Thanks to everyone who commented on the post about technological unemployment.
Not saying I necessarily think this is what is going on, but one simple possible explanation for why technological unemployment could happen now when it never happened much in the past could be quite simply the greatly accelerated pace of change.
For most of history, technological change was very, very slow. The past few hundred years we’ve moved increasingly to a place where each new generation has to learn to function in a world different from the one their parents grew up in. We could now be moving to a world where each generation has to learn to function in multiple worlds over the course of a lifetimes, which may stretch the limits of human adaptability.
I don’t know much about economics, so the following is just conjecture, but: I think one reason for this state of affairs might be the reduced pace of technological innovation. In the past, when an industry (such as horsemanship, agriculture, switchboard operation, etc.) was automated, the technological advancements that led to this automation also led to the creation of several new industries. To use an extreme example, the same technology that eliminated switchboard operators led to the massive paradigm shift that was the Information Age.
However, in the modern Western world, new industries are created a lot less frequently; and thus workers whose jobs got automated have nowhere to go but down. This could be due to a combination of factors:
1). Automation today is increasingly the result of incremental advances in the field, as opposed to paradigm-shifting discoveries. Thus, most of the new industries that could’ve been created, already have been.
2). Technological progress in general has slowed down significantly, because…
3). …All the easy stuff has already been discovered, and/or…
4). …A combination of the socio-political climate, as well as extremely aggressive copyright/patent law, is causing a chilling effect (at least, in the Western world).
5). Western countries are suffering from resource depletion; it’s harder to innovate on zero budget.
Graphs comparing college to high school over time need to show percent of the population each group is over time. (i vaguely remember this coming up on the blog before)
This is definitely true. For example, suppose that in 1950, only the richest 10% of people went to college. But in 2050, all but the poorest 10% of people go to college. We might discover that we had gone from most uncolleged people having decent jobs to most uncolleged people having terrible jobs – not because the importance of college had changed, but only because the poorest 90% of people might usually have mostly decent jobs, and the poorest 10% of people usually have terrible jobs. When I was doing some of this research, I asked Economics Tumblr whether this was a big deal, and I was told that some people had done the relevant calculations and they had been found not to affect any of these conclusions very much.
Chris Williams on why people on disability might claim not to want jobs even if they do:
Part of this is also the way the social safety net is setup. I know a dude I went to high school with years ago. He’s bipolar and a schizophrenic (sp?). Unmedicated, he basically goes all Charlie Sheen crazy from time to time. It makes him unable to function in society.So he’s considered disabled. However, he responds very well to medication. Now that he’s declared disabled and gets medication and doctor visits/therapy for free through Medicaid, and is on meds, he’s turned into a completely normal guy. 95% normal anyway. It’s completely turned his life around. Now, he would be perfectly able to hold down a job, or go to school, or learn a trade. That’s the rub though. If he gets a job or goes to school it serves as evidence he’s not disabled and could get him kicked off of food stamps, public housing and medicaid. So it’s a safer deal for him to stay home and play xbox than to contribute to society. He’d rather be “normal” in the head and sitting around than being crazy – which are basically his choices. If our social safety net was better designed, we could get some of those people back in the workforce.
There’s some disagreement and further discussion in the responses to his comment.
From Wrong Species:
It’s bugging me that you’re comparing total number of manufacturing jobs with percent of men in the labor force. It should either be both percent or both total.
This is a really important point which I messed up in the original version of the article. I originally thought it wouldn’t matter that much (US population hasn’t grown that much over time, has it?) but when I finally found the percent manufacturing jobs graph, it looked totally different and changed all my conclusions. I briefly panicked, then realized that probably what was going on was women were entering the labor force, generally into non-manufacturing jobs, and so the percent of workers in manufacturing was gradually going down over time. I can’t find a graph that adjusts for that, so I’m just going to trust my existing number of jobs graph for now.
