Maja Pawinska Sims 30 Oct 2024 // 5:20PM GMT
WASHINGTON, DC — A trio of newsroom veterans told PRovokeGlobal this week how their publications are using generative AI tools to free up reporters from the aspects of writing and editing that are “drudgery”, with Politico global chief technology officer Jeremy Bowers saying “robots can’t get scoops, let machines do what they are good at, and humans do what they are best at.”
In a session on the challenges and opportunities of gen AI, led by FleishmanHillard SVP Caitlin Teahan, Bowers said it would have “wild downstream effects” and every part of the organisation would be using gen AI tools in the next five years, but in these early days of its application “we want fast, cheap failure so we can learn from it.”
He said: “We’re using a bunch of interesting tools in the newsrooms, around tagging, taxonomy, spellcheck, style and grammar – all the things that are frustrating, we’re finding ways to sand down those corners.
“Gen AI is our first draft of everything. Going from 0-1 is the hardest step, so we’re using it to degrease the wheels and get this thing rolling. A rough first draft is by far the single most valuable thing that tech has for us now.”
In terms of internal “guardrails” that the news outlets are putting around use of AI, Bowers said: “Early on we felt it would be smart to set up some things that make it easy to experiment and feel more confident and comfortable. For instance, if a reporter is working on a draft of a story, don’t put interview notes directly into ChatGPT, be careful about unpublished content, but here’s an internal version you can use at that stage.
“If we come with a whole bunch of ‘don’t do this, don’t do that’, then no-one will try it out. If people are scared about what they can and can’t do, they won’t use it.”
Nevertheless, Bowers also said there was still no substitute for human testing of AI output: “One of the first tools we wanted to build was summaries of legislation. We’ve made some really great refinements, with additional training steps, that have helped us feel confident in those outputs, but there is no substitute for good old-fashioned testing. We had folks in newsroom and product teams read extracts, and helped familiarise our teams with what it can and can’t do. Ai is not magic and can’t do everything; it took some time and iterations and internal group testing to get it right.”
At Axios, chief operating officer Allison Murphy said the organisation was also taking a human-centred approach to gen AI: “We want to put effort into things that have to be human, like deeply-sourced reporting and expertise. If we can use AI, as long as we are transparent about sources and how content is created, we can bring our influence to more places.”
Like Politico, Axios is playing around with multiple AI tools, she said: “We’re looking at how tools can come into play across the newsroom and opportunities to try something different. It’s very much in spirit of experimentation, without any pressure on being right. Can it make us faster at aggregating news sources, can it help us write the intro or outro? It’s not about tools we can implement everywhere now, but finding the best spaces to experiment.”
And Murphy said that guardrails needed to be part of the learning and adoption process: “Everyone can imagine the worst possible scenarios that could come out of using AI in journalism, but there’s a chance to put in place guardrails and smart steps along the way. There is no way we can anticipate all the things that can go right or wrong. When you have guardrails, then you can have a conversation about what was hard, and create an open, safe space to ask questions and learn so we can be as responsible as possible.”
At the same time as everything is changing, she said, publications needed to stay “crystal clear and focused” on what is most important to the brand: “All the things we’re doing with AI are great, but we still have to – as job number one – make sure our expert reporters are out there, known and trusted.”
At Business Insider, chief technology officer Harry Hope said the “most powerful guardrail of all” was each reporter and editor’s byline: “There is an author attributed to that story, and that’s a powerful thing. A human being stands behind this story and their work, and it’s doubly true now in the world of gen AI; it shows accountability and ownership.”
The trio said that their publications were experimenting with generative AI across a huge range of functions, from enhancing the search experience for readers and audiences, to instant translations of content into different languages, to internal processes such as performance management and training, to image generation.
Discussing whether it was worth news publications building proprietary large language models (LLMs), Murphy said: “For us it’s not completely clear yet what will do the best job for our audience, whether we build an LLM ourselves, work with partners, or give it more time. Keep learning and trying, but don’t feel compelled to immediately do something.”
Bowers added: “We were looking at whether we needed an LLM to help us understand legislation and find out the most important thing for ‘me, my business and my industry’. The simplest tech we brought to the table was the best; the most fancy tech isn’t always the best solution. You have to be really comfortable with the idea of trying things out – the product is working well and doesn’t need all the bells and whistles.”
Hope agreed that organisations probably didn’t need “40 new AI engineers or a VIP of AI”: “The truth is we work with smart, talented people who are great at picking up new things and not afraid to fail and that’s the right mentality. But it’s not always the smartest person in the room who figures out the best solution or the best application of technology, it could be the 23-year-old who’s just joined your company. Everyone has a part to play in exploring tech innovation.”
The panel also talked about how generative AI might impact relationships between communications professionals and newsrooms. Hope said “nothing tech does can supersede the value of direct person-to-person communication between two human beings, and this was echoed by Murphy: “Relationships still matter, and so does newsiness. If you’re pitching a story to a reporter, it still matters that they know who you are. It’s still going to be about who you can get in front of and making it clear that what you have to say is important and meaningful.”
Bower’s closing message for the comms industry was to “use tech for the things tech is good at,” such as first drafts of press releases, network building and maintaining, working out subject matter experts, and testing messaging with virtual target audience groups: “Don’t just use the machine, teach it as you go.”