[I am not an economist or an expert on this topic. This is my attempt to figure out what economists and experts think so I can understand the issue, and I’m writing it down to speed your going through the same process. If you have more direct access to economists and experts, feel free to ignore this]
[EDIT: Commenters pointed out that I was confusing the trend in total manufacturing jobs with the trend in percent jobs in manufacturing. This would make me less confident that there is no technological unemployment, except that I think the trend of percent jobs in manufacturing represents women entering the workforce and overwhelmingly going into nonmanufacturing jobs. Unclear that this requires a major change in conclusion.]
Technological unemployment is a hard topic because there are such good arguments on both sides.
The argument against: we’ve had increasing technology for centuries now, people have been predicting that technology will put them out of work since the Luddites, and it’s never come true. Instead, one of two things have happened. Either machines have augmented human workers, allowing them to produce more goods at lower prices, and so expanded industries so dramatically that overall they employ more people. Or displaced workers from one industry have gone into another – stable boys becoming car mechanics, or the like. There are a bunch of well-known theoretical mechanisms that compensate for technological displacement – see Vivarelli for a review. David Autor gives a vivid example:
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). With the growth of ATMs, what are all of these tellers doing?
Bessen observes that two forces worked in opposite directions. First, by reducing the cost of operating a bank branch, ATMs indirectly increased the demand for tellers: the number of tellers per branch fell by more than a third between 1988 and 2004, but the number of urban bank branches (also encouraged by a wave of bank deregulation allowing more branches) rose by more than 40 percent. Second, as the routine cash-handling tasks of bank tellers receded, information technology also enabled a broader range of bank personnel to become involved in “relationship banking.” Increasingly, banks recognized the value of tellers enabled by information technology, not primarily as checkout clerks, but as salespersons, forging relationships with customers and introducing them to additional bank services like credit cards, loans, and investment products.
This kind of thing has been remarkably consistent – so much so that arguments that today is the day technological unemployment happens should be treated with the same skepticism as arguments that today is the day we build a perpetual motion machine that works.
The argument in favor: look, imagine there’s a perfect android that can do everything humans do (including management) only better. And suppose it costs $10 to buy and $1/hour to operate. Surely every business owner would just buy those androids, and then all humans who wanted to earn more than $1/hour would be totally out of luck. There’s no conceivable way the androids would “augment” human labor and there’s no conceivable way the displaced humans could go into another industry. So at some point we’ve got to start getting technological unemployment. Here the vivid example comes from Gregory Clark
There was a type of employee at the beginning of the Industrial Revolution whose job and livelihood largely vanished in the early twentieth century. This was the horse. The population of working horses actually peaked in England long after the Industrial Revolution, in 1901, when 3.25 million were at work. Though they had been replaced by rail for long-distance haulage and by steam engines for driving machinery, they still plowed fields, hauled wagons and carriages short distances, pulled boats on the canals, toiled in the pits, and carried armies into battle. But the arrival of the internal combustion engine in the late nineteenth century rapidly displaced these workers, so that by 1924 there were fewer than two million. There was always a wage at which all these horses could have remained employed. But that wage was so low that it did not pay for their feed.
There may be some point at which we too stop being worth more than it costs to replace us. And the decline of manufacturing, the increase in labor force nonparticipation and despair in rural Rust Belt communities, etc, suggest that point is fast arriving.
This is a look at which of those arguments is right. Part I will investigate whether unemployment is getting worse. Part II will investigate whether that is because of technology. Part III will investigate what longer-term trends we should expect.
As usual, this is very long.
Is unemployment actually getting worse?
Officially it’s at historic lows.
Source: Bureau of Labor Statistics. Just assume this for any graph that looks like this.
But the real concern isn’t about unemployment per se, but the labor force participation rate (from here on: LFPR). Unemployment measures how many people are looking for work; LFPR measures how many people are out of work and not looking. If people are so discouraged that they’ve given up looking for work, this would not show up in unemployment but would show up in LFPR. Here’s LFPR:
Remember, higher means more people working
What’s going on here?
