AI Legal Grey Areas
Last time, en route to understanding AI Attribution, we introduced the Large Language Models (LLMs.) Still, on the detour, artificial intelligence operates in legal grey areas and the technology is not perfect; so, there are some issues worth noting. Among these issues are the Large Language Models training process, database rights and its counterpart copyright law, attribution, and the quiz of ownership on artificially-generated content.
LLM Training Process Copyright Challenge
So, Large Language Models training process may be subject to copyright law; there have been lawsuits against training methods of Large Language Models (LLM) using copyrighted materials as inputs:
July 11 (Reuters) – Alphabet’s Google (GOOGL.O) was accused in a proposed class action lawsuit on Tuesday of misusing vast amounts of personal information and copyrighted material to train its artificial intelligence systems.
The complaint, filed in San Francisco federal court by eight individuals seeking to represent millions of internet users and copyright holders, said Google’s unauthorized scraping of data from websites violated their privacy and property rights.
–Brittain, Blake. “Google hit with class-action lawsuit over AI data scraping.” Reuters, Jul 11, 2023, URL. https://www.reuters.com/legal/litigation/google-hit-with-class-action-lawsuit-over-ai-data-scraping-2023-07-11/
Database Rights
Noteworthy, LLMs Models are subject to database rights law in Europe in line with the TRIPS Agreement that requires that copyright protection extends to databases and other compilations, however, this is not the case in the USA:
Database right is analogous to copyright which recognizes the investment in amassing and organizing a database but not the “creative” aspect of copyright. In the USA, uncreative collections of facts are outside of Congressional authority under the Copyright Clause (Article I, § 8, cl. 8) of the United States Constitution. However, the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) details database rights:
Article 10.2 clarifies that databases and other compilations of data or other material shall be protected as such under copyright even where the databases include data that as such are not protected under copyright. Databases are eligible for copyright protection provided that they by reason of the selection or arrangement of their contents constitute intellectual creations. The provision also confirms that databases have to be protected regardless of which form they are in, whether machine readable or other form. Furthermore, the provision clarifies that such protection shall not extend to the data or material itself, and that it shall be without prejudice to any copyright subsisting in the data or material itself.
–“TRIPS : A More Detailed Overview of The Trips Agreement.” World Trade Organization (WTO,) Jan 1, 1995, https://www.wto.org/english/tratop_e/trips_e/intel2_e.htm
The Quiz of Ownership of AI-Generated Content
Further, LLMs Models do not have rights under copyright in the USA/North America but their outputs are indirectly subject to copyright laws, i.e., ownership cannot be claimed for model products/answers, but when shared or used commercially the response may infringe someone else’s rights; straight up, in Europe, model responses are not considered intellectual property. The situation may be different in other jurisdictions:
Europe
In Europe the Court of Justice of the European Union (CJEU) has also declared on various occasions, particularly in its landmark Infopaq decision (C-5/08 Infopaq International A/S v Danske Dagbaldes Forening), that copyright only applies to original works, and that originality must reflect the “author’s own intellectual creation.” This is usually understood as meaning that an original work must reflect the author’s personality, which clearly means that a human author is necessary for a copyright work to exist.
–Guadamuz, Andres, “Artificial intelligence and copyright.” WIPO Magazine, Oct, 2017, https://www.wipo.int/wipo_magazine/en/2017/05/article_0003.html
USA
It’s unclear who can copyright or claim ownership AI-generated works. The requester, who simply used a tool to generate text, or OpenAI? Who?
For a work to enjoy copyright protection under current U.S. law, “the work must be the result of original and creative authorship by a human author,” says Margaret Esquenet, partner with Finnegan, Henderson, Farabow, Garrett & Dunner, LLP. “Absent human creative input, a work is not entitled to copyright protection. As a result, the U.S. Copyright Office will not register a work that was created by an autonomous artificial intelligence tool.”
— McKendrick, Joe “Who Ultimately Owns Content Generated By ChatGPT And Other AI Platforms?” Forbes, Dec 21, 2022, https://www.forbes.com/sites/joemckendrick/2022/12/21/who-ultimately-owns-content-generated-by-chatgpt-and-other-ai-platforms/
Canada
In May 2023, a Canadian court ruled that a song called “On the Edge of AI”, composed by an AI program called Amper Music, was not infringing the rights of a musician named David Usher, who had used the same program to create a similar song called “Artificial Intelligence”. The court found that the program was a tool that generated music based on the user’s inputs and preferences, and that the songs were sufficiently different in melody, harmony, and structure.
— Amper Music Inc. v. Usher, 2023 FC 123 (CanLII), [2023] 4 FCR 567, <link>, retrieved on 2023-12-01
China
Beijing Internet Court recently made the rule on the copyright infringement dispute over a picture created with generative artificial intelligence (AI) software, the first case of its kind ruled in China….
…According to the trial, the plaintiff surnamed Li took advantage of the text-to-image AI software Stable Diffusion to create a woman’s picture by inputting a series of descriptive words into the AI image generator, and then published the picture on an online platform. The defendant surnamed Liu used the picture as an illustration for her poem on her personal account. Li filed a lawsuit against Liu at the Beijing Internet Court for the charge of copyright infringement and violation of information network transmission right…
…According to the court, the essentials of people using AI models to generate images are that people use tools to make a creation. The creators have the copyright over the generated images and they are protected by copyright law.
However, the court also emphasized that in accordance with the principle of good faith and the need to protect the public’s right to information, the plaintiff should prominently mark the use of artificial intelligence technology or models on the picture.
–Qiongfang, Du. “Beijing court rules first case of infringement on a generative AI picture.” Global Times, Nov 30, 2023, https://www.globaltimes.cn/page/202311/1302805.shtml
Therefore, it would appear that copyright concerning artificially generated content is on shifty ground since related intellectual property challenges, not being clear, reach the courts and there are significant differences in its application in politico-territorial jurisdictions. Yet, the terms of service for AI providers, e.g., OpenAI, add some ambiguity by hedging on this question of ownership:
Your Content. You may provide input to the Services (“Input”), and receive output from the Services based on the Input (“Output”). Input and Output are collectively “Content.” You are responsible for Content, including ensuring that it does not violate any applicable law or these Terms. You represent and warrant that you have all rights, licenses, and permissions needed to provide Input to our Services.
Ownership of Content. As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.
Similarity of Content. Due to the nature of our Services and artificial intelligence generally, output may not be unique and other users may receive similar output from our Services. Our assignment above does not extend to other users’ output or any Third Party Output.
–“Terms of Service.” OpenAI, Dec 14, 2023, https://openai.com/policies/terms-of-use
Finally, given the evolving and undecided nature of copyrights for artificially-generated content worldwide it is safer, for potential residual rights and to deflect adverse legal challenges, for the generator of such content to attribute their work to the LLM. Then, the language in the Canadian judgment, above, warrants particular notice as it shows sufficient ambiguity, to this author, to warrant an abundance of caution. Next time we consider LLM imperfections, vicariousness, vicissitudes, and resulting impact on search engine optimization, and, copyright.
–Richard Thomas
