ANI has claimed that OpenAI is violating its copyright by taking content directly from ANI’s website and using it to train ChatGPT.
New Delhi: OpenAI, the company behind the AI chatbot ChatGPT, has informed the Delhi High Court that it cannot be held responsible for copyright infringement in India. OpenAI’s argument is based on the fact that its Large Language Model (LLM) is trained and its data is stored outside India.
ALSO READ: [ANI vs. OpenAI] OpenAI Challenges Delhi High Court’s Jurisdiction in Copyright Case
The statement was made by Senior Advocate Amit Sibal, who represented OpenAI in the case of. The case is being heard by Justice Amit Bansal.
Amit Sibal told the court that OpenAI’s LLM training and data storage do not happen in India, and therefore, the Indian Copyright Act does not apply to it.
“The training of the LLM does not take place within India and storage of training data also does not take within India and thus the Indian Copyright Act does not extend to it,” he said.
He explained that Indian copyright law applies only within India and does not have any effect outside the country. ANI has alleged that OpenAI is storing its news content and using it to train ChatGPT, but Sibal argued that even if this were true, it would not count as copyright infringement in India.
“Any action that is claimed to be infringement must take place in the four corners of this country. If it doesn’t take place in the four corners of this country, that particular action can’t be infringement. My lords are entitled to decide whether it has taken place in this country or not. If my lords come to the conclusion that it hasn’t taken place in the four corners of this country, then it’s not infringement and that plea would have to be rejected,” Sibal stated.
He further clarified that this argument only applies to the storage of data and that he would address the issue of ChatGPT’s accessibility in India separately.
Sibal pointed out that there is no evidence that OpenAI stores ANI’s data in India. Since the data is stored outside the country, Indian laws do not apply to it.
He also explained that an LLM learns through training, and just because it has “memorized” information, it does not mean that the data is stored permanently.
“There is no expressive use of their (ANI’s) content, no evidence of repository of stored content,” he said.
He also emphasized that ChatGPT’s training is based on a vast collection of data rather than a single source.
Sibal also informed the court that no court in the world has so far found ChatGPT guilty of copyright violation.
“The storage is only an intermediate step toward use of non-expressive elements of text corpus. Copyright Act only protects expression, it does not protection ideas or facts,” he added.
ANI has claimed that OpenAI is violating its copyright by taking content directly from ANI’s website and using it to train ChatGPT. The news agency has also accused OpenAI of scraping content shared with ANI’s subscribers.
ANI argues that OpenAI is using its original content to make profits by improving ChatGPT’s responses to user queries.
Sibal countered ANI’s demand for payment by saying that ChatGPT is trained on “hundreds of millions of works” and does not directly copy ANI’s content.
“The plaintiff says that he is entitled to a license for that expression that is ultimately based on training from linguistic structures and grammar in the public domain and owned by all and derived from millions of persons other than the plaintiff,” he said.
Sibal argued that ChatGPT does not reproduce content exactly as it is and that ANI has not provided proof of regurgitation or memorization.
“There is no regurgitation that has been shown and no memorization that has been shown in this particular case and it won’t be because today regurgitation is actually a failure of the system. It is extremely rare. It is what we don’t want to happen, because the whole purpose of the large learning mode, is to generate new responses of its own, and not to generate other people’s responses, and that is why regurgitation is frowned upon,” he told the Court.
Sibal will continue his arguments on the next hearing date.
The court had earlier issued summons to OpenAI on November 19, 2023. It had also appointed Advocates Adarsh Ramanujan and Dr. Arul George Scaria as amici curiae (friends of the court) to provide legal guidance. They have already presented their arguments.
The court is examining four key legal questions in this case:
- Does OpenAI’s storage of ANI’s news content for training ChatGPT amount to copyright infringement?
- Is OpenAI violating copyright by using ANI’s data to generate responses for users?
- Can OpenAI claim ‘fair use’ under Section 52 of the Copyright Act, 1957?
- Does the Delhi High Court have jurisdiction over this case, considering OpenAI’s servers are located in the USA?
