AI is forcing transformational change on every newsroom, but what will be its enduring impact?
Will it be measured in better efficiency, or an entirely new form of journalism and business?
That’s the challenge Sophie Huet as lead for AI at Agence France-Presse must navigate.
She joins us on Scotch & Watch to share her insights as she manages helter-skelter change in one of the largest newsrooms in the world.
She says it’s already helping filter intelligence from more than 150 to help her and her team know where to assign editorial teams and resources.
But it’s also asking fundamental questions, such as whether it will be newsrooms, consumers, or machines that ultimately end up being AFP’s primary customer.
“These will change the way we work,” she says. “Atomic information. Adaptive, fluid, elastic content.”
She joined Chris Duncan and I for a fascinating chat. Hope you enjoy.
Transcript:
Ricky: I’ve worked closely with AFP throughout my career and spent some with you in Paris in September. It’s a huge news gathering operation.
Sophie: “AFP is one of the big international news agencies. We have 1,700 journalists in 150 covering news in text, photo and video.
“We have people reporting constantly in countries where a lot of other media are not. We were first in Syria when Assad was toppled for example.”
Chris: Agencies are different to mastheads like Le Monde or The Times. Please explain the relationship.
Sophie: “We have 5,000 clients around the world. They are from print, web, TV, radio - all kinds - and we send them news and information via our wire service.
“AFP’s job is news gathering. Our team is on the ground, who know the country where they work, who report exactly what’s going on.
“When there’s breaking news somewhere, we report it to media who don’t have that network.”
Ricky: What I appreciate about AFP is how open-minded it is to new ideas. Is that a deliberate strategy?
Sophie: “Yes. We see journalism as linking humans to other humans from one part of the planet to another part of the planet.
“To achieve that, you need journalism doing its job. To do that, you need to go see these people. That’s what our journalists love to do. Maybe that’s AFP’s identity.
“And we’re developing that for the AI era. We’re enthusiasts for that. Every experiment is possible.”
Chris: Wire services are the unsung heroes. Everyday people don’t realise the depth of information that comes through the wires.
Sophie: “Journalists in every media are doing great jobs with investigations and more, but it’s important to have the breaking news to start this. That’s our job.
“We’re like the emergency room for journalism.”
Ricky: I was eager to get you on the pod because of AFP’s reputation for innovation, and I know you’re thinking hard about AI. Where do you see it fitting into newsgathering?
Sophie: “I was editor-in-chief. Now I’ve been heading AI in the newsroom for a year. We were working on it before that for three years.
“Most news organisations are developing tools to help newsroom productivity. We are too. We have tools for translation, transcription… all extremely useful.
“My role has been to understand the pain points for journalists, where AI can help them focus on adding value, and not to lose time doing non-journalistic tasks.
“We began an AI adoption programme last spring and recruited 12 of our journalists who were already advanced on AI or very interested.
“We gathered them at our Paris HQ to design training modules - created by journalists for journalists who know AFP’s culture.
“We’ve trained 350 journalists, and the plan is to train all the newsrooms and make it mandatory.
“There are also workshops where they can discuss their fears and understand the future value of their work.
“This month we’re implementing more AI within our workflow (and) while we do that, we’re redefining newsroom management, with new governance and evaluation.”
Ricky: Let me break that down. If I’m a reporter on the ground, in a desert somewhere, the benefit is speed and research efficiency. In the newsroom, it’s about automating manual tasks.
Sophie: “It’s helping us augment our journalism and capacity. The reporter on the ground now has real-time language translation for example.
“But AI can also turn data into journalism, because reporters can now do a better job digging in. Journalists who are not data journalists can now get these insights.
“We’re also developing a data journalism department. We used to focus on data for visualisations, but now we are using it to create real-time breaking news.
“We use it to produce news from Ukraine showing areas that have been taken by Russia, or the other way around. That comes directly from data analysis.
