Since its launch in November 2022, ChatGPT has taken the artificial intelligence (AI) community by storm.[1] The New York Times described it as “the best artificial intelligence chatbot ever released to the general public.”[2] Indeed, its ability to provide a detailed and coherent response to any topic is unparalleled in the competition of AI-based chatbots developed by other companies.[3] It can write a rap song, produce working computer code, explain scientific concepts, and write essays for college students.[4]
ChatGPT is based on a generative pertained transformer (GPT) language model trained to generate human-like language based on input.[5] It is designed to engage in a dialogue where it can “answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.”[6] This is the latest technology in the field of generative AI, which has seen tremendous growth in 2022;[7] another such software, DALL-E 2, allows anyone to create an image with a textual prompt.[8] This ability of generative AI to develop novel content rather than copy or mimic existing content has raised many concerns in the legal world, including accidental copyright infringement, plagiarism, defamation, risks to privacy and data, and ethical problems of AI perpetuating existing human biases.[9]
In particular, there are important copyright questions to answer as these technologies using generative AI become more and more prevalent. For instance, can authors copyright any works created by an AI system? Recently, the U.S. Copyright Office partially revoked a registration granted to a comic book author after discovering that the images used in the copyrighted comic book were generated by a text-to-image AI called Midjourney.[10] This copyright registration issue highlights one crucial factor in obtaining copyright protection: the degree of human input.[11] If a generative AI model requires very little human input, such as a simple textual prompt, the output of the model would not qualify for copyright protection.[12] Copyright protection will likely be granted, however, when an artist produces a version of the image with generative AI and then manipulates it through multiple iterations to add stylistic choices.[13] However, it is uncertain how much human input crosses the threshold for obtaining copyright protection.[14]
Another crucial question is whether companies can use copyright-protected data to train generative AI models. It is unclear whether AI models—which are trained upon billions of copyrighted works across the internet—infringe the copyrights of the original works’ authors.[15] In November 2022, a class-action lawsuit was filed in federal court against Microsoft, GitHub, and OpenAI, alleging that GitHub’s Copilot software, which is trained on public repositories of code scraped from the internet to provide a real-time auto-completion feature for writing code, violates the Digital Millennium Copyright Act because it does not follow open-source licensing guidelines such as attribution to the author of original code.[16] In response, the defendants have filed two motions to dismiss for lack of standing and failure to state a claim, arguing that the plaintiffs merely alleged theoretical harm and failed to “plead violations of cognizable legal rights.”[17] If these motions fail, the defendants may rely on the fair use doctrine as a defense.
The fair use doctrine allows artists to use copyrighted works without permission in certain limited situations, and the proponents of generative AI often cite this doctrine to argue that the generative AI models should be allowed to be trained using copyrighted works for their artistic creation.[18] Courts have yet to determine whether the fair use doctrine can be used as a defense in cases involving generative AI.[19] Authors Guild v. Google may be somewhat instructive;[20] there, the Court found Google’s unlicensed scanning of millions of copyrighted books for its book search project covered by fair use.[21] The case can be distinguished, however, because generative AI models scrape billions of images and books across the internet for the data, not the expressive content.[22] The fair use defense is also controversial because commercial companies hire academics to test their AI technologies on public images, and this practice has already coined the term “AI data laundering.”[23]
How copyright law will evolve in response to generative AI is still uncertain. In the best-case scenario for proponents of generative AI, the law may recognize fair use and allow generative AI to be used for commercial purposes. In the worst-case scenario, the law may prohibit any use of copyrighted works when training AI models, significantly hindering the progress and commercialization of generative AI. The outcome of the pending GitHub lawsuit will have an impact, as well as more cases to come. In the meantime, here is what ChatGPT has to say about granting copyright protection to its works:
“The question of whether copyright protection should be granted to works created by generative AI is a complex and evolving issue that raises a number of legal, ethical, and practical considerations.
At present, the law does not explicitly address the copyright protection of works created by generative AI, and different jurisdictions may have different views on the matter. However, in general, copyright law grants protection to works that are original and fixed in a tangible medium of expression.
In the case of works created by generative AI, the question of originality may be particularly challenging. While the algorithms used by generative AI may be unique and innovative, the output they produce may not necessarily be considered “original” in the same way that a work created by a human author might be. Additionally, it can be difficult to determine who the “author” of a work created by generative AI actually is.
Another consideration is the potential impact of granting copyright protection to works created by generative AI on the development of AI technology and the wider creative industry. Some argue that granting copyright protection to such works could incentivize innovation and investment in AI technology, while others argue that it could stifle innovation and creativity by limiting access to the underlying algorithms and data sets.”[24]
[1] See Kevin Roose, The Brilliance and Weirdness of ChatGPT, N.Y. Times (Dec. 5, 2022), https://www.nytimes.com/2022/12/05/technology/chatgpt-ai-twitter.html.
[2] Id.
[3] See Christopher C. Archer & Charlotte Roxon, ChatGPT: The Arrival of a Disruptive AI Tool, Morgan Lewis (Dec. 28, 2022), https://www.morganlewis.com/blogs/sourcingatmorganlewis/2022/12/chatgpt-the-arrival-of-a-disruptive-ai-tool.
[4] Id.
[5] Id.
[6] Id.
[7] See Jacob Bourne, Generative AI Made Its Public Debut in 2022—It Could Be an Internet Earthquake in 2023, Insider Intelligence (Dec. 22, 2022), https://www.insiderintelligence.com/content/generative-ai-made-its-public-debut-2022-it-could-internet-earthquake-2023.
[8] See Isaiah Poritz, Generative AI, Andy Warhol ‘Fair Use’ Lead 2023 Copyright Issues, Bloomberg Law (Dec. 28, 2022, 5:05 AM), https://news.bloomberglaw.com/ip-law/generative-ai-andy-warhol-fair-use-lead-2023-copyright-issues.
[9] See Archer & Roxon, supra note 3.
[10] See Blake Brittain, AI-Created Images Lose U.S. Copyrights in Test for New Technology, Reuters (Feb. 22, 2023, 8:41 PM), https://www.reuters.com/legal/ai-created-images-lose-us-copyrights-test-new-technology-2023-02-22/.
[11] See James Vincent, The Scary Truth about AI Copyright Is Nobody Knows What Will Happen Next, Verge (Nov. 15, 2022, 10:00 AM), https://www.theverge.com/23444685/generative-ai-copyright-infringement-legal-fair-use-training-data.
[12] See id.
[13] Id.
[14] Id.
[15] Id.
[16] Compl. at 33-36, Doe v. GitHub, Inc., No. 3:22-CV-06823 (N.D. Cal. Nov. 3, 2022).
[17] Defs. GitHub & Microsoft’s Mots. to Dismiss at 2-5, Doe v. GitHub, Inc., No. 4:22-CV-06823 (N.D. Cal. Jan. 26, 2023); Def. OpenAI, Inc.’s Mot. to Dismiss 2-4, Doe v. GitHub, Inc., No. 4:22-CV-06823 (N.D. Cal. Jan. 26, 2023).
[18] See Poritz, supra note 8.
[19] Id.
[20] Authors Guild v. Google, Inc., 804 F.3d 202, 229 (2d Cir. 2015).
[21] Id.
[22] Poritz, supra note 8.
[23] See Vincent, supra note 11.
[24] ChatGPT, Answer to “Should we grant copyright protection to works by generative AI like ChatGPT?,” ChatGPT (Mar. 20, 2023), https://chat.openai.com/chat.