Between 2004 and 2008, I worked as a video editor in Hollywood. This description comes with plenty of qualifiers: It wasn’t a job with any artistry or excitement (it was, as I described it to curious/confused parties back then, “industrial editing”). The motto of the company I worked for was “Know Better,” and its logo looked eerily like an Illuminati eye; it sold itself as a way for corporations/PR companies/freelancing-citizens-full-of-vanity to amass knowledge about their place in the business world through the monitoring of media. To do this, it recorded TV and radio broadcasts from around the country, and resold commercial-free chunks of said broadcasts to the as-yet uninformed. Need to see how your opposition is being perceived by CNBC’s Jim Cramer? We got what you need. Want a shot of you catching that foul ball at the game? Give us a call.
When I was first hired, there was an editing staff of twelve. We edited how teens did in the nineties: two VHS decks, lots of trial-and-error. Already, by 2004, the cracks in the company’s model were already evident, as the introduction of live-streaming and TiVo meant ambitious interns could accomplish what we were charging top dollar for. When YouTube hit, you could remove the ambition, and the intern, entirely. If an exec wanted to see how local Topeka news was covering his sex scandal, all he needed was an Internet connection and a few keystrokes. (Since the branch I worked at was based in Hollywood, in the same building where Larry King’s suspenders were filmed on a nightly basis, the most common requests were from movie studios wanting compilations of quips and anecdotes and fluff pieces that aired during the promotional round of their latest blockbusters. We also catered to Ron Jeremy, who’d forgo the custom of splurging for a delivery service and personally schlep into the office, once a month or so, to collect a mixtape of his mentions; personally, I worry for his surely-hermetic state following the invention of Google Alert.)
Midway through my term, the company came to understand its place in the world, and dumped the VHS decks for an expensive digital encoding system. Soon, we were no longer wasting time fast-forwarding through tapes; we wasted time surfing the net as the combination of fewer orders and quicker output times led to six hours of downtime per shift. Among other things, this allowed me to become adept at flicking playing cards into a hat. In 2011, the company filed for bankruptcy. It had outlasted its usefulness, and was summarily dispatched by the cold force of free market evolution. This is how progress works.
I first learned how to organize words into some kind of structure that people might want to read by writing recaps for my fantasy baseball league. For those who haven’t delved into the great wide world of fantasy sports — if you haven’t, you’re quickly becoming a minority — the basic concept is this: Fantasy owners pick real-life players from professional sports teams to be a part of their “fantasy roster.” When those players do something in real-life games, like hit a home run or throw a strike-out, the fake team reaps the benefits.
The average league consists of between eight and twelve “owners”; the specific league I wrote recaps for tended towards the higher end of that spectrum. It also utilized a head-to-head scoring system, meaning that each fantasy team played a different team each week, and whichever one earned the most points won. (This is in direct opposition to the “roto” variety of fantasy sports, where everyone plays everyone else at all times, like a golf or bowling tournament; this is better at determining the actual value of teams because it removes luck as a variable.) Head-to-head is the annoying ingrown hair of an option that’s been grandfathered in by meatheads who think trash-talking their friends is an enjoyable use of time, have literally nothing of substance to talk to their extended family about, or are scared of numbers because they make their heads hurt. This makes it the ideal venue for personalized recaps regarding an owner’s idiotic managerial decisions and, thusly, their presumed genital malfunctions.
Every Monday morning, I’d sit down for three hours and offer my witty take on the previous week’s fake games for the other people in my fantasy league. Each recap ran upwards of two thousand words — at least seven hundred of which were boner jokes and Billy Madison references. I wrote a bunch, edited very little, and, subconsciously, came to an understanding, however small, of how to write. So I was somewhat disturbed when this email from CBSSports.com showed up in my inbox earlier this year:
More Than $50 Bucks Worth had a week to forget in the WHIP category, notching a 1.56 WHIP. That was the highest WHIP of any team this week. Zach McAllister and Edinson Volquez were the captain and first mate on the good ship failboat. Together they produced a 2.36 WHIP in 14 innings.
The group of words, written by some kind of automated program, were one section of a larger recap that looked at how the teams in my current league were performing. It wasn’t half bad: the stats provided accurate analysis — in this case, that the stat “walks and hits per inning pitched,” which is used to determine how effectively a pitcher has thrown, was dominated by the team “More Than $50 Bucks Worth.” (Another aspect of fantasy sports: the ability to name your own team, which leads to plenty of jokes that are so insidery it makes Jimmy Fallon look like Charlie Rose.) And while a quip like “good ship failboat” isn’t the most advanced joke around, it’s quite good for a robot. As someone who makes money turning letters into words into sentences into things beyond that, it’s also unnerving.