Same data as before, only disaggregated by gender. From 1950 to 2000, workforce participation rises as women enter the workforce. But throughout this time, men are leaving at almost the same rate, leaving only a moderate net participation gain (and incidentally answering the question that confused me here). Around 2000, all the women who want to be in the workforce are there already, and the declining male trend takes over for a net decline. Because women’s increasing workforce participation is shaped by unrelated cultural trends, most of the rest of this article will focus on male LFPR (from here on: MLFPR).
A big fraction of declining MFLPR is Baby Boomers retiring. If most people are young, workforce participation will be high. If most people are old, workforce participation will be low. Economists adjust for this by taking something called prime age male labor force participation rate (from here on: PAMLFPR; if you think all these acronyms are getting annoying, I guarantee you it’s more annoying to read papers that just keep saying “prime age male labor force participation rate” again and again). Here it is:
Source for this and subsequent similar-looking graphs.
In the 1950s, ~97% (!) of prime-age men had a job. Today that number is more like 88%. This is the decline people are worrying about when they talk about technological unemployment or any other threat to work, and it seems to be happening across the Western world:
Frickin’ Germany, always making everyone else look bad.
The next few paragraphs are based on data from a Scott Winship’s Mercatus Center report on this. Their conclusion is contrarian, and they’re a libertarian-ish think tank which means they have some risk of bias. I’m citing them anyway because they have really fascinating data presented much better than anyone else, but keep this in mind.
Winship’s first point is that the decline in PAMLFPR doesn’t seem to be caused by people who can’t find jobs:
The first graph shows officially designated “discouraged workers”, people who were previously unemployed and looking for jobs but eventually gave up. The definition changed around 1990, but they never seem to be more than about 10% of prime age male labor force nonparticipators (from here: PAMLFNPers).
The second graph shows what percent of PAMLFNPers claimed to be looking for jobs, based on a survey that was only given until 1993. It shows only about 20% of them were interested.
The third graph is using slightly differently-parsed data to try to continue the trends after 1993. It’s sort of unfair, because it separates out “disability” into a separate category and assumes none of them want jobs. But about 30% of people on disability do say they want a job. It’s unclear exactly what they mean (are they just saying the wish they weren’t disabled? or that they might be willing to come off disability if a properly non-physically strenuous job became available?) but even if we count them, the percent of PAMLFNPers who want a job never goes above 30%.
If not discouraged workers who can’t find jobs, what’s going on? Here’s Winship’s answer:
Prime age non-working men are mostly on disability. But some are also in school (despite having to be above 25 to be included as “prime age”), retired (despite having to be below 55), or homemakers (remember, these are all men). Again, only about 1% (out of the total of 12%) say they can’t find work.
If we were very optimistic, we could paint a rosy picture of what’s going on here. The increase in disability represents improving social safety net that allows disabled people to be better supported. It’s great that more people are financially secure enough to retire early. It’s great that more people are pursuing a graduate education that has them in school after age 25. It’s great that gender stereotypes are decreasing and more men feel comfortable as homemakers, perhaps supported by a working spouse.
This is basically Winship’s account, although he is concerned that increasing disability benefits are discouraging work. He cites a bunch of papers to that effect which you can find in his footnote 42, and which aren’t super relevant to the question at hand.
But what would the pessimistic interpretation look like?
The next few graphs and some of the analysis below comes from Brookings Institute, another potentially biased think tank
In 1970, educated and uneducated men were about equally likely to be PAMLFNPers. The rate for educated men didn’t change. The rate for uneducated men shot up.
And I won’t show you graphs, but there are similar trends for poor people, ex-convicts, blue collar workers, and minorities. These are not the sort of people who are likely to be able to retire early, pursue graduate school, or defy gender norms. But they are the sort of people who might have trouble finding work. This is pretty suspicious. Also:
Labor force nonparticipation is increasing primarily in poor and lower-middle-class people without a lot of good options, just as their remaining options get much worse. Surely this suggests something worse is going on.
The easiest place for this to happen is disability. It doesn’t require disability fraud, per se. It just requires some people on the threshold of disability who are motivated by marginal cost/benefit analysis.
