The latest development in AI (or “deep-learning”) models is ChatGPT, a chatbot whose responses to natural language prompts are so detailed and articulate that anyone who’s used it is probably wondering whether their job is now at risk.
While we’ve been familiar with chatbots for some years now, the impressive thing about ChatGPT is its ability to engage in seemingly human-style conversation, processing large amounts of information to produce (in its own words) “high-quality research output”.
We decided to put it through its paces. At first, we thought of asking it to write an article assessing the opportunities for commercial law firms of using ChatGPT to write articles. But that led on to a more interesting question – if ChatGPT did produce such an article in response to our prompts, whose article would it be and who could use it? What copyrights and other legal rights are at play in generating, adapting and publishing AI-scripted output?
Of course, we could have asked ChatGPT that question instead – and we did. It gave us a very high-level precis of copyright law, in a reply which can best be summarised as “it depends”. (Some might say this sounds all too much like a lawyer’s answer - maybe our jobs really are under threat!) However, to be fair, we had specified that the article should be written in the style of a junior lawyer, so perhaps we shouldn’t have expected an in-depth analysis. Nevertheless, when we asked ChatGPT to try again in the style of a highly experienced practitioner, the results weren’t much better. Perhaps we’d have to do our own dirty work…
To search for a more specific answer, we went right back to basics. The starting point is that copyright protects original creative works, where the threshold of originality is fairly low – as long as a work has not been slavishly copied, it does not need to be particularly novel or unprecedented.
ChatGPT works by stringing together appropriate words, facts and phrases into a plausible answer, based on its corpus of knowledge (essentially, the internet as at 2022). This is an orderly, derivative process of selecting and joining snippets to form an output. The process involves no actual analysis, understanding or thought (indeed ChatGPT has no way of knowing whether its responses are accurate or appropriate). In other words, it could indeed be characterised as “slavish copying” and may therefore not be original enough to attract its own copyright (or to displace any copyrights in the individual snippets used). In any case, under Australian law, copyright can only exist in works authored by a human.
Assuming the output can attract copyright, whose copyright would it be? ChatGPT’s terms and conditions of use gloss over this question and either license or assign their interest in any copyright to the user (while retaining a limited licence over it), and require that the user takes ultimate responsibility for the end result. There’s a good argument that, whatever the copyright status of the original output, by the time we’d carried out all necessary reviews, edits and revisions, the end result would be a new original work which we’d be able to claim as ours anyway. But, depending on the circumstances, it might also belong to OpenAI (the start-up behind ChatGPT), to a third party, or even to no-one. We’re back to where we started – it all depends!
The issue of AI-generated works is extremely topical but one the law is yet to fully address. As of 2022 the Full Court of the Federal Court has decided that an AI can’t be an inventor for patent purposes (a similar decision was also reached in the UK and EU). On the question of copyright, Australian law requires a human author before a work can be protected (and makes no specific mention of computer-generated works, unlike the situation in e.g. UK, NZ and Eire). Both these positions are likely to be challenged as the AI sector grows and expands into new areas, and we can expect further legal consideration of how intellectual property applies to AI.
So, going back to ChatGPT’s original answer – we have to agree with it that the copyright status of any work it produces does, indeed, depend. It’s a complex and knotty problem, and one that will be ultimately down to the courts and legal experts to determine. In the meantime, however, that lack of clarity should not dissuade businesses from investigating this fascinating and useful tool.
“Moulis Legal generated this text in part with GPT-3, OpenAI’s large-scale language-generation model. Upon generating draft language, we reviewed, edited, and revised the language to our own liking and we take ultimate responsibility for the content of this publication.”
This memo presents an overview and commentary of the subject matter. It is not provided in the context of a solicitor-client relationship and no duty of care is assumed or accepted. It does not constitute legal advice.
© Moulis Legal 2023