By Jackson Pittman
With contributions from Chat GPT (see text in blue).
Note: The U.S. Copyright Office’s official guidance for AI-generated works was discussed at the McCarthy Institute Copyright Symposium in Los Angeles on March 16, 2023.
In March 2023, the U.S. Copyright Office released official guidance concerning the copyrightability of AI-generated works. [1] The key takeaways from this guidance are as follows:
- Human authorship requirement: To qualify for copyright registration, a work must be created by a human author. AI-generated works, which are created autonomously by artificial intelligence, do not meet this requirement and therefore are not eligible for copyright registration.
- Creative contributions: If a human makes significant creative contributions to an AI-generated work, such as providing input or shaping the output, the human-authored portions of the work may be eligible for copyright registration. However, the AI-generated elements will still not be copyrightable.
- AI as a tool: The guidance emphasizes that AI systems can be considered tools in the creative process and works created with the help of AI may still be eligible for copyright registration, provided that the human author’s contribution is original and significant.
Overall, the Copyright Office’s guidance emphasizes the necessity of human authorship for copyright registration and encourages creators to consider the role of AI as a tool in the creative process.
The USCO’s statement was published amidst controversies surrounding attempts to copyright AI-generated works. [2]
These discussions stem from the rapid development and increasing capabilities of artificial intelligence in creative fields. The controversies often involve debates over authorship, ownership, and the scope of copyright law.
At least, that’s what GPT-4 said when I asked it to summarize the USCO’s recent policy statement. [3] And it’s mostly right, though somewhat generic. Perhaps that highlights both the capabilities of predictive AI language models and concerns about their use. See, GPT-4 doesn’t actually have information related to the recent guidance; its “knowledge” is limited to pre-September 2021 information. But after tweaking my prompts enough, GPT-4 put out some text I thought was good enough for an intro. Fine, I guess I’ll write the rest myself.
GPT-4 was right; the USCO’s primary focus in its statement was explaining when the “human authorship” requirement for registration is met despite the use of AI technology in creating a work. The USCO noted that this determination is “necessarily a case-by-case inquiry.” It recognized that in some circumstances, AI is a tool like a guitar pedal or photo-editing software that merely assists an author in the creative process. However, in other situations, the USCO believes that AI technology is truly behind the “expressive elements of its output,” and in such cases the work “is not the product of human authorship.” The USCO explained that a human may select, arrange, or modify “AI-generated material in a sufficiently creative way that ‘the resulting work as a whole constitutes an original work of authorship.’” It summarized its stance by stating that what matters in each case “is the extent to which the human had creative control over the work’s expression and ‘actually formed’ the traditional elements of authorship.”
These principles drove the USCO’s partial cancellation of Kristina Kashtanova’s registration for Zarya of the Dawn. [4] The comic book, which features text written by Kashtanova and images generated by Midjourney, was initially issued registration in September 2022. The USCO reviewed the registration after learning about AI’s role in the work’s creation. The Office concluded that Kashtanova could register the arrangement of text and images in the comic book, but not the images themselves. Viewing this decision through the lens of the USCO’s recent statement, Midjourney controlled the “expressive elements” of the images themselves, and therefore the “human authorship” requirement was not met. On the other hand, Kashtanova exercised creative discretion when deciding the arrangement and selection of text and images, thus making the final product “an original work of human authorship.”
Although the guidance is one of the USCO’s first official statements regarding artificial intelligence, the relevant concepts have existed for many years. In 1884, the Supreme Court noted that a photograph may be the product of an author’s “original intellectual conceptions,” or it may simply be a “mechanical reproduction.” [5] The use of a tool, whether a camera or AI software, is not dispositive of copyrightability then. What matters is that the work is a result of some creative spark of the author. Over 30 years ago, the Supreme Court held that a compiler of uncopyrightable facts may nevertheless arrange those facts in a sufficiently original way to create a copyrightable result. [6] Since then we’ve seen rapid innovation in technology, but the applicable legal principles appear to be the same.
We can speculate on the copyrightability of this article. Had I simply asked GPT-4 for 1000 words, ignored its glaring inaccuracies, and slapped my name on the byline, the result likely would not be copyrightable under the USCO’s guidance. The statement is clear that “when an AI technology receives solely a prompt from a human,” a resulting written work is not the product of human authorship. Instead, whether through sheer brilliance or laziness, I delegated some writing to GPT-4, arranged AI-generated portions with my own writing, and edited for accuracy. Could this article be registered? Probably so. The content of this article is mostly my original writing, and GPT-4’s contributions were prompted and arranged according to my creative vision. However, the USCO did not provide a bright-line rule about when a work is sufficiently human-authored such that it is copyrightable. There’s no magic ratio of human work to AI-generated work that qualifies for registration. It remains to be seen whether there is some minimum amount of human involvement to qualify for registration, or how the USCO will approach situations in which the distinction between human contributions and AI-generated ones is less clear.