Suppose that you have bad back pain. You work in the auto factory, like your father and his father before him. Your back pain flares up pretty often, but you know your foreman pretty well and he gives you an easy shift until it passes, and the union makes sure that nobody gives you any grief about it. You like your company and your coworkers and you want to make them happy. Also, if you didn’t work, you would starve to death.
Now suppose that your factory closes, and the only job available is being a home health aide. This involves a lot of bending over and puts you in constant almost-unbearable pain. And it’s run by a giant faceless corporation which always seems to be trying to screw you over. Also, you live in West Virginia and are very manly, and changing diapers in nursing homes seems like undignified women’s work. Also, the pay is half what you’re used to. Also, the government just passed a new law making disability benefits much more generous and easier to get. So…
The graph shows that PAMLFNPers generally have terrible health. So real disabilities must have something to do with this. But Winship presents a lot of evidence that illnesses and chronic pain haven’t gotten worse over time, so it can’t fully explain the rise. The “gradually worsening job pushes person with serious disability over the edge” hypothesis has a lot going for it. Also: although 96% of people on disability say they’re out of work because of a health problem, 46% also say they’re out of work because there are no good jobs available (source).
The problem with this is: disability really doesn’t represent that much of the rise in PAMLFNP since 1960. Looking at our graph above, it goes from 2% of workers in ’68 to maybe 6% of workers now. And surely some large fraction of those people are actually disabled in ways that have nothing to do with their social circumstances. We’re talking like 2 percentage points tops.
Okay, fine. Let’s say you’re our West Virginia factory worker again, only now you can’t get on disability. Now what?
Maybe you choose to retire. And maybe you’re 53 years old and this isn’t the most reasonable financial plan, but you own your house, you get food stamps, and you can do odd jobs around your friends’ farm to make some extra money.
Or maybe you choose to go to that ridiculous Coal Miner To Coder school that got profiled on NPR a little while ago, in the hopes that you can have a pathway to a new career, or just so that you have something to do.
Or maybe you choose to stay at home with your kids, while your wife does the home health aide thing, and if anybody asks, you’re a “stay-at-home dad”.
And then when economists look at the statistics, they say “Oh, look, there’s no problem here, it’s just a combination of retirees, students, homemakers, and the disabled.”
I realize this is a stretch, especially since you would expect such a person, unless they were very self-deluded, to identify as “looking for work”. But the only sense I can make of all this is a model where, the more miserable your work is, and the more decent options you have available to you, the more likely you are to leave work.
If you’re very conservative, you might say – aha, I knew that people were just unemployed because they’re lazy!
But if you’re more progressive, you might ask – exactly how miserable do you have to be before you stop working? Should people with broken legs literally drag themselves on all fours to their workplace, just because it’s not physically impossible? I know that “I refuse to do this job because it’s too undignified for me, let me go on the public dole” doesn’t really win you a lot of social credit. But maybe conservatives could find it in their heart to be sympathetic to our hypothetical West Virginia factory worker with a bad back, who’s proud of having worked his whole life, but who feels like having to pivot at age 53 to changing diapers in nursing homes for minimum wage isn’t his cup of tea?
(remember, the other thing that’s way up among this same demographic is suicide – and probably for the same reason)
But even if we assume half the increase in disability plus a quarter of the increase in the other things is due to employment issues, employment issues still really only explain about 3 pp of the 10% increase since the 1960s. I can’t think of any reasonable assumptions where they explain more than half.
I like Derek Thompson’s discussion of this question, because he’s the only writer who seems to share my sense of puzzlement. There are all these men who seem miserable and who have vanished from the labor force. We all know it’s true. But the statistics don’t really seem to reflect or shed light on it. Somehow we, as a country, have managed to just lose several million working-age men. Maybe Donald Trump is going to look behind the White House couch one day, and find a large portion of the male population of the Southeast under the cushion. I don’t know.
In the next part, we’ll talk about whether automation explains the decline in labor force participation. But let’s keep in mind that the argument that there is a significant meaningful decline in labor force participation to explain, aside from people going to school and having more access to disability benefits and things like that – is not on super-solid ground.
Recently, US manufacturing jobs collapsed. US manufacturing is still doing just fine in terms of number of widgets produced. It just no longer employs that many people.
