One morning, you wake up from a vivid dream and feel inspired to create something from it. You decide to illustrate the fascinating creature that you dreamt of using an AI program that can generate images from natural language descriptions. After typing in your instructions, you pick your favourite result, but you aren't completely satisfied with the final work. So you refine the work with text prompts and manually correct any imperfections using a popular graphics editor tool. You are thrilled with the result and decide to mint these pictures into NFTs (Non-Fungible Tokens) and offer them for sale via NFT marketplaces. At the same time, you want to take your creativity a step further by writing a novel about this cute creature. You seek help from an AI writing tool, which helps you to plan and outline your novel before you start writing so you know what you need to include and can stay on track. Whenever you hit a creative wall, this tool provides you with inspiration and ideas. Finally, you finish your novel and decide to publish it along with the AI-generated images…
***********
Generative artificial intelligence (AI) is transforming the way we create and consume art, literature, and other creative works. With the help of various AI tools, it's easier than ever to bring our wildest dreams to life. From generating beautiful illustrations to helping writers plan and outline their next novel, these AI tools are opening up new avenues of creativity for people around the world.
In this article we explore the intricate intersection of generative AI and copyright law, primarily focusing on the French and European Union (EU) perspectives.
We first tackle the complex challenges surrounding the determination of authorship and ownership in works generated by AI, considering the varying degrees of human involvement. We then shift our focus to the nuanced realm of copyright exceptions in connection with AI training.
Deciphering Authorship and Ownership in AI-Generated Works
The primary challenge is to determine who is qualified or entitled to own the new works that have been autonomously created by generative AI. In other words, who is the author of the images created by AI, with minimal involvement or input from human creators (i.e., by typing simple instructions such as "cute rat with a red scarf")?
In some jurisdictions, the term "creator" under copyright law designates a natural person, which emphasizes the central role of humans in the creative process. Under French law, the attribution of intellectual property rights to a natural person is explicitly mentioned for certain types of work, such as collaborative or audio-visual works. Moreover, in a landmark decision in 2015 (Cass. 1re civ., 15 janv. 2015, n° 13-23.566), the French Supreme Court implicitly affirmed that authorship is exclusively reserved for natural persons. This ruling reinforced the notion that the legal concept of authorship, as it currently stands, is confined to human creators. Similarly, in the United States, the Copyright Office issued a statement of policy in March 2023 applying a human authorship requirement for copyrightability (see our coverage on Engage here). In the notable decision Telstra Corporation Limited v Phone Directories Company Pty Ltd [2010], the Australian Supreme Court declined to grant protection to a database generated solely through an automatic process by an AI system.
As per the French copyright case law and doctrine, the notion of originality is interpreted as requiring the imprint of the author's personality or intellectual effort. The French Supreme Court has held that a mere "implementation of automatic and constraining logic" without a genuine personalized effort falls short of copyright protection (Cass., ass. plén., Babolat c. Pachot, 7 mars 1986, n° 83-10.477). Similarly, the Court of Justice of the European Union has held, notably in its Infopaq decision (C-5/08 Infopaq International A/S v Danske Dagbaldes Forening, that copyright protection applies only to a "work" and that an object will be a "work" if it is original because it is the "author's own intellectual creation".
Considering the "natural person" and "originality" criteria described above, the copyright protection of works autonomously created by AI systems has long been a topic of discussion, notably given the lack of clear regulations.
Under French law, some authors suggest dealing with the ownership of AI-generated works by invoking the ownership mechanism under property law (i.e., "accession par production" set forth in Article 547 of the French Civil Code). From this standpoint, AI-generated works could be viewed as the "fruits" or outputs of the AI itself, implying that the rightful ownership would be attributed to the right holders of the AI system involved in the creative process.
The United Kingdom's Copyright Designs and Patents Act 1988 (CDPA) provides for a distinctive category for "computer-generated works" that allows for copyright protection in cases where there is no human author involved. This specialized category grants a protection term of 50 years, deviating from the "life plus 70 years" protection typically provided to works created by natural persons, and does not extend moral rights protection to computer-generated works. It is worth noting that section 9 of the CDPA designates the human author of computer-generated works as "the person by whom the arrangements necessary for the creation of the work are undertaken". According to some authors, the human author as defined in the section 9 above seems most likely to refer to the developers or right holders of the computer system that autonomously creates works.
