The Associated Press (AP) said Thursday that it had reached a deal with ChatGPT-maker OpenAI to license AP’s archive of news stories to help train the artificial intelligence company’s systems.
Under the arrangement, OpenAI licensing will be “part of AP’s text archive, while AP will leverage OpenAI’s technology and product expertise,” AP and OpenAI said in a joint statement.
It was not immediately clear exactly what technology and product expertise The Associated Press will be able to access.
“Both organizations will benefit from each other’s established expertise in their respective industries, and believe in the responsible creation and use of these AI systems,” according to the joint statement.
To train and improve their AI systems known as large language models, OpenAI and other AI technology companies ingest aggregations of written works like news stories, books and social media posts.
AP emphasized the company does not use AI in its news stories and “continues to look closely at standards around generative AI,” according to the joint statement.
OpenAI expressed support for the “vital work of journalism” in the joint statement.
“We’re eager to learn from The Associated Press as they delve into how our AI models can have a positive impact on the news industry,” said Brad Lightcap, chief operating officer at OpenAI. “The AP continues to be an industry leader in the use of AI; their feedback—along with access to their high-quality, factual text archive—will help to improve the capabilities and usefulness of OpenAI’s systems.”
The deal comes amid news of the Federal Trade Commission’s probe into OpenAI over potentially engaging in deceptive privacy practices or harming consumers, The Washington Post reported Thursday.
Congressional leaders on both sides of the aisle have raised concerns over exactly how these AI large language models are being trained and if they are expressing false statements.
Following the Senate’s first-ever classified briefing on artificial intelligence Tuesday, Sen. Elizabeth Warren (D-Mass.) said the AI systems training large language models are not designed for accuracy.
“That creates real threats that AI can be used to sound sensible while it perpetuates one wrong answer after another,” Warren said Tuesday.
Book authors have also become increasingly vocal on having their trove of works used for AI training. Some authors are requesting they be compensated for the use of their works, according to AP. A group of 4,000 writers wrote a letter last month to the CEOs of OpenAI, Google, Microsoft, Meta and other developers about their concerns.
Comedian Sarah Silverman sued OpenAI for copyright infringement this week, claiming she didn’t give permission for OpenAI to ingest a digital version of her 2010 book to train its AI systems. In the lawsuit, the comedian said it was likely stolen from a “shadow library” of pirated works.
In an effort to better understand both the benefits and risks of AI, the Senate Judiciary Committee heard testimony from OpenAI CEO Sam Altman in May. Altman told Congress government intervention will be crucial in lessening the risk of powerful AI systems.
Neither AP nor OpenAI disclosed the price of the deal.
The Associated Press contributed to this report.