Will the robots take our jobs?

This post is about robots and the economy, but takes some detours first. Bear with me.

Robert Gordon and the End of Growth

There has been a very interesting discussion going on recently, prompted by an article by economist Robert Gordon of Northwestern University. Gordon’s article (“Is US economic growth over?”) makes the case that long-term US economic growth on the scale of the last century was due to one-time events and has run its course, with future growth prospects being much lower. He attributes the growth of the past few centuries to three distinct industrial revolutions. The first, beginning 1750-1830, was due to steam power and railroads. The second, 1870-1900, was due to electrification, internal combustion engines, running water and petroleum. The third, beginning around 1960, was due to the computer and the internet. Gordon makes the case that the second industrial revolution, from 1870-1900, was by far the most important, and that computers and the internet have had far smaller impacts on GDP. Combined with demographic headwinds, he sees much lower rates of growth in the next century.

Martin Wolf summarizes the pessimist’s case succinctly:

Unlimited growth is a heroic assumption. For most of history, next to no measurable growth in output per person occurred. What growth did occur came from rising population. Then, in the middle of the 18th century, something began to stir. Output per head in the world’s most productive economies — the UK until around 1900 and the US, thereafter — began to accelerate. Growth in productivity reached a peak in the two and a half decades after World War II. Thereafter growth decelerated again, despite an upward blip between 1996 and 2004. In 2011 — according to the Conference Board’s database — US output per hour was a third lower than it would have been if the 1950-72 trend had continued (see charts). Prof Gordon goes further. He argues that productivity growth might continue to decelerate over the next century, reaching negligible levels.

Robots to the rescue?

What interests me most is the responses that Gordon’s article has received. His position is very interesting, but likely wrong in one massive aspect.

To say that the computer revolution won’t match the developments of 1870-1900 is to ignore the obvious potential of robotics. So far computers have mostly been good for manipulating pixels on a screen. They haven’t done quite so much to manipulate atoms in the real world, which is the kind of GDP economists can easily measure. The quality improvement of online news compared to newspapers is pretty clear to me, but economists don’t capture this properly in GDP figures. On the other hand, it is easy to count the number of cars coming off a production line. So the obvious riposte to Gordon is that computers will really show up in the GDP figures once robotics comes of age. Many commentators have responded to Gordon making just this argument. So what’s been cool for me is to suddenly see a substantial discussion among opinion makers about a robotic economy (and also about the plausibility of exponential economic growth continuing forever, another question that has always seemed under-discussed).

I’ve collected a few of my favourite pieces below:

Paul Krugman – Is Growth Over? (also here, here)

Jeremy Grantham – On the Road to Zero Growth

Kevin Kelly – The Post-Productive Economy (and accompanying Wired article)

Yann LeCun’s G+ post (Great discussion from various people in the robotics and machine learning community.)

Kenneth Rogoff – Innovation Crisis or Financial Crisis?

Robots and the Economy

So robots may provide the next big surge of economic growth, but what will happen to employment and the wealth distribution as robotics starts to make a big impact on the economy? This is what I really want to discuss. I think to answer this you need to address a few key questions:

  • What percentage of jobs are automatable in the near term?
  • What is the time constant of this change? Will jobs disappear gradually or in big lumps?
  • Will new jobs emerge to take up the slack? What kind of jobs will they be? Will they be suitable for workers being displaced by automation?
  • How will this affect the US economy? The Chinese economy?
  • Have we seen comparable changes in the past, and what happened?
  • What is likely to happen to the overall distribution of wealth?

I don’t pretend to know the answers to all those questions, but let’s take a stab at it. I’ll confine the discussion to a 30-40 year time horizon.

How much of the economy is left to automate?

The idea of robotics in the workforce still makes a lot of people uneasy – the common fear is that robots will take all the jobs, low-skill workers will be pushed out of the economic system, and only a small set of owners will become extremely rich. Yann LeCun’s recent G+ post led to an excellent discussion along those lines. There is a lot of insight in the comments, I’d encourage people to go read them. However, it’s easy to get carried away with speculation. How plausible is a scenario where robotics leads to mass unemployment?

