Generative AI and copyright: The story so far

The use of artificial intelligence (AI) to generate images, text, video and music is causing quite a stir in the creative and legal world. Valerie Annan provides an overview of recent generative AI and copyright disputes and asks what might happen next. 

Generative AI is a type of artificial intelligence (AI) that uses models or algorithms to create content such as images, texts, videos and music. These models are trained using huge amounts of data, enabling them to “create” new content without the need for human intervention, other than an initial prompt. This content could include anything from the design of a new chair to a song in the style of Billie Holiday  

Generative AI and copyright: Trouble brewing 

The recent buzz around generative AI has generated several legal concerns; in particular, when it comes to the relationship between generative AI and copyright. 

Firstly, there is the question of protection for AI-generated texts and images. In principle, only original works created by human authors are eligible for copyright protection, meaning AI-generated texts and images fall outside the scope of copyright law. However, there are nuances to this conversation (read more here). 

Secondly, there is the issue of infringement. Various authors and artists, including Paul Tremblay and Sarah Silverman, have filed lawsuits against different tech companies on the grounds of alleged copyright infringement. In September, a lawsuit was filed by another group of authors, including George RR Martin (whose novels are the inspiration for the TV series Game of Thrones), backed by the American Authors Guild. The authors allege that the AI-generated summaries of their works are so detailed that their works must have illegally been used as training data. This not only infringes their copyright, but it also tampers with their market. 

Generative AI and copyright: What constitutes fair use? 

OpenAI, the organisation behind ChatGPT and DALL-E, has responded to several lawsuits by claiming fair use. For example, in the legal challenge brought by the author Paul Tremblay among others, OpenAI argues that the authors have misinterpreted the scope and limits of copyright law. 

American copyright law has a ‘fair use’ exception, which means that copyrighted works are free to use under specific conditions. OpenAI believes this exception is there to encourage innovation, such as ChatGPT. In its view, the purpose of generative AI models is not financial gain, but rather to save users time. In addition, copyright law is meant to encourage the development of arts and science. That's why underlying ideas, facts and statistical information (eg, word frequencies, syntactic patterns and thematic markers) all fall outside the scope of copyright protection. On the basis that these are the elements that ChatGPT uses, OpenAI filed a motion to dismiss the case. 

As we wait to see how the court will respond to OpenAI’s motion, the company’s argument does highlight another important consideration, namely: how training materials for generative AI models can and should be collected. European copyright law does not have a similar fair use exception, so it will be interesting to see how courts and lawmakers will respond to similar cases in the EU.  

In the meantime, generative AI model providers have already started to seek a middle ground. For example, DALL-E=3 will offer users an opt-out option, enabling authors and artists to stop their works used as training materials. Time will tell, however, whether this is a good enough solution for copyright holders. 

Generative AI and copyright: What’s next? 

Despite the many (legal) questions, AI is continuing to develop rapidly. DeepMind cofounder Mustafa Suleyman recently predicted that the future of AI will no longer be generative, but interactive. Interactive AI is designed to interact with humans and other systems in real-time, by having conversations, assisting with various tasks or controlling autonomous systems.  

As opposed to the ‘static’ nature of generative AI, interactive AI is dynamic and will be able to understand and respond to human emotions. We will likely see interactive AI technology implemented in various industries over the coming years, including media and education. 

Interactive AI is likely to have legal implications too. For example, if an original text or image is produced using interactive AI, will it count as joint copyright ownership? Will interactive AI fall under the same category as generative AI in IP law? Will it be necessary to review the requirement of a human author again?  

Subscribe to our newsletter to keep up to date with generative AI and copyright considerations, including future developments. 

Valerie Annan works at the Knowledge Management department of Novagraaf in Amsterdam.