The Office’s statement also included advice for potential copyright applicants seeking registration for works involving AI-generated content. It explained that applicants are obligated to disclose such contributions when applying. Individuals seeking registration can use the Standard Application, and identify which portions were human-created and which were AI-generated. The statement clarifies that applicants should not list the AI technology as an author; perhaps feeding a prompt to an AI is like handing a camera to a monkey. [7] It also explained that previously submitted and pending applications should be corrected if applicants insufficiently disclosed AI contributions. Registrations that are not updated may be cancelled or disregarded in an infringement action. [8]
There still exists an inclination for many to think of AI as some impending technology, sure to turn our world upside down, but not quite yet. I remember playing with DALL-E Mini last summer and laughing at the Uncanny Valley-esque results it produced. [9] I thought that in five years, AI would be able to make some truly realistic images. Just a few weeks ago, convincing images of Donald Trump being arrested circulated Twitter. [10] Earlier this year, a deepfake of Joe Rogan promoting supplements he’s never mentioned went viral on TikTok. [11] Make no mistake – the age of AI is now. Though some may disagree with the USCO’s seemingly open-ended approach to copyright of AI-generated works, it evinces an intent to actively respond to new innovations. In a field that has historically been slow to adapt to developing technology (see basically any congressional hearing with a tech company), the fact that the U.S. Copyright Office is taking such an approach seems like a good sign.
Before I let GPT-4 rest, I realized I was forgetting something important. “GPT-4, what’s your billing rate?”
“As an AI language model created by OpenAI, I do not charge any fees for providing information or answering questions. However, it is important to note that I am not a licensed attorney, and my responses should not be considered legal advice. For specific legal questions or concerns, it is always best to consult a qualified attorney who can provide accurate and up-to-date information tailored to your particular situation.”
There you have it, folks – our profession is safe (for now).
Endnotes
[1] Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 37 C.F.R. § 202 (2023).
[2] James Vincent, The Scary Truth About AI Copyright Is Nobody Knows What Will Happen Next, Vᴇʀɢᴇ (Nov. 15, 2022, 8:00 AM), https://www.theverge.com/23444685/generative-ai-copyright-infringement-legal-fair-use-training-data.
[3] GPT-4, OᴘᴇɴAI, https://openai.com/research/gpt-4.
[4] Tiffany Hu, AI-Generated Comic Art Not Protected, Copyright Office Says, Lᴀᴡ360 (Feb. 23, 2023, 9:05 PM), https://www.law360.com/articles/1579038/ai-generated-comic-art-not-protected-copyright-office-says.
[5] Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58-59 (1884).
[6] Feist Publications, Inc. v. Rural Telephone Service Co., Inc., 499 US 340, 357-59 (1991).
[7] SeeNaruto v. Slater, 888 F.3d 418 (9th Cir. 2018) (holding that a monkey lacked standing to bring copyright infringement action under the Copyright Act).
[8] See 37 CFR 201.7(c)(4); see also 17 U.S.C. 411(b)(1)(A).
[9] Maria Parazo Rose, The Dall-E Mini Image Generator’s Ridiculousness Might Be Its Main Appeal, Pᴏᴘᴜʟᴀʀ Sᴄɪ. (June 15, 2022, 7:02 PM), https://www.popsci.com/technology/dall-e-mini-image-generator-explained/.
[10] Isaac Stanley-Becker & Drew Harwell, How a Tiny Company with Few Rules Is Making Fake Images Go Mainstream, Wᴀsʜ. Pᴏsᴛ (Mar. 30, 2023, 11:57 AM), https://www.washingtonpost.com/technology/2023/03/30/midjourney-ai-image-generation-rules/.
[11] Liam Ho, Insane Deepfake Sends Joe Rogan Viral for Promoting a Product He’s Never Discussed, Dᴇxᴇʀᴛᴏ (Feb. 12, 2023, 10:58 PM),https://www.dexerto.com/entertainment/insane-deepfake-sends-joe-rogan-viral-for-promoting-a-product-hes-never-discussed-2059439/.