Robot Journalists Are Closer Than You Think

We all know robots are stealing our jobs doing physical stuff—like manufacturing, nursing, driving, yadda yadda. But as the semantic web comes out of infancy, it’s becoming possible to automate not only actions, but insight. Thinking. Creativity! Which means it may only be a matter of time until bots take over information industries, like media, too.

Using automation to find and read online content is nothing new—news aggregators and loathsome content farms are old hat—but bots that actually create the content is a more chilling ballgame. One such company, Narrative Science, just got an influx of cash from the CIA’s venture firm, In-Q-Tel, All Things D reported today. The government is interested in the startup’s artificial intelligence technology, Quill, which is able to extract human-like insight from data and use it to create content.

If Narrative Science doesn’t sound like a news media company, that’s because it isn’t—yet. Right now the bulk of their work, aside from finance reporting, is in sales, generating insights for brands and content for their customers. Moving into the traditional news space is likely; when the startup first came on the scene it generated sports news stories by analyzing sports stats, and talked a big talk about being the future of journalism. Then-CEO Kristian Hammond—whose background is in artificial intelligence—bragged that it would only take five years for a computer to win the Pulitzer Prize. How long until newsbots replace newsboys entirely?

It's coming on quicker than you may think. Today there’s more information on the web than ever before—ridiculous amounts of data—and machines are smarter than ever at analyzing it. The semantic web opens up the flood gates for automation. Google's making advancements here, expanding its Knowledge Graph to add explanation, not just aggregation, to search results. By extracting context and insight from a better-linked web, Google wants to make search into a conversation.

Combine that with the untapped potential of big data and it's not hard to imagine how newsbots could be useful for journalists. Think: data visualizations, explainers. The optimists among us might even say it could free up humans' time for valuable things like investigative reporting. The pessimists assume that it's all going to turn into automated churnalism.

Up until recently, news sites have relied crowdsourcing to dissect, collect and analyze large amounts of information like legislation or maps. The guys behind Rap Genius, for instance, recently launched News Genius, which publishes news articles with crowdsourced annotations to give deeper context.

Total automation may not be far behind; at some point, the line between hundreds of human brains and a machine blurs. Netflix successfully leveraged data from its millions of users to decide to create House of Cards. Summly, the news summary app that Yahoo forked over $30 millions for, uses bots to find news stories and automatically generate brief summaries.

For now though, automation’s effectiveness is limited. In the human-algorithm spectrum, most of the news industry falls somewhere in the middle—relying on “hybrid intelligence," as a recent Wired article put it. Obviously there’s a limit to what a machine can understand. The New York Times engineer behind Times Haiku, an algorithm that finds haikus in the Gray Lady’s articles, points out that “The machine has no aesthetic sense. It can't distinguish between an elegant verse and a plodding one. But, when it does stumble across something beautiful or funny or just a gem of a haiku, human journalists select it and post it on this blog.”

Narrative Science addresses this by employing professional journalists to teach computers how to find an story angle in a mass of data—not unlike how an editor or professor would train a young writer. They look at the computer-retrieved data, add their human analysis, and give this input to the computer, which, like a good student, learns.

Narrative Science's experts also provide the machine with a template for how a story is formed—kind of like mad libs, only instead of “insert adjective” it’s “insert data point.” When you break it down like that, it’s not so different from what reporters are already doing. After all, the humans are the ones writing the algorithms anyway, right? The way the Vancouver Sun’s digital editor Ken Schwencke, who also developed an algorithm to automatically write news stories, figures it, “Whether you write the code that writes the news or you write it yourself, the rules are still the same.”