Copyright in the Age of AI

The rapid development of artificial intelligence (AI), and generative AI in particular, has created a maze of novel copyright issues. These issues primarily concern the way in which AI models are being trained and whether or not the output of these models is sufficiently personal as to entitle them to copyright protections. 

Copyrighted Training Material

Developers of AI train models by providing vast amounts of content in order to improve the model’s prediction capabilities. However, much of this content is copyrighted material and it has been argued that training a model on copyrighted material is itself copyright infringement, even if the model does not produce the exact text as part of its output. 

Questions including whether users should be able to prompt these tools with direct reference to copyrighted material (and trademarked works by name) of other creators without their permission have also been raised. There are certainly risks of infringement especially where contracts and terms of use are silent on AI usage. If a business user is aware that training data might include unlicensed works or that AI can generate unauthorised derivative works not covered by relevant exemptions (e.g. fair use), there could be a claim for wilful infringement. 

This is an issue that jurisdictions across the world continue to grapple with, resulting in differing judicial treatment and an elusive regulatory baseline. American computer scientist and author Jaron Lanier offers one possible answer with his idea of data dignity, which distinguishes between training (or “teaching”) a model and generating output using a model. Mr Lainer argues that the former should be a protected activity.

Whilst a solution to this issue is yet to be found, AI developers should ensure that they are in compliance with relevant laws regarding data capture for use in training their models, which may require making arrangements with those who own the intellectual property (IP) rights that developers seek to add to their training data. This could entail licensing or revenue sharing arrangements. The creators (i.e. the customers of AI tools,) should ask providers whether their models were trained with any protected content, review the terms of service and privacy policies and avoid generative AI tools that cannot confirm that their training data is properly licensed. 

We are also likely going to see more content creators and brands that create content looking for their work in compiled datasets or data lakes, including visual elements, such as logos, artwork and textual elements, such as image tags. Content creators and brands creating content are also likely to be monitoring digital and social channels for the appearance of works that may be derived from their own. 

Copyright for AI Generated Works

Whether AI generated works qualify as sufficiently personal creations to be eligible for copyright protection is another topic of discussion which consequently raises questions around the levels of human creativity that went into the prompts that generated the work.

One position is that works created through text prompts and lacking any further creative input by the human user, as in the case of generative AI tools, would not be copyrightable because these prompts function more like instructions to a commissioned artist. In 2023, a court in Washington D.C ruled that only works with human authors can receive copyrights as human authorship is a "bedrock requirement of copyright" based on "centuries of settled understanding." Likewise, in India it was decided that a work must involve a minimum degree of creativity and not be a product of only skill and labour. Therefore, output produced by AI may not satisfy the requirement of ‘‘creativity’’ required for copyright protection, if viewed as a collection of data compiled from already existing sources without any infusion of creativity. In this sense, Indian and US copyright law agree that a certain class of AI-generated works would not qualify for copyright.

Interestingly, in China the Beijing Internet Court’s decision in Li v. Liu makes a distinction between ‘‘straightforward’’ AI generated output where the human author simply takes and uses the output “as is” without any creative involvement and AI generated output where the human author keeps experimenting and adding various prompts, including negative prompts and tech parameters, until they receive a satisfactory result. In the later scenario, the Beijing court determined that such “AI-assisted work” (meaning output where aesthetic choices were exercised and there was personal judgement in the final rendition) would be eligible for copyright protection. 

Future Considerations 

The evolving legal landscape of generative AI has certainly revealed shortcomings in current global copyright regulations. While the questions that this new technology raises are undoubtedly interesting, we are likely to see many more intellectual property infringement cases in the near future.