The world’s biggest tech companies are in talks with leading media outlets to strike landmark deals on using news content for training in artificial intelligence technology.

OpenAI, Google, Microsoft and Adobe have met with news executives in recent months to discuss copyright issues surrounding their AI products such as text chatbots and image generators, according to a report. Some people are familiar with the negotiations.

These people say publishers including News Corp, Axel Springer, The New York Times and The Guardian have discussed with at least one of the tech companies.

Those involved in the discussions, which are still in the early stages, added that the arrangements could involve media organizations being paid a subscription-style fee for their content to broadcast. develop the technology that underpins chatbots like OpenAI’s ChatGPT and Google’s Bard.

The talks come as media groups have expressed concern about the threat to the industry posed by the rise of AI, as well as concerns about OpenAI and Google’s use of their content without an agreement. Several companies like Stability AI and OpenAI are facing legal action from artists, photo companies and programmers, who allege contract and copyright violations.

Speaking in May at INMA, a news conference, News Corp chief executive Robert Thomson summed up the industry’s outrage, saying “[media’s] Collective intellectual property is at stake and we should argue vigorously for compensation.”

He added that the AI ​​is “designed so that readers will never visit a journalistic website, thus seriously undermining that journalism”.

An agreement will establish a blueprint for news organizations in their dealings with major AI companies around the world.

The Financial Times, which is also discussing the issue, said: “Copyright is an important issue for all publishers. “As a registered business, we need to protect the value of our journalism and our business model. Engaging in a constructive dialogue with relevant companies, like us, is the best way to get there.”

Media executives want to avoid the mistakes of the early internet age, when many offered free online articles that ultimately undermined their business model. Big Tech teams like Google and Facebook then accessed that information to help build multi-billion dollar online advertising businesses.

As the popularity of artificial intelligence (AI) grows, so too must the news industry be concerned with the technology’s ability to produce convincingly human-like text.

Google recently announced a general search function that returns an AI-written infobox above its traditional weblink format. It has launched in the US and is preparing for a worldwide release.

Some of the discussion currently involves trying to find a pricing model for news content to be used as training data for AI models. According to one industry executive, one figure that has been discussed by publishers is $5 million – $20 million a year.

Mathias Döpfner, chief executive officer of Axel Springer, owner of Politico, who has met with leading AI companies such as Google, Microsoft and OpenAI, said his first choice was to create a “determining” model. volume” similar to the model developed by the music industry for radio stations and nightclubs. and streaming services that pay record labels every time a track is played. That would first require AI companies to disclose their media content usage — something they don’t currently do.

Döpfner, whose Berlin-based media company also owns the Bild tabloid and the German wide Die Welt newspaper, says an annual agreement for unlimited use of its content a media company would be the “second best choice”, because that model would be more difficult for small areas. or local news outlets to take advantage of.

“We needed an industry-wide solution,” says Döpfner. “We have to work together on this.”

Google has led negotiations with UK press agencies, meeting with the Guardian and NewsUK. The Alphabet-owned company has longstanding partnerships with many media organizations to use data from content such as articles to ensure it is optimized to appear in search engines. mine. The company used the data to train its large language models, according to two people familiar with the arrangement.

An executive from a newspaper group said: “Google has put a licensing agreement on the table. “They accepted the principle of paying. . . but we haven’t gotten to the point of talking about zeros yet. They’ve acknowledged that there’s a money discussion that we need to have over the next few months, that’s the first step.”

Google will not comment on financial discussions. However, the search company says it is having “ongoing conversations” with news outlets large and small in the US, UK and Europe, and has trained its AI on “information available”. public”, which may include fee-walled sites.

The Silicon Valley giant added another option it’s considering is a way to give publishers more “choice and control” over whether their content becomes part of the web. training dataset for AI or not, similar to how it allows websites to opt out of using their content in search.

Since launching ChatGPT in November, OpenAI director Sam Altman has met with News Corp and the New York Times, according to people familiar with the discussions. The company admits it has held talks with publishers and publishing associations around the world about how they can work together.

According to publishing leaders, developing a financial model for using news content to train AI will be extremely difficult. A senior executive at a major US publisher said the news industry is working retroactively because tech companies have launched these products without consulting them.

“There was no discussion, and so now we have to try to get paid after it happens,” the executive said. “The way they launch these products, in complete secrecy, the fact that there is no transparency whatsoever, no communication before it happens, there is reason to be quite pessimistic.”

Media analyst Claire Enders said the negotiations were “very complicated at the moment”, adding that, as each organization has its own approach, a single commercial agreement for media groups is unlikely and can be counterproductive.

Enders added: “Chatbots would not be a reliable tool if they were trained primarily in the literal sewers of misogyny and racism that make up much of open, accessible text. close by.”

Tech companies that build AI are keen to focus on its utility in boosting efficiency in newsrooms and enhancing journalism, and are willing to pay millions to maintain relationships. long-term with the industry, people involved in the talks said.

It was “in the early days of talking to media and publishers, and part of it was just helping people learn about how models were trained,” said Brad Smith, vice president of Microsoft. “.

“I think our bigger opportunity is actually working with publishers first to think about how they can use AI to generate more revenue,” he added.

Adobe CEO Shantanu Narayen said he has been meeting with Disney, Sky and the UK’s Daily Telegraph over the past few weeks to discuss how they can develop custom models for companies to use. Its general AI for images.

Adobe’s models are trained on images in their own available image library, as well as public domain and open-licensed content when copyright has expired. Narayen said separate deals and pricing will depend on the company, but customers can add their own exclusive content to the tool.

Axel Springer’s Döpfner expressed optimism that agreements will be reached as both media organizations and policymakers have grasped the scale of the challenge faster than with a major wave of tech disruption. past.

AI companies “know that regulation is coming and they fear it,” he said, adding: “It is in everyone’s interest to come up with a solution for a healthy ecosystem. If there is no incentive to create intellectual property, there is nothing to crawl. And artificial intelligence will become artificial stupidity.”