The two graphs differ a bit in emphasis. The first, in raw numbers, makes it look like there was a big discontinuity around 2000. The second, in percent, suggests that percent manufacturing has been shrinking for a long time. This could represent a long-term decline in manufacturing – but more likely, it represents women entering the workforce; women are much less likely than men to have manufacturing jobs. The reality is probably somewhere in between the two.
Is the decline in manufacturing – whichever timescale we choose to look on – due to automation, or to other causes like globalization? A Ball State University report argues for the former:
Had we kept 2000-levels of productivity and applied them to 2010-levels of production, we would have required 20.9 million manufacturing workers. Instead, we employed only 12.1 million.
The report gets summarized in a few places as “13% of job loss is due to trade, 87% is due to increasing productivity/automation”, which seems like a fair summary of some of its claims. Although some commenters raise doubts, its numbers are in the same direction as the conclusions of economists Autor, Dorn, and Hanson, who published a series of papers finding that that “import competition explains one-quarter of the contemporaneous aggregate decline in US manufacturing employment”.
Hasn’t productivity actually been growing more slowly than usual recently? See for example this this Brookings paper, which notes that:
The past decade has seen slowdowns in measured labor productivity growth across a broad swath of developed economies. Aggregate labor productivity growth in the U.S. averaged 1.3% per year from 2005 to 2015, less than half of the 2.8% average annual growth rate it sustained over 1995 – 2004. Similarly sized decelerations were observed between these two periods in 28 of 29 other countries for which the OECD has compiled productivity growth data
The drops in productivity growth have struck some as paradoxical, given the seemingly brisk pace of technological progress and plethora of new products that have been introduced and diffused throughout the world during the slowdown period. Indeed, many have suggested that the slowdown is substantially illusory, a figment of the inability of current economic statistics to capture the true rate of technological advance in standard productivity metrics.
If the productivity slowdown were illusory, that would help explain the apparent speedup in automation, and all those job losses. Sounds promising, but what do all the economists in the world think?
Justin Fox has some more in-depth analysis here and also concludes productivity is not that great.
Are we allowed to say “that’s just how things work”? Like, agricultural productivity increased for millennia, but didn’t lead people to abandon agriculture en masse until the Industrial Revolution – when it did exactly that. In 1790, 90% of Americans were farmers, even though agricultural productivity had been improving for ages. Today, 2.6% of Americans are. Maybe manufacturing just had the same kind of moment. Advances in technology can put farmers out of work (but shift them to manufacturing) – surely it can put manufacturers out of work (and shift them to ???).
So, if 70% – 80% of manufacturing job losses were due to automation, might automation be responsible for the decline in PAMLFPR, thus revealing the elusive technological unemployment?
The time trend in absolute number of manufacturing jobs reveals a sharp drop around 2000, which doesn’t match change in PAMLFNP. The trend in percent reveals a gradual decline, which does sort of match the PAMLFNP trend, but because the percent is likely to indicate women, who are not included in PAMFLNP, it might not be as significant. But there are two other reasons why these don’t seem to match very well.
First, disability represents the main route by which people could plausibly be leaving the labor force to become PAMLFNPers. Here’s the graph:
There is a general trend of increasing disability since 1985, but a paper by Autor and Duggan suggests this is almost entirely due to a reform of the disability system around 1984 which made it easier to get benefits (also, the size of the workforce increased due to more female participation, meaning the pool of potential disabled people increased too).
Second, it is hard to explain why now. Whether we’re worrying about the past twenty years of automation (collapse in raw numbers of manufacturing jobs) or the past sixty (decline in PAMLFPR, decline in percent manufacturing) – neither corresponds to the several hundred years in which things have been getting more automated. Why didn’t previous eras of improving automation result in job loss?
Let’s go back to all the economists in the world and see what they think:
Economists very strongly believe automation has not historically reduced employment. But they do believe automation is making wages stagnate right now. I don’t really understand what’s going on here. Are they saying that automation can depress wages, but not reduce employment? Surely (given the existence of a minimum wage) that doesn’t make sense. Or are they saying that automation never caused any problems before, but it is causing problems now?
