AI tools can create text and images in a matter of seconds based on large amounts of data that already exists. Can that happen just like that though, and is it permitted? What is the situation regarding copyright and the content produced by AI – is the result protected, or can the output infringe on the rights of others? There is still much to be clarified in this field at the moment. There is also the question of how courts will deal with AI-related copyright cases. Here’s why.
How AI learns: machine learning and text and data mining
Under the AI bonnet of ‘engines’ such as the chatbot (program) ChatGPT, for example, there is a process called machine learning happening. This is a technique whereby computers ‘learn’ from large amounts of data. AI models analyse the data and learn (or rather, are trained) to recognise patterns occurring in data. The European AI Act states that machine learning can be seen as ‘text and data mining’, or TDM.
TDM
TDM is a term derived from other EU legislation, specifically the Digital Single Market (DSM) Directive. That Directive states that works may be used for text and data mining, where there is question of ‘lawfully accessible works’ and where the makers (rightholders) have not made any express reservations regarding that use. Any such reservation must have been made in a clear and appropriate manner.
One exception to that is scientific research. If the lawfully accessible works are used for text and data mining for scientific research, authorisation from the maker (rightholder) is not required.
Does machine learning infringe on copyright?
There is a great deal of discussion about whether machine learning should or should not be permitted. Machine learning is regarded as text and data mining. The data (works) used for machine learning may be protected by copyright. It means that machine learning is in principle permitted where legal text and data mining requirements are met. However, it is not clear what those legal text and data mining requirements mean in practice. It is therefore also unclear whether the data which will be or is used to train AI models meets those requirements. Makers (rightholders) of works that are protected by copyright would therefore be wise to make a specific and appropriate reservation in the future.
Is AI output protected by copyright?
In order for something to be protected under copyright law, a person’s creativity is required in the making of it. AI tools operate based on tasks (prompts) given by users. Those users may give a combination of different prompts. They may also thereafter select part of the AI output, and perhaps also go further and modify it. The question is then whether those actions generate sufficient human creativity. Together, the creative combination of different prompts and the necessary creative selection and modification might, in fact, generate a work that is protected by copyright.
At present it remains unclear what the rules in the EU will look like. Greater clarity will come when EU courts deliver verdicts on the matter. The rules in this regard will likely focus in particular on the extent to which there is sufficient human creative contribution – with AI’s role being therefore that of an assistant – or whether the creative contribution originates in particular from AI.
Can AI output infringe on copyright?
The output generated by AI is based on works that exist. That then also means it is possible the output may infringe on copyright – whether or not by chance. The risk of this happening increases many fold if users control it with prompts, intentionally or not, to make the output look like an existing work. As an example, there is a high risk that the following prompt would generate output that would infringe on copyright: ‘make a new version of ‘Portrait of an Artist (Pool with Two Figures)’ by David Hockney’.
Good to know
It is perhaps useful to know that styles in any case are not protected by copyright. For example, if someone designs a lamp in boho-style, others may also design lamps in boho-style as well. What those others cannot do is incorporate specific, original elements of the original design, because that can then indeed infringe on copyright. That is also true for AI. There still remains much to be clarified on this point also though.
