ChatGPT and Legal Theory

*this article was first published on Medium on Mar 19, 2023

It couldn’t skip my attention that the majority of my contacts have posted, re-posted or react to posts about ChatGPT’s mind blowing abilities. In addition to my full-time job, I spend my past 4 years studying and researching the topic of AI, especially in the boundaries of law, governance and ethics. Working towards my PhD, I have also managed to get my second masters in this discipline. So, I just had to add my 5 cents on this matter, especially from a legal theory perspective.

ChatGPT is an artificial intelligence language model. And despite the fact that it is currently one of the most advanced publicly available language model AIs on the market, it is still considered being a weak AI. Meaning an AI that has been designed to perform a specific task. In the case of ChatGPT, it would be answering questions, generating text, or performing language-based tasks. It has, however, no abilities to reason, learning from or interacting with the environment, in a way a human would. But even if ChatGPT does not reach a level of a so called artificial general intelligence (strong AI), it will disruptively alter business models, education and society in general.

So how does legal theory respond to this development?

A computation legal theory and legal informatics cluster the usage of AI in three main groups: AI as a tool, AI as a representative, and AI as a legal person.

ChatGPT is a weak AI and falls currently under the terms of tools with a tendency to become a representative. One of the examples would be integrating ChatGPT into a contract drafting platform, hence replacing lawyers or legal departments for contract drafting activities in the nearest future. Keep in mind, that more advanced large multimodal model GPT-4 has already passed a bar exam.

AI as a tool can be regulated to some extent within the existing legal framework (e.g. GDPR) and legal instruments such as copyright or licensing. But already there we see a big regulatory gap, as a rapid development of this technology and the human level results of machine’s drawings or texts stretch the law to its limit. There are tons of questions that remain yet to be answered — questions about intellectual property rights, liability, or the right punishment for a naughty AI under the criminal law.

As AI is trained on big data and uses principles of machine learning, it is not legally clear to what extent a developer or the provider can be held responsible for AI’s actions. So, for example, if ChatGPT insults you, was it because the developer programmed the chatbot in that way, or was it because the AI learned to insult people while interacting with users? Did you know that the threshold for insulting a chatbot is quite low? Machine learning principles are based on data and, therefore, garbage in, garbage out. Or, as in this case, insult in, insult out.

But the most exciting question, in my opinion, remains the question of the legal personhood of the AI. The research in this regard started not that long ago, but it constantly gains momentum with scientific fields of law, philosophy, computer science and engineering, helping to form the idea of AI as a person and hence the legal personhood of the AI. The reasoning in favour of granting a legal personhood to the AI is mainly but not exclusively (see Sophia, the first robot citizen) reserved for the strong AI.

But who knows how fast technologies such as ChatGPT will develop? Undoubtedly, the more advanced these technologies will get, the more tangible and less abstract the entire idea of AI as a legal person will become. And maybe already in the near future, humanity can face a machine in court as equals.