Description
The past months have witnessed a spate of lawsuits (primarily in the US) alleging copyright infringement by Generative-AI (G-AI) systems. Authors claim that the unauthorized use of their works for the training of G-AI systems threaten the sustenance of human creativity by; (a) undermining incentives to engage in creation (as claimed in the Getty Images dispute) and, (b) substituting human creativity with machines (a key concern voiced in the SAG-AFTRA protests in the US). Stringent copyright protection of works used for training G-AI systems has been advocated as a strategy for safeguarding human creativity.Using an interdisciplinary (law and computer science) approach, this presentation explores the potential legal implications of the use of copyright protected content for the training of G-AI systems [both Generative Adversarial Networks (GANs) and Diffusion models] generating visual output, from the perspective of EU law. Such an interdisciplinary approach is necessary since the way these two architectures utilize their input while generating their output differs substantially, and this may lead to different legal implications from a copyright law standpoint. For instance, the generator, which is the responsible part of a GAN to generate artificial images, does not have direct access to the input of the network, whereas such a separate generator does not exist in a diffusion model.
While acknowledging the validity of the aforementioned concerns raised by authors, we argue that the adoption of a stricter approach to copyright protection is not a viable strategy for safeguarding human creativity in the age of AI. We elucidate that this would involve an ever-more expansive interpretation of the scope of copyright’s exclusive rights; particularly the exclusive right of reproduction, to also cover non-expressive, incidental uses of copyright protected works in a way that goes against copyright’s essential purpose and function and could stifle the potential offered by G-AI for the democratization of creativity and culture.
We assert that a purposive interpretation of copyright’s exclusive economic rights is required in order to ensure that non-expressive uses of copyright protected content for the purpose of training and operating G-AI is permitted. We finally propose that, alternative legal regimes such as personality rights, unfair competition law, contract law and labor law offer more appropriate legal tools for addressing the concerns outlined in (a) and (b) above in a way that could achieve a fairer balance between the interests of authors and users of G-AI.
| Period | 18 Apr 2024 |
|---|---|
| Event title | BILETA Annual Conference 2024 |
| Event type | Conference |
| Location | Dublin, IrelandShow on map |
| Degree of Recognition | International |