The rapid advancements in generative artificial intelligence are fueling both excitement and apprehension, particularly in the realm of intellectual property protection. The ability to share creative outputs has never been easier, driven by decades of technological evolution; from the emergence of computers and the internet to the rise of mobile devices. These innovations have progressively lowered barriers to the dissemination of information and ideas. However, as generative AI becomes increasingly capable, growing concerns are surfacing about whether existing IP protections are effectives against abuses made by generative AI.1 Traditionally, the use of AI was highly supported among the creative community, as a way to gain inspiration for future works. In today’s climate, AI has gained autonomous characteristics in being able to create its own works.2 The spreading use of generative AI poses two main issues with regard to IP protections, namely relating to the usage of protected works as training data and the ambiguous extent of IP protection afforded to AI works. Therefore, as AI continues to develop, it simultaneously blurs the understanding of traditional IP tenets, such as creativity, authorship and ownership.3
What is generativere AI?
Generative AI is increasingly pervasive in today’s society, with its innovations being actively utilized across various domains and its outputs widely consumed by the public. Throughout the years, the technology has amassed popularity among different sectors, including medicine, art and education. However, there is currently no universally agreed upon definition for ‘generative AI’.4
What most of these AI systems have in common is their functioning on the basis of generative deep learning models.5 These are algorithms which essentially try to simulate the 5 David Foster, Generative Deep Learning (p. 16, 2nd edn, O’Reilly 2023).
decision-making processes of humans.6 The starting point for these models are training data sets, which are used by the algorithm to detect patterns and eventually put out its own outputs in a probabilistic manner. Simply put, the old data aids the algorithm in generating a new set of data.7
However, it must be noted that this technology existed for over fifty years, while many seem to think it is a new technology. This confusion arose following the release of ChatGPT, which marked the first encounter many individuals had with generative AI technology.8
Can AI generated works be protected?
As it stands today, there are currently no direct regulations on international or domestic levels which address the IP protectability of AI generated works. There is a large divide on affording protections to AI works. On the one hand the question arises whether AI output is actually original if it is based on other people’s works? On the flip side, protecting generated works could result in various economic benefits. According to Dr. Faye Wang, these benefits could include the incentive for innovation and investment ‘to the benefit of economic development, efficiency and the advancement of human society’.9
Under current EU law, it is difficult and arguably impossible to accord any protections to AI works. In order for any work to qualify for copyright protection, it must fulfil certain requirements.
Firstly, the work must be characterized as a protected subject matter. According to the Berne Convention, literary and artistic works shall fall within the ambit of protection.10 The test from the CJEU Painer case can be used to effectively determine whether a specific subject is entitled to copyright protection. The Court established that in order for intellectual property to be considered an author’s own creation, it must reflect the author’s personality.11 Accordingly, an author’s personality is reflected when they are able to reflect their creative abilities in the production of the work by making free and creative choices.12
The next criterion which must be fulfilled is the notion of ‘originality’. The current EU acquis communautaire does not effectively address this concept, however CJEU case law has harmonized its approach to determining originality. The CJEU in its Infopaq judgement has concluded the threshold for the originality requirement.13 As such, a work may be considered original if it reflects the author’s own intellectual creation.14 An author’s intellectual creation can be deducted from their creative choices, combinations and arrangements.15
Already at first glance when applying the Painer test, the works generated by AI do not fall within the criteria of protected subject matter, as they lack the reflection of an author’s personality. Personality is connected to human intellectuality, which generative AI aims to duplicate, however it is not authentic. Secondly, it is difficult to determine the originality of a work generated by AI considering that it is based on previous works which have been fed to the algorithm. Therefore, as it currently stands, AI-generated works cannot be protected under the EU’s copyright regime.
Does the Use of Protected Works as Training Data Infringe on Author’s Rights?
Under the current EU copyright framework, the use of protected works for purposes of training AI can be exempted under specific circumstances. The Directive on Copyright and Related Rights in the Digital Single Market contains an exception related to Text and Data Mining (TDM). The Directive defines TDM as ‘any automated analytical technique aimed at analyzing text and data in digital form in order to generate information which includes but is not tied to patterns, trends and correlations.16 Furthermore, the technique is most commonly used by developers to extract data and it primarily depicts patterns, correlations and other core information from data sets.17 Thus, frequently TDM is used on IP-protected data, to later generate new works. Such activities may pose a risk on the copyright vested with the author against reproducing their works.18
The European Union’s Digital Single Market strategy aims towards ‘better access to digital content’, which explains the mandatory exceptions related to TDM.19 The directive provides for two exceptions; for purposes of scientific research and; for reductions and extractions necessary for TDM.20 The second exception has a more general character, making it more easily applicable to activities of AI developers. Specifically, the article provides for an exception ‘for reproductions and extractions of lawfully accessible works and other subject matter for the purposes of text and data mining’. Additionally, the article’s primary aim is to advance and foster initiatives related to big data and artificial intelligence. Therefore, AI developers may legitimize their usage of protected works by arguing that their TDM activities fall within the scope of the exception.
