It is no secret that AI is everywhere these days. It’s certainly all over the news, although the smart phone in your pocket is also packed full of face and voice recognition technologies that would have been considered AI just a few years ago as well. The current cutting edge isn’t in recognising faces, though, it’s in the realm of human creativity: art, music and the written word. Modern generative AI services (think ChatGPT, DALL-E, Midjourney and the like) have a slightly worrying capacity to create art of all sorts and are already giving educators a headache by being able to churn out plausible essays on almost any topic. There are legitimate economic, social, political and technological concerns to unpack here, but these are best left to economists, sociologists and philosophers. Instead, what concerns us today are the legal issues: who owns the output of a generative AI? Is it the person who told it what to do, or the person who made it in the first place?  

Interestingly, the owners of these AI services themselves haven’t yet come to a firm conclusion on the question. Some, such as Midjourney, assert that they own all the assets you create with their platform (but with the proviso that you can become the owner if you are a paid subscriber). Others, such as ChatGPT, punt on the issue by assigning to you any ownership that they may or may not have to the outputs from their model. And others, such as Stable Diffusion, grant a non-exclusive licence to all comers for the model itself (subject to certain use restrictions) and claim no rights to the output generated by the model. In this rather confused environment, anyone who wants to use a generative AI service would be well advised to hunt down and read the associated terms and conditions. 

With the AI companies seemingly in disagreement on ownership, what does the law itself say? Here the trend in international law is also somewhat confusing, with a lot of the discussion occurring in light of US copyright laws and practices that are not broadly shared by other countries. Indeed, one of these US cases (the “monkey selfie” case) seems to indicate that non-human authorship may be an automatic no-go for copyright in the US, at least for now. However, in other jurisdictions the situation is more nuanced, and may consider the work put into setting up a system to produce an artwork even when the system then runs autonomously. And in the case of South African law, our legislature and courts may have inadvertently been prescient in dealing with the issue all the way back in the early 1990s.  

This is because of the Copyright Amendment Act 125 of 1992 which, among other amendments inserted a provision governing the definition of authorship into our Copyright Act. Specifically, the author of a “literary, dramatic, musical or artistic work or computer program which is computer-generated means the person by whom the arrangements necessary for the creation of the work were undertaken”, as opposed to the ordinary conceptions of authorship which normally relate to that kind of work. So not only is a valid copyright to a work assumed, but our Copyright Act (as amended) provides some guidelines as to who the author should be when the work is computer-generated as opposed to being a human creation. 

The meaning of “computer-generated” has also been parsed out by our courts on at least two occasions (Payen components SA Ltd v Bovic CC and others, and Haupt v Brewers Marketing Intelligence (Pty) Ltd), with the consensus being that there is a distinction between “computer-assisted” works (where the computer acts as a tool to assist the artist, and ordinary considerations of authorship apply) and “computer-generated” works in terms of our Copyrights Act. On contemplation, our courts have also decided that computer-generated works must be created with little to no human intervention at all – the newer case law being stricter in this regard than the older.  

The upshot of this for users of generative AI services, at least in South Africa, is that the more detailed the input they provide, the more likely that they (instead of the owners of the service) would be considered the authors of the copyrightable work so produced. The use of techniques such as prompt engineering (where careful construction of input prompts is used to create a desired output) may also help make this more likely by pushing the generation of copyright further away from merely “pressing a button”. 

As always, knowledge of the law and its application in the real world can assist creators in managing ownership of their creations. For this we of course recommend the services of a qualified and experienced IP attorney to help you in understanding the legal effects of the AI that now seems to be everywhere.