Large language models provide a wide range of services but the way they synthesise their information may infringe copyright laws.
The advent of large language models, including ChatGPT, which was launched by Open AI in November 2022, is remarkable.
ChatGPT is an AI chatbot system that interacts with users in real-time and in a conversational way. No topic is off limits, and, if it is asked, the bot will happily prepare a rap song about your least favourite colleague, solve complex mathematical problems, help you to plan a party or give advice in all kinds of fields, including the law and medicine.
ChatGPT's feedback, like every office gossip, despite sounding convincing, is often incorrect, misleading and laced with inherent bias.
The tool is trained to create text based on what it has seen. It mines enormous volumes of information from the internet to respond to questions. The process can involve, for example, finding a paragraph of text, deleting a few words and asking the AI to fill in the blanks.
It is this process that raises copyright red flags in many countries around the world, including in South Africa.
What is copyright?
Copyright is a type of intellectual property right that protects the originality of works produced by individuals for a period of time. The rationale behind copyright laws is to provide copyright owners with the ability to exploit and profit from their copyrighted works.
In South Africa, copyright vests in literary, artistic, and musical works, computer programmes, sound recordings, cinematographic films, broadcasts, programme-carrying signals and published editions, provided that the creation of the works meets certain requirements. The general principle in copyright law is that the author of the work typically owns it, although in South Africa there are certain exceptions. Importantly, the Copyright Act 98 of 1978 (Copyright Act) grants the owner of the copyright work the exclusive right to do certain things or prevent the doing of certain things in relation to the copyright work.
The Act contains specific exceptions that can be raised as defences to copyright infringement. The best-known defences are that the copyright work was used for: (i) private use or for the purposes of research or private study; (ii) criticism or review; or (iii) reporting current events. These are typically referred to as the "fair dealing" defences.
What's the catch?
The way that ChatGPT generates written content could potentially infringe upon the copyright of existing works, particularly since OpenAI does not obtain consents or licences from the owners of copyrighted works that it mines for information. Under South African law, copyright infringement could occur if ChatGPT generates content that amounts to a reproduction or an adaptation of existing copyrighted material. This raises an interesting question: whether the users of ChatGPT, in addition to OpenAI, could be liable for damages when their reproduction or adaptation (or further exploitation) of the copyright work falls outside the ambit of the "fair dealing" defence.
Another interesting question arises in the context of who the owner of the copyright in the content generated by ChatGPT would be (assuming that such content is original). As stated above, the model is trained on a vast amount of text from numerous sources, which means it could be argued that the creators of the source material have some claim to the copyright in the ChatGPT generated content. It could, however, also be argued that OpenAI should own the copyright, since their bot generated the content.
ChatGPT and similar AI tools have the potential to fundamentally change the way we access information. The more this technology is refined, rolled out and used by individuals and organisations across the globe, the more important it is to flag the evolving copyright-related considerations, as they are not clear cut. The South African Copyright Act, which is now 45 years old, certainly did not contemplate artificial intelligence technologies when it was drafted. Even though there are proposed amendments to the Copyright Act which should, in theory, bring the law into the 21st century, it will be interesting to see how the legislation, courts and organisations will balance the multitude of competing rights.