When asked to imagine how major broadcasters might use generative AI in coming months, most people conjure up dystopian visions, from deepfakes of politicians to Minority Report-style ads that follow them around the city. However the reality, at this relatively early stage in the AI game, is far less sinister – and the biggest players are hoping it will prove far more valuable, too.

Take the BBC’s live text pages: a popular if unglamorous service that turns unfolding events, such as breaking news or sports fixtures, into real-time blogs. Automatically translating audio commentary of football matches into text isn’t exactly worthy of an episode of Black Mirror, but it is one of the AI applications the world’s leading public broadcaster is most excited about, as it experiments with a dozen pilots across the organization in an attempt to explore the technology’s potential for the 318 million or so people it reaches weekly across the globe.

“[The audio to text project] opens up so many potential use cases for us to bring our audiences more value through some quite simple things like the reformatting of content,” explains Peter Archer, the BBC’s programme director of generative AI. “That to me is just really exciting. It might sound slightly prosaic, but if you can get that working at scale in a way that mitigates the risks, and still hews to the BBC’s editorial values, it is such a big win.”

Freedom to experiment

Other possible wins being trialed include a chatbot for internal knowledge management, tools for headline generation and translation, and a public-facing BBC Assistant focused on educating kids.

But there is “a much more ambitious story here” too, according to Danijela Horak, head of AI research at BBC R&D. For “a very long time” her team has been fine-tuning an open source Large Language Model using proprietary data, introducing increasingly ambitious tasks as it becomes ever more fluent in the BBC’s workflows, tone and style. “There [are] a bunch of things we’re doing with AI to improve efficiency, save money, improve recommendations, transcribe stuff, search our archives; a whole set of tools that aims to reduce complexity in a way that actually helps people do their job more effectively.”

Staff across editorial, business affairs, legal, and policy teams have also been given individual freedom to explore how ChatGPT, Microsoft Copilot, and Runway might add value to their roles. “This is a whole-organization issue”, Archer says. “You’ve got to let people experiment, you’ve got to let people play. It’s so new, we’re all learning together, we’re ramping this up with as much internal engagement as we can.”

Move steady, don’t break things

So far, so rosy. But as a publicly funded organization with a remit for “impartial, high-quality and distinctive” output, the BBC is also highly conscious of the need to corral employees’ enthusiasm within strict guardrails. Last October, BBC director of nations Rhodri Talfan Davies laid out a set of AI principles emphasizing the importance of transparency, human creativity, and social responsibility, while staff, freelancers, and partners must stick to stringent editorial guidelines.

Archer sees this as an opportunity as much as a constraint. In a landscape where U.K. audiences are spending more time watching BBC TV or iPlayer on average per week, per person, than they spend on Netflix, Disney+, and Amazon Prime combined, he believes that the corporation has a unique chance to lead best practice in the broadcasting space. “Nothing we’re doing at the minute is automation without a human,” he insists. “The license fee is a privilege so we do have a responsibility to how we use new technology. We absolutely have to be on the front foot.”

After all, in the aftermath of the ‘move fast and break things’ era, which surely broke as much as it fixed, surfing the top of the innovation S-curve doesn’t always make sense, for commercial as well as moral reasons. “We have what we call our Gen AI tech stack,” Archer explains. “We’re trying to think about what the architecture of that stack needs to be, which specific capabilities in the organization we need, which LLM models we need, how we make sure we are building in the most consistent way possible and exposing that through APIs to the rest of the organization.

“The opposite is that you’ve suddenly found you’ve got nine different contracts with four different cloud providers, you can’t keep track of cost, someone’s just run up a multimillion pound bill for the use of ChatGPT. So from the start we’re going to build it in a particular way and it may be slightly slower, but that’s okay.”

Frenemies under pressure

Another challenge lies in the dynamic between media organizations and tech companies, a special relationship as fraught as any geopolitical equivalent.

In April, Google launched U.K. trials of AI-generated search, and Pedro Cosa, a senior data expert who has led teams at Channel 4, WarnerMedia, and News UK, considers the tension between discoverability and discretion to be one of the most critical issues broadcasters face.

“The media companies have a very ‘frenemies’ relationship with the technology platforms,” he says. “These days people will often check their news on Facebook or Google or Apple, and what shows up first is what makes you money or gets you clicks. So media companies really depend on these platforms to get audiences, and the platforms rely on them to provide high-quality content, and AI is really making the media companies think harder about how to balance that.”

But perhaps the greatest challenge that the BBC – or any organization – faces when it comes to AI is the scale of the unknowns.

Archer emphasizes that this is a time for leaders to question everything, stay humble, and leave themselves open to being surprised. “This is a long-term fundamental transformation of the media market,” he says. “Just like with the smartphone and the internet, we must both look for the things that provide immediate value, but also the things that stretch us and stretch the industry in terms of our understanding about where this technology could go.”

Of course, for those broadcasters without license payers to answer to, AI may well go places that Auntie might balk at.

Personalized TV shows that adapt to your viewing history and preferences already exist, while LA-based startup Channel 1 plans to stream AI-generated news, presented by multilingual bots, by the end of the year. For football fans across the world, however, those spiced-up text pages may well have a bigger impact on their day-to-day lives. And nobody’s going to lose their job in the process.