The terms and policies employed by certain AI vendors also explicitly provide that users (especially those using the AI tools for free) do not possess ownership rights over the AI-generated content but rather are granted a non-exclusive, non-sublicensable and non-commercial licence to use it. It is therefore crucial for individuals involved in these activities to carefully review and adhere to the terms and conditions of AI platforms to ensure compliance and avoid potential disputes.
Determining ownership becomes more complex when dealing with AI-generated works that arise from interactive collaborations between AI systems and human beings (e.g., writing a novel with the assistance of AI). In such cases, ownership attribution becomes intertwined with the level of human creative input and involvement. The delineation between the contributions of the AI system and the human participant becomes blurred, arguably making it difficult to determine the sole authorship or ownership of the resulting work. It's essential to carefully consider the nature and extent of the human's creative contributions and the role played by the AI system in the collaborative or iterative process.
At the EU level, where copyright protection automatically applies to original works upon their creation without the need for registration, we anticipate that case law will eventually provide clarity on attributing authorship or determining ownership in such scenarios. In the meantime, for users of generative AI, it is imperative to establish evidence of their human involvement in the creative process right from the outset to the extent possible. In France, authors have the option of registering works with the French National Institute of Industrial Property (INPI) for probative purposes, even though the value of such evidence in court remains uncertain, given the lack of substantive examination conducted by INPI on the registered works.
Some authors also advocate the importance of establishing new sui generis rights, similar to the ones granted to database producers, to acknowledge the significant contributions and efforts made by AI users in refining and improving the output of AI systems (e.g., refining images with text prompts and manually correcting any imperfections).
Copyright Protection and AI Training
The training of generative AI may involve the processing of copyrighted works. Under EU copyright law, AI vendors can mainly rely on the "text and data mining" (TDM) exceptions, as outlined in Articles 3 and 4 of the Digital Single Market (DSM Directive) of 2019, to make (at least temporary or indirect or transient) copies of copyrighted works for the purposes of AI training. TDM is defined by the DSM Directive as "any automated analytical technique aiming to analyse text and data in digital form to generate information such as patterns, trends and correlations".
Article 3 of the DSM Directive establishes a mandatory exception for TDM conducted by research organizations and cultural heritage institutions for scientific research purposes. Article 4 permits various beneficiaries to engage in TDM for any use, including commercial purposes, as long as copyright holders have not expressly reserved their rights (the "opt-out" approach).
The "opt-out" approach means that copyrighted works can be used for AI training unless right holders proactively object in advance. Right holders can object to TDM through machine-readable processes such as hashing, watermarking, metadata, fingerprinting, or by specifying terms and conditions on websites hosting the content. For new works, opting out can be relatively straightforward, such as attaching "Do Not Train" credentials using provenance technology. However, opting out may become more challenging for older works (e.g., a book published in the twentieth century) that are still protected by copyright up to 70 years after the author's death.
The recent amendments to the EU Artificial Intelligence Act, as adopted by the European Parliament on 14 June 2023, introduce supplementary transparency requirements, which compel providers of AI models (such as Generative AI systems) to publish summaries of copyrighted data used for AI training. The advocates of this transparency requirement believe that, if this disclosure obligation is ultimately implemented, it could assist copyright holders in identifying the enforceability of their opt-outs and taking the measures necessary to prevent the unauthorized use of their works for AI training.
Next steps
As we stand at the crossroads of technological advancement and artistic expression, it is clear that redefining copyright in the generative AI era requires a multidimensional approach. Taking into account the interests of AI algorithms, human creators, and those of the wider society is an intricate task that requires ongoing dialogue, legal frameworks that adapt readily to emerging technologies, and collaborative efforts between stakeholders.
In the meantime, let's stay updated on the latest regulations and legal developments. Keep an eye out for upcoming articles in Hogan Lovells AI series to learn more about the intersection of generative AI and copyright.
Authored by Ying Lou, Maria Rozylo and Cemre Ercakir.