The site offers some of the economists the chance to explain what they meant, and a lot of them seem to be saying that automation has temporarily caused problems in the past, but they always resolved with time as new industries open up. Maybe we’re just in a temporary bad period? Likewise, one economist who agrees that automation caused wage stagnation says that “it may have a short-run impact but there is no reason to believe that it is permanent.”
All of this is a mess. But the impression I get from this mess is that there is little sign of technological unemployment happening today in a historically unique way, or even picking up pace. I get this from a few sources.
First, the official unemployment rate looks great, so if we were going to make this argument we would have to do it off of PAMLFPR.
Second, Winship’s optimistic take on PAMLFPR is hard to easily refute. PAMLFNPers pretty clearly say they’re not looking for jobs, and they’re just perfectly innocuous students, retirees, etc. We have trouble believing them, especially based on their demographics. But it’s very hard to look at the increase and see a place where unemployment issues could have slipped in.
Third, PAMLFPR has been getting worse gradually since about 1960, with no sign of any recent worsening. It is hard to explain why technological unemployment would have started around that time – at least if we limit our explanations to the nature of technology alone. And it doesn’t seem to match the more sudden decline in manufacturing around 2000.
Fourth, most economists appear to remain doubtful of the possibility of long-term technological unemployment.
I realize this goes against common sense. Maybe I’m missing something and totally wrong here. But if I am forced to interpret the data as I see it, I just don’t see the signs of technological unemployment. It’s just not there.
And in my defense, this also seems to be the opinion of David Autor, the main economic expert on this subject. In an interview with The Economist, he said that there was “‘zero evidence’ that AI is having a new and significantly different impact on employment”.
This doesn’t mean everything is great. As the IGM panel shows, even if robots aren’t putting people out of work, they may be causing wages to stagnate. The people getting kicked out of manufacturing jobs may have other jobs available to them (and so not end up affecting the PAMLFPR numbers), but those jobs may not be as good or pay as well. This isn’t “technological unemployment”. But it might be technological underemployment.
Most people expect that technological unemployment will hit the least skilled first, but that doesn’t seem to be entirely true. This chart and some of the following analysis are going to be from the Heritage Association – another potentially biased think-tank, but hopefully without much reason to obfuscate these issues.
The best-paying jobs – managers, professionals, and the like – are doing fine. The lowest-paying jobs, like personal care and food, are also doing fine. It’s the middle-paying jobs that are in trouble. Some of these are manufacturing, but there are also office and administrative positions in the same categories.
This is potentially consistent with a story where the jobs that have been easiest to automate are middle-class-ish. Some jobs require extremely basic human talents that machines can’t yet match – like a delivery person’s ability to climb stairs. Others require extremely arcane human talents likewise beyond machine abilities – like a scientist discovering new theories of physics. The stuff in between – proofreading, translating, records-keeping, metalworking, truck driving, welding – is more in danger. As these get automated away, workers – in accord with the theory – migrate to the unautomatable jobs. Since they might not have the skills or training to do the unautomatable upper class jobs, they end up in the unautomatable lower-class ones. There’s nothing in economic orthodoxy that says this can’t happen.
David Autor and his giant block of citations agree:
Because jobs that are intensive in either abstract or manual tasks are generally found at opposite ends of the occupational skill spectrum—in professional, managerial, and technical occupations on the one hand, and in service and laborer occupations on the other—this reasoning implies that computerization of “routine” job tasks may lead to the simultaneous growth of high-education, high-wage jobs at one end and low-education, low-wage jobs at the other end, both at the expense of middle-wage, middle education jobs—a phenomenon that Goos and Manning (2003) called “job polarization.” A large body of US and international evidence confirms the presence of employment polarization at the level of industries, localities, and national labor markets (Autor, Katz, and Kearney 2006, 2008; Goos and Manning 2007; Autor and Dorn 2013; Michaels, Natraj, and Van Reenen 2014; Goos, Manning, and Salomons 2014; Graetz and Michaels 2015; Autor, Dorn, and
The fall of middle-skill-level jobs has led to a corresponding fall in middle-income jobs:
Note that, contrary to an extremely pessimistic picture, this would suggest that most people who leave middle-paying jobs go to higher-paying jobs
And with a corresponding decline in the fortunes of the middle class:
Note that this does not really back up the optimistic picture from above.