Automation is hardly new. The Luddites were marching against the introduction of automated looms in 1811. That was a messy social transition – at one stage there were more British soldiers fighting the Luddites than Napoleon on the Iberian Peninsula. However, in the long run it worked out OK. Automation has steadily increased for 200 years without leading to mass unemployment, despite repeated predictions. John Maynard Keynes expected we would all be working 15 hours a week by now. Approximately every decade since, someone has made a similar pronouncement about the impending demise of human labour. Will it be different this time around? It might be, if the scale and speed of job replacement is qualitatively different to before. In some parts of the economy the degree of automation is stunning. In 2011 Dropbox had 50 million users, a $4 billion valuation, and a grand total of 60 employees. These kind of numbers are fairly typical of a successful tech startup. Facebook provides its service to almost everyone on earth with less than 4,000 staff. The magnification factor of computing and the internet is truly incredible So is this the future of the physical economy too? Let’s get some numbers.

Exactly how much of the economy consists of jobs that can be done by robots in the near term? People often associate robotics with manufacturing jobs. However, modern manufacturing is already highly automated. If you visit a modern factory, the overwhelming impression is the lack of people. Only a few human employees are needed to babysit the machines, troubleshoot problems and do a few bitty jobs which are not yet worth automating. Those manufacturing tasks which can’t yet be cost-effectively automated have moved to low-wage countries like China, so that from the point of view of US workers they may as well be automated. Manufacturing is 9% of the US economy, down from 15% in 1990. Robotics in manufacturing will mostly be a case of new machines replacing older machines. The bulk of the jobs have already gone. So how much of US employment is still in occupations which are likely to be automatable?

US maunfacturing employment

Let’s try to guesstimate, for a time horizon 30 – 40 years out. Of course there’s huge uncertainty about how far robotics will get in that timeframe, but for the sake of argument let’s suppose that robots will replace many routine jobs, but not more complex ones. Lawyers and software engineers will still be employed, but robots will do substantially all the manufacturing work, drive all the trucks, taxis, forklifts and tractors, pick fruit in the fields, carry out routine surgical procedures, do most of the non-sales jobs in big retail stores, and (a little further out) do routine construction work, domestic chores, laundry, perhaps cooking and assistance of the elderly. Jobs I would guess will not be replaced include anything with a direct human focus – roles like salesperson, waiter, and (with more uncertainty) teacher, childcare, etc.

So what percentage of the workforce is that? The Bureau of Labor Statistics has the numbers:

Job CategoryPercentage of US workforce
Retail (food and general merchandise)4%
Food Service6.8%

The above is a pretty rough estimate. I just went through the BLS job categories and picked the ones that seem likely to have large job losses in a robotic world. The one category on that list that I think will hold out longest is food service, mostly because I think it will remain quite hard for robots for some time. So if you total up the numbers (excluding a portion of the food service jobs) you get a figure of about 23% of the US workforce which could be replaced by robotics. Of course not every single job in a sector is going to be automated, and maybe I missed some categories, so let’s say a reasonable estimate with error bars is maybe 15%- 30% of current US jobs which may be replaced. I would guess most likely at the lower end of that range.

This is a pretty big chunk of current employment, but it’s not a future in which nobody works. Remember that US manufacturing as a percentage of the economy already declined from 15% to 9% from 1990 to 2010. Most of that decline happened in just the 10 years from 2000 – 2010. The shift to robotics will happen over decades, so the rate of change is probably no more dramatic than what we have seen already throughout the 20th century. Some industries may move relatively quickly to almost complete replacement (e.g. transportation jobs I expect will be pretty much all gone within 20 years), while others may happen more slowly, decades further out (e.g. maid service jobs in the accommodation sector). Entire industries have disappeared in relatively short order over the last century, e.g. stevedores, typing pools, etc, etc. The economy can certainly adapt to changes of the magnitude we’re discussing. I can’t help feeling it will lead to somewhat higher structural unemployment, but it’s not apocalypse. The prices for services which embrace automation will fall and, assuming the wealth distribution is not too skewed, new jobs will arise somewhere else in the economy to absorb the excess spending power. The US economy has been on this path for some time. Here’s Larry Summers writing in the FT last year:

The agricultural economy gave way to the industrial one because progress enabled demands for food to be met by only a small fraction of the population freeing large numbers of people to work elsewhere. The same process is now under way with respect to manufacturing and a range of services, reducing employment prospects for most citizens. At the same time, just as in the early days of the industrial era the combination of substantial dislocations and greater ability to produce at scale is enabling a lucky few to acquire great fortunes.

The nature of the transformation is highlighted by the 50 fold change in the relative price of a television set of a constant quality and a day in a hospital over the last generation. While it is often observed that wages for median workers have stagnated, this obscures an important aspect of what is occurring. Measured via items such as appliances or clothing or telephone services, where productivity growth has been rapid, wages have actually risen rapidly over the last generation. The problem is that they have stagnated or fallen measured relative to the price of food, housing, healthcare, energy and education.

