What are news media really doing with AI?

“AI itself is not a strategy, we had to find how it fit on top of our core strategy”: that was media innovation expert Ezra Eeman’s warning and the underlying theme of the News in the Digital Age session on artificial intelligence. Moderated by Madhumita Murgia, the Financial Times’ Artificial Intelligence Editor and author of a new book on AI, the panel provided first-hand accounts of how publishers are using AI and Large Language Models (LLMs) in their operations as well as the dilemmas that they face in doing so.

Eeman, who is currently Director of Strategy & Innovation at Dutch public broadcaster NPO, opened the discussion by emphasising the importance of grounding AI initiatives in guiding principles that align with an organisation's core values. His team does this through three primary lenses:

• Operational efficiency: focusing on time-consuming and resource-intensive tasks like adding metadata to archives, reformatting images, and creating thumbnails.
• Augmentation: enhancing inclusivity, accessibility, and personalisation of content; for example, making stories accessible in different languages to reach diverse audiences.
• Innovation: experimenting with telling stories in new formats such as the JKF - The Secret Archives, which used AI to recreate the voice of journalist Willem Oltmans.

While audiences have generally been receptive, Eeman noted that there had been some pushback, such as when an image created using a Generative AI tool was used in the 7 o'clock news without full transparency. Labelling audio to let audiences know that it has been synthetically created, he said, was also increasingly important.

Lea Korsgaard, co-founder of Danish outlet Zetland, also explained the backlash that her team faced when they tried using DALL-E to illustrate stories. “I think we had to try it to learn the lesson. The customers were concerned about the relationship [with AI],” she explained. “They want to know that there’s a human being behind [our] words and that there’s [clear] intention”.

That hasn’t stopped Zetland from aggressively experimenting with AI tools in the newsroom. Korsgaard is quick to advise her team to use these tools as a "creativity machine and a really efficient secretary." This fail-fast mentality led to an engineer building a transcription tool that was built out into a fully-fledged product — Good Tape — which now has customers in every country in the world bar North Korea and has been spun off into its own company. It’s popular with Zetland’s journalists too: Korsgaard said that, during a demo of the product, one journalist said “I have now seen God” because of how much time it saved them.

The Financial Times have yet to build any standalone products but it has been carefully infusing AI into its digital products, according to Matthew Garrahan, the FT’s Head of Digital Platforms. As well as using AI to make the comment sections under articles more civil, the product team are rolling out automatic alt text generation functionality in Spark, the FT”s content management system, which should save journalists and editors time as well as improving article SEO.

Garrahan emphasised the importance of the newly created cross-departmental team that works across product, tech and editorial to validate AI use cases. “The testing and robustness is really important when we’re using this stuff,” he explained. “We really need to make sure it’s giving us what we want. It’s slow and methodical”.

All panellists made it very clear that AI is not a one-size-fits-all solution but rather a powerful tool that when used thoughtfully, can enhance both the journalistic process and its end product.