🎶 Yoshimi Battles the Pink Robots by The Flaming Lips
“Always describing and then discarding their
throwaway world
and then leaping and listening to the tiny slippage
between real and technical time, I heard them
persistently telephoning and glorying in their
lightness
saying ‘Singing is who we are in this place.
we are made of digital sounds, we are seeking
to be slightly more precise than is possible,
whizzing around, trying to unconceal things
literally momentary.’”
-Alice Oswald, “Five Fables of a Length of Flesh”
A few weeks back we explored what a ten year career might look like, which made me realize I’m in the second arc of finally landing in a ten year career that I love. But I got there after quite a few fits and starts, including ejecting from a PhD program and wandering the maze of Seattle tech companies as the first tech bubble popped, leaving me chewing that piece of Juicy Fruit like I could still squeeze some flavor out of it. There’s a Tim Urban piece on picking a career that maps to my own process for career-hopping in agonizing detail. But really, the idea of choosing a career at all is a hell of a privilege. One that fewer and fewer of us are going to have.
The whole “what do you want to be when you grow up?” question is so broken from the get-go. My 7-year-old daughter’s reaction to this still strangely pervasive question lobbed at kids for millennia is that she’ll “be a teacher on Tuesdays and Thursdays, a nurse on Mondays and Wednesdays, and work on her professional mountain biking career on Fridays and maybe over the weekend.” So at least she’s starting with rejecting the notion that you have to pick one. But all this is still predicated on the outdated canon that whatever a bright-eyed child picks would almost certainly provide security, a good living, and a healthy pension you could retire on at the end. The reality is much more likely my daughter having to juggle three possible careers just to get by. As someone who has now navigated the tech industry into a second decade, I start contemplating all the ridiculous options: should she learn to code so she can oversee the machine workforce? Should she optimize her career-juggling choices for an insane career path that ensures she’ll always be one step ahead of the robots?
Almost on cue, I find out about this piece from The Pudding, “Why the tech sector may not solve America’s looming automation crisis.” In it, Jordan Dworkin and Ilia Blinderman lay out a data-heavy and visually impressive case for how “learn how to code” is a ridiculous answer to the vexing problem of automation eliminating a swath of American jobs. They lead with this persuasive bit of data: “5% of America’s employees may be replaced by robots.” The data stem from this McKinsey report and this OECD working paper, but at the root of both of those (and many, many other reports and articles on automation) is this 2013 paper from Oxford University. It ultimately claims that “up to 47% of total US employment is at risk [of computerisation].” If you think that’s hyperbolic, that’s because it is—this study the Typhoid Mary of automation research, and it has problems.
You can’t read anything about robots and automation right now without someone taking a swing at these kinds of numbers. But the problem is, they’re all built on a sandcastle of assumptions (some might call them myths) about automation that fail to understand certain fallacies about how automation really works.
Envisioning the future is a precarious enterprise that is subject to biases. As past work has shown, claims about the effects of future technology change are underspecified, ungrounded, and overconfident, whereas new risks are missed, ignored, or downplayed.Woods and Dekker*
At the core of the Oxford study is the Substitution Myth, and it’s pervasive to current attitudes and predictions about automation. It’s the belief that machines can take on tasks previously done by humans—that automation can be substituted for human action without any larger impact on the system in which that action or task occurs, except to increase output. In particular, it assumes that automation needs no human oversight once implemented. And y’all, that’s just not how it works. In part because this myth is rooted in an old, broken idea that humans are better at some things, and computers are better at others (aka HABA/MABA).
Anyone who has built, and especially operated, a system with any sort of automation has a long list of stories about when it didn’t behave as expected, and humans had to get involved to figure out why. (In some cases, entire new teams spring up solely to manage all the new realities of these automated systems.) It’s easy to write these off as corner cases and figure you’ll eventually nail all those corners down, but that’s not how complex systems work, especially when those complex systems are navigating the natural world. It also presumes that we’ll be able to think of or iteratively find all those cases enough to effectively rule humans unnecessary at some point, but how do you know you’ve gotten to that point? The reality is that we’ll always be in the loop, an ouroboros of human-machine cooperation that’s defining a new hybrid of work. This new hybrid means both receding and emerging jobs, which is… complicated.
The TL;DR on this is to be very skeptical of any numbers you see floated around about how many people or jobs might succumb to automation, and by when.
Back to Pudding. What this is all about, really, is jobs. The authors ask, “Is it necessary for workers to move that far across the job landscape to avoid automation?” If you scrap the fear-mongering about automation, the reality is that stable, long-term jobs are harder and harder to come by for a variety of reasons. If you do need to make a move, how would you even know where to start? If you’re a long-haul trucker and you’re worried that Elon Musk is about to start slicing away at your hard-earned middle class way of life, do you spend days or months examining your subconscious via your Yearning Octopus and figuring out what intrinsic and extrinsic motivators really drive you, as Tim Urban suggests? Hell no, you try to figure out what you can realistically do that will also earn you close to the same salary and not require you to go back to school for years. Dworkin and Blinderman use Department of Labor data to show what a transfer of similar skills and income levels might look like, which short-circuits Urban’s career navel-gazing with a simple tool that anyone can use.
So before you send that tool to your friends who are panicking about the robots coming for their jobs, it’s worth considering that we should start by… not panicking. As Jill Lepore writes in her take-down of this very same topic, “Panic is not evidence of danger; it’s evidence of panic.” If you read nothing else about automation, please read her piece. Our fears about this are not new, and the tactics of the people stoking them are time-honored traditions of low-key terror.
If all you read is the Economist or HBR or many mainstream newspapers, it’s easy to believe the robots are coming for us tomorrow. Here’s some people I suggest following on Twitter who are at the pointy end of the stick when it comes to building, running, and researching automated systems.
That’s Good Work for this week. Looking forward to what’s next.
Courtney and the Holloway Team
Good Work is written and curated by Andy Sparks, Courtney Nash, Dmitriy Kharchenko, Hope Hackett, Joshua Levy, and Rachel Jepsen.