Are Machines Replacing Humans? Or: Am I a Luddite?
Update: You can find a followup post including some brief answers from Robin Hanson (and my commentary on same) here.
My gentle readers will undoubtedly remember a question I’ve asked repeatedly: as technology steadily increases productivity, will we (have we) come to a point where a large portion of workers can’t do “valuable” enough work to earn a decent living? Where only the technologically adept, and owners of technology, “merit” liveable earnings?
My thinking is not particularly original. As Robin Hanson points out in the presentation I’m about to discuss, it goes back at least to David Ricardo. who went after this subject quite rigorously in 1821 (The Principles of Political Economy and Taxation, 3rd Edition).
Many will say (not without grounds) that my thinking is simplistic, amateurish, and silly. The most powerful argument that my concerns are groundless or stupid: invoking history in the form of The Luddite Fallacy:
If the Luddite fallacy were true we would all be out of work because productivity has been increasing for two centuries.*
If technology is making human labor less valuable, how do you explain the steady, seemingly inexorable rise in wages/earnings/GDP per capita (choose your measure) over the last century or two? (Yes–if you look globally–even over the last few decades.)
That, by my lights, is a bloody strong argument. How does one respond to it?
It’s a question I’ve struggled with mightily, not only to protect my pet theories but because of a feeling that something was missing. And I think I’ve found an answer in Robin Hanson‘s “Economics of Nanotech and AI” presentation at the Foresight 2010 conference. (Liberal-bashers take note: Robin is a quite devoted if decidedly idiosyncratic libertarian.) If you’re a reader like me you’ll find much of the matter, more quickly, in Robin’s IEEE Spectrum article, “Economics Of The Singularity“–though if you’re feeble-minded like me you’ll also need to view the presentation and the slides (PowerPoint) to understand his insights.
This is probably old-hat to many who are better-versed than I. But it’s something of an aha! for me, and may be likewise for some of my readers.
Here’s the heart of his matter, a debate economists have been having for centuries: do machines, does technology, complement (I would prefer “augment”) or substitute for (I would prefer “replace”) human labor? If all technology does is augment human labor–making each person more productive–that increased productivity is purely good and all boats rise (at least over the long term, though obviously with local disruptions, both temporal and geographic, that societies might want to address and ameliorate.) If technology replaces human labor, then some portion of the population, over time, will get squeezed out, with no opportunity to “earn” a share of the increased production. (And/or, wages for labor will decline.)
So which is it–does technology complement or substitute for human labor? Augment or replace? My immediate, uninformed answer is “Yes. Both.” But I’ve had no idea how to quantify those effects, or characterize their interactions.
The economic consensus (as I’ve discerned it and as Robin reports it) is that it’s all complement/augment–substitution is only local and temporary. And two centuries of rising wages certainly give a great deal of weight to that position. But it still seems more likely to me, even obvious, that both occur.
Which leads me to wonder: How do the two effects interact? What’s the ratio of the two? Does that ratio change? Most interestingly to me, has that ratio changed in any steady way over the decades and centuries?
Happily, it turns out that Robin is a Yes man like me:
I want to help you understand how they can both be right. (22:32)
He explains it with the following waterline model.
Machines do (obviously) take over human tasks, replacing human labor in those tasks, as machines gain a “better relative ability” to do those tasks than humans. No doubt about it. That’s the point where the waterline meets the shore, where a task can be equally well/efficiently done by a machine or a human. And the waterline is obviously rising.
But (not shown) those machine tasks complement the (more cognitive/creative) tasks that humans still do–and give humans time to do those tasks–increasing the humans’ productivity. (Robin explains: “The tasks themselves are all complements–and this is a very robust, standard thing [in economics]–the better the world economy does any one thing, the more valuable doing all the other things becomes.”) This makes the human effort steadily more “valuable,” so the humans can produce more with machines’ help, and because the humans’ work is valuable, those humans can claim their share of the increased production. Sounds great.
Another way to think about it: humans can only migrate so far above the waterline. They need that vast body of technology–that body of water at their back–to expand into new territories.
It’s worth pointing out that there’s an unstated presumption here: that the graph continues up and to the right, that humans can climb to ever-higher levels–leaving lesser tasks to machines while we tackle tasks that deliver ever-greater value per hour–ever-greater productivity. (Designing more efficient cars instead of building cars.) Assuming that everyone shares in that greater prosperity (as everyone has, in the long historical picture though certainly not in detail), this truly is an all-boats-rise scenario.
