Job-Killing Robots Are Neither Robots Nor Killing Jobs
If you’re using machine learning to do a task that’s currently performed by people, you’re going to face a version of the “Jobs” question. When posed by your workers, it boils down to this: When you increase automation at our company, what will happen to my job? That’s the sanitized version of the question. The more visceral version is more like: Is a robot coming to make me irrelevant?
Bringing more machine learning and automation to your company could frighten your employees — or make them extremely happy if it reduces the tedium involved with their job. It all depends on the choices you make and how you communicate them.
Maybe your employees are explicitly asking you questions on this topic. Maybe they’re keeping it to themselves. But in an era of headlines like “Automation could kill 73 million jobs by 2030”, you can safely assume that this is a topic you need to address.
‘Job-Killing Robots’ Are Not Robots
For starters, the “robots” are not really robots. In the context of most white-collar job automation, we’re usually talking about computer software. The software may manifest itself as a user interface on a screen, but otherwise, it takes no physical form. You don’t walk into an ML company and see Rosie from The Jetsons or Mr. Data from Star Trek sitting at keyboards that were once occupied by humans. You see people working at computers — much the same as you’d see in any modern office (to be fair, you might find Mr. Data on a t-shirt or two). Robots are generally a metaphor, a tangible way to talk about a largely intangible process.
I assume, of course, that you already know this. But writing about automated journalism a few years ago, Slate’s Will Oremus said he “wouldn’t be surprised, at this point, if the word robot in headline after headline had left a lot of folks genuinely confused.”
To be fair, it’s not just headline writers who want to give AI software a more corporeal form. “Olive” is an AI solution for healthcare providers that has not only a name, but a gender. The company positions “her” as a digital employee with a manager and an email address. One of the most famous AI personifications is IBM’s Watson, which defeated humans at Jeopardy! in 2011. In these and other cases, however, the anthropomorphization of the machine is more of a marketing choice than a technical description.
Headlines and graphics about machines “killing” jobs can conjure up ideas that automation is some kind of Terminator, sent from the future to destroy gainful employment. But despite what you see in the movies, many “robots” are actually software tools. And their impact on jobs so far has been nowhere near-apocalyptic.
‘Job-Killing Robots’ Are Not Killing Jobs
In 1987, economist Robert Solow famously wrote that “You can see the computer age everywhere but in the productivity statistics.” Today, you can see the automation everywhere but in the unemployment rate. The US unemployment rate is at a low, having fallen every year since 2010 across two very different presidential administrations. To borrow another classic phrase from the 1980s: Where’s the beef?
Look at Google search interest for “machine learning” over the five year period from 2014–2018 and the unemployment rate trend for the same period is telling. The “robots” may be coming, but businesses are still very keen to hire old-fashioned humans.
Of course, the economic situation will change eventually. But we don’t foresee the kind of sudden, massive, near-term unemployment that some seem to be imagining. One could argue that the data is wrong and that a lot of unemployment or underemployment is going unaccounted for. One could argue that, while the situation seems ok for now, a switch will soon flip and rapidly turn the world upside down.
But I’m going to go with the data in front of us. While it’s undoubtedly true that automation will replace some jobs and reshape others, I don’t believe hyperbolic scenarios are a helpful guide to what’s happening. Imagining a scenario in which half of all workers are suddenly ousted by machines is, for now, a better exercise for a science fiction writer than an executive.
Automation does not necessarily take jobs. For example, take Automated Insights, a natural language generation company I founded in 2007. Its Wordsmith software automatically turns structured data like a spreadsheet into stories that appear to have been written by a person. Wordsmith produces everything from Yahoo Fantasy Football write-ups to financial articles for the Associated Press. In the latter case, the software not only took over work that human reporters had done, but it also multiplied the article output by nearly 15.
Because of these and other projects, reporters would often ask if Wordsmith was a threat to their jobs. But even after fielding that question for several years, I never knew of a single job that Wordsmith actually displaced. That’s because Wordsmith didn’t replace people — it replaced a specific, rather monotonous writing task that was better suited to software. As the Automated Insights story shows, job losses are not guaranteed, even when the software in question (a “robot writer”) seems at first glance like an obvious human replacement.
Robbie Allen is a Senior Advisor to Infinia ML, a team of data scientists, engineers, and business experts putting machine learning to work.