The Directive does provide for additional protections that rightholders may take. Although written in broad terms, an ‘opt-out’ provision can be found under Article 4(3). Rightholders may require developers intending to utilize their works to obtain prior authorization before
engaging in TDM activities. The opt-out must be concluded ‘in an appropriate manner, such as machine-radiable means in the case of content made publicly available online’. This is a highly relevant limitation to authors as they still retain the right to stay in control of third-party usage of their works, and limit the exploitation of their works by AI developers.21
Conclusion
The rapid development of generative AI presents significant challenges to traditional frameworks of intellectual property protection. While these technologies hold immense potential for innovation and creativity, they also disrupt longstanding principles of authorship, originality, and ownership. As this analysis highlights, AI-generated works currently lack the necessary human intellectual and creative input to qualify for protection under the EU copyright regime. Similarly, the use of protected works as training data raises concerns about infringement, though certain exceptions, such as those provided under the EU’s Text and Data Mining provisions, offer limited safeguards.
The legal framework must strike a balance between fostering technological advancement and protecting the rights of original creators. While the existing copyright regime accommodates some aspects of AI-related activities, it remains insufficient to address the broader implications of generative AI. Moving forward, clearer regulations and updated legal definitions will be necessary to ensure that intellectual property laws remain relevant in an increasingly AI-driven world.
1 Ron Schmelzer, ‘What is the Future of Intellectual Property in a Generative AI World?’ (Forbes, 18 July 2024) <https://www.forbes.com/sites/ronschmelzer/2024/07/18/what-is-the-future-of-intellectual-property-in-a-generat ive-ai-world/> accessed 4 December 2024.
2 Jan Smits et al, Law and Artificial Intelligence (p. 324, Vol 35, Springer 2022).
3 Jim Holloway, Milton Chong and Julia S. Dickenson, ‘Will copyright law enable or inhibit generative AI?’ (World Economic Forum, 13 January 2024)
<https://www.weforum.org/agenda/2024/01/cracking-the-code-generative-ai-and-intellectual-property/> accessed 4 December 2024.
4 Francisco Jose Garcia-Penalvo and Andrea Vazquez-Ingelmo, ‘What Do We Mean by GenAI? A Systematic Mapping of the Evolution, Trends and Techniques Involved in Generative AI’ (2023) GRIAL Research Group Paper <https://reunir.unir.net/bitstream/handle/123456789/15134/ijimai8_4_1.pdf?sequence=3&isAllowed=y> accessed 4 December 2024.
6 Cole Stryker and Mark Scapicchio, ‘What is Generative AI?’ (IBM, 22 March 2024) <https://www.ibm.com/topics/generative-ai> accessed 4 December 2024.
7 Giorgio Francheschelli and Marco Musolesi, ‘Copyright in Generative Deep Learning’ [2022] p. 2 4 Data & Policy
<https://www.cambridge.org/core/services/aop-cambridge-core/content/view/C401539FDF79A6AC6CEE8C52 56508B5E/S2632324922000104a.pdf/copyright-in-generative-deep-learning.pdf> accessed 4 December 2024. 8 Adam Zewe, ‘Explained: Generative AI’ (MIT News, 9 November 2023)
<https://news.mit.edu/2023/explained-generative-ai-1109> accessed 4 December 2024. 9 Faye F. Wang, ‘Copyright Protection for AI-Generated Works: Solutions for Further Challenges from Generative AI’ [2023] p. 92 5 Amicus Curiae Series 2 <https://journals.sas.ac.uk/amicus/article/view/5663> accessed 4 December 2024.
10 Berne Convention for the Protection of Literary and Artistic Works (adopted 9 September 1986) 828 UNTS 221 (Berne Convention) art. 2.
11 Case C-145/10 Eva-Maria Painer v. Standard Verlags GmbH and Others [2013] EU:C:2013:138, para 99.12 Mireille van Eechoud, The Work of Authorship (p. 95, Amsterdam University Press 2014). 13 Tatiana Synodinou, Codification of European Copyright Law: Challenges and Perspectives (p. 100, Kluwer Law International 2012).
14 Case C-5/08 Infopaq International A/S v. Danske Dagblades Forening [2009] ECR I-06569, para 37. 15 Synodinou (n 13) p.100-101.
15 Synodinou (n 13) p.100-101.
16 Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC [2019] OJ L130/92, art. 2.
<https://link.springer.com/content/pdf/10.1007/s12689-020-00089-5.pdf> accessed 4 December 2024. 20 Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC [2019] OJ L130/92, arts. 3 and 4.
21 Joris M. Roos, ‘Artificial Intelligence: Copyright & Consequences’ (Thesis, Utrecht University 2023) p. 32.
Comments