But here’s where things get interesting–when Robin changes the shape of the shoreline:
The land contour is actually a graph of the change in the complement/substitute ratio. Sometimes one effect dominates, sometimes the other. On the left, machines’ takeover of human tasks does more to complement human effort than it does to substitute for that effort–resulting in economic growth and human prosperity. (To repeat: this rosy view reflects a big-picture, long-term orientation. Some individuals can’t make the steep climb, and they drown. But humanity moves ever higher.)
But then that ratio changes–the shoreline flattens out, and the scenario shifts radically. For every increase in machine abilities, there’s more substitute, and less complement. But the water keeps rising.
It looks to me like a lot of people just drowned, while a lucky few remain perched on the precipice.
Here’s Robin’s explanation:
There’s both a substitution and a complementary effect. And which dominates depends on the shape of this curve. Down here where it’s very steep you have very little substitution and a lot of growth. The machines getting better basically means people get richer, wages rise. But we could also reach a point [slide change] where there’s a large flat region in principle, and we could have the wave of water coming in, to a point where most income in the world is going to the machines, and a relatively small fraction is going to the people, and depending on the shape of this shoreline it might simply flood the entire region.
Unlike the future that Robin’s envisioning, though, in our world machines have no claims on earnings–they aren’t “people,” so they don’t get income. Their owners do. So he’s describing a situation “where most income in the world is going to the [owners], and a relatively small fraction is going to the [workers].”
Robin says this quite explicitly in his Spectrum article:
Wages could fall so far that most humans could not live on them.
This is, to my understanding, exactly the situation that the Luddites (and many others since) were so concerned about.
Robin is rather blithe in his statement that “in principle” there could be a large flat region. His talk has heretofore argued that periods of rapid growth (the steep parts of the curve) occur, highlighting the agricultural and industrial revolutions. And he asserts that those periods are getting closer together. He also asserts that another one is imminent–he thinks in the next hundred years.
Which would suggest that we are currently in one of those flat periods, where you see a lot more substitution relative to complementing–where a lot of people are being squeezed out of the economic system (drowned), without commensurate gains via complementarity (machines’ increased ability to produce things–and help humans produce things–that humans value).
Let’s hope the hill keeps going up to the right, and that humans have the capacity to keep climbing.
But here, perhaps, is the issue: many don’t. While measured IQ has been increasing over the decades since it was first measured (they keep having to recalibrate the test to achieve the 100 median), it’s not increasing anywhere near as fast as machines’ abilities. It’s possible that a large group of humans–at least in advanced, knowledge-driven societies like the U.S.–are being left below the waterline.
There’s much more I’d like to say on this topic, but I find myself out of time. (And I’m thinking you might be as well.) So I’ll just leave you with the questions I posed for Robin on his blog (slightly modified and expanded here).
Your augment/replace waterline model is, IMHO, profound. To bring it down to our current situation:
Have we reached that plateau? Perhaps sometime in the 70s, give or take?
Is it related to the limits of (aggregate) human cognitive capacity? IOW, since 50% of people have an IQ below 100, and “valuable” knowledge-worker tasks are requiring ever-greater cognitive skills, can the ameliorating effects of education continue to maintain the augment/replace ratio, as they did for much of the twentieth century?
Since machines currently are not people, but are owned by people (so the machines’ earnings go to the owners), could this explain the increasing wealth and income disparities (labor vs. capital, wages versus rents) in recent decades?
Re: your much-less-than-satisfying answer to the question about Germany’s success (highly industrialized, but with major social programs): is it possible that in order to maintain demand for ever-more-efficient productive capacity, government redistribution is a necessity? No–not at 1,000 times some imagined level, but somehow relative to per-capita shares of production?
Given that individual utility functions (as measured by “happiness”) seem to flat-line at about $15K in annual income in developing countries, about $60K in the U.S. (yes, iffy stuff, but the threshold/flat-line seems likely at some level), can the demand from a small cadre of owners–who don’t “value” most goods very highly–provide the demand necessary to keep the economic log rolling?
Could the absence of this widespread demand–making it difficult for capital to find productive investments that pay a good return–explain the massive increase in “casino investing” over recent decades? A desperate search for returns in a world where demand does not reward valuable production?
Could this situation also explain the downward pressure on secondary-education budgets? A vague sense of (impending) declining returns to education?
Could it also explain the rapidly increasing lengths of “jobless recoveries” since the 70s?
Is it possible that the current…difficulties are like a wave crashing on that plateau?
IOW, because of the limits of human capability, could the Luddites (finally) be right? Even a stopped clock…