Rethinking IP in a generative era

Discussion of the opportunities, challenges and questions around generative artificial intelligence (AI) has been an inescapable part of the discourse about the future of business and working life since its more mainstream introduction a couple of years ago (despite its inception in the 1960s). Its impact on the legal landscape is no exception. Where the AI system generates new content or outputs, what are the IP law implications of this?

This is the first article in a series that hones in on the intellectual property (IP) issues arising from the development and use of generative AI.

The focus of this article is on the ‘generative’ aspect of generative AI – where the AI system generates new content or outputs (for example, think of the response generated by ChatGPT to a prompt you input). What are the IP law implications of this?

Do IP rights exist in generative AI outputs?

A key question to which there is currently no clear or easy answer is whether outputs of generative AI are actually capable of being protected by IP rights. The most relevant IP right is copyright, which in the UK protects works that are “original”, and this is generally interpreted to mean that the work is the author’s own creation.

This raises a couple of further points. First, the originality of generative AI outputs is up for debate. Indeed, it is the subject of current litigation in both the UK and the US in the context of allegations of generative AI being “trained” with copyrighted online content and then reproducing it in its outputs. We will consider the ongoing litigation in further detail in a future article, but it seems reasonable to assume that if a generative AI output does reproduce existing copyrighted content, it would not be considered original. 

Second, production of an original work requires there to be an author of that work. Unlike many other countries, UK legislation (though it predates generative AI) does provide copyright protection to computer-generated works in circumstances where there is no human author. The computer program itself is not considered to be the author. Instead, the author is “the person by whom the arrangements necessary for the creation of the work are undertaken”. 

This of course raises a further question of who is responsible for making the “arrangements” for the creation of generative AI outputs? For example, in relation to use of ChatGPT, would this be OpenAI as the developer of the system, or would it be the individual user who has prompted the system to generate the output? The reality is that this question is currently untested, but it is possible that it would be interpreted with reference to the level of input by the human user. Where a generative AI system requires input and control by a human user beyond minimal basic text prompts (for example, by extensive detailed prompts or uploaded material), there is likely to be a greater argument that the human user is the author.

All of this is to say that the law does not currently offer certainty in terms of both whether generative AI outputs attract IP rights at all, and if they do, who the author and first owner of those rights is.

Why does this matter?

Uncertainty around the IP ownership position of outputs comes into sharp focus for organisations exploring use cases for generative AI within their business. The rubber hits the road in relation to use cases where an organisation has an expectation of exclusive use of the outputs in question. One example of this is the use of generative AI to produce software code that is of value to the organisation, and which they would wish to prevent third parties from replicating. Uncertainty around the IP ownership position means there is uncertainty as to whether the organisation has any such exclusive right.

The uncertainty is reflected in the terms and conditions on which many generative AI systems are made available by providers. Some providers will be unwilling to provide an assignment of any IP rights in the outputs, on the basis that they are unclear as to whether they own any rights capable of assignment. Other providers may take the view that it is possible for their generative AI systems to produce the same or highly similar outputs for different users, and that as such, they cannot assign IP ownership rights to any one user but can provide a non-exclusive licence to use the outputs.

In navigating this uncertainty, organisations will need to ensure in relation to each use case that they are clear on how they intend to use the outputs of the generative AI systems they deploy, and consequently how important it is for them to own IP rights in those outputs.

Written by Richard Danks, Senior Legal Counsel, The Royal Bank of Scotland