A flaw in the way smart people think about robots and job loss


People talk about robots coming for their jobs quite a bit. It’s a big topic, and I have thoughts about a few different aspects of it. But, for today, I want to focus on a particular logical error I see smart people making all the time.

Thinking about robots replacing people one at a time

Consider this quote from Matt Levine:

If you are a human employed at an investment bank and worried that robots are coming for your job, I recommend that story about UBS AG’s forays into artificial intelligence, which I found extremely soothing. There are two products involved. One is a boring office-automation thingy, which “scans for emails sent by clients detailing how they want to divide large block trades up between funds” and then does the dividing, “doing a task that would normally take a person about 45 minutes in only about two minutes.” Yeah look no one is losing their job over that; that is a pure win for the junior person who would otherwise have been doing that allocating.

Now, Matt Levine is a smart guy, and his posts are often very insightful (seriously, read his blog). But this last sentence is a perfect example of the flawed thinking I have in mind. The implied claim is that if a robot can’t replace 100% of a particular job (or at least a very large fraction of that job), then “no one is losing their job” because of that robot.

And that’s just not how it works.

As Matt Levine knows very well (he used to work there), Goldman is really big: they hire nearly 10,000 people in a given year. Anything that saves 10,000 people 45 minutes saves the equivalent of 3.75 full time jobs.

Of course, that’s not exactly right—it’s not like all 10,000 new hires do exactly the same thing, and I’m sure this new tool wouldn’t save them all 45 minutes. On the other hand, I’m sure that it’s not a task that people do just once a year. Let’s say it’s a task that 100 peopleFN 1 each do once a week. That means it saves an average of 75 hours/week.

Now, in the short run, I’m sure Matt Levine is right that it’s a “pure win for the junior person.” No one is going to get fired because they have 45 minutes less work to do. On the other hand, when there’s 75 hours less work to do between 100 people, well, I bet that there will be 99 people doing that work the next year. And at a place that hires hundreds of people every year, it may be that no one can ever point to the “guy who got fired because of that invention.” But that doesn’t mean that the invention isn’t reducing Goldman’s demand for junior labor going forward—or that there’s no one who would have worked there who won’t now, just because of the new invention.FN 2

A case study

This seems to be exactly how technological change has caused job loss in many industries. Take secretaries as a prime example: in the vast majority of companies, there’s no single invention in the last 30 years that you can point to and say “yes, that’s the reason we fired all those secretaries; as soon as that invention came out, we knew we didn’t need as many people.” Instead, it’s been a slow but steady increase in productivity-enhancements, all of which let an ever-shrinking number of people do the same amount of work.

Here’s how this plays out, according to a 2015 New Republic article: Every invention makes existing secretaries more efficient, which mostly doesn’t get anyone fired—it just gives them easier jobs (and probably reduces hiring). Then, there’s some sort of bad recession, and companies look around and realize they employ a lot of people doing jobs that could be done by many fewer people. Layoffs ensue. After the dust settles, the company discovers that they’re getting along just fine with fewer people, and those jobs never come back.

As the New Republic Notes,

This phenomenon is somewhat new. Routine jobs like administrative support used to take the largest hit during recessions in the 1970s and early 1980s, but then they would experience strong gains afterward, fully recovering, according to a recent report from Third Way. But after each of the three recessions we’ve endured since 1991, employment in routine jobs fell and never bounced back. “In fact, employment in this occupational group never recovers,” the paper notes; “these occupations are disappearing.”

And the reason those occupations are disappearing is technology.

The assistants, for their part, report that they’ve had to support more and more people. According to surveys of the membership of IAAP, more than half worked with one or two executives in 2005, while about a quarter supported three to four and 11 percent supported five to ten. Today, less than 46 percent work with one or two executives, while more than 27 percent support three to four and 15 percent work with five to ten. “That tells me that fewer people are needed to support more people,” Allen said.

Less dramatic, but more important—and not going away

This framing (that technology dives job loss by incremental improvements rather than revolutionary change) isn’t as dramatic. It’s a lot more striking (and makes for better copy and more clicks) to write headlines like Driverless trucks: economic tsunami may swallow one of most common US jobs or SELF-DRIVING TRUCKS ARE COMING TO SAVE LIVES AND KILL JOBS. And, sure, there may be some technological changes that present an “economic tsunami” that results in thousands or millions of job losses at the same time.FN 3

But the reality is that, for every Uber-invents-self-driving-car,-thousands-now-homeless story out there, there will be dozens or hundreds of small-scale, minor inventions that help the a smaller number of people do more with less.

And, when it comes to these small-scale changes, I don’t think we’re anywhere near having exhausted the low-hanging fruit. I don’t have the statistics to back this up in a fully rigorous way, but I do have some anecdotal experience I find personally compelling.

As I’ve mentioned before on this blog, I’m learning to program. And as is pretty obvious from the one project I’ve posted so far, I’m in the very early stages of learning how to program. I’ve probably only spent a few dozen hours learning, and have basically just been picking up the most rudimentary aspects on an as-needed basis.

But, last Saturday, I sat down and spent 30 minutes writing a tool that will automate a task I need to do. Using that tool, I turned a 3+ hour task into one that I can finish in under an hour. If I had this tool three months ago, I would have saved nearly a full workday; if I’d shared it with the three other people on my team doing the same tasks, I bet we would all have saved nearly that much.

And I’m just taking my first tentative baby steps into programing. I look around at work, and see good, intelligent, hard-working, and efficiency-minded people—all of whom are spending a lot of time on tasks that could be automated. and I don’t think I’m the only one who sees it. The corporate world is still full of low-hanging fruit, and the chance to do ever more with ever less is not going away.

And that’s where the real changes are going to come from, not sweeping changes from robots that replace people wholesale.

(Well, at least for the next decade or so. As they say, predictions are hard, especially about the future. Maybe in 50 years the Artificial General Intelligence will be a reality; I’m not yet willing to venture a guess. But for the—very short—foreseeable future, my money is on the incremental change that comes from doing more with less.)

Related Posts

2 thoughts on “A flaw in the way smart people think about robots and job loss”

  1. I heard an interesting version of your take on this on Planet Money a while back, talking about the invention of the spreadsheet.
    http://www.npr.org/sections/money/2017/05/17/528807590/episode-606-spreadsheets

    As they said in the promo, “Any calculation made on a spreadsheet was done by hand, and these could take days to complete for an accountant or bookkeeper. It was tedious. One little adjustment to these calculations meant a whole day of erasing and filling the same boxes out again.”

    I think the “truck driver” issue you mention above originates in the people most likely to lose their jobs to technological advancement (think factory laborers who lose out to automation) being the least likely to benefit from industries relying on high tech workers rather than laborers. Because there’s a very large segment of America that lost work and couldn’t find it again due to changing technology, they’re very visible. Especially in light of their highly unionized environment and the political power of said unions. It rather sets the tone of the discussion on the benefits of these things.

Leave a Reply

Your email address will not be published. Required fields are marked *