From the same comment, on Autor’s example about bank tellers:
[Autor wrote]: “Consider the surprising complementarities between information technology and employment in banking, specifically the experience with automated teller machines (ATMs) and bank tellers documented by Bessen (2015). ATMs were introduced in the 1970s, and their numbers in the US economy quadrupled from approximately 100,000 to 400,000 between 1995 and 2010. One might naturally assume that these machines had all but eliminated bank tellers in that interval. But US bank teller employment actually rose modestly from 500,000 to approximately 550,000 over the 30-year period from 1980 to 2010 (although given the growth in the labor force in this time interval, these numbers do imply that bank tellers declined as a share of overall US employment).”
I find it odd how people just gloss over that [statement in parentheses]. The right story is the simple one, where ATMs reduced employment among tellers. Rise through population growth doesn’t count.
The Bessen (2015) ATM study is a great example of the conclusion not at all matching what the actual data says. Go look at the plots in that paper. What I see is that 10 years after ATMs are introduced employment of tellers plateaus and stays there with very little increase until the present. Virtually all the job growth in the sector occurs within the first 10 years after ATMs are introduced. Eyeballing it, from 1970-1980 tellers increase by ~5% a year and 1980-2010 by 0.6% a year. In other words, it takes approximately 10 years for the ATMs to become widespread and impact tellers jobs.
I believe it was Gwern who made the argument that even figures such as PAMLFPR were poor ways of trying to understand total employment, since such categorization is very much an invention of the modern era.
As I understood their argument, even 100 years ago ideas such as retirement and compulsory education simply didn’t exist (note: am not historian, I don’t know when exactly compulsory schooling or retirement would have become widespread); anyone who wasn’t very wealthy would start working as soon as they were physically able, and continue until they were either dead or physically unable, more likely the former given health outcomes in those times. Thus, technological unemployment did occur following the industrial revolution, because it allowed for a huge number of former workers to stop working. However, economists did not recognize this as a decline in employment since new categories of education and retirement were introduced to absorb them.
Part of the reason the statistic only measures prime age people (between 25 and 55) is that they’re at an age where they’re probably not in school or retired. I don’t know if the decrease of working years through the lifespan can fairly be called “technological unemployment” any more than the entrance of women into the workforce can be called “technological superemployment”. They’re just supply-side trends.
From Grey Enlightenment:
Regarding the horse example, the US horse population is 9 million. The demand for horses never went away despite automation and has remained steady. If one tried to extrapolate the early 20th century horse population trend to the future, there should presently be no horses alive.
Wikipedia says there were 20 million horses in the US in 1915, falling to a low of 4.5 million in 1959, and increasing to 9 million now. So the overall story of technological unenhorsement survives this objection.
Ricraz on whether perfect cheap androids would really make human workers obsolete:
Significant parts of many jobs are driven by social interactions and status considerations. The job of doorman at fancy hotels may be both the most easily automated job in existence, and also the last one to actually be automated. In this particular scenario, it depends on how people relate to androids. If society has the general idea that interacting with androids is low-status and not as meaningful as human interactions, then there would be an explosion in customer-facing jobs, as it becomes the easiest way to differentiate your product. Also, everything produced by androids would be so cheap that people wouldn’t need to work very long hours in those jobs to earn a living.
In other words, whether or not humans are practically useful, they’ll always be useful for aesthetic and signaling purposes. This provides the missing piece of the technological unenhorsement story; machinery made horses mostly useless, but aesthestic and signaling purposes were enough to maintain them at 25% to 50% of their maximum population.
I find this worrying for the same reason I find the lack of technological unemployment worrying; it means there might never come a time when we’re forced to really confront our values and decide whether we want to be properly post-scarcity. Even in the world with perfect android laborers, instead of letting the androids labor and letting humans live in leisure and share part of the pie, we’ll have the opportunity to sleepwalk into letting a few android-owners get super rich and giving everyone else jobs as hotel doormen. What do you mean it’s time for a universal basic income? You should be grateful for the hotel doorman job you have!
AnteriorMotive with one last unenhorsement related point:
I think the horse example is deceptive: the bottleneck for horses is human labour. Horse breeders, handlers, etc, found better uses for their labour and transitioned to other sectors of the economy. Put another way, horses were priced out because they need to hire humans to be able to do anything, and humans’ opportunity cost got so high they could no longer afford to outbid it.