CBSSports.com is one of the most popular destinations for owners to host their fantasy league. A host takes care of the automated grunt work of fantasy leagues, like tracking which player is on what team and calculating up-to-the-second stats. There are many, many, many free options available, but CBS gets “millions of players” to spend upwards of a hundred and eighty dollars a year to essentially do the same thing by offering premiums with their product, like the ability to use their server to perform a start-of-season online draft in real time, or their private chat rooms and message boards to “trash talk” your opponents. One of the newer perks is automated recaps.
“We’re taking data and creating millions of stories,” Jonathan Dube, the General Manager of CBSSports.com, told me. “That would be impossible for individual journalists to do.” (Not to mention, boring as hell.) “The idea was to use technology to examine the data and produce effective ‘journalism,’ so you can read about it the same way you’d read about a real sporting event,” he continued. There’s even an option for a fantasy team owner to email answers to questions, which are spit into the algorithm as “quotes,” producing fake interviews. “What we do is take fantasy sports and make it real through storytelling,” Dube said.
That storytelling looks like this, a more recent recap from CBSSports.com’s fantasy football service, which shows the evolving capabilities of the program:
After losing three straight games, Coach Rick Paulas’ squad finally had enough. Oakland Upsides got past Pico Burros, winning 111 to 105.
Oakland Upsides got off the schneid with this win. They now sit at 1–3. The loss drops Pico Burros’ record to 2–2.
Blair Walsh tweeted this week that he “really does like winning”, and helped Coach Paulas to share the feeling. Although Oakland Upsides had plenty of players step up in the win, Walsh was just a bit ahead of the others, putting up 17 points. That was the most that any kicker collected this week.
They’re not only putting my name in there, they’re pulling quotes from Twitter. Couldn’t the algorithms and “interviewing” methods used to create these fantasy reports be used for real games too? I asked Dube. “It’s certainly possible,” he said. But Dube, who once worked as a freelance reporter for the New York Times, doesn’t think that real writers have anything to worry about. “Automated stories could not approach the full creativity or expertise of human writers,” he said. “I don’t envision a day when an algorithm is named Fantasy Sports Writer of the Year.”
In 2011, the computer software company Narrative Science released a program called StatsMonkey. It took simple box scores from sporting events and created accurate and informative prose recaps, without human assistance. Behold, the machines:
UNIVERSITY PARK — An outstanding effort by Willie Argo carried the Illini to an 11–5 victory over the Nittany Lions on Saturday at Medlar Field.
Argo blasted two home runs for Illinois. He went 3–4 in the game with five RBIs and two runs scored.
Illini starter Will Strack struggled, allowing five runs in six innings, but the bullpen allowed only no runs and the offense banged out 17 hits to pick up the slack and secure the victory for the Illini.
StatsMonkey even once indisputably wrote a better recap than a human: In 2011, Will Roberts, a pitcher for the University of Virginia, threw a perfect game against George Washington University, only the eighth such performance in NCAA Division I history since 1957. Yet, the recapper for GWSports.com didn’t mention the twenty-seven-up-twenty-seven-down feat until the penultimate graf of his story. When Deadspin read the recap, it thought the deep interment of the lede was possibly a failure of robo-recapping software. Turns out, it was simply a human trying to get a jump-start on his copy and failing to rewrite after the narrative of the game shifted. Narrative Science took a stab at their own recap by plugging in the game’s box score; their robot mentioned the rare perfect game in the first sentence.
When Kristian Hammond, the co-founder of Narrative Science, was asked by Wired if his company’s tech would make human recappers obsolete, he used the same explanation as Dube to downplay the possibility: The tech’s only intended to perform menial jobs with such low levels of interest they wouldn’t be worth sending out actual human resources. For instance, Little League games:
“Have you ever seen a reporter at a Little League game? That’s the most important thing about us. Nobody has lost a single job because of us.”
While sports — baseball, in particular, where an entire game can be condensed into the heralded space of the box score — certainly lend themselves to a certain kind of journalism-by-raw-number-crunching, they’re not exceptional in that regard; substitute dollars or drone strike casualties or political polls for RBIs and WHIP, and you essentially have many of the kind reports you’d find in Bloomberg, AP, or FiveThirtyEight.
The recapping technology used by CBSSports was developed a few years back with a Chicago-based narrative generator startup, Fantasy Journalist. Earlier this year, it changed its name from the niche moniker to the more sweeping infoSentience, which reflects its newly heightened ambitions to creater a wider ranging artificial intelligence with broader applications. “Our initial goal was to just do a great job creating recaps for fantasy sports,” Steve Wasick, the president of the company, told me. “However, the technology that powers these recaps can be applied elsewhere. The question is: what business and organizations have large amounts of data that they would like to have instantly summarized? We think the answer covers a ton of potential applications.”
In 2013, Narrative Science has released Quill, billed as a way for companies to forgo the confusion/boredom that comes with trying to deliver data to their employees. Sell paper clips? Plug your spreadsheets into Quill, and it’ll regurgitate a narrative report suggesting which markets to ignore and which to focus on, in relatively plain English. An even newer version of the program, Quill Engage, takes data from Google Analytics and outputs a report in understandable prose that tells the webmaster, or whatever we now call the people in charge of websites, how many Kim Kardashian posts did well last month and, therefore, how many should be scheduled this month to hit traffic goals. (Just imagine two tiny clusters of robots: one producing algorithmically generated content and one analyzing traffic, each feeding into other, producing a perfect feedback loop of infinite linkbait. It’s not very hard.)
While algo-writing technology may seem brand new, Narrative Science and infoSentience already have plenty of competition. Automated Insights provides a program that is, according to their site, “like an expert talking with each user in plain English.” (In June, the Associated Press began using the program to produce “a majority” of their U.S. corporate earnings stories.) Unlike other producers of robo-content, who sheepishly suggest human creativity can’t be disrupted, the Dallas-based Yseop bills itself as “disruptive technology” — that is, technology which will displace an existing market. Yseop provides executive summaries, product descriptions, fact sheets, and even biographies based on a person’s LinkedIn account, all written in milliseconds, and in languages other than English, if you’d like.
The standard defense when it comes to algo-writers not taking human jobs — the ones extolled by Dube and Hammond, that machines will never supplant humans because we have consciousness, is great for writers to use to tuck ourselves in at night. But none of us can be entirely sure of the consciousness of even other humans. That the consciousness of a human can’t be proven, nor can the consciousness of a computer be disproven, highlights the falseness of a “conscious” writer having a leg up on an algorithm. Words on a screen don’t have to be created with consciousness, or with expertise and creativity. They just have to display the qualities of expertise and creativity — which is to say, the organization of letters and words and paragraphs in a specific structure in a way that seems like thinking. Computers have always been good at organizing things, and they’re only getting better.
In 1950, mathematician Alan Turing developed his famous test to figure out when artificial intelligence has reached the point of being indistinguishable from human intelligence. It works like so: A handful of judges get together and text-only chat with one human and one computer. They then have to decide which is the human, which is the computer; once a third of the judges are fooled by the machine, legitimate artificial intelligence has been created. A few months ago, a computer finally passed it. Oh, sure, there are all sorts of caveats to it — including that the program pre-declared it was a 13-year-old boy and English was “his” second language, which sets the bar low for judges’ interrogative techniques — but even if you want to hold onto the uniqueness of humanity for now, the test will certainly be defeated within your lifetime.
Maybe this isn’t such a bad thing. There hasn’t been a day since I left my Hollywood editing job when I’ve missed it. Oh, sure, the social aspect was nice, and having healthcare and full dental was pleasant. But the job sucked, even with the occasional Ron Jeremy sighting. Maybe all the robo-writer rise will do is give us human-writers more time to work on more artistic ventures, like lengthy novels, or whatever the hell this piece is. Maybe we’ll be forced to leave the comfort of our home offices and develop legitimate skills with actual earning potential, like first aid or the ability to teach the next generation. Maybe swapping out thousands of “industrial writers” who produce technical documents, or long lists about TV shows from twenty years ago, for those who can actually assist the progress of humanity isn’t the worst thing to happen.
Or maybe our jobs will just shift: If these robo-writers are a digital update on that famed room of infinite monkeys pounding on their infinite typewriters, a human presence will still be needed to sift through the mountainous pile and find the Shakespeare. Update your resumes accordingly.
This article was written by a sex robot.