For a few months now, the Associated Press has been publishing corporate earnings reports written entirely by computer programs. Except no one’s calling them “computer programs.”
“AP’s ‘robot journalists’ are writing their own stories now,” the Verge wrote last week. “Robot-writing increased AP’s earnings stories by tenfold,” reported Poynter, an organization typically known for its thoughtful approach to journalistic word choice. No less a wordsmith than New York Times media columnist David Carr marveled on Tuesday at seeing an AP report on his own employer’s earnings that was “written by a robot.”
I followed a link and read an earnings story about the NYT ... written by a robot. http://t.co/zgWXMVxYRf Gee.— david carr (@carr2n) February 3, 2015
The robo-reports look a lot like the human-written earnings briefs that the AP and other news wires have been publishing for years, if perhaps a tad drier. They’re now generated almost instantaneously by software provided by a Durham, North Carolina-based startup called Automated Insights, which also provides automated fantasy football recaps for Yahoo and automated car descriptions for Edmunds. The AP announced its deal with Automated Insights last summer, at which time I wrote about it at length. As Poynter notes, the move has allowed the AP to cover 10 times more companies’ earnings than it did before.
If I were a “robot” journalist, I would deftly fill the rest of this post with updated facts and numbers describing the scope and pace of robo-journalism. Instead, my quixotic human brain has led me to focus on a question of semantics: Are the computer programs that are writing the AP’s news stories really robots? And if not, why do even the most careful of writers keep calling them that?
“It’s funny,” said Robbie Allen, Automated Insights’ CEO, in a phone interview. “It seems like every single article that covers what we’re doing has some sort of picture of a robot or robotic hand typing on a computer. I have no idea why. I assume people know that’s not what’s actually happening. I mean, they know it’s all software, right?”
There was a pause on the line, as if Allen were expecting me to reassure him on this point. I, after all, am among the many writers who have cheerfully deployed the term robot in headlines describing products like his. And, yes, I too have illustrated said posts with cheesy stock images of what look like shock troopers seated at computer keyboards.
Of course, I wanted to reassure him. Everyone knows we’re talking about lines of code here, and not a fleet of journalistically inclined Terminators.
Do they, though? I assume most readers are savvy enough to take those stock illustrations in the whimsical spirit intended. But I wouldn’t be surprised, at this point, if the word robot in headline after headline had left a lot of folks genuinely confused. Automated Insights’ own public relations department was concerned enough that it recently published a half-serious explainer video answering the question “Does Automated Insight use robots?” You can see the entertaining answer in the clip below:
There is something intrinsically amusing about the word robot, and still more so about the idea of a Rosie Jetson-type character pulling up a chair in a newsroom and hopping on a corporate earnings call, pen and notebook at the ready. Very soon, though, the novelty of automated journalism is going to wear off. And when it does, I suspect we’ll realize that robot was the wrong term for it all along.
Determining what is and is not a robot is complicated by the absence of a widely agreed-upon, scientific definition. It derives from the Czech word robota, which means work, and was popularized by Karel Capek’s 1920 science-fiction play R.U.R., in which humans manufactured cyborgs to serve as slave labor. (Inevitably, the roboti rebelled and exterminated the human race.)
Today, the Oxford English Dictionary defines robot as “a machine capable of carrying out a complex series of actions automatically, especially one programmable by a computer.” At first glance this would appear a broad enough definition to include journalism software—and all kinds of other software, for that matter. That is, until you notice that the OED further defines machine as, in part, “an apparatus using or applying mechanical power.” For a human, writing a news story might incidentally involve the application of some mechanical power, but not for Automated Insights’ software.
Reference books can’t tell us everything, however, especially when it comes to terms that describe fast-evolving technologies. So I went out and asked some prominent figures in modern robotics for their own definitions of robot, and whether it encompassed newswriting programs. Their definitions differed slightly, but on the second point they were unanimous.
“I don’t think those are robots,” said Chris Anderson, co-founder and CEO of 3D Robotics and former editor-in-chief of Wired. A robot, he said, can sense the world around it. It possesses some form of intelligence, however rudimentary. And it can act on the world around it. So toasters and dishwashers aren’t robots, because their operations are limited to what transpires within their own confines. Autonomous drones and self-driving cars are robots. But newswriting software is not, because it neither directly senses nor acts upon the world around it.
Asked what he would call the programs that write earnings reports and the like, Anderson suggested bots. The word derives from robot but has come to mean something different: Per Wikipedia, it’s “a software application that runs automated tasks over the Internet.” That sounds more like it.
The chief technology officer for iRobot, maker of the Roomba, agreed on the basic criteria for robot-hood. “In my book, a robot needs to have perception, thinking, and action,” Paolo Pirjanian told me. So, I asked, is a newswriting algorithm a robot? “No,” he replied. “It’s a software agent.”
