There’s plenty of debate these days about whether or not robots and AI will take our jobs. I think that they will, but not in the ways we think, and that the difference is important.
Will we lose our jobs?
The debate around automation and employment today goes something like this:
- We’ll be unemployed, because robots can do work better than humans, and algorithms can reason better than humans. Therefore, they’ll take our jobs. Consider exhibits A and B:
Pancake time.
I am so glad this wasn’t around when I was eleven.
- No, we will still have jobs because all this has happened before (for example, the Cotton Gin) and historically, new technologies created more jobs than they consumed. The automation of things is just the most recent example of Jevons’ Paradox: More efficiency means more new uses means more consumption.
- Yes, but this time is different. The industrial revolution was about replacing muscles with steam; this is about replacing neurons with electricity. Machines can’t upgrade themselves; algorithms can, and the singularity is near. Tim Urban has explained this amazingly well:
Compounding is the most powerful force in the universe.
- No, because machines are special-purpose and narrow. Humans are generalists. So the machines will actually augment us, making us better at what we do.
Work will change
So we come to the current state of this debate, namely, “the nature of work will change.” Kevin Kelly has written eloquently about the post-productivity economy. He sees the future coming, but thinks we’ll redefine work when the things that robots and algorithms do are cheap and abundant.
Meanwhile, the Pew Research Center has pointed out the huge disconnect between how many jobs will be replaced by machines, and how many of us are worried about it.
Head. In. The. Sand.
Is the future good or bad?
Personally, I’m relatively optimistic about automation, as long as we do the following things:
- Avoid the concentration of data and algorithms among a powerful few, the way capital has concentrated in the hands of the 0.01%.
- Change the metrics by which we measure how good a society is to make it less about productivity (GDP) and more about creativity or quality of life.
- Help people redefine what they consider a “good” career, emphasizing art, open-ended science, and so on.
- Find a way to ensure that just because an algorithm paints an “optimal” path, we aren’t forced to take it, because that means the road less taken is only open to those who can afford to ignore the optimum.
So, yeah, I’m not worried at all.
Death by a thousand cuts
Anyway, one of the most common objections voiced by those who think employment will continue is that machines can do some things well, but not all things. The mainstream media has done us a disservice here: Every news story about the latest AI breakthrough tends to hide the fact that it’s specialized. AlphaGo is as bad at Jeopardy as Watson is at driving cars as Tesla is at playing Go.
Heinlein famously said, “specialization is for insects.” Which is great, until you’ve played Starcraft.
This is how you use a swarm of narrow-minded insects to overthrow humans.
I’m going somewhere with this.
General contractors and nail robots
Imagine, for example, you’re a talented generalist construction worker. You’re good at lots of things—from framing and lumber, to wiring, to plumbing, to drywall, to painting. “Sure,” you think, “machines can perform tasks. I like it that way! The circular saw and the drill and the nailgun changed my job, and I’m busier than ever!”
Now suppose that someone introduces a very narrowly focused machine, one that’s good at, say, framing 2x4s on job site. You might think you’re competing on how good you are at hammering nails. But you’re not. You’re actually competing with how good it can be within the whole set of tasks related to framing.
If I made a robot that could frame walls by hammering 2x4s, I probably wouldn’t stop there. I’d probably add some form of record-keeping, and analytics, so I’d know where nails had been placed, and where the wood was. This, in turn, would be provided to the bank (for cost analysis); to insurers (to demonstrate compliance); to the homeowner (as a form of 3-dimensional digital blueprints); and so on.
When you think this way, you realize that the construction worker isn’t competing with a robot that hammers nails and holds planks. The worker is competing with a robot that lets a homeowner put on AR goggles five years from now, and see where all the joists are, so that homeowner can hang a painting perfectly without a stud finder.
Generalist humans are only competing with a small part of their replacements’ advantages.
In other words, we didn’t know a construction robot was for hanging paintings five years later, but it is. And it’s hard to compete with that kind of change.
There’s a similar narrative for software. A digital accountant, or lawyer, or diagnostic technician isn’t just good at reviewing records; it has unintended benefits—collective baselines, chain-of-evidence recording, visualizations, etc. Software will consume jobs faster than hardware, because software is quickly, infinitely self-upgradeable. It keeps getting better. Hardware still needs retooling.
In the case of both robots and algorithms, this boils down to: Humans are bad at taking notes; machines have no choice but to do so; and the notes have a value far beyond the task itself once analyzed.
Notice we haven’t even delved into the supply side of the workforce. Robots and algorithms don’t call in sick, don’t need to sleep, don’t become inattentive, don’t make for ugly headlines in industrial accidents. And the trickle-down consequences for insurance, architecture, and so on are vast.
I think the jobs we have today—the jobs we tell people like Pew we believe will be around in fifty years—are going to vanish more quickly than we expect. But we aren’t paying attention, because these jobs won’t die a big, generalist death. Instead, they’ll die by a thousand specialized cuts.