I learned that robots could do my job just as I was about to graduate from college, in 2010. My university’s computer science department had created a new class for journalists during my senior year, and I had enrolled in a last-ditch effort to acquire marketable skills.
At that point, my classmates and I were used to hearing bad news about our future careers. Most of us had declared journalism majors in 2006, the same year that The Economist wrote, “while newspapers have not yet started to shut down in large numbers, it is only a matter of time.” We expected to be unemployed upon graduation.
So it only seemed fitting when, a few weeks into the class, my professor explained he was working on a software program that could write news articles—sans humans.
Technology had destroyed the media business model by bringing news online, and now, it seemed to us at the time, technology would take the few jobs that remained, too.
This was our early introduction to the pandemonium that has launched an entire panel circuit around the “future of work” and the role that automation may play in making many jobs obsolete. Like manufacturing workers before them, knowledge workers across industries have been learning that new technology can accomplish at least some portion of their jobs faster, more consistently, and with fewer errors than they can. Technology is willing to work around the clock, doesn’t need to be paid, and never gets sick or hurt. The inevitable conclusion is that, as a human, it will be hard to compete.
“Journalists, beware!” read one recent story about a Google-funded software project that aims to automate local news articles. Similar warnings have been given to other knowledge workers: Investment bankers aren’t safe. Auditors aren’t safe. Insurance brokers aren’t safe. Therapists aren’t even safe. Nobody’s job is safe from the machines.
But what’s less clear than this message are the details of exactly how technology might replicate the responsibilities of human beings. To describe the technologies that research has predicted will take over as many as half of all US jobs, the media often uses vague terms like “digital workers,” “artificial intelligence,” “digitization,” and “algorithms.”
Automating a newspaper article sounds like a big deal, but even a couple of months ago, I still didn’t know exactly what it meant. When I asked companies that sell and employ labor-saving technologies to describe them, they often returned vague responses and more buzzwords.
So, a few months ago, I changed the way I approached research into the various robots that purportedly threaten to replace me. Instead of inquiring about the big picture, I asked the consultants who sell automation services to the biggest companies in the world—the people who are deploying all of those contract-scanning, customer service-handling, process-improving machines—the question that I believe drives the sudden fascination with the future of work. What could their technologies do to automate me? And finally, I got some specifics.
The first company I ask to automate my job, a British technology company Blue Prism, says it has installed a “digital workforce” of more than 300 robots at a UK utility company, and that those robots do the work of around 600 people. As I schedule the meeting, I can’t help imagining an army of machines sitting in the same chairs as the humans who they’ve replaced, tapping away at their keyboards.
But though I give Blue Prism CEO Alastair Bathgate a hypothetical unlimited budget for the hypothetical task of replacing me, he doesn’t describe a robot journalist. Instead, he asks, “What do you hate most about your job?”
This is where the mystical quality around the concept of “digital workers” begins to shatter. I tell Bathgate that wading through massively distributed press releases is the worst part of my job. As soon as I delete one from my inbox, another appears. Four more “just following up” emails follow, and sometimes a phone call. Responding and deleting doesn’t add any value to my work – it’s just another task I need to get through before I start my actual job.
Digital workers can’t help me with this. The less exciting name for what companies like Blue Prisim do is called “Robotic Process Automation” (RPA), and even that probably makes it sound too entertaining. RPA involves no robots, in the sense of the mechanical moving things that are fun to watch. Rather, it uses software to run pre-defined steps on computer modules. It’s “like a piano that plays itself,” explains Bathgate. It only works with structured processes that can be written down as a flow chart. It can’t make decisions, such as which press releases to delete, and it can’t handle uncertainty (though really, who among us can).
If you could see a “robot” at work on a computer monitor, it would look like a ghost had developed an interest in re-formatting columns in spreadsheets or moving data from one database to another.
Most of what RPA can actually accomplish revolves around mindless tasks. At the utility business where Blue Prism has installed 300 robots, for instance, those robots have handled jobs like adding contract end dates to invoices, notifying customers that don’t receive regular invoices about emergency contact numbers, and identifying customers whose contracts are going to expire. An online insurance sales unit used Blue Prism’s RPA technology to process insurance applications and apply the monthly charge to the account. A hospital used it to update patient records.
If anything goes off-script—something as small as a button in a computer program moving to a different part of the screen during an update—RPA needs to be adjusted (which is part of why it requires firms like Blue Prism to orchestrate it). It definitely can’t help me sort the good pitches from the bad.
