Another AI Copyright Registration Rejected

A few weeks ago, we took a look at the case of artist Stephen Thaler, who filed a lawsuit against the US Copyright Office (USCO) after the office declined to register an AI-generated work that he had submitted to them.

However, as I noted then, the case wasn’t particularly important as it didn’t actually weigh in on issues related to AI and copyright. The reason is that Thaler did not list himself as the author of the work, instead, he attempted to claim that his AI system was the author.

That argument was never going to work. Regardless of any point he was trying to make about the creativity and/or sentience of AI, the USCO only registers works of human authorship, as we saw in the “monkey selfie” case. Even if the USCO conceded that the AI was sentient, the outcome would have been the same.

However, earlier this week, the USCO issued its decision in a second request for reconsideration of registering a different AI-generated work, Théâtre D’opéra Spatial, which was generated by the Midjourney image AI system by artist Jason Allen.

This case is very different from Thaler. Where Thaler listed the AI as the author, Allen claimed the authorship for himself. However, the USCO has ruled that Allen failed to meet the de minimis standard and needed to both disclose and disclaim the AI-generated portions. 

Allen failed to do that and, as such, the USCO has denied the registration.

It is unclear if Allen is planning to follow Thaler’s lead and file a lawsuit against the USCO. However, whether or not he does, this could have major implications for AI when it comes to determining whether its output can or cannot be protected by copyright.

Background of the Case

Allen, along with Théâtre D’opéra Spatial, rose to fame and controversy in September 2022 when he submitted the work to the Colorado State Fair’s annual art competition and won.

It was one of the first major headlines about the growth of generative AI, and took place more than a month before the launch of ChatGPT in November 2022. 

However, that publicity ended up working against Allen in this case. When Allen filed his application for copyright registration that same month, he did not list Midjourney as the author. In fact, he didn’t mention AI at all. 

According to the USCO’s official decision, the use of AI was not disclosed at all and, instead, they were only aware of it due to having “garnered national attention”.

As such, the person handling the registration on the USCO’s side sent a letter to Allen’s lawyer asking for clarification on the process that was used to create the image. Allen, for his part, highlighted a multipart process that included:

1. At least 624 text prompts and revisions of text prompts
2. The use of Photoshop and other image editing software to “remove flaws” and “create new visual content”
3. The use of Gigapxel AI to “upscale” the image.

However, those elements were not convincing to the USCO and the office denied the registration in December 2022.

The next month, Allen filed his first request for reconsideration, but the office quickly rejected that, saying they would only accept the registration if Allen excluded content generated by Midjourney and Gigapixel.

He declined to do so and, instead, filed a second request for reconsideration, which is what the USCO issued this week. 

To that end, Allen didn’t fare any better, with the USCO once again refusing to register the work. However, this time, the USCO cited the Thaler decision four times, focusing on how copyright protection necessitates a human creator and how the portions generated by Midjourney and Gigapixel did not qualify for that.

As such, the USCO denied the registration for a third time, setting up a possible legal battle over the issue in the coming months.

If that legal battle does happen, it will be an important one to watch because the questions that Allen raises are much more relevant to the copyright issues around AI than any of the ones asked in Thaler.

Why This Case is Important

This case is important because it follows the more expected arc in an AI-generated work.

Allen didn’t attempt to claim that the AI was the author. Instead, he claimed that he was the author and the USCO only learned otherwise due to news coverage. They analyzed the process for creating the work and concluded that, unless the AI portions can be and are disclaimed, that the registration cannot be processed.

This is despite the fact that Allen did quite a deal of work in creating the piece. In addition to over 600 different prompts and prompt revisions, additional edits in Photoshop and other image editing tools and using an additional AI to upscale the image.

However, the USCO makes it very clear that all this work does not make Allen the author of the portion generated by either AI. Though he may have provided hundreds of prompts to the AI, prompts that may be protectable as literary works separately, the USCO said those prompts were not “direct instructions”. 

As such, the USCO ruled that the copyright-protectable elements in the AI-generated portions were the product of the AI system. This means that those elements were not created by a human author and, thus, there is no copyright protection.

This is in keeping with the USCO’s policy of requiring filers to disclose and disclaim any content that “contains more than a de minimis amount of content generated by artificial intelligence.” De minimis, in the legal field, means that it must be an amount so small or unimportant that the law will not consider it. The USCO claims that the AI portion of the painting is more than de minimis, and thus, it doesn’t qualify for copyright protection.

This is because, in the United States, there is no “sweat of the brow” doctrine when it comes to creating a copyright-protected work. Just because a work took a great deal of effort to make doesn’t mean that work qualifies for copyright protection. If the work has no requisite level of creativity, such as a phone book, or the creativity was provided by a non-human, as in the monkey selfie case, it enjoys no copyright protection.

To that end, the USCO ruling that prompts that are provided to an AI system are not direct instructions may be the most damming part of this decision. It makes it so that, no matter how creative the input is, the output is not protectable. Even if the input can be protected as a literary work, the output of the AI, whatever it may be, cannot.

If that position holds, it would be a major blow for anyone who is hoping to use and protect AI-generated content, regardless of what form it may take.

Bottom Line

Obviously, this is an early position. It’s likely that we’re going to get court decisions and additional USCO opinions on this topic. Nothing is set in stone at this phase.

However, the USCO’s position is extremely influential on such matters. Courts and legislators routinely look to the USCO for guidance on issues such as this. 

This is why it is important to participate in the current public comment round the USCO is holding on the topic of AI. The USCO’s opinions carry an immense amount of influence on US copyright policy, and US policy often leads global policy.

So while this is just a single decision by the USCO on a very specific case, it’s also a clear indication of where the USCO stands on the issue of AI. While that position can change, with the pace AI is moving, we could be in a very different position when and if it does.

In short, the USCO is going to shape the early path of AI, and this ruling is a clear indication of how it wants to do it.