It’s hard to argue with the accuracy of software agent, but it isn’t the catchiest term. To this human journalist’s ear, bot looks more like headline material. But the people who build the software in question aren’t convinced that’s quite right, either.
“To me, bot puts the emphasis slightly in the wrong space,” said Kris Hammond, chief scientist at Narrative Science and a professor of computer science and journalism at Northwestern University. It’s better than robot, he added quickly. But it still reflects an impulse to reify or personify what is really a technological process, not an individual agent.
“If we were writing baseball stories and that’s all we did, I would say, ‘Sure.’ That’s actually a fine domain for personification,” Hammond said. Narrative Science first made its name with software that turned box scores and play-by-play data into narrative recaps of college baseball games. But the company has more recently branched out into a wide array of business-intelligence applications, such as analyzing stock portfolios and generating reports on why they have or have not been performing well. The key to this type of software, Hammond said, is that its workings must be comprehensible to the humans who use it. Bot, to him, implies a sort of technological black box that could go rogue without warning.
OK, I asked Hammond: What would you call the type of product your company builds? “We use the relatively generic term artificial intelligence platform,” he said. “It’s about intelligent situation assessment and explanation. But we don’t have a catchy niche term for it. Maybe we should come up with one.”
Desperate for a term that would fit better in a headline than intelligent situation assessment and explanation, I put the same question to Allen, the Automated Insights CEO. “Say you’re an editor and you see the headline ‘Robot journalists are writing AP stories,’ ” I said. “To what do you change robot journalists?”
“I’d just call it Automated Insights’ Wordsmith platform,” Allen said. “Or, a natural-language generation platform called Wordsmith.”
I sighed. It was almost as if these robotics experts were more concerned with accuracy than with getting bored Facebook users to click on their stories, as we online journalists have been programmed—er, well-trained—to do.
“Isn’t there anything shorter?” I pressed Allen. After some haggling, we compromised on software. Software is writing AP stories. Not exactly the stuff of viral traffic gold, but it’s better than some of the alternatives. That said, the eagle-eyed among you may notice that software is not the word that ultimately found its way into the headline of the story you’re reading right now. (Would you have clicked if it were?)
Allen said he doesn’t know for sure how the “robot journalist” meme got started. But he does have some theories as to why it has stuck. “Everybody needs some sort of image in their story. When they think about what kind of image would be representative of this technology, I guess a robot doing what a human used to do is the closest thing you could find.”
I think both he and Hammond are onto something. All the hoopla about “robots stealing our jobs” has led people to assume that if some new technology performs a task traditionally performed by humans, it must be a robot. But the term carries a lot of pop-cultural baggage that risks clouding our understanding of what we’re really talking about when we talk about automation. The roboti of our imagination are bumbling today, but could turn deadly tomorrow. They’re cute yet alien, endearing yet inscrutable. They have minds and bodies of their own, and maybe lasers too.
The sort of automation that is actually revolutionizing the economy has little to do with Rosie Jetson, the Roomba, or Boston Dynamics’ Big Dog. It’s about software, not hardware. And when it works, it tends to quietly ingratiate itself into our everyday lives, rather than rising up in armed rebellion. If newswriting software is a robot, then so are Facebook’s news feed, Netflix’s recommendation algorithm, and Pandora.
So why don’t headline writers refer to any of those as robots? Anderson has another theory. “Robots are things that don’t work right,” he said. “The moment they work, we call them a vacuum cleaner, or a washing machine, or a flying drone, or something else.”
By that definition, too, it’s time to retire the term robot when we talk about the sort of software that automates the transformation of spreadsheets into prose. It’s clear from the AP earnings reports that it is already working.
In the long run, newswriting software may or may not usurp reporters’ jobs. (I’ve argued before that it probably won’t anytime soon—not the parts of their jobs worth preserving, anyway.) The Associated Press says it has not in fact replaced any of its editorial staff with Automated Insights’ software so far. It has just altered their roles to focus more on the sort of creative journalistic work that software can’t yet perform.
But when and if automation does reshape the news business, as it has reshaped and displaced human labor in so many other fields, thinking of it in terms of robots will leave us ill-prepared to recognize the change. We’ll be waiting for something to roll into the office one day, take our seat, and start typing on our keyboard. And we won’t notice the 50 desks that have already been vacated without any visible sign of a replacement.
This article is part of Future Tense, a collaboration among Arizona State University, New America, and Slate. Future Tense explores the ways emerging technologies affect society, policy, and culture. To read more, visit the Future Tense blog and the Future Tense home page. You can also follow us on Twitter.