RPA may not be sophisticated enough to be my pitch filter, but it’s not useless, either. “Maybe you want a chat bot,” Bathgate offers. Just like the customer service bot that picks up the phone when you call an airline, my bot could filter PR pitches. “Is your pitch related to these topics? Press 1.” RPA often works with other technologies, like this chatbot, to coordinate between them. It could take the pitches that pass the chat bot’s interview, for instance, and file them into a spreadsheet. Which would be helpful, but still leaves a long way to go before I can call my job automated.
Next I ask Benjamin Pring, the leader of the “future of work center” at Cognizant, a professional services firm with 256,800 employees, to help me find the right technologies to replace me at work. I expect him to refer me to the latest AI technology that I can hook into my RPA system, to employ artificial intelligence that will auto-magically respond to my emails and fact-check my stories. Maybe the technology can even interview my sources for me, and write a draft of a story on my behalf.
Instead, Pring starts almost the same way as Blue Prism’s Bathgate: By asking me for a list of the tasks in my job “that seem most stupid to you—the things where, every time you do them, you think, there’s gotta be a better way to do this.”
To address these problems, he recommends a set of technologies that I already knew about, but have categorized as productivity tools rather than automation. When I explain that part of my job involves keeping constant tabs on the news, including scanning Twitter, he recommends Google News and Apple News, which are automating the collection of news sources. For dealing with email, he recommends x.ai, which makes a virtual assistant that will handle back and forth scheduling for meetings. To use it, I just need to cc “Amy,” who has access to my calendar and can confirm and book an appointment. Companies like Dragon, a transcription service, can transcribe my interviews for me.
“Little examples like these, maybe they’re boring in isolation,” Pring concedes. “But in aggregate, apply that to your job, that’s maybe an hour of your day you have back to you.”
It’s around this time that I realize that my employer is already automating me.
Last November, Quartz won a grant from the Knight Foundation to build something called the “bot studio.” The idea is to create bots that are helpful to journalists, or that relay the news to readers in novel ways. Before this point, I’d thought about the bots as a cool project. I hadn’t linked them to the type of automation that is happening across other industries.
“These projects are not replacing you, they’re enhancing you,” John Keefe, the head of the Quartz bot studio assures me. At least in this case, the promise appears to be true. The bot studio has already built a couple of bots that use our Slack channel as an interface. There’s a bot that will Slack message a reporter screenshots of websites. Another will Slack a reporter if a specific website —say, a company’s SEC filings page—updates in any way, so that the reporter doesn’t need to check for changes as part of a routine.
A reporter who writes stories about airline computer glitches worked with the bot studio to create a bot that alerts him when something unusual happens within a Twitter list he’s created for the purpose of tracking those developments. He no longer has to manually open the list and check it. And after the WannaCry ransomware attacks encrypted the files on computers across the globe, demanding that their owners pay $300 to $600 in Bitcoin to unlock them, another Quartz reporter set up a bot to watch the blockchain transactions and alert him when the attackers cashed out. The rest of us thought he was very smart for setting up this system. We didn’t worry about his job.
The same seems to be true of technology experiments in other newsrooms, where most experiments with “automation” treat technology as an assistant for reporters, not a replacement. Buzzfeed’s Buzzbot, developed to gather information at the Republican National Convention in 2016, runs in a Facebook Messenger window and conducts interviews that reporters can use for tips or crowdsourced stories. Reuters built a technology that assists in the sometime mind-scrambling task of scanning Twitter for breaking news stories. It assigns breaking news stories one score to indicate how newsworthy they are, and another to indicate how likely it is to be authentic. Reporters take it from there.
We call some of these technologies tools and others, like the interviewing bot, automation. But the divide between the two has always been a moving one. In the 1860s, The Metropolitan Record, a newspaper, scolded “very many young women” for fearing the arrival of French’s Conical Washing-Machine. “This machine will lighten the labor, save the hands, and relieve many of the wearing and disagreeable features of hand-washing,” it argued, “but is not designed to, and will not, take the place of a single young woman at service, we feel confident.”
It’s hard to say whether a “single young woman” ever lost her job to the washing machine, but we’re no longer afraid of the prospect. Same for the spreadsheet, which was once considered dangerous, accountant-automating, job-threatening technology. Today the accounting profession is projected to grow faster than the average occupation, and spreadsheets are just a boring piece of office software.
Though rarely discussed, it’s also a possibility that today’s technologies follow a similar path from frightening to dull.
Some robots that have completely taken over the writing of news stories—but they tend to have very uninteresting beats. The Washington Post built a tool called Heliograf that combines databases with editor-created story templates to generate stories. During the 2016 election, it created more than 500 articles about house, state, and gubernatorial races in every single state. The AP partnered with a company called Automated Insights to write thousands of stories about financial quarterly earning reports, and the Los Angeles Times has used bots to write about earthquakes and homicides.