You can tweak the metaphor to make the intended comparison more applicable, but it rapidly starts losing its rhetorical power. I recommend the discussion in the comments of: http://slatestarcodex.com/2017/02/06/notes-from-the-asilomar-conference-on-beneficial-ai/
“This wasn’t to say that they aren’t an argument for human obsolescence, just that when you exchange the horse analogy for the structurally identical ‘cassette tape’ or ‘slide-rule’ anology, suddenly it’s lost its rhetorical punch. As I see it, this suggests a misleading analogy, hinging on the listener inappropriately anthropomorphizing horses.”
Good point. This is why I previously got accused of “thinking horses are LITERALLY humans”. I apologize for the error. But I’m not going to back down on them basically just being elongated cows.
vV_Vv asks if the existence of a middle class is just a weird historical blip:
For most of recorded history, almost every human was a subsistence farmer, likely under-using their cognitive abilities, yet no middle-paying jobs materialized to use their full potential. Then, during the first century or so of the industrial revolution, most people in industrialized countries were assembly line workers, jobs even less cognitively demanding than subsistence farming, and still no middle-paying jobs appeared until most of that low-skilled factory work could be automated. There is nothing in standard economic theory that predicts that jobs that allow average people to use their full cognitive ability should exist and pay middle-class wages. In fact, there is nothing that predicts that a middle-class should exist.
U.S. educational and occupational wage differentials were exceptionally high at the dawn of the twentieth century and then decreased in several stages over the next eight decades. But starting in the early 1980s the labor market premium to skill rose sharply and by 2005 the college wage premium was back at its 1915 level. The twentieth century contains two inequality tales: one declining and one rising. We use a supply-demand-institutions framework to understand the factors that produced these changes from 1890 to 2005. We find that strong secular growth in the relative demand for more educated workers combined with fluctuations in the growth of relative skill supplies go far to explain the long-run evolution of U.S. educational wage differentials. An increase in the rate of growth of the relative supply of skills associated with the high school movement starting around 1910 played a key role in narrowing educational wage differentials from 1915 to 1980. The slowdown in the growth of the relative supply of college workers starting around 1980 was a major reason for the surge in the college wage premium from 1980 to 2005. Institutional factors were important at various junctures, especially during the 1940s and the late 1970s.
Looking at your numbers we may have 2 different things happening here. Maybe from 1960-2000, the decrease in male employment was caused by the increase in female employment; and from 2000-today, female employment has been more flat, and the decrease in male employment has been caused by automation? Does that correlate better to the numbers?
A similar point from Proyas:
Also, it might be the case that the entry of women into the workforce starting 50 years ago permanently depressed wages and drove down the labor force participation rate. Is there an economic theory stating that the number of jobs must keep growing and their wages must also grow or stay the same, regardless of how many new workers are added to the system?…Picking a random example, if the number of people with J.D.’s in America doubles because women are no longer barred from entry, is there any reason to assume that the economy will automatically expand to absorb them all, with the number of attorney jobs doubling and salaries staying the same? “Is there an economic theory” that says such?
Alef is more (less?) subtle:
So in 1950 we introduced (or rather, used far more widely than just its former niches) a new technology that was an essentially perfect substitute for manpower. To pretty much exactly the extent it became increasingly deployed, the ‘workforce’ declined (an almost perfect offset for 25 years, a bit looser thereafter). In the past, new technologies have (after some painful adjustment time) lead to new jobs, new opportunities, and a recovery or gain in ’employment’; but in this case we’ve waited a couple of generations and still seen absolutely nothing in the way of recovering from this offset. Is this experience not relevant, as robots and AI will very shortly do this again? For a while, they won’t be remotely as capable or flexible substitutes for manpower as in the former case; but they will also be cheaper.
I’m sure using women as a metaphor for robots will greatly increase this blog’s popularity among the feminists. But this is a good point – why shouldn’t wages have crashed when the labor supply doubles?
The only good paper I can find on this claims it didn’t happen. I agree that’s surprising and that it needs more thought.
Probably the same arguments about how immigrants (and new graduates) entering the workforce don’t necessarily drive down wages should apply here also. The only caveat is that women are already around and consuming, so they might not create as many new jobs by entering the workforce as (say) a new immigrant arriving in the country. But there’s also the effect where somebody has to do whatever they were doing before (housework? child-rearing?) so that could possibly balance it out.