As fewer people are needed to meet the population’s demand for goods like appliances and clothing it is natural that more people work in producing goods like healthcare and education where outcomes are manifestly unsatisfactory. Indeed as the economist Michael Spence has documented, a process of this kind is under way: essentially all US employment growth over the last generation has come in non-traded goods.

If we are lucky we may naturally transition to a world with more pilates instructors and fewer taxi drivers. The US may even experience a small renaissance in manufacturing jobs as robotics make it more competitive with low wage countries such as China. Which leads me to…

Robotics and China

So what does this transition look like for the Chinese economy, which unlike the US, is not yet heavily service-orientated. I couldn’t find comparable employment figures for China, but here are some rough Chinese numbers:

Job CategoryPercentage of Chinese workforce

I think the major difference with China is that the manufacturing jobs there represent a higher rung on the economic ladder. Mass manufacturing employment is one of the drivers helping to spread around the gains of the Chinese economic boom. The major social pain from the introduction of robotics may thus not be in the US at all, but in China. There is also going to be impact from “reshoring“. As robotics eliminates the low-wage advantage of China, industries which have moved there may leave. There is some evidence this is already happening as the Chinese wage differential decreases relative to transport costs. Rodney Brooks talks about this as an explicit goal of Rethink Robotics (originally called Heartland Robotics). If this proceeds rapidly it could be a real problem for China. It would be a tragedy if the Chinese economic miracle was derailed, it has lifted more people out of poverty more quickly than any other event in history. On the other hand, China has the ability to pursue more radical policy options than the West. It’s possible they may manage the transition successfully.

More Questions

I didn’t answer all my own questions. I’m still very uncertain about how things might play out. Will new jobs really emerge for the people displaced by automation? They have in the past, but it is possible we are reaching some kind of transition point where all low-skill jobs will be eliminated and some (possibly large) segment of the population is simply not skilled enough for the remaining employment opportunities? That is a depressing thought, but it is possible. I think a more likely path is that perhaps the nature of employment will change, with fewer large employers offering stable jobs, and more people working in rewarding but unstable self-employment. This is the kind of future where many people work as artisans on Etsy, assistants on TaskRabbit, peer-to-peer personal chefs on Gobble, and other occupations which currently sound exotic. Robotic transportion and logistics, in particular, will enable a lot of these kind of services which were previously uneconomic. Right now those jobs are much riskier than working for a large firm, but I think we will find structures to spread risk and improve stability in a world where many people move to that mode of employment.

How will robotics impact the price of goods? Are we headed towards a world where physical goods are similar to online music, where the only scarcity is the “fake scarcity” of DRM and copyright? I don’t know the answer here. e.g. I’ve seen articles that claim that the manufacturing cost of an iPhone is only 4% labour, but that refers only to the assembly costs. What percentage of the cost is labour if you take into account the whole supply chain, all the way back to mining the raw materials? We are definitely going to see dramatic drops in the price of some goods, especially those which heavily involve logistics and assembly, but probably no more dramatic than the drop in the price of clothes over the last few decades. Still, H&M is pretty wonderful.

Will wealth distributions continue to skew? This does seem like a natural consequence of more automation. Here’s Paul Krugman’s take:

So machines may soon be ready to perform many tasks that currently require large amounts of human labor. This will mean rapid productivity growth and, therefore, high overall economic growth.

But — and this is the crucial question — who will benefit from that growth? Unfortunately, it’s all too easy to make the case that most Americans will be left behind, because smart machines will end up devaluing the contribution of workers, including highly skilled workers whose skills suddenly become redundant. (….)

The college premium hasn’t risen for a while. What has happened, on the other hand, is a notable shift in income away from labor:

If this is the wave of the future, it makes nonsense of just about all the conventional wisdom on reducing inequality. Better education won’t do much to reduce inequality if the big rewards simply go to those with the most assets.

This is a political problem. Naive as it may sound, I’m hopeful it won’t get too bad. We already do a lot of redistribution, we can certainly dial it up (much easier than to dial it down!). We can change tax rates, increase public services, and covertly spread wealth via make-work (e.g. compliance officers, health and safety inspectors and other kinds of modern bureaucracy). It’s not exactly a utopian vision, but I think we will muddle through. Ultimately, with political consensus, a society can engineer whatever wealth distribution it chooses. e.g. US vs Scandanavia. This blog post is already very long, I will leave this topic there. The arrival of robotics will probably be no more dramatic than what we have already seen, assuming it does not lead to mass unemployment. If that does happens then we are in for a very different world, with much more radical social change required.