These are stories that, even when written by humans, are already structured like a game of Mad Libs. Most of them wouldn’t be written at all if they weren’t written by robots. For a human with limited capacity, it makes more sense to focus on stories that seem particularly interesting to a wide group of people rather than try to cover every single earnings report or athletic event.
Technology “has the ability to write, but they don’t have ability to think about where it can create human interest,” says David Poole, the CEO of Symphony Ventures, a consulting firm that bills itself as a “future of work” expert. “They don’t know what is interesting to a human, so they will write anything.” In other words, an entire paper written by bots would be very boring.
If I want to apply artificial intelligence to more complicated stories, Poole says, it will probably need to work in conjunction with human intelligence. The LA Times, for instance, used a computer program in an investigation that found police had misclassified an estimated 14,000 serious assaults as minor offenses in an eight year period. The program used data from records the Times had manually sorted into serious and minor categories to learn keywords associated with each, and then applied those to eight years worth of incident data to understand how often those records had been misclassified. Technically, the algorithm automated hundreds of hours of work for the Times reporters, though I don’t think anyone saw it that way. The computer did the grunt work.
Kris Hammond, the professor who taught my computer science for journalists class, doesn’t remember the course exactly the way that I do.
What I remember is a sense of urgency around learning to code. To save ourselves from impending obsolescence, we were trying to become full-fledged software engineers. But the syllabus grew shorter as it became apparent that journalism students, for the most part, had little natural aptitude for coding.
When I reach Hammond in August to ask him about automating my job, however, he says that he never wanted to teach my class how to code in the first place. “The people in the class demanded we teach you how to code,” he says. “We said, ‘Ok, but you’re not going to end up coding in your life. You’ll end up working in areas where being informed of nature of tech is important, but you’re not going to write code.’”
What he wanted to teach us, he says, was how to work together with computer scientists and technology. “We don’t need two computer scientists on a team. You need computer scientists and journalists.”
Hammond cofounded a company that automates news articles in January 2010, shortly after I took his class. Called Narrative Science, last year it automated the writing of four million newspaper-style stories about youth baseball, softball, and basketball games, in partnership with GameChanger, a score-keeping app. But most of its clients are businesses like MasterCard, PWC, and Credit Suisse that need to extract information from large data sets (and have bigger budgets than newsrooms for doing so).
The company’s tool, called Quill, can for instance, look at a fund manager’s decisions and generate a report that explains them to her investors. Narrative Science works with 13 of the top 25 largest asset-management firms to automate the process of writing portfolio commentary.
If Hammond had to automate my job, a thought exercise to which he somewhat reluctantly agrees, this is the part that he would automate. “Imagine if you had the smartest data scientist available who could work at blinding speeds,” he says. “And you could pose hypotheticals and it would run around gather all the information and come back to you and talk about it.”
If I took every suggestion that my consultants offered me, I would still have plenty to do at my job. I’d take that time I saved transcribing interviews and responding to emails and use it to interview more people and write more articles. There are a lot of jobs that are in similar positions. McKinsey estimates, for instance, that 25% of a CEO’s job can be automated, including tasks like analyzing reports and preparing staff assignments. We’d still need CEOs, but they’d have more time to develop strategies and mentor employees.
This argument is easiest to make in the realm of knowledge work. It’s harder to imagine how jobs inside factories and trucks would shift toward new tasks. Even in these cases, though, it’s not always as simple as firing an employee and installing a robot in her place. Most of the time, robots replace a portion of a job, rather than everything a human worker does during the day. Sometimes, work will change altogether as the world changes.
“We expect that in a few years, EY will do all sorts of things that you wouldn’t have expected,” says Chris Mazzei, EY’s “global innovation technologies leader” at a New York press event. Auditing, one of the company’s core human activities, is ripe for automation, and it has started to expand technology that automatically extracts appropriate information from contracts. Theoretically, EY can offer more services to customers using the hours its human employees save scanning through contracts. But he can’t say can’t say exactly what those new things will be, how many people they’ll require, when those jobs will appear, or whether the people who lose jobs will be qualified to take the new ones. Technological changes is moving too fast, and the only sure thing is that EY and everyone else will need to evolve in some way.
It’s not that workers have nothing to fear from automation, but rather that companies will have a fair amount of choice over what they want to do with the extra efficiencies that technology will bring.
This was supposed to be the point of Hammond’s computer science for journalists course. “You have to use technology to do what you want to do,” he says. “The more you know how to use the technologies and the more you understand what you want, the better the world will end up being.”
And, on the plus side, he emphasizes, “Something you’re in partnership with doesn’t replace you.” I hope he’s right.