A.I. and Authorship: Thaler Decision is Just the Beginning

Thaler Decision

United States District Court Judge Beryl A. Howell recently ruled that artwork that is entirely generated by artificial intelligence (A.I.) does not qualify for copyright protection, noting that “human authorship is a bedrock requirement of copyright [1].” The related lawsuit had been brought forward by Stephen Thaler against the U.S. Copyright Office after it had rejected his attempts to copyright an artwork titled “A Recent Entrance to Paradise,” which was created by a generative A.I. he calls the Creativity Machine. The image was generated in 2012 as part of a series where Thaler’s team trained the A.I. to generate fantasy imagery while they tailored the algorithm “to simulate a dying brain [2].” Thaler stated that his goal was not to prevent infringement of the image but to establish that machine-created works could receive copyright protection. Thaler argued that he should be awarded the copyright for this machine-created work as the owner of the machine since he considered the image generation a work-for-hire [1].

"A Recent Entrance to Paradise” – Stephen Thaler and/or Creativity Machine [1]

The Copyright Office initially declined to award the work copyright protection in 2019, and an internal board reviewed and upheld that decision in 2022 [1]. The Copyright Office states that copyright law only protects “the ‘fruits of intellectual labor’ that are ‘founded in the creative powers of the mind [3],’” and there is judicial precedent for disallowing copyright protection for works created by non-human authors. In the well-known 2015 People for the Ethical Treatment of Animals lawsuit, the U.S. 9th Circuit Court of Appeals ruled that a monkey cannot own the copyright to the ‘selfie’ it took on a photographer’s camera [4]. Nevertheless, Thaler chose to file a lawsuit against the Copyright Office and called their refusal to grant his copyright “arbitrary, capricious, … and not in accordance with the law [1].” Thaler already has plans to appeal the decision of his case, and the wide-ranging effects of this and potential future decisions across industries and mediums warrant discussion.

The effect of A.I. in our world has continued to grow at an staggering rate. While its benefits for analytical and logical uses are obvious, experimentation with applying A.I. to creative purposes has been a surprising success. As generative A.I. models quickly improve, the quality of their output in some areas is approaching levels indistinguishable to that of a human’s. By harnessing A.I. as a tool to assist in generating portions of creative works, authors in many fields are finding new efficiency in their work (see this article on A.I. Code Assistants written by my colleague Madeleine B. Weber). The ability of generative A.I.s to create artworks with little to no human input has also encouraged some individuals and companies to use A.I. tools as an alternative to hired artists – saving them significant money in the process. If the resulting works can’t be protected by copyright, however, it may discourage the use of generative models and allow traditional artists to retain more leverage in these discussions.

Copyright Office A.I. Guidance

The Thaler decision is consistent with an official copyright registration guidance document from earlier this year, which states that “if a work’s traditional elements of authorship were produced by a machine, the work lacks human authorship and the Office will not register it [5].” The Copyright Office aims to distinguish between works of human authorship that use technology as a tool versus works that are conceived by machines. Defining a clear separation between those ideas could prove arduous, and the Copyright Office acknowledges this difficulty by noting that differentiating authored and authorless works is “necessarily a case-by-case inquiry [5].”

Consider this example: just last year the Copyright Office granted protection to a comic book titled “Zarya of the Dawn.” The eighteen-page work by Kristina Kashtanova was awarded a standard copyright registration in September of 2022. However, this registration was reviewed after the Office later learned via social media platforms that A.I. tools were used to create the book [6]. While Kashtanova planned and wrote all text in the comic book herself, she used the image-generating application Midjourney to create all of the book’s many images before making a few tweaks with other photo editing software. The review resulted in changes being made to the granted copyright protection. The Office ruled that the work as a total compilation was still protectable, and the same protection remained for the text within the comic book. The individual images, however, were deemed “not original works of authorship protected by copyright [6].”

page from the comic book “Zarya of the Dawn” – Kristina Kashtanova https://twitter.com/icreatelife/status/1617216573365800961/photo/2

Detecting A.I. Creative Influence

According to the U.S. Copyright Office, applicants have “a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author’s contributions to the work [5].” Kashtanova is unlikely to be the only author who does not follow this guideline, whether intentionally or otherwise. Will the Copyright Office be able to differentiate original works from those generated by A.I.? Unfortunately, existing tools for detecting A.I.-generated content have much room for improvement. Look no further than OpenAI, the creators of ChatGPT. With the software’s sudden rise in popularity – especially for generating written essays – educational institutions are wary of the new potential for cheating on assignments.

To address this concern, OpenAI announced last January that they were launching detection software that could be used to distinguish human-created textual works from A.I.-generated ones. As the creators of the most popular A.I. model, OpenAI might seem well-equipped to detect material generated by their product and others. However, the product “could only correctly identify 26 percent of A.I.-written text [...] according to OpenAI [7].” Further, OpenAI’s detection software had a nine percent false-positive rate [7], meaning it incorrectly flagged human-created work as generated by an A.I. The product was quietly discontinued in July without an official announcement. With similar issues plaguing competing products, some universities’ leadership, such as the Teaching Center at the University of Pittsburgh, have announced that they “[do] not endorse or support the use of any AI detection tools [8]” due to inaccuracy.

The Copyright Office may have to rely on the same flawed detectors or find some other method of discerning which works submitted to them are human-authored. Even as more detection tools are being developed and refined, the models that they are attempting to recognize are becoming smarter and less predictable. Recognizing machine-generated content appears to be a losing battle at least for the moment, leaving the Copyright Office seemingly with few effective options.

What comes next? As mentioned above, Thaler intends to appeal this decision. It would not be surprising for such an influential case to make its way all the way to the Supreme Court, where the plaintiff has already brought a similar case. Last April, the Supreme Court ruled against Thaler in finding that A.I. cannot be granted patents [9]. Concurrently the U.S. Copyright Office is in the middle of an initiative to study the policy issues presented by A.I. and what ought to be considered in future regulation. To encourage collaboration on this process, they have already held multiple listening sessions and webinars to hear professional and public opinions on the topic [10]. Anyone who would like to make their voice heard regarding A.I. authorship can submit a comment responding to the Copyright Office’s notice of inquiry from August 30, available at this link. Comments are due by October 18, 2023.

[1] https://www.theverge.com/2023/8/19/23838458/ai-generated-art-no-copyright-district-court
[2] https://imagination-engines.com/cm_art.html
[3] https://www.copyright.gov/comp3/chap300/ch300-copyrightable-authorship.pdf
[4] https://www.cnn.com/2018/04/24/us/monkey-selfie-peta-appeal/index.html
[5] https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence#footnote-9-p16191
[6] https://www.copyright.gov/docs/zarya-of-the-dawn.pdf
[7] https://observer.com/2023/07/openai-shut-ai-classifier/
[8] https://www.utimes.pitt.edu/news/teaching-center-doesn-t
[9] https://www.reuters.com/legal/us-supreme-court-rejects-computer-scientists-lawsuit-over-ai-generated-2023-04-24/
[10] https://www.copyright.gov/newsnet/2023/1017.html

A.I. and Authorship: Thaler Decision is Just the Beginning

Levi D. Winkle

IP Litigation, Technology Strategy & Analysis

September 18, 2023