Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The BBC is making plans to build its own artificial intelligence models, while holding talks over selling access to the broadcaster’s vast archives to Big Tech groups developing the cutting-edge technology.
The UK’s national broadcaster is seeking to use its decades-worth of journalism to “train” a generative AI model — systems that can quickly produce text, audio and code — according to people familiar with the moves. This could then power products, such as tools to help journalists produce stories, that can be used in-house.
The people said the BBC had also held separate talks with tech companies such as Amazon about allowing its content to be used to train general purpose “foundational” models being used by developers to build new applications.
Any deal over selling access to the BBC’s archives could be lucrative. Media groups including Axel Springer and Le Monde have struck multimillion dollar deals with OpenAI to use their journalism to train its AI models, while others such as Thomson Reuters have struck multiple such agreements with AI developers.
The talks come ahead of a strategy presentation next week by the BBC, which is expected to reveal plans for further commercial opportunities as it seeks to diversify its income away from a reliance on the national licence fee.
The development of BBC AI products through its own R&D team would probably involve utilising “open source” — or publicly available — AI models, said a person close to the situation.
The BBC’s content archives are potentially valuable to train AI models given a relatively unbiased stance in much of its output. Data scientists are concerned that training AI on content with either right- or left-leaning bias will produce models that likewise have issues.
However, BBC executives are worried that many AI companies are already in effect using its content to train models despite efforts to deny them access, with much of the BBC’s content freely available online rather than behind a paywall.
Referring to the plans to develop its own AI, Rhodri Davies, the BBC’s director of nations, told a House of Lords committee this month that the broadcaster was looking “at whether we might do that in partnership or unilaterally”.
He added: “We will need to do a cost assessment on it as well. Clearly, the BBC holds almost 80 years of audio and video archive as well as a huge text archive online. We are actively looking at whether that might be an option for us.”
The BBC is already experimenting with the use of AI in its newsroom, although it has set in place rules to ensure that anything published will have active human oversight. For example, the technology could be used to recommend headlines on a story, but the decision rests with the editor or journalist.
The BBC said it “has no agreement with any organisation to use its archive to train their large language models that power generative AI tools for commercial use. We are looking at a wide range of issues such as potential bias within such models and how the BBC either in partnership or unilaterally can address these.”