There’s little room for women entering the workforce to contribute to unemployment, since Part I of the original post explained that there is not really an unexplained unemployment increase. I wish I had a better answer to the wage stagnation question, and will be eternally grateful to anyone who can direct me to more literature or analyses on this.
Surprised to see no mention of videogames. The career prospects of young men are humiliating or tediously circuitous to puruse. So they just stay at home and play Xbox. See http://marginalrevolution.com/marginalrevolution/2016/09/labor-force-participation-video-games.html
Again, I want to emphasize that this is solving a non-problem. There is no unexplained jump in unemployment or nonparticipation. You can all stop trying to come up with explanations for it.
The data clearly shows (I can link if desired) that the larger trend happening over past thirty years is the middle class is shrinking primarily by moving into higher tiers or retiring, the lower class has grown slightly, but not if you exclude thirty to forty million immigrants. And here much of the statistical increase has been caused by lower numbers of married families (and thus fewer jumps to middle income due to failure to get hitched).
Given the categorization of jobs by type, there’s presumably a hell of a lot more low paying jobs than high paying jobs. So a smaller percentage increase in low paying jobs could presumably account for a lot more workers than the larger percentage increase in high paying jobs.
I don’t necessarily think that most people think that technological unemployment will hit the least skilled first. It should hit the jobs where it’s most profitable first. How cheap would a robotic server in a restaurant have to be to actually get a return on investment? Replacing a unionized assembly line worker makes you money a lot faster even if it’s more expensive. As we get better at replacing middle cost jobs, the cost of automation should go down and that should allow us to replace the unskilled.
I have two comments: first, the automation of manufacturing may now have meaningfully affected the PAMLFPR nationwide, but surely the localized effects may tell a different story? The economies of manufacturing heavy States like Ohio are in crisis, which would strongly suggest that either PALFPR or underemployment would go up in these areas. Increases in non-manufacturing jobs elsewhere in the nation could smooth those numbers in the US and hide regional inequities.
Yeah, the lack of obvious jumps in unemployment and PAMLFNP have to be squared with the universal common-sense perception that parts of the Rust Belt are an apocalyptic mess right now. The idea that these areas are getting anomalously worse while others (Big cities? The coasts? Silicon Valley?) are getting anomalously better might be one explanation for how the statistics could all even out.
An argument that often gets overlooked is that housing costs in high productivity cities are crowding out potential increases in employment and productivity.
What if we are as good as ever at fundamental technological aspect of productivity innovations but have gotten worse at moving people to places where those innovations are occurring?
Interstate labor mobility rates are at an all time low since the 1960s and housing costs in our most productive cities are at an all time high.
Hsieh and Moretti (2017) argue in “Housing Constrains and Spatial Misallocation” that increases in housing costs due primarily to zoning have reduced aggregate GDP growth by 50% since 1960. That is huge. Even if they are off by an order of magnitude, it could still help explain several pp of the reduction in prime age workforce. Additional workers have to come from somewhere…
In San Francisco the median housing price is over $1 million. Not everyone needs to move here and be a programmer there’s lots of other work to be done. Unfortunately, most people can’t afford to live in the city so they sleep in cars or commute in 2 hours just to work. Imagine how many more people would work if they could just live here.
The Economist argued that if not for housing costs the population of the Bay Area would be around 30 million (can’t find the link) instead of ~7 million. That’s just one region where the employment growth would impact things on a national scale.
Imagine what would have happened to Detroit if in the early years housing prices spike to over $1 million. Or if in the 1860s when millions were migrating to the United States they arrived to discover that the median housing cost $1 million. They would write home and tell friends not to come.
Housing costs in the last few decades have finally reached a point where they have become a constraint on growth that was not true during previous periods of economic expansion in the US.
What to do about housing costs? Obama’s outgoing team of economic advisors published a paper with policy recommendations which can be see here. Basically if calls for increased density and by right zoning (ie you can’t put someone’s bulding permit through 3 years of discretionary review and comment).
An unexpected place for the housing crisis to show up. At what point do I have to end all these posts with “ceterum autem censeo domuum numerum augmentum esse”?