Robot Reporters Will Write News Tailored to You
AI keeps its readers in mind, even when it's targeting an audience of one.
Image: Lindsey Turner/Filckr
The robots are stealing our jobs.
Well, not your job, perhaps, but mine. As if journalists didn't have enough to worry about with unpaid internships and the slow demise of print media, we must now contend with artificial intelligence.
Quill is AI software designed to produce data-based news stories and corporate reports in seconds. Created by Chicago tech company Narrative Science, it's already used by clients such as Forbes, the US intelligence community, England's NHS, and Credit Suisse, to create financial reports and sports news stories in seconds.
It produces millions of articles for these clients every year. As a writer, I'm a little bit scared.
Bots are already monitoring Wikipedia for stories and even producing listicles for Buzzfeed. But Quill develops a superhuman level of efficiency, taking seconds to condense data into comprehensive articles devoid of spelling errors.
I spoke to Kris Hammond, chief scientist at Narrative Science, over email to find out whether I should give up already, go back to college and learn coding.
"The reality is that Quill is going to be everywhere. Wherever there is a spreadsheet that people are struggling to understand, Quill can provide the explanation of what is happening in that data to help people make informed decisions," he said.
Its uses are myriad and versatile: It can crunch figures for financial news outlets or create personal reports for individuals, depending on the data its fed.
Quill is forever diplomatic with the truth
"Do you want Fitbit numbers, or would you prefer the story of your health and fitness in a form that you can read and use to guide your actions? You need to know about crime in your neighborhood, but don't know how to interpret the numbers and graphs? You could have Quill explain what is happening, how your neighborhood relates to others in your city—all in a second," said Hammond.
To me, it seems the need for Quill was born out of the internet itself: our short attention spans, our thirst for micro-updates via the Facebook ticker and tweets. Quill even caters to Twitter users with Quill Connect, which can compile a report for you on your own followers and their interests, letting you know how to connect with them more efficiently. It is a very competent, very clinical take on the "optimisation" of one's own social media presence: a kind of "executive summary" of the self.
The idea is not simply to produce objective writing, but to produce efficient, functional journalism which is tailored to the reader. Quill factors the audience into what it produces, customising its work down to a micro level. Though a Quill article is robotically simple and free from opinion, its composition is determined by its audience, whether they are left-wing or right-wing, young, old, supporters of the winning or the losing team… Quill is forever diplomatic with the truth.
"The power of Quill is its ability to generate stories from large data sets for specific audiences," Hammond elaborated, "including an audience of one. While it works well in the arena of media where it can provide breadth of coverage, it shines brightest in those instances where it can generate specialized stories for many instances of small audiences."
What Quill can improve is the very same news that journalists hate writing: dry figures and statistics
It pre-empts the sentiment of the reader in its cold, robotic way: a story is compelling not due to the quality of the prose, but the relevance, timeliness and context of the article.
Hammond alluded to Quill's early use in sports journalism such as GameChanger, a hybrid news app which provides a constant stream of statistics on any sports team. GameChanger uses Quill to produce over four million "stories" per year, and even the relatively straightforward narration of goals scored and penalties called can be artificially humanized.
"A baseball story for the home team is going to look different than the same story for the visiting team, because each side is interested in their own players and how they are performing." The same principle applies to finance: "If a company has suddenly lost value, for example, it will be uninteresting to an investor who has no stock in that company."
One thing which strikes me is that what Quill can improve is the very same news that journalists hate writing: dry figures and statistics, aggregated into frequently turgid prose.
The hype around Data Journalism exemplified by Nate Silver and his ilk has come and gone, and those figures and data visualisations remain for the most part as boring as ever.
I asked Hammond's opinion on news stories which make a feature of data. "The problem with data-driven journalism is that it doesn't scale," he said. "A journalist with a national data set about crime, government services or education can write a couple of stories: a national story about trends and a local story if they live in, for example, New York City, Los Angeles or Chicago. But there isn't anyone who is going to write the stories for each of the small towns and cities in the country, because it costs too much and too few people will read the story, making it cost prohibitive."
The use of outsourced, cheaper labour has been exposed in recent years with reports on "hyperlocal" news writers in the Philippines. These companies kill off jobs at small-scale publications in favour of a slower, human version of Quill's newswriting on autopilot.
AI offers no recourse for those unemployed local news writers, but it does propose an apparently flawless replacement. Hammond cited as an example their work with ProPublica, a not-for-profit guilty of "data-hoarding:" They had amassed information on the American school system but had not put it to use. "We configured Quill to write over 40,000 stories about individual high schools, assessing what they could provide under the economic conditions that surrounded them and how they compared to similar and other schools in nearby districts."
I asked him whether they test articles on readers to see if they can tell that they were written by bot authors. "The reality is that we have never had anyone approach us and complain that our stories read like they have been written by machine,"he said. "Because they don't."