The automation debate is flaring up again! Having a candidate like Bernie Sanders talking about labor in ways that have been neglected for years has put globalization back in the forefront. And it seems that those who work on AI risk are looking at the problem closely again. In line with the tradition of Malthus, Keynes, and many others, I’m seeing predictions that robots will steal our jobs. Or work for us, if you’re an optimist.
Here’s the thing. We already had an automation revolution. It was in the 1800s. Machines took human jobs. People jammed their wooden shoes into the machines. Machines still took their jobs.
And then they got other jobs.
In the jargon of economists, automation puts deflationary pressure on consumer prices. The only reason to automate is to make production cheaper because if you have to pay more for automation than you are for labor* you will stick with labor. This means that in a reasonably competitive environment, workers (and everyone else) end up paying less because of automation.
Workers end up needing less in wages—in such an economy, keeping the amount you are paid in wages constant is actually like a raise since things are getting cheaper. In Great Britain, between 1800 and 1900 prices fell so that a pound of goods was worth .68 pounds. Part of this** was the industrial revolution making production cheaper. That wages rose during this time is a testament to the relationship between capital and labor being linked, not antagonistic.
This means that robots work themselves out of jobs. By putting downward pressure on prices, they make it less attractive to hire robots and stick with inferior meat-humans. The very thing that makes automation attractive limits itself.
Now, here comes some important qualifiers.
First, this can be hugely disruptive. “Oh, yeah, things got cheaper during the 1800s” smooths over the century that eroded the peerage and European social order, set the stage for WWI, saw colonialism peak, was riddled with socialist and anarchist agitation, and saw abject poverty move from the country to arguably worse slums. While I stand by my sanguine analysis that automation is probably going to continue to be self-limiting overall, we will need policies designed to smooth the waters.
Second, this assumes that the cost of capital will remain high. I do not believe that AI advances are the main threat to labor. Software will plausibly chip away at some margins, but it is the hardware that is likely to be the limiting factor. Durable materials—plastic, metal, silicone, etc.—are probably going to increase in price as demand for them continues to rise. There is a scarcity of raw material and they are relatively difficult to work with. Innovations on that front, I believe, would do more to make capital attractive and squeeze out labor. We’ve seen a lot of progress, but things like Moore’s “Law” are breaking down. At any rate, this would not completely overcome the benefits of downward pressure on prices.
The history of labor and capital so far has not suggested that capital is especially detrimental to labor. The age of capital accumulation—the 17th century to present—have seen labor’s general improvement. (Those who would point to the faltering middle class in America miss the larger picture; Americans have lost at the margins to labor else where. See the first qualification above.)
I boldly predict that centuries of precedent will largely hold.
*Provided you get the same output. I mean, obviously, it can make sense to move to machines if they do better, and even worse, but that just adds qualifications without clarity.
**Here there be dragons. There is wide consensus what I wrote is true. There is no consensus on what fraction of that decline was from capital. It’s complicated.