Generative Artificial Intelligence ("AI") is transforming how text, music, art, and other works are created. From DALL-E 2, which enables the generation of lifelike images and artwork, to ChatGPT, which generates human-like text in response to various inputs.
In today's landscape, generative AI raises a lot of concerns about copyright. Should these works have been the result of only human effort, copyright protection would be a straightforward matter. However, the landscape of creation has shifted. It is becoming increasingly difficult to make a distinction between human-made and non-human made works.
The emergence of generative AI particularly raises the question whether AI systems can claim authorship under Belgian copyright law.
Copyright protection
Under Belgian copyright law, copyright protection is granted to literary and artistic works that constitute a concrete and original expression by the author(s). This concept must be understood in a much broader manner considering that any creation of the human mind can be protected by copyright. Copyright protection entails that numerous uses of a work are prohibited without the permission of its author (or lawful copyright owner).
The requirement of a concrete form entails that the protection of copyright is not available to mere abstract ideas, principles, or methods. As a result, ideas, principles or methods should always be expressed in a concrete way or form which can be perceived by the senses, e.g. a book, a painting, an image.
The requirement of originality, which is not defined in Belgian copyright law, is based on the established case law of the Court of Justice of the European Union (“CJEU”). Originality in this regard is not synonymous with ‘new’, i.e., a creation that has never been seen or created before. Originality signifies that the work must be an intellectual creation of the author(s), reflecting their personality. This will be the case if the author is able to express free and creative choices. The choices, however, need to transcend the mere technical interpretation.
Once the abovementioned conditions for copyright protection are fulfilled, such protection shall apply automatically, without the need for registration.
Generative AI: input v. output
At the core of generative AI lies a very consistent approach. AI models take in a vast amount of data (‘input’), analyze such data, and then produce or generate new data (‘output’).
On the input side, questions arise on how copyright law addresses the matter of infringement and who will be responsible in case of an infringement. However, these questions fall outside of the scope considering that this article will focus on the output side.
The term ‘output’ refers to the responses or answers that the AI system provides, referred to as works created by generative AI.
Can AI systems claim authorship?
From a copyright perspective, a distinction must be made between weak or narrow AI and strong AI.
Weak or narrow AI
In this context, an AI system acts as a technical instrument to assist a natural person or human. The AI system solely functions as the executor without the capacity for creativity. In this sense, the AI system is really nothing more than a step in the evolution of technology. Such scenario refers to an AI-assisted content.
For weak or narrow AI or AI-assisted content, no new legal issues arise considering that there is a human involved. As long as the human chooses the creative form, the output of the AI system can be qualified as the product of human creativity. Therefore, if the output meets the requirements of copyright (i.e., concrete form and originality), the human involved in the AI system will be qualified as the author and not the AI system itself. However, there is much discussion about who the human author is considering that there are many people involved in the process of creating output through an AI system, such as the user, the programmer, the owner of the data, the legal entity that employs them or some form of co-authorship between these actors.
For example, Adobe photoshop has an AI tool - Adobe Firefly - to help edit pictures in a quick and easy manner. Given that the user provides their own image, it might be argued that the user is the author of the final output generated by Adobe Firefly (Adobe Firefly, in that regard, is nothing more than a tool to help edit an image, much like a physical brush used by a painter would be considered a tool). However, this is less apparent for DALL-E 2, considering that the AI system creates the image based on the input of the user. In this instance, Open AI, the company behind DALL-E 2, has attempted to resolve the discussion by transferring all intellectual property rights to the user. The fact that a transfer is explicitly stipulated, suggests that Open AI sees itself as the original copyright holder.
Strong AI
Within this context, the AI system acts like a human (‘no human in the loop’) or even superior to a human (‘artificial superintelligence’). In such a system, the intelligent system autonomously works towards an outcome that was not predetermined in advance by a human. Such scenario refers to an AI-generated content.
However, some experts seem to believe that such an AI system acting like a human or superior to a human does not yet exist. Others argue that such an autonomous AI already has occurred or that it will soon.
When an AI system autonomously generates an output with no involvement from humans, the question also arises whether such a smart AI system can claim authorship? The response to this question should be negative, given that copyright can only be attributed to a human creator.
Under copyright law, initial copyright is granted to its author. Such author has traditionally always been considered a physical, natural person. This is, for example, demonstrated by the fact that the term of copyright protection is calculated from the author's death, or the importance placed on moral rights protection, which encompasses the author’s right to integrity and paternity. The need for human intellectual authorship is ingrained in our system, which excludes authorship for non-human entities.
An additional reason to exclude AI authorship is that currently, an AI system as such is not considered as a legal entity and therefore cannot possess any legal rights, including ownership rights.
Conclusion
Based on the above analysis and as the law currently stands, an AI system cannot claim any authorship in Belgium given their absence of both physical and legal personhood. Only the human actors behind such AI system can claim authorship. However, there remains uncertainty on which actor involved in the AI system can claim the actual authorship. It is likely that such uncertainty will be resolved through contractual agreements among the concerned actors. Perhaps the rapidly evolving landscape surrounding generative AI will offer more insights in the future.
