Ken Schwencke, a journalist and programmer for the Los Angeles Times, was jolted awake at 6:25 a.m. on Monday by an earthquake. He rolled out of bed and went straight to his computer, where he found a brief story about the quake already written and waiting in the system. He glanced over the text and hit “publish.” And that’s how the LAT became the first media outlet to report on this morning’s temblor. “I think we had it up within three minutes,” Schwencke told me.
If that sounds faster than humanly possible, it probably is. While the post appeared under Schwencke’s byline, the real author was an algorithm called Quakebot that he developed a little over two years ago. Whenever an alert comes in from the U.S. Geological Survey about an earthquake above a certain size threshold, Quakebot is programmed to extract the relevant data from the USGS report and plug it into a pre-written template. The story goes into the LAT’s content management system, where it awaits review and publication by a human editor.
Here’s Monday morning’s initial Quakebot report:
A shallow magnitude 4.7 earthquake was reported Monday morning five miles from Westwood, California, according to the U.S. Geological Survey. The temblor occurred at 6:25 a.m. Pacific time at a depth of 5.0 miles.
According to the USGS, the epicenter was six miles from Beverly Hills, California, seven miles from Universal City, California, seven miles from Santa Monica, California and 348 miles from Sacramento, California. In the past ten days, there have been no earthquakes magnitude 3.0 and greater centered nearby.
This information comes from the USGS Earthquake Notification Service and this post was created by an algorithm written by the author.
Read more about Southern California earthquakes.
Not exactly Pulitzer-worthy—but then, the first news reports on an earthquake never are. The algorithm’s goal, Schwencke says, is not to write a compelling or insightful story. That’s up to the LAT’s human staff. Rather, it’s to “get the basic information out” as quickly and accurately as possible. That way, “Everybody else can go out and find out: Was anybody hurt? Was anything damaged? What do the people at the USGS actually have to say?”
Which is just what the paper’s reporters proceeded to do. By noon Pacific time on Monday, Schwencke told me Quakebot’s post had been updated 71 times by human writers and editors, turning it from the squib above into this in-depth, front-page story.
Quakebot isn’t the first bot of its kind at the LAT. Schwencke and his colleagues on the paper’s data team modeled it on a similar bot that generates automatic reports about homicides in the paper’s coverage area. Again, it’s up to the humans to decide which ones merit further reporting.
Robo-journalism is often hyped as a threat to journalists’ jobs. Schwencke doesn’t see it that way. “The way we use it, it’s supplemental. It saves people a lot of time, and for certain types of stories, it gets the information out there in usually about as good a way as anybody else would. The way I see it is, it doesn’t eliminate anybody’s job as much as it makes everybody’s job more interesting.”
Having spent some years as a local news reporter, I can attest that slapping together brief, factual accounts of things like homicides, earthquakes, and fires is essentially a game of Mad Libs that might as well be done by a machine. If nothing else, a bot seems likely to save beleaguered scribes from scouring the thesaurus for synonyms for “blaze.” (Lacking an ego, Quakebot does not concern itself with elegant variation.) And in the case of earthquakes, an algorithm may actually be better at judging the newsworthiness of a particular small quake than your average gumshoe reporter or editor. Quakebot knows, for instance, that a magnitude less than 3.0 means it’s probably not worth freaking out about, a lesson that over-eager wire reporters don't always grasp.
At the same time, Quakebot neatly illustrates the present limitations of automated journalism. It can’t assess the damage on the ground, can’t interview experts, and can’t discern the relative newsworthiness of various aspects of the story. Schwencke notes that it sometimes generates a report based on a false alert or glitch in the USGS system. (Like many of its human counterparts, Quakebot doesn’t double-check its facts before publishing.) And this might be nitpicking, but only the densest of sentient journalists would see fit to include the detail that the epicenter of Monday’s shaking was “348 miles from Sacramento, California.” Compare the bot's bare-bones brief with the LAT’s updated, human-penned lede:
Seismologists say Monday's magnitude 4.4 temblor near Westwood could mark the beginning of the end for L.A.'s years-long "earthquake drought."
Typically, they would expect a 4.4-sized earthquake about once a year in the Los Angeles Basin, but that hasn't happened for years.
“We don’t know if this is the end of the earthquake drought we’ve had over the last few years, and we won’t know for many months,” said Caltech seismologist Egill Hauksson.
It’s reasonable to expect that robo-journalism will improve over time as companies like Narrative Science refine their algorithms. And it’s remarkable that we’ve already reached a point where LAT readers can expect to encounter the phrase, “This post was written by an algorithm.” Just don’t expect the Quakebots of the world to provide human-level context or news judgment anytime soon, let alone a bon mot or a narrative yarn. It’s true that a lot of newspaper jobs are in danger, but that has nothing to do with news-writing robots. Blame that on a business model punctured by readers’ and advertisers’ move from print to the Internet.
If anything, helpers like Quakebot might save a few journalists’ jobs by freeing them to focus on the type of work that can only be done by a local reporter, on the ground, with a brain.
Previously in Slate: