Generative AI’s copyright conundrum

AI tools which produce realistic content have burst onto the scene over the past 12 months. From the images generated by OpenAI’s DALL-E 2, to Github’s Copilot software programming assistant, these tools produce realistic content on demand with wide-ranging applications.

Their commercial impact is obvious. Microsoft recently agreed to a “multibillion-dollar investment” in OpenAI, the developer behind DALL-E 2 and ChatGPT. Missteps can also be costly, as Alphabet experienced when a mistake by its Bard chatbot in its first public demonstration caused a market reaction that wiped $100bn off its market cap.

To generate realistic content, AI tools need to be trained through machine learning, which involves a vast volume of data. This data, whether text, images or sounds, is likely to be subject to copyright protection. However, the behind-closed-doors nature of AI development poses a headache for copyright owners, who may be unaware that their works are being used in this way.

Enforcement against the operators of AI tools to date has been limited. Getty Images recently took a copyright infringement claim against Stability AI in the UK and US. It was able to assert that millions of its images had been copied without its consent, as the images generated by Stability appeared to include the Getty Images watermark. However, for most rightsholders identifying the use of their works in an AI training dataset will likely be all but impossible.

It may be that market-driven change bridges the gap between content creators and AI operators. Image libraries and other content platforms could act as intermediaries that collect content from individual creators and bundle it into datasets. By granting licences to these datasets, it would allow both AI systems access to large datasets and rightsholders (big and small) to collect royalties for the use of their protected works.

In June 2022, the UK Government announced its intention to introduce an exemption to copyright law. That exemption would have permitted the use by AI systems of copyright-protected works without the consent of the copyright owner. However, it recently announced a U-turn in the wake of a backlash from content creators. In the UK, at least, the copyright conundrum is set to remain.