“We’re also using AI and data to identify where news is happening, so we get there to report it. Mixing data and on the ground reporting with AI is very interesting.”
Ricky: It’s helping with deploying resources - so somebody gets in a Jeep and off they go. Is it also helping you anticipate where news will break?
Sophie: “This is something AI will help us to do more, and this is where we want to go next.”
Ricky: This makes me incredibly jealous. My happiest days were running newsrooms. Dawn. 300 things to do. Highly caffeinated. Madness, shouting, urgency, danger.
I’m sure you feel the same, so let me ask you this. AI is already changing the way news is produced. Will it change the way news is delivered to consumers?
Sophie: “That’s a big question, and I struggle to answer because the answer doesn’t exist yet. We need to find it.
“It’s a liquid journalism question. What are atoms of information going to look like with AI?
“As a global news agency, the impact will be anticipating the metadata we provide. It can sound boring, but our clients need it.
“It also enables us to adapt the way we send and format our content; short for breaking news, others get longer articles. Will this lead to extreme personalisation?
“Then, does news need to be delivered in a way that a machine is going to read it? Is our future customer a database that needs to be fed, or a human?”
“These are big questions that will change the way we work. We’re working on that over the next month. Atomic information. Adaptive, fluid, elastic content.”
Chris: Wire content has typically been designed to be used by a human to enable broader news coverage.
Do you now end up with two streams? Sending articles but also mass streams of data and metadata?
Sophie: “I’ll supply all the information we have, because I’m not only giving it to a human now. I’m also giving it to a newsroom’s AI interpreter. AI agents need all the information.”
Chris: I hadn’t thought about that, but it changes the supply side.
Sophie: “Humans will want stories for humans. That’s not dead, but we will need to create new feeds for machines.
“It’s always like that with transformation and innovation. You need to service the legacy, while preparing for the future.”
Ricky: Let me share an insight. I became close to your management when I spoke about AI at the Reuters’ AGM in London two years ago.
Sophie: “I was there.”
Ricky: I was pitching Airgap, my concept of pay-per-crawl that’s now becoming fashionable with Cloudflare and Microsoft.
I predicted that we would be licensing content to machines, and agents, but I made a big mistake. I thought that journalists would not want to train AIs.
And my presentation, your team came to correct me. You said the agency’s role was to report news to enable the narrative of news for the world.
For agencies, training AI was a good fit for your reason to exist - you’re raison d’etre in French. It changed my perspective. I hadn’t appreciated that nuance.
Today, I predict that there will be entire news organisations that emerge now with no other role than to train AIs. Look at Reddit.
Chris: Does AFP have a direct-to-consumer product?
Sophie: “AFP is a newsroom product. We also provide fact-checking for Meta and some platforms, and corporate companies.
Chris: OK. Ricky and I have discussed how OpenAI should be doing deals with every national newspaper right now.
Talking to you though, why would it not go right over the top of them and do a deal with you and other wire services? I assume you’re getting a lot of calls.
Sophie: “AFP has obviously had talks with AI companies, but we have not sold our content for training.
“We provide content as data to be ingested by AI systems to provide fact-based answers to public requests.
“But this is extremely interesting and important for the years to come. Who is going to survive in an AI-mediated era?
“Lots of companies gather and provide data, but newsgathering is critical because AI systems still need to be fed facts to end the hallucinations.
“News organisations, including our clients, produce fact-based journalism, especially at the local level, which matters, but what’s the right mix? Who will provide it to AI?
“The challenge for media is to build on their audiences and community. Will that work for everyone? Probably not. What happens if people rely only on ChatGPT?”
“These are big questions, and nobody knows how it ends, but what’s vital is linking and deepening relationships with audiences and maintaining trust.”
Chris: There’s a perception that AI is neutral, but it isn’t. Models must be trained and guided. That’s critical in breaking news which means models must weight sources.
Models that overweight opinion, rumour, and disinformation are dangerous to society. I’m keen to see how that plays out.
