It’s the summer of bots, after a spring of bots, probably before an autumn of bots. At the Republican and Democratic conventions, news outlets trotted out their new experiments: BuzzFeed’s Facebook messenger bot collected news, The Washington Post’s bot on wheels rolled around live streaming video, and CNN reported the news to teenagers via a Kik bot. Now, at the Rio Olympics, another non-human reporter is making its debut.
Heliograf is the Post’s automated storytelling software, and while it’s been in use since the primaries, the Olympics is its first public appearance. Heliograf takes data, like scores and tallies from sports or elections, and outputs templated stories. “It’s like reverse Mad Libs,” with the data used to fill in the blanks of an already written story, says Jeremy Gilbert, director of strategic initiatives at the Post.
The software was named after the heliograph, a nineteenth-century invention that used sunlight and mirrors to transmit Morse code faster than a telegraph. The Post’s twenty-first century iteration uses data and code to transmit stories nearly in real-time.
Heliograf will pull data from the International Olympic Committee to produce three types of stories during the Olympics: it will report the gold, silver and bronze winners of each competition, the medal tallies by country, and the day’s schedule. Heliograf will feed those stories (“stories” is a rather generous term; notifications would be more apt) to a Twitter bot created especially for the occasion and to the Post’s live blog, where they will be interspersed with live coverage from Post reporters.
“It’s an attempt to take the most mundane part out of the lives of our journalists,” says Gilbert, “and free even more journalists up to do great original reporting that only they can do.”
— Post Olympics (@wpolympicsbot) August 8, 2016
The software has some built-in flexibility to allow for situations that don’t fit the templates created for it. For example, if a score is too far outside the bounds to be possible (even as a world-breaking record) as defined by the developers, an update will be sent to an internal Slack channel, and a reporter will look into it.
Heliograf was designed with election night in mind, and the Olympics are serving as a pilot for the real rollout in November. The idea, according to Gilbert, is to use Heliograf to simultaneously report the close of 500 races taking place on Election Day, and deliver related stories and results to users tailored by location. Accomplishing that will require combining multiple data sources and doing real-time analysis, taking it a step up from from plugging data points into a static story.
Heliograf is hardly the first software to produce automated storytelling. Companies like Automated Insights and Narrative Science have developed their own automated storytelling programs and produce news stories from earnings reports, real estate listings, and box scores for outlets including, most prominently, The Associated Press.
However, while storytelling software continues to develop, it remains limited to very particular types of stories. “It works in domains where there’s going to be a lot of data,” says David Riordan, the Chief Technology Innovation Officer at the Columbia Journalism School’s Brown Institute. “And that data is available to you.” For an event like the Olympics or the general elections, there’s a ton of investment in data infrastructure to get computable data to news outlets, says Riordan.
However, as both the software capabilities and the availability of computable data sources grow, it will become a distinct advantage for a newsroom to have an in-house program such as Heliograf. “If you’ve got the resources to hire a data science team to build a product like this,” says Riordan, “it means you can do things that other papers, and other outlets that are using third-party software, can’t do. You can do the customization that they can’t do.”
This comes at a time that news outlets are experimenting with other kinds of automated storytelling, primarily through messaging apps. There’s the slew of chatbots on Facebook’s Messenger platform since it opened to developers in April; various attempts at figuring out text messaging like Purple, the app that sends election updates via text; and Quartz’s news app, which launched earlier this year and resembles a messaging app.
At the Olympics, the Post will be deploying its own Facebook messenger bot, which launched last month. It allows users to request updates on the same type of data Heliograf will be generating: scores, tallies, and schedules, though it’s hard to see how it’s preferable to simply Googling the information. “It’s a very very basic bot,” but they intend to build it out as the platform matures, says Joey Marburger, director of product at the Post.
The New York Times has launched a two-way text messaging platform for the Olympics. If you opt in, one of the reporters on the ground will send periodic updates and occasionally ask users questions. BuzzFeed’s bot is taking a hiatus during the Olympics after a busy couple weeks at the conventions, says Amanda Hickman, a senior fellow at BuzzFeed’s Open Lab, where the bot was developed.
While the software required to turn data into stories and to have a conversation with a human in a Facebook messenger app are not the same, they are both attempts to reach more readers on more platforms more efficiently (with minimal human labor).
“We look at it more as a tool than a bot,” says Marburger of Heliograf. “It allows reporters, editors, producers to go dig into the bigger stuff, the more original stories.”