ChatGPT Unravelled

: A Deep Dive into the AI Language Model and its Copyright Implications

Abstract

While most people familiar with the topic knew that OpenAI had been pushing the boundaries of large-language models in recent years, many were still astonished by ChatGPT's capabilities. For the first time, an extremely powerful tool, capable of understanding and producing human-like text in a conversational context was freely available to the public. It hit like a bombshell in the media, generating an exponential amount of attention around the subject... Then, when the dust settled, the legal questions followed. This thesis particularly focuses on the interplay between ChatGPT and the European copyright regime. In order to fully understand what is at stake, it is crucial to explain the underlying technical aspects of ChatGPT. As such, this thesis first focuses on explaining machine learning, natural language processing, and the particularity of generative pre-trained transformers, which are the underlying architecture of ChatGPT. We then move on to the legal part of this thesis, concentrating ourselves on the implications that ChatGPT may have on an author's economic rights. Here, we first cover not only the implications during the training process, but also during the user-prompted content-generation process. Afterwards, special attention is paid to eventual arguments that could be used to defend ChatGPT. Finally, this thesis ends with a brief outlook into the future. Is ChatGPT compatible with the European copyright framework? And if not, could there be a way out? The present thesis attempts to find an answer to these questions.

Date of Award	30 Aug 2023
Original language	English
Awarding Institution	University of Namur
Supervisor	Benoit Michaux (Supervisor)