genre, but its exact design and the characters within may be protectable.

Who can create a copyrighted work? (United States)

Now that we know what a copyright is, let’s talk about how things change when AI is involved.

A copyright is granted to the “ author” of a work. But who is the author of a work, for copyright purposes, when AI does the heavy lifting?

In the United States, nobody is able to claim authorship in works solely generated by AI. The US Copyright Office only will grant registration to art created with “the fruits of intellectual labor” that “are founded in the creative powers of the mind.” In other words, only human ingenuity can create a copyrightable work at this time.

This is largely due to the way AI “creates” art (at least how it currently works). A human types in a few key words or phrases into a text box. From there, the program analyzes prior works to learn a specific style and creates an image in line with what it has learned.

AI cannot be “inspired” in the same way humans can, needing to rely on direct copying or the creation of derivative works. This is banned under the Copyright Act and is therefore unprotectable.

Since there is no creative intervention by a human author in this process, the US Copyright Office has consistently refused to grant AI-generated works copyright registration. Due to that, a work wholly created by artificial intelligence cannot be protected from infringement.

Similarly, the Copyright Office has not granted copyright to selfies taken by animals or works purportedly created by the divine or supernatural. This relies on centuries of precedent dating back to 1879. The Copyright Office has, however, hinted that AI-generated works with significant human creativity and input may be copyrightable.

It is unclear the degree to which AI involvement may render works created using AI unprotectable by copyright law.

If you make a piece of art with AI assistance, I encourage you to speak to a lawyer to determine whether there is sufficient human ingenuity to allow for copyright registration.

Who can create a copyrighted work? (United Kingdom)

Some nations actually grant copyright protection to “computer-generated” works. Even in these jurisdictions, the application of copyright law to AI-created works is unclear. The United Kingdom, which explicitly protects “ computer-generated” works with “no human author,” provides useful insights into the applicability of copyright law to AI works.

First, in the UK, the computer program itself is not granted copyright in its created works. Instead, the “person by whom the arrangements necessary ” for the works’ creation is granted authorship status. It is not entirely clear whether this is the user or the developer of the program.

Second, works will not obtain copyright protection in the UK without “originality.” Originality requires a work to be the result of the “ author’s own intellectual creation,” exhibiting a “personal touch.” The UK Copyright office grants that applying originality as it currently stands is unclear.

Third, the length of the copyright’s term is likely to be shorter. For instance, in the UK, copyright for computer-generated works vests for only 50 years as opposed to the typical 70. The UK has suggested a new right that would specifically protect AI-generated works for 25 years. After objections from other nations, little movement has occurred.

Owners of AI programs that create art should speak with a lawyer about writing into their terms of use that all copyrightable material created by the AI stays with the website’s owner – not the user (assuming they do want to retain ownership of the works).

Are there Fair Use issues with AI-created works?

Since AI uses prior art to create similar works, there may be fear of copyright infringement. The more similar a piece of art is to what preceded it, the more likely a court will find infringement.

While there are four fair use factors a court will analyze, the outcome often hinges on whether a work was sufficiently “ transformative.” The more transformative a work is, the more it furthers the purpose of free expression that copyright was meant to protect, creating a safe harbor from liability.

A creator will have to look at whether something new is added to the underlying work that was used to create the new art. Parodies and modifying works to fit into wholly different uses are examples of transformative use, but it’s never as simple as it seems, in my experience.

Before relying on fair use, I’d recommend speaking to a lawyer. You can also check out my eBooks, one of which is all about Fair Use (over on the sidebar to the right!).

Relevant Cases

As AI increases in popularity and ability, lawsuits on the scope of copyright as applied to their works will likely increase. These cases build on a long history of copyright litigation in the game and media industry — for a concrete example, see our breakdown of the Capcom Resident Evil copyright lawsuit. There are several AI-related cases pending in courts throughout the United States. We’ll be watching three in particular.

Anderson, et al., v. Stability AI LTD., et al.

Several individual plaintiffs sued Stability, Midjourney, and DeviantArt for copyright infringement, violations of their rights of publicity, and violations of unfair competition laws.

Stability AI is an artificial intelligence company aiming to create market-leading open-source models in image, language, and audio creation/production. The company is best known for its image generator, Stable Diffusion, which uses machine learning to create works of art. In a sense, prior art is used as an inspiration for the image generation system.

In its most recent model, artists were unable to opt in or out of their data collection software. There is no compensation for artists whose images were used in image generation.

Stability AI, Midjourney, and DeviantArt all use Stability AI’s Stable Diffusion software to create images from user text input. DeviatArt additionally hosts millions of images, some of which allegedly were created from the Stability Diffusion software.

Plaintiffs are artists who claim that the images created by the Stable Diffusion software are necessarily derivative works, particularly when they mimic the “style” of the underlying artists. Defendants reportedly believe their images constitute fair use because the new images are sufficiently “transformed” from the original(s).

Getty Images, Inc. v. Stability AI, Inc.

Similar to the above case, Getty Images is suing Stability AI for copyright infringement. Additionally, Getty is suing for trademark infringement, both directly and through dilution. There are other bases for suit in Getty’s complaint, including unfair competition and deceptive practices.

As stated above, Stability AI is an artificial intelligence company aiming to create market-leading open-source models in image, language, and audio creation/production. Getty Images is a company selling high-quality digital stock images to consumers, corporations, media entities, and other organizations.

Getty Images alleges that Stability AI copied “ more than 12 million photographs” without compensation. Getty further states that it has previously licensed its photos to other organizations for “purposes related to artificial intelligence and machine learning.”

Additionally, some of the images from Stability have Getty’s watermark, albeit in distorted form. Other images are alleged to have removed the mark entirely.

Left: Original; Right, Stability

Experts say Getty’s case is strong, and that we don’t really know how a court will rule in the matter. Fair use is the likely defense from Stability. In the meantime, the case could take years if taken to trial.

Thaler v. Perlmutter

Thaler attempted to register an AI-generated work from a system he created with the Copyright Office. The work had no other human input. His work was denied copyright protection. As a result, Thaler sued the Copyright Office and its Register: Shira Perlmutter.

The Copyright Office has maintained a human authorship requirement in order to grant copyright protection. A work without human ingenuity or creativity, therefore, is currently uncopyrightable.

Thaler claims that this requirement is unsupported by law and does not support the Constitutional purpose of “promot the progress of science and the useful arts.” Thaler further posits that the originality and authorship requirements under Feist are met.

The Copyright Office has pointed to case law preventing non-human entities, like supernatural beings or monkeys, from having a copyrightable interest in a work.

Conclusion

In conclusion, a work solely generated by an AI program cannot be copyrighted in the United States. If there is sufficient human intervention, then a copyright may be granted, but this remains murky. You should speak to an attorney as the law shifts to see what the current situation is.

If you create an artificial intelligence program, you may be able to protect the underlying algorithm via copyright or patent registration. However, the piece of art created itself is likely unable to be copyrighted under current US copyright law.

In other jurisdictions, such as the United Kingdom, AI works may be copyrightable. It is important to seek counsel with an experienced attorney in your jurisdiction.

Looking to integrate AI into your game development projects and have a legal question? Head over to my contact page to set up a consultation or sign up below for my game development eBook collection and more!