OpenAI: Gen AI 'Impossible' Without Copyrighted Material

January 9, 2024

Responding to a U.K. Parliament probe on large language models, OpenAI said the company uses three primary datasets for training data for its AI system, including ChatGPT. The three are publicly available information, data licensed from third parties, and information that ChatGPT's human trainers provide.

Since copyright right law today covers "virtually every sort of human expression" such as blog posts, photographs and software code, the company said, "it would be impossible to train AI models without using copyrighted materials."

The company defended its use of copyrighted material, stating that current copyright law does not forbid training data. The company said it has put in place provisions to allow content creators to opt out inclusion of their images in OpenAI's DALL∙E training datasets.

Regardless of these measures, there is "still work to be done to support and empower creators," OpenAI said when probed about the AI copyright issue.

The comment from the company comes as it is fighting a copyright lawsuit filed by The New York Times against OpenAI and close investor Microsoft.

OpenAI on Monday accused the Times of using "intentionally manipulated prompts" to trick its chatbot into producing identical materials published by the newspaper and of "cherry-picking" the content produced by the chatbot to back the publisher's copyright claim.

AI startups Stability AI and Midjourney also face similar copyright lawsuits in the U.K. and in the United States. Last month, the U.K.'s Supreme Court sided with the country's Intellectual Property Office to deny patent rights to an AI-generated idea citing limitations in the existing statute.