Social media is abuzz with posts about images, art and writings created by generative AI. They inspire awe and at times also a sense of apprehension. The companies that own these AI models are valued at billions of dollars. Open AI that is behind ChatGPT is valued at nearly 90 billion USD, while Microsoft, its partner has a market capitalization of nearly 3.1 trillion dollars.

The AI models of these companies have been trained on data – whether text, images, videos or music. However, the issue being raised is that no permission was taken from the authors of this data before training the models on it. The work that the AI models train on is made by humans, and in most cases that creative human work is protected by copyright laws. OpenAI’s disclosures in the past show that ChatGPT 3 was trained on sources, including from the internet, which comprised large amounts of copyrighted data. It is being alleged that generative AI models were built with stolen goods.

This has led to a spate of litigation against the companies that own the AI models. Leading newspapers are suing OpenAI and Microsoft, a large music label company is suing Anthropic for using its song lyrics without permission and one of the biggest web image libraries is suing Stability AI for using its stock of images.  The tech companies deny the allegations. The argument being that much like humans absorb material already in existence to create something new, AI too trains on available material to create something different.

While legal battles continue, the changing world is trying to adapt to the challenge thrown by technology. Many companies are licensing their data to generative AI companies for their models to train on. Many owners of copyrighted material are trying to build their own AI tools that will train on internal data. Big media houses and broadcasters like BBC are exploring these options.

This on-going tug of war between right holders and owners of generative AI models is also throwing up questions of the adequacy of the copyright laws as they exist. Will copyright extend to AI made content? In India in 2020, the copyright office rejected an application which listed AI (RAGHAV) as the sole author for an artwork. However, a second application was filed where a natural person and an AI (RAGHAV) were named as co-authors, and the registration was granted. Subsequently a withdrawal notice of the said registration was also issued. The next question that arises is who owns the content that AI creates. ChatGPT’s terms and conditions stipulate that while the platform initially owns the content generated, ownership is transferred to the user whose prompts led to content generation. If ChatGPT produces the same output for multiple users, neither can claim rights to it due to the nature of machine learning. However, the legal framework as it exists does not clearly answer such questions.

Every area of life is getting transformed due to generative AI, and the issues being thrown up are complex. The debates, litigations and discussions will settle eventually, once laws and notions of ownership and creativity transform in human lives. 

Published in Dainik Bhaskar in May 2024