Finding purpose for AI in the newsroom

News organisations need to link practical applications of AI to their company values.

That was the main takeaway of the News in the Digital Age: A Future with AI event held by FT Strategies and Google News Initiative today.

Artificial intelligence has only just been vaulted into the mainstream, but the news industry has been experimenting with machine learning for years. It is the arrival of generative AI and large language models (LLMs) that is the source of much hype and concern.

Our social media feeds are real-time testimonies of what is possible with this emerging technology: from song covers performed by dead musicians to anime filters on Snapchat. AI can comprehend and create a wide range of content for a wide range of purposes.

Finding purpose in the possibilities

The technology is not just introducing new efficiencies and augmentations, it is getting more advanced with the "right brain" creative functions, says Joseph Teasdale, head of technology, Enders Analysis. But it cannot create it from thin air.

How these models are trained on existing datasets will inevitably continue to surface new legal and ethical questions on ownership and copyright, editorial judgments, transparency guidelines, and so on.

Newsrooms need to have compelling answers to these moving goalposts. Not least because on a day-to-day level, there is little time to run questions up the flagpole, adds Louise Story, journalist, consultant and former senior editorial strategist at the Wall Street Journal and The New York Times.

Those on the shop floor need working AI guidelines that they can use with confidence. That comes from having data specialists and generative AI task forces that can collect, analyse and act upon all company data in real time.

News companies exist in this tech ecosystem whether they like it or not. Is it now up to them to decide what to do about it. They should explore how the technology serves the business and establish the "red lines they are not going to cross", says Lucky Gunasekara, co-founder and CEO, Miso.

Though AI offers publishers plenty of possibilities, more interesting is the reverse prospect, argues Teasdale, referencing a new deal between Associated Press and OpenAI's ChatGPT platform to train the model on news stories.

If LLMs do truly reach the limit of pre-training, then suddenly high-quality, exclusive, factual news content becomes a hot commodity. Teasdale retains some scepticism, as opinion and commentary, archived news and general trending news reporting are unlikely to be of value - and that represents a large chunk of what news organisations bring to the table.

There is little doubt though that artificial intelligence will influence audience behaviour and expectations. As readers want less friction, news products will have to keep up. Gunasekara imagines a world where audiences can immediately email back their favourite newsletters, have greater text-to-speech functionality within podcasts, or find fewer blue links on a Google Search result.

Test, iterate, strategise and repeat, is the advice from Story. But never lose focus on what company values are.

Connecting purpose to applications

For instance, public broadcaster Swedish Radio's mission is to make it easier to access trustworthy information. It has some wider "special" values around promoting diverse voices and on-location stories.

Automation can help to great effect, such as on its podcast recommendations where it has a "public service algorithm" to help audiences get a balanced intake of voices and stories. This works well because people have become accustomed to algorithmic recommendations generally, says Olle Zachrison, head of artificial intelligence, Swedish Radio.

With the zeitgeist shifting to generative AI, the broadcaster has a prototype chatbot that can respond to audience answers from an archive of 2m articles.

Zachrison says that the broadcaster is approaching the technology cautiously because of an error made by a fellow Scandinavian publisher that claimed in an AI-generated summary that the Norwegian footballer Erling Haaland was shot. This was a syntax issue that could have been avoided with more rigorous sub-editing.

Paige Bailey, lead product manager of generative models, Google DeepMind, says the tech platform is also worried about these "hallucinations" - when generative AI can outright make up facts.

It is working on a "retrieval technique" for Bard, its proprietary LLM, where users can go under the hood and see the source of any claim it is making. This operates on a traffic light system where green indicates high levels of confidence, yellow would be more dubious, and red would be a flag for editors to check.

Mediahuis is a Belgian media company that is also experimenting with AI, from personalising paywalls to news articles. It has a whole dashboard that can make recommendations to editors on house style, target audience, invoked emotion and relevant older stories.

To come back to company values, Mediahuis' mission is producing quality journalism for the general public and helping them understand complex stories, says Bram De Ruyck, head of data science and AI, Mediahuis Group.

A criticism of personalisation is that it risks creating filter bubbles. De Ruyck says that Mediahuis is working more on recommending other topics when the AI spots that readers are falling into this trap.

He argues that an algorithm can also recommend more positive news before audiences get too overwhelmed by negative news: "The alternative is they don't read the news, that's the worse option."

Internally, Mediahuis keeps a living document about the company's working protocol on generative AI, but De Ruyck says it is open to change given the quick advancements of this space.

The rule of thumb is that as a suggestion tool, generative AI has a place in the newsroom. But if you want to make the case for any wider adoption, show people how it helps. Or to use Bailey's mantra: "demos are better than memos".

Learn how to get the most from image, text and video creation tools at our next Newsrewired on 15 November at Reuters HQ in London. See the full agenda and grab your tickets now

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).