For years, the threat of a robot takeover has been kept at bay by the fact that they can’t mimic humans very well. Sure, they can walk and gesture and even make coffee, but their speech is weird and mechanical. Or at least it used to be. Now, software made by Narrative Science is aping human writing so well that publications like Forbes are using it.
The startup, which began as a Northwestern engineering and journalism research project, relies on algorithms to churn through data to create narrative articles. The process is only a few steps: “We ask what the basic facts are, what’s important, and what kind of story we should be writing,” says Narrative Science CTO Kris Hammond. Humans are involved in those tasks, but after that, algorithms take over.
Narrative Science’s first story was about a Northwestern volleyball game. More sports came–straightforward data make them relatively easy to “write”–followed by financial reports, market research, and more. The results are so good, Hammond says, that some clients still can’t believe the sentences came from a machine. Plus, stories come cheap: A 500-word article produced by the software costs just 2 cents a word.
That may be scary for writers, but Stuart Frankel, the company’s CEO, doesn’t think his machine is taking over. Instead, he says, it’s enhancing what reporters do, and allowing them to spend more time writing in-depth stories. He points to stories the company has made, using data input by coaches, about Little League baseball. “That content didn’t exist anywhere before,” he says. The same is true for the point-of-sale-data-turned-to-content Narrative Science is making for a chain of fast-food pizza restaurants (Frankel wouldn’t name names). It’s the kind of data that existed across the entire chain but is only now personalized for individual stores. Plus, Frankel says, “it’s in narrative form, so the information is easily digestible.” Perhaps no writers even want to write pizza-data-as-narrative on the cheap?