The story is familiar enough at this point that I could probably automate it myself:
[Prominent news organization] announced on [day] that it has [contracted with/acquired/partnered with] [tech company] to automate its coverage of [routine news event]. The move is part of an accelerating trend of software-generated journalism, also somewhat misleadingly called “robot journalism,” in which reports once written by humans (if written at all) are composed instead by specially designed algorithms that can produce them almost instantaneously.
In this case, the prominent news website is the Associated Press, the tech company is Automated Insights, and the routine news events are minor league baseball games. The technology comes from Automated Insights’ Wordsmith program, the same one that already writes corporate earnings reports and is expected to begin writing college sports recaps for the AP.*
The popular discussion around “robots writing the news” tends to presuppose that automation will displace human journalists. While that’s certainly possible in some cases, the AP says it hasn’t eliminated any reporting jobs as a result of its deals with Automated Insights. (In fact, it has added at least one: an “automation editor,” hired last year.) Rather, it says the automated earnings stories free journalists to focus on adding detail and context to the more newsworthy reports. And the automated sports recaps have allowed the agency to expand its coverage to events it had previously ignored. For instance, the AP did not previously offer recaps of most minor-league baseball games. Now its recaps will appear not only on its news wire, but on MiLB.com and on the official websites of the relevant minor-league clubs.
So, what does an automated minor-league baseball recap look like? About what you’d expect from a reasonably competent human reporter, minus any telltale signs of abject boredom, self-loathing, or stifled literary ambition.
Here’s one of the samples that Automated Insights provided to me:
Columbia beats Hickory 10-9 after Palsha induces double play
HICKORY, N.C. (AP) – Alex Palsha got Ti’Quan Forbes to hit into a game-ending double play with the bases loaded, leading the Columbia Fireflies to a 10-9 win over the Hickory Crawdads on Tuesday.
Hickory grabbed a 7-5 lead in the sixth after Chuck Moorman hit a two-run double as part of a four-run inning.
Trailing 9-7, the Fireflies took the lead for good in the eighth inning when David Thompson homered to bring home Enmanuel Zabala and J.C. Rodriguez.
Craig Missigman (1-1) picked up the win after he allowed two runs and five hits over two innings. He also struck out one and walked one. Omarlin Lopez (4-3) went two innings, allowing three runs and four hits in the South Atlantic League game. He also struck out two and walked one.
Dash Winningham doubled and singled three times, driving home two runs in the win. Thompson homered and doubled, driving home five runs and scoring a couple.
Ricardo Valencia had four hits, while Yeyson Yrizarri and Dylan Moore recorded three apiece for Hickory in a losing effort.
Read a few, and you’ll quickly gather that Automated Insights’ software is significantly more sophisticated than the Madlibs-style automation algorithm that I half-jokingly suggested in the first sentence of this post. The AP says its human baseball writers and editors worked with the company to customize the Wordsmith platform to conform to its house style for game recaps. And Automated Insights, which was acquired by the sports data firm STATS in 2015, has developed techniques for identifying and highlighting key plays and turning points in a given game just by analyzing the game data. That allows Wordsmith to write anecdotal ledes in some cases, making its stories sound less robotic.
Yet I’m still confident, as I wrote in 2014, that human journalists have little to fear from their mechanistic counterparts, at least for the foreseeable future. Programs like Wordsmith excel at quickly converting data into prose, making them well-suited to stories whose key elements can be quantified in a spreadsheet. (In fact, you can now try Wordsmith yourself, using data from your own Excel or Google Sheets documents.) But, generally speaking, they can’t make their own qualitative observations about the world, let alone pick up the phone and grill a source to get a scoop. Automated Insights’ CEO, Robbie Allen, acknowledged as much in an interview with Poynter in 2015, when his company began partnering with the AP to summarize college baseball games:
For instance, Allen acknowledges a computer won’t report an outfielder lost a flyball in the sun, allowing the winning run to score. “Our stuff is quantitative related,” Allen said. “We’re not able to make a statement about the quality [of a play].”
Perhaps I couldn’t have automated this story after all.
*Correction, July 1, 2016: This post originally misstated that the AP was already automating game recaps for certain college sports. In fact, the AP has not yet followed through yet on its plan to automate those game recaps, which it announced in 2015.
Previously in Slate:
