This article wasn’t written by a robot – but the next one you read might have been.
Robots are already in our midst, and you might be surprised what they can do. Robots are reporting on earthquakes, sports, and writing huge numbers of Wikipedia articles. Odds are pretty good you’ve already read articles written entirely by robots without realizing it.
I have such mixed feelings about this. On the one hand, I’m a tech journalist, and AI is awesome . I’m thrilled by the cool and futuristic – I love the idea of machines putting humans in their place and replacing messy meat sacks with cold, ruthless, efficient silicon.
On the other hand, I’m a tech journalist. I write words for money. I like my job. Nobody else is going to let me binge Wikipedia for four hours a day and call it a career. So the idea of facing obsolescence is scary. Right now, the robots aren’t great at what I do. These algorithms can only really do a handful of things. But they aren’t going to stay that way forever.
So, please, hang in there with me while I try to finish this piece before I get replaced by a bash script.
What the Enemy Looks Like
Right now, the enemy comes in the form of two major software libraries: Wordsmith, by Automated Insights, and Narrative Science, a technology created (in part) by Northwestern’s Medill School of Journalism, Media, and Integrated Marketing Communications.
The irony of a journalism school creating the technology to replace itself isn’t lost on me.
The robot, luckily, has yet to figure out how irony works.
Wordsmith is owned by Automated Insights, a company founded by CEO Robbie Allen after he left Cisco in order to work on a project that combined his passion for writing, science, and sports analysis. This project came to be known as StatSheet and was meant to turn data-heavy box scores and sporting stats into readable bits of content with a pre-programmed narrative. According to Allen,
“The traditional approach of hiring a lot of writers wasn’t attractive to me […] What’s exciting about sports recaps is that 90 percent of what you do is write about the numbers.”
This algorithmic approach excels in several specific areas, like creating fantasy football game summaries for millions of Yahoo users or turning box scores into game recaps for the Associated Press (AP).
The StatSheet algorithm did so well that Robbie Allen began to look for other data-heavy verticals in which to use the program. After changing the name to Wordsmith, Allen began to tweak the program to work with data-lead industries, such as finance. The goal was to turn spreadsheets, earnings statements, and – of course – sports scores, into readable bits of content that don’t require human assistance.
Where Automated Insights takes a humble approach to what its program – Wordsmith – is capable of, Narrative Science is decidedly more ambitious. CTO and co-founder Kristian Hammond predicts that computers will be responsible for writing over 90-percent of all news within the next 15 years.
Hubris? Delusions of grandeur? Call it what you want, but as you’re deciding on the wording it’s important to note that some analysts in this industry are predicting that a computer could win the Pulitzer Prize for Journalism within the next 20 years.
Kristian Hammond disagrees.
He thinks it’ll happen within five.
“In five years, a computer program will win a Pulitzer Prize — and I’ll be damned if it’s not our technology.”
How Does a Robot Write Content?
Hunter S. Thompson famously penned novels under the influence of near-lethal doses of whiskey, cocaine, and LSD, but it appears the content writer of the future might only need some shelf space and an outlet.
Before getting into a nuts and bolts approach, it’s important to understand that these programs have limitations. Both of the aforementioned companies and their algorithms are currently only capable of producing data-driven reporting, for areas such as sports and finance. If it’s a quantitative industry driven by numbers, the robot can probably handle it – but you’re not going to be seeing mommy blogs with algorithmic content in the near future.
A Data-Driven Approach
First, the algorithm collects data from high-quality sources. Then, it fits this data into its overall understanding of the subject matter. For example, the most goals in a hockey game wins, or higher earnings than last quarter are good. From there, it can turn this data into content fit for human consumption.
The algorithm relies on a team of meta-writers, which are trained journalists that build templates for the algorithm to work from. These templates are standardized in both form and nuanced use of language in tech-heavy industries, and feature both native vocabulary and industry-specific jargon, which makes the robot sound more human.
Once the robot analyzes the data, it finds ways to insert it into one of these pre-made templates in order to create a narrative that intends to match one of several pre-defined styles (sideline reporter, financial analyst, etc.).
It’s not always correct, but mishaps are becoming rare. Both algorithms mentioned have safeguards in place to alert humans to potential inaccuracies in data before compiling the story. For example, a 1,200-percent increase in earnings by a Fortune 500 company might be a red flag – ditto for a baseball game that ends in a tie.
