Generative artificial intelligence (AI) is a type of artificial intelligence system that can generate text, images, and other media.
The process of creating content with generative AI can be roughly divided into two steps:
– The initial source of inspiration is termed as the “input”. The generic creation is based on this initial data, either retrieved from data labelled by the user/creator of the AI or from data autonomously found by the AI, to subsequently produce an “output” or result.
– From the initial data (input), the AI generates a model, infers rules, and then applies these rules. This is the output.
As a result, generative artificial intelligence raises concerns about the protection of both upstream content and downstream output.
Upstream protection for content processed by generative AI
In the upstream, intellectual property rights may be infringed upon if data is copied by AI and then used without permission. Indeed, there is a reproduction of copyrighted content (texts, images, sounds) to feed AI databases.
These contents, however, are protected by intellectual property rights. To legally feed AIs without prior authorization, the European legislator created an exception for “text and data mining,” which suspends the exploitation monopoly under copyright and related rights for data mining purposes. This relates to Articles 3 and 4 of Directive 2019/790 of the European Parliament and Council on copyright and related rights in the digital single market, which was adopted on April 17, 2019.
Data mining involves collecting data to transform them. There is a mining operation, which means algorithmic processing in order to interpret results. Nonetheless, these mining operations violate intellectual and artistic property rights.
Text and data mining can also lead to acts protected by copyright, sui generis rights on databases, or both, especially regarding the reproduction of works or other protected items, the extraction of content from a database, or both, which is the case when data is standardised during the text and data mining process.
The Directive 2019/790 establishes two mandatory exceptions for copying, i.e., two exceptions on reproducing content protected by literary and artistic property rights overall (texts, sounds, images, for all copyrights, related rights, and sui generis rights).
– Article 3 is a mandatory academic exception, benefiting research bodies and cultural heritage institutions performing mining for scientific research purposes, which right holders can’t oppose.
– Article 4, on the other hand, is an exception for all uses, regardless of the purpose (including commercial), provided that the copyright holder has not expressed opposition. In this case, the right holders can oppose even though the exception is mandatory. This may seem contradictory, but in reality, it’s the only possible balance between the rights of intellectual property right holders and the rights of those reproducing the data.
While data mining is currently allowed, the debate on the protection of content used upstream by generative AI to produce content is not over. Appeals are multiplying. For instance, in Europe, right holders are mobilising through “position papers” since the context between the drafting of Directive 2019/790 in 2018 and the time when generative AIs are booming in 2023 is very different. Compensation for intellectual property rights holders for the exception on data mining is now part of the considerations of intellectual property specialists working on the topic.
Protection of Content Generated by Generative AI
While robots were once considered passive tools, advances in artificial intelligence research have raised concerns about their role in the creative process. Indeed, these advancements sometimes attribute a significant role to them in content creation, raising questions about the protection of content generated by generative AI.
Under copyright law, the principle remains indifferent to merit (CJEU 1st March 2012, Football Dataco Ltd et al. vs. Yahoo! UK Ltd et al., Case C-604/10). Yet, it remains contentious to assert that generative AI can be the author of a work it produces.
This brings up the question: Is the appearance of a work tied to a regime? Should we accept protection for works generated by AI?
The European Parliament’s report from 27th January 2017 on recommendations for civil law rules on robotics suggested that copyright criteria should be adapted to accommodate these new AI-generated creations.
What’s crucial here is distinguishing between creations assisted by generative AI and those autonomously generated by AI.
If generative AI is used as a tool assisting creation, there’s no debate. The work is protected by copyright, with the affiliated rights belonging to the human creator. In this scenario, the work clearly reflects the personality of its human creator.
But what about the opposite scenario? Do we lean towards copyright or other paths (common law, special law)?
In 2018, the Superior Council of Literary and Artistic Property conducted an economic and legal analysis of the options.
The idea of AI as an author is ruled out today. To qualify as a work under the European Union law, it must meet specific conditions:
– There must be a human author.
– The work must be original.
Given these conditions, today’s perceived autonomy of AI is exaggerated as human involvement is essential. The idea of creating an “electronic personality” is rejected, as it might disrupt existing legal categories, leading to a legal chimaera. Without a human entity, the work can’t fulfil the originality criterion, which in France refers to the “imprint of the author’s personality” and similarly in the European Union, denotes an “intellectual creation unique to its author”, both necessitating a human presence.
The link to a human author in copyright law seems to be an international requirement
suggested by the Berne Convention.
If literary and artistic property rights are chosen, determining which specific rights apply (copyright, sui generis rights) becomes essential. Also, the definition of the author (indirect author, work without an author?) must be clarified.
Some argue no protection is required, and the “opt out” would fall under the “common by design” regime.
While awaiting legislation specific to works produced by generative AI, it’s worth noting emerging case law in this field. Notably, the US District Court for the District of Columbia’s decision in Thaler vs. Perlmutter on 18th August 2023, which ruled that copyright doesn’t apply to creations made by AI tools, even if they’re trained by human intelligence.
Such discussions on highly current topics raise genuine legal questions. The rise of new technologies requires the legislature’s keen attention and continuous updates to stay relevant. It’ll be fascinating to see future measures and laws adopted to further regulate these smart new inventions.
To find out more about generative AI, read our article Generative AI: Balancing Innovation and Intellectual Property Rights Protection or visit WIPO Magazine and Blog Modérateur.