Reuters is building an AI tool to help journalists analyse data, suggest story ideas, and even write some sentences, aiming not to replace reporters but instead augment them with a digital data scientist-cum-copywriting assistant.
Called Lynx Insight, it has been trialled by dozens of journalists since the summer, and will now be rolled out across Reuters newsrooms. Reg Chua, executive editor of editorial operations, data and innovation at Reuters, says the aim is to divvy up editorial work into what machines do best (such as chew through data and spot patterns), and what human editorial staff excel at (such as asking questions, judging importance, understanding context and — presumably — drinking excessive amounts of coffee).
That differs from previous editorial tech efforts that sought to train AI to write entire stories, such as snippets about local sports teams or earthquake warnings. Reuters already tried that with financial stories, and that work has "informed" the new aim to build a "cybernetic newsroom", rather than a fully automated one, says Chua. "The real value is using machines to do what they're good at and then presenting that to humans — that's the best of both worlds."
The system will churn through massive datasets, looking for anything interesting: a fast moving stock price, intriguing changes in a market, or subtler patterns. Journalists are handed that information however they choose — in an email, messenger service, or via their data terminals when they sit down for a shift — alongside key context and background to help jumpstart their research if they think the story is worth pursuing. They can also enter a particular company into the system to get a quick overview, handy for background research and interview preparation.
Reuters is not the first news organisation to turn to machines for a helping hand. The Washington Post has run hundreds of machine-written short snippets using its in-house robot reporter, Heliograf, while the Press Association is working with Urbs Media for natural-language generation for local news with a project called RADAR, or Reporters and Data and Robots, and Yahoo turned to automated writing tool Wordsmith for everything from fantasy football snippets to Game of Thrones recaps. The Associated Press has increased the number of companies it covers by an order of magnitude using automation.
Read more: It's time you turned off all those breaking news push notifications
But it's not all spitting out machine-written stories — others have used AI to boost reporting. In NiemanLab's predictions for this year in journalism, the industry watcher predicts artificial intelligence will "scoop" human reporters, pointing to ProPublica's use of machine learning to analyse what topics US Congress spends its time on and BuzzFeed's algorithm for sifting through flight-tracking data to find spy planes.
Lynx Insight can write sentences, letting journalists drag-and-drop machine-generated copy into their own stories, but it's more about supplying data. In one example shared by Reuters, Lynx Insight would alert a journalist that Walmart shares fell by ten per cent, noting the last time the dropped as fast and why, by pulling in headlines from that day about the company. The demo also showed Lynx Insight offering the names of analysts quoted in previous Walmart stories, and explaining how the company has recently performed against rivals.
There's no question journalists could gather all that information themselves, trawling through databases and skimming quarterly earnings reports, but using machine learning is "going to give them a huge boost down the runway" in terms of research, says Padraic Cassidy, editor for news technology at Reuters. Chua adds: "The goal is it adds value to what they would done, and frees them up to pursue angles that would have taken them longer to get to."
Not all story ideas or data points will be useful, and journalists are able to flag the less insightful suggestions and tell developers what they've missed. That means that as the system learns, it can find more interesting stories.
The aim is to eventually remove the UI layer and have the system pop up suggestions, facts, and contextual background information as the journalist types — yes, it's Word's "Clippy" for journalists. "You appear to be writing a story about Microsoft... " laughs Chua. "I hate to use the Clippy analogy, but that's the idea: you'd get 15 facts on the topic you're writing about. It's on our roadmap."
Right now, Lynx Insights works with Reuters' ocean of financial data, but there are plans afoot to extend it to sports and potentially even the company's extensive legal databases, though it's less clear exactly how that information could be sifted by AI. Any dataset can be pulled in, be it government data for investigative reports or polls on the campaign trail, making it easier to spot why a candidate is trending up or down. "In theory it could be any data set we've got a handle on, that's got a good structure," says Cassidy. "We could pipe that in."
Thanks to metadata, Lynx Insight also works on videos and photos. Pick a bit of text, and it can find a photo that matches to pair up for social posts — it's essentially a financial news meme builder.
Alongside helping journalists do better work more quickly, Chua and his team believe the technology behind Lynx Insights could be used to personalise news stories. That could be for wire services, offering a sports roundup with the local team in the lead for each market, or even for individuals, with a paragraph explaining how a stock swing impacts their own portfolio.
Is this the future of journalism? David Caswell, fellow at the Reynolds Journalism Institute at the University of Missouri and founder of the Structured Stories news platform, believes AI can help relieve newsroom pressures as journalism revenues decline, especially for beats that don't bring in much cash, such as local reporting. "Other big benefits are increased scale, extreme customisation and personalisation, speed and accuracy and the ability to become systemic on reporting certain beats," he says. "There are also longer term benefits resulting from the accumulation of news as data over time that may be even more important."
Read more: How Peter Thiel and Hulk Hogan broke Gawker: UpVote 24
While Reuters is starting with financial data, Caswell argues automation can assist any area of journalism that has repetition, from crime to business deals, award shows to celebrity gossip. "Anything that is formulaic — which is a significant portion of journalistic output," he says. "Essentially the only barrier to all this is the availability of carefully structured data, which machine learning is helping to solve." And structured data is something Reuters, and its parent firm Thomson Reuters, has oodles of.
But we shouldn't lean too heavily on data-driven journalism, and limit reporting to what can be aided by machines. "We need to remember that this is a tool, not a replacement — data and algorithms just can't see, feel or know most of what we see, feel or know, so there may be a risk that we may narrow our view of journalism to only that which we can capture as data," he says. "That would be a grave mistake."
Indeed, Chua hopes Lynx Insight becomes just another tool in a journalist's arsenal, as revolutionary and mundane as the telephone or Google Search. "The goal is not that it becomes some 'gee whiz' new tool, but that it becomes just a part of their working day," he says. "In an ideal world, people wouldn't even notice this — it just works in the background, like a telephone or Google. If it just becomes part of what people do, that'd be a marker of success."
This article was originally published by WIRED UK