Why is this happening now when technological progress has been going on forever? This gets into the whole decline-of-the-middle-class argument, a giant political morass featuring de-unionization, regulation, automation, globalization, the 1%, and pretty much everything else. Is there also a role for today’s robots just plain being better than yesterday’s Rolodexes and whatever else the forefront of technology was? Or our education system being less able to cope with them? I’m not sure.
As far as I know, there is no economic theory stating that the number (or percent) of high-skilled jobs must always stay the same. I’m also not sure how to include fixed cognitive skills (eg some people are smarter than others) in this question. An optimist might argue that things will get better as today’s obsoletely-trained workforce retires and tomorrow’s trained-for-the-appropriate-jobs workforce graduates. But maybe this is better viewed as a race between two competing forces; generational churn producing students with the right set of skills, and technology making new skills obsolete. I don’t know why this should have increased recently, but it seems like – at least for the middle class – this is a race they are now losing.
Predicting the future is naturally harder than observing the present, since we have data about the present and not the future. But the data about the present is contradictory and incomprehensible and just makes things more confusing, so we might as well try going with the future and seeing how we do there.
We’ll start with those surveys of economists again, since they seem like the people most likely to know. Here’s a panel of top European economists on the future of technological unemployment:
In the same survey, 93% of economists with an opinion on the issue agreed that the economic benefits of robots will be so great that they could be used to compensate the workers who were negatively effected. But in a survey I conducted in my imagination, 100% of people who have not been living in a cave the past two hundred years agreed that this will never happen in real life.
So economists really have no idea about any of this. What are we paying them for, anyway?
Frey and Osbourne analyze what jobs are most susceptible to automation. They claim that “47% of total US employment is at risk”. This sounds suspiciously precise and it’s unclear their numbers have any relationship to reality. They do find “evidence that wages and educational attainment exhibit a strong negative relationship with an occupation’s probability of computerisation”.
Overall none of this seems to be making things much clearer.
I’ve cited David Autor something like five times already. He is the recognized expert in this area – I blame nominative determinism – and has written widely. His own opinion is:
I expect that a significant stratum of middle-skill jobs combining specific vocational skills with foundational middle-skills levels of literacy, numeracy, adaptability, problem solving, and common sense will persist in coming decades. My conjecture is that many of the tasks currently bundled into these jobs cannot readily be unbundled—with machines performing the middle-skill tasks and workers performing only a low-skill residual—without a substantial drop in quality. This argument suggests that many of the middle-skill jobs that persist in the future will combine routine technical tasks with the set of nonroutine tasks in which workers hold comparative advantage: interpersonal interaction, flexibility, adaptability, and problem solving. In general, these same demands for interaction frequently privilege face-to-face interactions over remote performance, meaning that these same middle-skill occupations may have relatively low susceptibility to offshoring. Lawrence Katz memorably titles workers who virtuously combine technical and interpersonal tasks as “the new artisans” (see Friedman 2010), and Holzer (2015) documents that “new middle skill jobs” are in fact growing rapidly, even as traditional production and clerical occupations contract.
This prediction has one obvious catch: the ability of the US education and job training system (both public and private) to produce the kinds of workers who will thrive in these middle-skill jobs of the future can be called into question. In this and other ways, the issue is not that middle-class workers are doomed by automation and technology, but instead that human capital investment must be at the heart of any long-term strategy for producing skills that are complemented by rather than substituted for by technological change. In 1900, the typical young, native-born American had only a common school education, about the equivalent of sixth to eighth grades. By the late 19th century, many Americans recognized that this level of schooling was inadequate: farm employment was declining, industry was rising, and their children would need additional skills to earn a living. The United States responded to this challenge over the first four decades of the 20th century by becoming the first nation in the world to deliver universal high school education to its citizens (Goldin and Katz 2008). Tellingly, the high school movement was led by the farm states. Societal adjustments to earlier waves of technological advancement were neither rapid, automatic, nor cheap. But they did pay off handsomely.
