Augmented Advocacy Series (Bermuda): Copyright infringement in the age of AI

Tools such as ChatGPT are used to generate just about anything and as powerful as these AI tools are, AI comes with complexities and important considerations.

AI refers to systems that can perform tasks that typically require human intelligence. At a high level, AI works through machine learning algorithms, allowing systems to learn from data.

AI models are trained on large data sets that allow it to recognise patterns and make predictions. After training, the AI model can generate new data – for example, systems such as ChatGPT or Stable Diffusion.

Such systems are intentionally programmed to search for, find, sample and reproduce relevant segments of existing works as an inherent part of its programming.

So that raises the question: with the use of AI-generated work, does the potential for copyright infringement arise as an unintentional mistake, or as an intended, purposeful design?

Copyright law protects the rights of creators against unauthorised reproduction of their work with its aim being to encourage and promote innovation through the creation of original works.

Bermuda’s copyright law, the Copyright and Designs Act 2004, does not provide protection for AI-generated work.

Copyright infringement occurs when someone reproduces the whole or a substantial part of an author’s work without the author’s permission and where none of the permissive exceptions to copyright law apply.

The courts have interpreted “substantial part” to mean a qualitatively significant part of the work, therefore even a small portion of work can be substantial.

As court cases continue to unfold, the copyright legal landscape begins to take shape and starts to reform UK copyright law, which means Bermuda is sure to quickly follow suit.

In the recent UK High Court case of Getty Images v Stability AI, Getty alleged that Stability AI infringed its intellectual property rights and copyrights, claiming that the defendants accessed and reproduced its copyrighted images without permission to develop the AI solution known as Stable Diffusion, an AI image generation tool that creates images based on text prompts.

Stability AI claimed that their tool functions as an AI model trained on publicly available data, deeming it “fair use” or “legally accessible” for use in training.

A copyright-protected work is intended to prevent unauthorised reproduction that is typically published or accessible online. The way AI tools work, a computer programme browses the internet in search of content, extracts samples from that content and then uses those reproductions, whether whole or substantial, to “create” new work.

Given this, is there scope to argue that this is copyright infringement, or can it be deemed as “fair use”?

The Getty case is part of an ongoing legal debate in the age of AI and the outcome of the case could have significant implications for how AI companies source their data and how copyright law should apply to AI-generated content to avoid the issue of copyright infringement.

Considering this quickly evolving phenomenon, it appears that the question of ownership of AI-generated work is not fully settled by the law.

Is it the AI tool developer, the individual user who input the query, or the AI tool itself?

Critics argue that the law needs to adapt to the realities of the future of AI and the risk of copyright infringement for using AI-generated content.

In the meantime, users should proceed with caution until copyright laws are updated to include AI-generated work.

As exciting as the future of AI will turn out to be, we cannot completely cut out the need for human intervention.

In the future of AI, humans will serve as the “redundancy plan” if AI fails or malfunctions. Further, without human intervention, there will be a more serious risk of copyright infringement.

AI systems are designed to simulate human intelligence and as such (for now) humans can combine their original content with new creative AI perspectives with less risk of committing copyright infringement.

However, ask yourself if you know how much of the documents, text or other content that you have had AI generate for you infringes anyone’s copyrights.

As AI continues to develop and play a significant role in the creation of content, the law surrounding copyright will be forced to evolve accordingly.

In the meantime, human intervention plays a crucial role in avoiding copyright infringement and serves as the balance between encouraging innovation while protecting originality.

Ultimately, taking steps to understand the legal parameters surrounding copyright will be essential for users of AI until the evolution of the law.

First Published in The Royal Gazette, Legally Speaking column, November 2024