What’s Next for the Robots?
Both algorithms are good at what they do. In fact, there are dozens like them that can do just about anything, although how well they actually do it is a subject that we’ll continue to debate.
The next logical step is to tweak the algorithms in order to venture into other niches; a step that’s already in the works at Narrative Science. After a rather successful foray into finance and sports, Narrative Science is now moving on to the food realm by creating short listicle-type posts based on reviews written by both humans, and professional critics.
The algorithm was tweaked in order to identify how to pick a quality restaurant. Things such as survey scores from Zagat (and others), notes on service, food quality, and quotes from customers compiled from Yelp, Google Places reviews, and others. This data then makes its way through the templating process in order to create short, digestible list posts such as “The 10 Best Pizza Joints in San Diego” or “The Best Sushi in Dubai.”
From there, the sky is the limit. Right now it’s about venturing into new niches, and finding ways in which to edit the algorithm to work with new data sets. In addition, these algorithms are getting more human as they gain better understanding of the data in which they are responsible for analyzing.
Why I’m Not Switching Careers Just Yet
Robot writers have a lot of upsides: fewer arguments with editors, less reliance on caffeine, less typos or grammatical errors, and even the ability to completely remove opinion from narrative. However, they also have some major flaws that have to be addressed before they’re an actual threat to legit journalists.
Limitations of Algorithmic Writing
These robot journalists excel in areas where data tells the story. This is a problem in most niches, as data isn’t the only story in a lot of cases. Even in one of the areas where these algorithms produce adequate content – baseball – the story is often more than the sum of the numbers from a nine-inning contest.
While the robot is more than capable of telling us who won, as well as identifying and reporting key facts, they aren’t able to recognize the subtle nuances that really tell the story. The bad jump on a fly ball that led to a double, the bunt single that a replay showed to be a foul ball – these are are all details that a human beat reporter relies on to deliver a compelling story. These nuances are why sports fans watch games, but they’re largely qualitative and beyond the scope of modern machine learning.
As a tech writer, my job is safe (for now), simply because – outside of a glorified spec sheet – a robot can’t really tell the whole story. They can’t tell you that a user interface feels clunky, or how certain devices feel when you hold them. They can’t form an opinion using incomplete data, and they certainly can’t check things out in person.
Good writers excel by doing the things that it’s hard to automate. We think on our feet, adapt to changing information, and fill in the gaps using sources, incomplete data, and experience in which to tell a story.
In short, we’re able to think outside of the paint-by-numbers approach of your average algorithm.
The Future of Robo-Journalism: A Cautionary Tale
There’s an idea, in journalism, that after decades of experience writing about a particular topic, a reporter sometimes gets so habituated that they become a sort of human Rolodex. A Rolodex reporter can just change a few particulars around and then write the exact same story they’ve written a thousand times, without a trace of thought or insight. It’s a cautionary tale that rings especially true today.
For content producers that rely on the Rolodex approach; your days are numbered.
Formulaic writing approaches and writing that features heaps of data are going to be the first to go. Robots are better at it than you are, and it won’t take long for this kind of mindless space-filled to be automated away. For the rest of us, the situation is more complicated. On some level, I want to write off the achievements of our new robot overlords as natural language party tricks. However, the futurist in me can’t believe that the tech won’t improve.
Most good journalists are safe – for now – but we shouldn’t forget that these algorithms are still in their infancy. It isn’t completely outside the realm of possibility to assume that they’ll get better, and more capable at picking up the subtle nuances that makes human writers, well, human.
Then there’s the question of quality. The quality isn’t great, but it also isn’t bad. You have to ask yourself: how much does anyone actually care? While there are many of us out there who appreciate good writing, it’s an open question whether the masses are more than happy to settle for “good enough.”
I, for one, am skeptical that robots can do what I do anytime soon. Will I eat my words one day? That remains to be seen. But tell Skynet I’m not going down without a fight.
Do you think robots can replace human journalists in the near future? Sound off in the comments below. I’d love to get your take on this impending clash of man vs. machine.
Image Credits: Human head via Shutterstock, Mr. Robot has some RAM by Chris Isherwood, Content writer by Ritesh Nayak, The FREE HUGS Robot by Ben Husmann all via Flickr, Human Head silhouette with binary codes via Shutterstock