During the U.S. national elections in November 2016, the Washington Post published hundreds of local articles tracking the results.
One article that recapped Republican Darrell Issa’s win over Doug Applegate, a Democrat, for California’s 49th Congressional District seat, read: “Republicans retained control of the House and lost only a handful of seats from their commanding majority, a stunning reversal of fortune after many GOP leaders feared double-digit losses.”
While the verbiage has the ring of a veteran political reporter tracking election results, the article was in fact produced by a robot—Heliograf, the Post’s artificial intelligence (AI) system, built in-house.
A New Team Member
The Washington Post first deployed Heliograf earlier in 2016 in order to more efficiently report on the the Rio Olympics. The intelligent system has since produced hyperlocal recaps of sporting events, such as this football game in Yorktown, New York and, as we saw last November, political coverage.
So, how, exactly, does Heliograf work?
Reporters create narrative templates for stories that include phrases for a variety of potential outcomes. In the case of election reporting, for example, a phrase could read, “Republicans retained control of the House.”
Before an event takes place, the reporter tunes Heliograf in to a certain structured data set (in the case of the election, it was hooked into VoteSmart.org). As the event unfolds, the software works to identify relevant data, marry the data with corresponding phrases, and then publish the story.
And while Heliograf may be one of the more sophisticated examples of AI in the newsroom, it’s part of a long list of how the emerging technology is transforming journalism.
Publishers from the Associated Press (AP) to InStyle have utilized some form of AI to inform the public. In September, AP produced a financial article on stock market forecasts, while earlier in 2015, InStyle produced a video highlighting the beef between Taylor Swift and Nicki Minaj. Each utilized some form of AI.
But what effect is AI having on journalism?
For publishers, it’s about growing audience and output in the most efficient way. For instance, in 2016 alone, Heliograf produced more than 850 articles. It’s election-related stories—of which there were more than 500—garnered more than 500,000 clicks. If that doesn’t sound like a lot, consider this: Heliograf published 85 percent more articles than the paper’s 2012 election-related coverage.
Local Coverage, at Scale
The Associated Press recently partnered with Automated Insights to analyze financial data and produce quarterly reports. To produce AI-driven stories, the AP looks to Automated Insights’ Wordsmith program, an AI-based system that can analyze large swaths of data and produce articles based on that information. The partnership has allowed the AP to scale exponentially—going from about 300 to 3,000 reports per quarter.
“News companies are looking for a sustainable model with all the disruption going on in that space,” Robbie Allen, founder and chairman of Automated Insights, said. “We find we’re playing a key role where newsrooms have to scale up their content production without scaling up their expenses.”
Yet AP’s automated coverage has so far been limited to formulaic types of reporting, like game reports for college sports and minor league baseball.
“At this point, automation and augmentation — in effect, AI for journalism — work in contexts where the work of humans is highly routinized and predictable, and where data is easily and consistently quantifiable,” Seth C. Lewis, the Shirley Papé Chair in Emerging Media at the University of Oregon, said.
“Finance and sports are the most obvious candidates for automated journalism because there are so many numbers behind the narratives,” Lewis explained. “In those instances, it’s quite clear that the so-called ‘robot reporter’ — really, just software — is good enough to produce reports that are almost indistinguishable from those produced by humans.”
And while so far it’s been mostly larger news organizations utilizing AI, it could potentially have the largest impact on smaller, more local outlets—or at the very least, expand local coverage. For example, Urbs Media, a UK-based company, recently brokered a Google-backed deal with Press Associated (the UK’s leading news agency), to generate 30,000 localized news reports every month.
“We started writing national news stories and selling them to national press in the UK, but in dealing with open data about London, we realized that much of it was segmented to a localized level,” Gary Rogers, co-founder and editor-in-chief of Urbs Media—which has published stories in The Times, Daily Mail, The Mirror, and The Sun—said. “London is divided into 33 areas, so with most data sets we could be writing not one London story but 33 more localized ones.”
For instance, on Urbs’ concept site, you’ll find a story about the estimates on the growth of diabetes over the next 20 years, which has 33 localized versions.
“It’s quite clear that the so-called ‘robot reporter’ — really, just software — is good enough to produce reports that are almost indistinguishable from those produced by humans.”
—Seth C. Lewis, Shirley Papé Chair in Emerging Media, University of Oregon
Working in Tandem
At Urbs Media, the AI system is able to cover any topic that contains structured data—for starters: health, crime, transportation, education, environment, social policy, demographics, and lifestyle. Urbs Media has written stories on the performance of every hospital in England—about 150 versions. They’ve looked at the car ownership statistics for each local government area of the UK, helping track diesel car ownership—a growing environmental concern among citizens. They even examined the location of every bike theft in England over 12 months to identify a series of local stories at street level.
Yet, despite the dystopian trope of “AI stealing jobs,” leaders in AI journalism assert that journalists need not worry. Even those publications and technologies that are perfecting AI-driven reporting rely on the human touch.“We’re talking about the most mundane and routinized stories, not the kind of features and human-interest stories that, almost by definition, require creation by humans,” Lewis assured.Rogers and his team at Urbs Media, agree.
“Urbs is still based solidly on the human journalist working with the automation, not being replaced by it,” said Rogers. “It is the journalist who identifies the data sets that might be interesting and who has the contextual knowledge to find the story and pursue the best angle. So far machines aren’t so good at that.”
“It is the journalist who identifies the data sets that might be interesting and who has the contextual knowledge to find the story and pursue the best angle. So far machines aren’t so good at that.”
—Gary Rogers, Co-Founder and Editor-in-Chief, Urbs Media