Sophie: “Sam Altman’s latest statement about opening ChatGPT to erotic content suggests fact-based content isn’t the priority.
“There’s also a big issue with data privacy - the intimate, private questions that people ask chatbots without understanding what happens to that data.
“What we saw with social media will be amplified. How these companies use this huge amount of data… So far, we have no idea.”
Ricky: When the internet began to emerge in newsrooms - maybe 1998 - I remember there was an order to how innovations rolled out.
News distribution started very early. So did erotica. Then there was sponsorship, advertising, then retail, but AI feels different from previous technology leaps.
It’s faster than the pivot from print to digital, and web to mobile, and direct sales to programmatic, and to social. AI is a whole new digital.
That makes me think that AI companies are going to start to buy large media orgs to turbocharge their growth and training.
That was a prediction Chris and I made on stage in Madrid, and being provocative, OpenAI could buy The New York Times for the price of a toilet in its data centres.
We’re fortunate though that some media ownership is not reliant on shareholder-focused boards. How is AFP funded?
Sophie: “We have a mission of general interest. The French state contributes funding, but two-thirds of our revenue comes from clients.”
Chris: And The NY Times has the Sulzbergers. News Corp has the Murdochs. These owners don’t need another dollar.
Ricky: But is selling bad if AI’s the emerging platform? Why is AI owning media a bad idea?
Chris: Should a few huge US companies - propping up the world economy with enormous political influence - control journalism to deliver returns on CAPEX?
No. Great newsrooms have a mission to be sustainably profitable while informing the world.
That’s why many countries limit foreign ownership of media, and police interference in that. Each nation must control its media environment.
Selling out to an unproven, oligopolistic tech owner isn’t right. Sometimes you need to do the right thing, not seek the fastest return.
Ricky: Beautifully delivered. And you were the MD of The Times of London, and CEO of Bauer. You know that I agree, but I wanted to test the hypothesis.
Chris: It’s also different because of speed.
In 2004, the innovators were startups that had to earn adoption. Now, the biggest AI players are already the biggest digital players.
Their investments have instant guaranteed reach. Google Gemini doesn’t need to fight because it’s injected into default search.
It also removes consumer choice. In 2007, people chose to visit Facebook. Now AI is injected into habitual interactions without a voluntary choice. That’s why it’s different.
Ricky: That describes antitrust. You can inject your project into your existing monopoly and be certain to win.
Another test for monopoly is that you can make the product worse and still grow.
Chris: And Google just had its first $100 billion quarter, after being convicted three times as a monopolist and telling the court it was under intense AI competition.
Ricky: Are you optimistic Sophie?
Sophie: “You must be, but the big challenge is speed. The digital transition took ~30 years so we adjusted. AI is moving far faster, with huge money behind it.
“We must be smart, stick to fundamentals, believe that journalism matters, communicate better what we do, and keep to our core purpose.”
Chris: Tech always finds its place, and people always need trustworthy news. My bet is that AI companies will be pushed towards products that are societally valuable.
Ricky: My prediction is that we’ll face a rough period of disinformation as bad actors exploit a wild west web as tech firms chase growth, leaving others to police the mess.
But then a major event will force a pause and reset. History shows us that’s what happens, but I am most worried about America’s over-reliance on these companies.
Just five companies are 32 per cent of the total S&P index. A failure of any one of them - let alone all - would be catastrophic to the US and the global economy.
I also worry that engineers view the world in binary terms; one or zero, good or bad, profitable or not, while societies and people are shades of grey and highly complex.
Journalism brings nuance and wisdom to that. It can be a vital guide to the builders of these powerful machines to achieve better societal outcomes.
If AI can be trusted because it’s trained by journalism, public doubts will soften. If the scale leads to a breakthrough in cancer because of AI, adoption will grow.
Thanks for joining this latest episode of the Scotch & Watch podcast.
You can get alerts to join in live and ask questions live by downloading the app.
You can also tune in on all the major podcasting channels and YouTube.
