AI vs Copyright

It seems that the whole world has been talking about artificial intelligence (AI) over the past few months. All aspects are being discussed and scrutinized in the court of public opinion. Platforms like OpenAI’s ChatGPT in particular have received a lot of air time. ChatGPT is a ‘large language model’ (or LLM) that ‘learns’ the relationship between words by processing huge amounts of text, and can then go on to produce phrases, sentences and paragraphs of text based on that massive volume of ‘learned’ data. This is how ChatGPT can compose songs and write wedding speeches and draft legal contracts, and all the other things you’ve heard about recently.

Whilst AI can process large quantities of data more easily than humans, problems potentially arise because AI lacks human-like judgement or ethical decision making. Problems also potentially arise because ChatGPT’s training data comes from small group surveys where users are asked to rate ChatGPT’s answers to questions. This obviously gives rise to the risk of ChatGPT processing incorrect and possibly biased information.

From the perspective of giving/receiving legal advice/assistance, great care should be taken if intending to rely (even tangentially) on ChatGPT output. ChatGPT frequently ‘hallucinates’ and produces output that is factually inaccurate, ambiguous or lacking in context. This is double problematic because it’s output is usually only moderated through an algorithm. This poses a risk to private individuals who seek legal ‘advice’ from ChatGPT themselves, but it also poses a risk to legal professionals who intend to use ChatGPT or other AI technologies as part of their professional practice (potentially exposing themselves and their clients to legal, ethical and reputational risks).

From a copyright/IP perspective, platforms like ChatGPT potentially also give rise to copyright issues because, as above, the technology learns by processing massive volumes of data from various sources e.g. text from books, websites and other works protected by copyright. This is how the technology learns to structure sentences and paragraphs. However, it’s not clear whether all the responses from ChatGPT are unique and/or whether such output is likely to constitute an infringing work.

Open-AI’s own terms of service and privacy policy explicitly provide that any text/content input or uploaded may be used by the company to improve the service, although it is possible to opt out where such use would give rise to IP/confidentiality issues. However, clearly regulation will be required to address the various risks, as well as the copyright/IP issues mentioned above.

Whilst AI is not new, it certainly seems to have received a new lease of life in 2023, with more than 100 million people now using the ChatGPT. As time goes by, we will learn more and more about the benefits/advantages/efficiencies, as well as the problems/risks/issues, associated with ChatGPT and AI generally. But will law-makers and regulators be able to keep up?

In the meantime, if you have a business, product, service or idea involving AI, and you would like to discuss the potential IP/copyright and/or other legal issues involved, please do not hesitate to get in touch and one of our lawyers will be happy to offer you a consultation.

Written by Éamon Chawke – Partner and Priyanka Vithlani – Work Experience