Generative AI tools are in demand due to their ability to generate content at the press of a button. The tools are able to develop sophisticated visuals or compelling textual works within seconds. They are developing at a rapid rate and continue to cement their place within the creative sector, with brands such as Coca-Cola recently releasing an advert containing AI-generated content.
The AI systems are trained on extensive datasets of text and images. The training data will almost certainly contain works that are protected by copyright and other intellectual property rights. If the EU’s Artificial Intelligence Act is approved, companies will be required to publish summaries of the copyright materials used to train the AI.
Copyright and Trade Mark Infringement
Issues of copyright infringement have been raised around the world. A class action brought by a group of artists in California alleges that the AI systems: Stability AI, Midjourney and Deviant Art, have infringed many thousands of copyright works, both in the input and the output of the AI systems. The claimants argue that their works have been “scraped” and thereby copied by the defendants in order to train their AI systems, without obtaining a licence. The claimants allege that the defendants embedded and stored compressed copies of the training images, and further processed the compressed copies when generating output, in response to text prompts. The claimants’ second argument is that the works created by the AI systems are derivative of the claimants’ works. End users are able to prompt the AI system to create a work “in the style” of an artist, resulting in alleged vicarious copyright infringement. This may be particularly distressing for artists who risk being under-priced and outpaced by AI competition in the creative sector.Getty Images also brought a claim against Stability AI for copyright and trade mark infringement. Getty Images state that without obtaining a licence Stability AI copied over 12 million images from its website, including the metadata describing the images. Getty Images states that it has agreements with other AI companies to license such use: something which may become more common between rightsholders and generative AI companies. Getty Images also claims that the output of Stability AI has infringed its trade marks, as some images produced by Stability AI contain the Getty Images Watermark, which they claim may cause consumer confusion.Opt-Out Requests
Recently, Mr Kneschke, a German photographer, requested that LAION, an AI database, remove his images from its database. LAION’s database is used by companies such as Stability AI for training purposes. LAION responded to the request stating that it does not store any of Mr Kneschke’s images in its database but instead store links to locations where his works are available online. The letter also contained an invoice for €887, stating that Mr Kneschke had made an unjustified claims of copyright infringement. Mr. Kneschke argues that during the collection of the dataset, LAION downloaded his images in order to evaluate them to obtain information. Mr Kneschke claims this collection process amounts to copyright infringement.
Streaming services removed the song “Heart on My Sleeve” at the request of Universal Music Group. The song was created by an AI system trained on copyright content, including songs by the musicians Drake and The Weeknd. The AI system was able to generate a synthetic copy of their voices. Universal has also asked streaming services to block AI systems accessing songs which Universal owns, to prevent AI systems using that content for training purposes. This represents another developing legal battleground between rightsholders and AI companies.
Data Protection Hurdles
AI companies faced further issues regarding their datasets when ChatGPT was temporarily suspended in Italy following a data breach, and concerns over the lack of an age verification mechanism. The Italian Data Protection Authority commented that no information is provided to users or data subjects whose data are collected by the AI companies. They also found no legal basis for the massive collection and processing of data for training purposes. OpenAI confirm that it has addressed or clarified the issues raised by the Italian Data Protection Authority. However, the company could face further challenges following similar investigations in Canada.
The above cases reflect the developing legal landscape surrounding generative AI tools, as different jurisdictions seek to tackle these issues. It is clear that existing legislation was not intended to consider the modern technical processes used by AI tools, and how they may result in infringement. The outcomes of these cases will hopefully help clarify whether AI companies will be required to compensate creators in order to train their AI on copyrighted material. It will also clarify whether creators can opt out of the reproduction of their work, used to train AI systems, and how this can be achieved.
It's understandable that creators would seek to enforce their rights against the AI companies that could force them out of their own market, but it remains to be seen how the Courts will balance their rights against the development of AI technologies.