Do we really have evidence that compulsory schooling was a result of increasing automation? If so, could we tell a story where the gradually increasing length of schooling – from minimal, to primary school, to high school, to “you better get a college degree or you’ll regret it later”, to increasing pressure to go to graduate school – is a reaction to automation and the threat of technological unemployment? Could this be the reason why automation finally seems to be causing problems – a financial and cultural inability to extend schooling any further than it’s already gone?
Or – inspired by Bryan Caplan’s The Case Against Education – could we tell the opposite story? One where increasing credentialism makes it harder for people whose jobs have been automated to switch careers the way they did before? Higher-paying jobs no longer just require skills, they require a college degree in a relevant field – which is very hard for a mature worker to get. Being an office manager and being a nurse are both middle-income jobs – but in the past, an office worker would have needed about six months of inexpensive training plus a lot of on-the-job apprenticeship to become a nurse, whereas now they would need a four-year Bachelors of Science in Nursing from a university whose price tag has dectupled for no reason over the last half-century. A now-unemployed office manager might have been able to afford the first even if middle-class; the second might be well beyond her reach. Unable to shift into another middle-class job, she is forced to take a lower-class job as a fast food worker or something.
I am not entirely sure how differences in cognitive ability fit in here. My guess is to a first approximation they don’t – if standard economic theory is correct, it should be possible to create middle-paying jobs that use the full potential of people with any amount of cognitive ability, taking advantage of various human cognitive skills that are difficult to automate. Although some naive takes like “everyone should just become a programmer” fail to understand this, I don’t think the entire argument is based on misunderstanding of this point, or that it forms a particularly strong counterargument.
Anyway, if Autor’s prediction is “we will be able to weather this danger as long as our education system is able to rise to meet the challenge”, I’m just going to round this off to “we’re super doomed”. But I think his methodology – of noticing that we always met the challenge before, and trying to figure out what might be different this time – is a promising one.
Finally: we’ve been talking about economists a lot here, but what about the roboticists? Aren’t they relevant too? Grace et al survey top AI researchers on when AI might be able to replace humans in various things; these researchers don’t necessarily know anything about economics, but they at least know something about progress in robotics. On average, they believe AI will reach human performance at truck driving, retail selling, translation, transcription, and bipedal running all before 2030. Whether those robots will be affordable, widely adopted, or able to deal with the long tail of real-world situations like blackouts or vandals or bad weather – is a different story. The point is, roboticists are pretty sure they’ll have their contribution to the economic takeover ready pretty soon.
(they do say robots won’t be writing bestselling novels until 2050, so JK Rowling’s job is safe for now)
They also say that robots will be able to do all human tasks, including novel-writing, science, and further AI research – sometime between 2050 and 2150. At that point, obviously, all bets are off, and we have a lot more than unemployment to worry about.
Here are some tentative conclusions:
1. Technological unemployment is not happening right now, at least not more so than previous eras. The official statistics are confusing, but they show no signs of increases in this phenomenon. (70% confidence)
2. On the other hand, there are signs of technological underemployment – robots taking middle-skill jobs and then pushing people into other jobs. Although some people will be “pushed” into higher-skill jobs, many will be pushed into lower-skill jobs. This seems to be what happened to the manufacturing industry recently. (70% confidence)
3. This sort of thing has been happening for centuries and in theory everyone should eventually adjust, but there are some signs that they aren’t. This may have as much to do with changes to the educational, political, and economic system as with the nature of robots per se. (60% confidence)
4. Economists are genuinely divided on how this is going to end up, and whether this will just be a temporary blip while people develop new skills, or the new normal. (~100% confidence)
5. Technology seems poised to disrupt lots of new industries very soon, and could replace humans entirely sometime within the next hundred years. (???)
This is a very depressing conclusion. If technology didn’t cause problems, that would be great. If technology made lots of people unemployed, that would be hard to miss, and the government might eventually be willing to subsidize something like a universal basic income. But we won’t get that. We’ll just get people being pushed into worse and worse jobs, in a way that does not inspire widespread sympathy or collective action. The prospect of educational, social, or political intervention remains murky.