Major deals could signal emerging partnership between the arts and AI

Artists, writers, singers, publishers, and content creators have expressed concern that their copyright is being infringed by generative AI models, for example ChatGPT, Stability AI, and DALL-E. In order to produce the weird and wonderful images and passages of text that such AI models have become famous for, they need to be trained on and taught to recognise patterns in pre-existing data – whether that be prose, photographs, drawings, or songs – which may well be covered by copyright.

AI developers ‘scrape’ this content from thousands of online sources, but they often do so without the rightsholders’ permission and without a licence in place. This could constitute copyright infringement and deprive the rightsholders of valuable revenue streams.

The backlash from content creators has been so great that in February 2023, the UK government rowed back on its proposal to extend the text and data mining (TDM) exception in the 1988 Copyright, Designs and Patents Act, which would have allowed AI developers far broader access to a whole range of copyrighted material without necessarily having to pay any traditional licensing royalties. Such an extension would have limited the ability of rightsholders to control that material and reduced any revenue they might have gained from it.

Content creator deals with OpenAI

However, the tide may now be turning, and a more positive relationship could be emerging between AI developers and content creators. Associated Press (AP), an American non-profit news agency, recently formed a two-year partnership with OpenAI, the company which created ChatGPT, perhaps the most famous generative AI chatbot of them all. The partnership involves AP licensing its text archive, which extends back almost 40 years, to OpenAI.

The deal will provide a wealth of high quality data to OpenAI to allow it to further train its AI tools and improve its product. In return, AP will be given access to OpenAI’s technology and product expertise, helping it leverage its existing AI operations, which include transcribing live audio and video, and automated story previewing. AP is not OpenAI’s only partner, however.

In July 2023, OpenAI finalised a six year deal with Shutterstock, under which the stock images company agreed to license photos, videos, music, and meta-data to OpenAI. Developers will then be able to use the licensed content to train DALL-E, its text-to-image tool. Shutterstock’s library of content will prove a rich treasure-trove of training data for OpenAI, once again affording it the opportunity to hone and improve the output of its AI tools.

Similar to the deal with AP, in return Shutterstock was granted “priority access to the latest OpenAI technology” and would “continue to leverage DALL-E’s ground-breaking generative text-to-image capabilities directly into the shutterstock.com platform”. The technology will enable Shutterstock customers to use AI to create new content and streamline the editing of existing content. Shutterstock has also signed deals with NVIDIA, Meta, and LG to develop generative AI tools to be used with 3D images and text.

Continued resistance to AI

It appears, therefore, that an increasing number of organisations are embracing rather than shunning generative AI, as long as they are able to control the terms on which the AI developers use their content. However, other organisations are not so keen to implement generative AI into their operations, echoing the hostile reaction from content creators earlier this year to the expansion of the TDM exception.

In September 2023, Immediate Media, the owner of BBC Good Food, blocked ChatGPT from scraping recipe data from its website, fearing that recipe books and cooking websites could soon disappear if individuals ask AI tools for recipes instead. There are also concerns that AI tools’ ability to produce incorrect and inaccurate information could prompt them to create recipes that are dangerous for users – or which ignore allergens and dietary restrictions.

No unified approach to AI partnerships

These two contrasting approaches from content creators suggest that there is currently no unified approach for organisations deciding how to implement AI into their business operations. The current climate we are experiencing around the fast rise of AI suggests that content creators will have to embrace the technology sooner or later, and one way or another. 

There may well be merit for them in forging a closer relationship with AI developers, as the partnerships between OpenAI, AP and Shutterstock give the latter companies valuable access to cutting-edge AI technology which could transform their business operations, as well as affording them a lucrative opportunity to commercialise and exploit their expansive IP portfolios. In return, AI developers gain legitimate access to a wealth of valuable training data for optimising their AI tools, without running the risk of copyright infringement and potentially expensive lawsuits. Content creators will need to be comfortable with their works being exploited in this way.

UK Intellectual Property Office code of practice

However, there is no one-size fits all solution. Content creators still devising their response to AI may be particularly interested in the code of practice due to be published by the UK Intellectual Property Office. The objective of the code is to strike a balance between the need for AI developers to have broad access to creative works as training data, and the need for content creators to realise the financial benefits of exploiting their works in return for royalty payments.

Although exactly what the code will say is as yet unclear, it may help content creators to devise their AI strategies by helping them understand the extent to which they are able to control use of their works by AI developers and what a best practice relationship with an AI developer might look like. The code is due to be published imminently.

Co-written by Concetta Dalziel of Pinsent Masons.