News Stay informed about the latest enterprise technology news and product updates.

'Robot writers' adept at natural language generation, but they're no Melville

Robots aren’t taking over the world (just yet, anyway), but the idea that robots could take over the world is certainly taking over the conversation among techies. Earlier this week, I published a story out of the MIT Sloan CIO Symposium titled Automated systems: Dehumanizing the workplace. The story featured comments from a distinguished panel, including Robbie Allen, CEO at Automated Insights (Ai).

You may not be familiar with the company name, but you’re familiar with its work if you read the news. The Associated Press, for one, started using Ai’s “robot writers” to craft earnings reports, many of which are placed directly on the wire. Ai doesn’t get a byline, but the robot writers do get credit. At the bottom of an Ai story, a tagline states that the piece was generated by Automated Insights.

My story prompted James Kotecki, manager of media and public relations at Ai, to get in touch. (Kotecki, by the way, has a 21st century media story all his own to tell.) The “dehumanizing the workplace” headline caught his eye, he said in an email, because he had recently published his thoughts on how a company like Ai (and I’m quoting from his piece, here) “makes the world a better place to be a human.”

He offered an opportunity to chat, and I took him up on it. While some of the conversation veered toward the philosophical (an easy trap to fall into when talking about robot writers), the more relevant aspects of the conversation for CIOs had to do with the technology.

Ai’s main product is its Wordsmith platform, patented technology that specializes in “natural language generation.” Kotecki described it as “almost the inverse” of natural language processing, a technology at the heart of IBM Watson.

The Wordsmith platform takes data (such as business intelligence data, or, specifically for marketers, Google AdWords and Google Analytics data, and even personal fitness data), analyzes it for trends, measures it against the aggregate as a way to add historical context, and then turns that data into a narrative — that is, into natural language.

If businesses want to program in a specific tone, Ai can do that too. Yahoo Fantasy Football uses Ai to generate individual weekly reports for participants. Last year, according to a press release, Yahoo requested more snark — and got it — to the apparent delight of Fantasy Football fans, who shared plenty of Ai’s comments on Twitter, according to Kotecki.

“We know we’re engaging with people in a way that just a raw set of numbers and charts would definitely not be able to,” Kotecki said.

But, as Kotecki was quick to point out, machines can’t do it all — and likely won’t be able to. “A lot of folks ask us, ‘Can you write fiction? Can you write a novel,'” he said. “And the answer is, typically, we rely on structured data to do what we do and there’s a configuration process that works with the output for any given client.” When it comes to fiction, what kind of data is necessary for the Wordsmith platform to produce the next great American novel?  “I don’t even know how you’d start,” Kotecki said.

Even more to the point for journalists, Kotecki believes augmentation is a better, stronger alternative than automation. “Humans are, frankly, much, much better at [adding context] than machines. And machines are better at doing that data processing and number crunching than humans,” Kotecki said.

He pointed to a recent AP story about the Disney Corp. (like this one), which includes more contextual information than your typical earnings report. At the bottom of the story, the tagline disclosed that only elements of the story were generated by Ai. “That’s usually the tell-tall sign that humans have worked on it as well,” Kotecki said.