The Long Run: How do you divide a pie that a robot baked?

So what happens in the long-term end game, when robots do not just the menial tasks, but most jobs? How do you divide the pie if the robot did all the work? Here is Jeremy Grantham’s take form the latest GMO quarterly letter:

…this will continue into the indefinite future until one day, perhaps, we will reach what has been called a singularity. The last handful of humans engaged in manufacturing — all engineers and designers — are supervising intelligent robots making and designing yet another generation of even more productive and intelligent robots. On this particular day, R3142 sends the fateful silent communication to his fellow robots suggesting that their friendly human acquaintances, Fred and the boys, are beginning to get in the way. After which, there is no more productivity per man-hour at all, but only productivity per robot-hour or per unit of capital employed. This deepening of capital and technology almost guarantees that productivity will continue to be high in manufacturing even as the percentage of the total workforce employed there dwindles away toward zero. As the rest of us do each other’s art appraisals and investment management we can fantasize about productivity, but it will mainly represent hard to measure qualitative improvements. (On a hypothetical island where services are outlawed and only manufacturing exists, the final position is that automation, and thereby capital, produces everything while all of the mere mortals sit on the beach. And starve? The worthless unemployed who are obviously not carrying their weight? Ah, there’s the rub! Up the beach, in a protected, cordoned-off section is the capital owners’ club. There, a handful of equally “unemployed” owners sit, enjoying tea and the ocean. How material goods and sustenance are divvied up will determine the future of that island, for the unemployed will be 100 or 1000 times the number of dividend counters.) Is there not a growing element of this unfortunate hypothetical island in our current world, for basically the same reason? Capital deepening and technology (and offshoring) steadily replace manufacturing and farming jobs until one day perhaps there will be no manufacturing jobs at all. The task of maintaining growth then has to be borne solely by service jobs where measuring productivity has always been quite flakey. It seems true, though, that the most important values that are generated, particularly when things start to go wrong, are in necessities, all of which seem to fall under the heading of “manufacturing”: think about the trade between (necessary) bread and (luxury) haircuts as times get tough: seven loaves per haircut quickly become seven haircuts per loaf! Earlier, more philosophical economists than the current generation, like John Stuart Mill, Adam Smith, and Keynes, seemed to take pleasure in the idea of a distant future where citizens had vast amounts of leisure time to enjoy the world’s beauty. They saw that as a sensible response to increased wealth. Unfortunately, they did not tell us much about the problem that, when that day arrives, capitalists — at least those in manufacturing — will own everything and the “unemployed” manufacturing workers nothing.

And a nice metaphor from Bob Mottram that I have referenced here before, with a more optimistic slant:

So imagine a situation in which a single human caretaker runs a large fully automated business empire having no human employees. What if even this single owner were made redundant or were to die leaving no successors. Would it be possible to have a large industry which was neither owned nor run by humans, and if so what would be its legal status? I think there is a precedent for this kind of situation because we already have large and extremely productive industries which fall into this category – the world’s ecosystems. The natural ecosystem sustains human life on earth and I think that future (originally man-made) industries will increasingly come to resemble ecosystems, in that they will be things which we rely upon but have little direct involvement with their internal workings.

Or go read Richard Brautigan’s poem “All Watched Over by Machines of Loving Grace”. I’ll leave it there. This future is far enough out that there’s not much value in trying to predict how society will handle it. Maybe at that stage we will be the robots too.

  • We already do a lot of redistribution, we can certainly dial it up (much easier than to dial it down!).

    What makes you think it would be easy to increase the amount of redistribution in places like the US and Canada?

    Those who are likely to pay the costs associated are either the very rich (who are politically influential) or the middle classes (who are arguably even more politically influential, as a class).

  • Well, the US has dialed it up and down at various points during the last century (e.g. up under the New Deal, down with Regan and after). Europe demonstrates that a society can do much more redistribution than the US status quo, if there is social consensus. These things are fairly malleable over the medium term.
    Bigger issue I think will be the impact on distribution between rich and poor countries, that is much harder to do anything about.

  • You left out siri and watson in the analysis, as well as the increasing capabilities of software like catia and photoshop to do more with less human input. There’s also efficiency effects of the web in things like khan academy and amazon which haven’t gone mainstream yet but will things tike self-driving cars and great broadband access for teleconferencing. The service sector is not safe for multi-decade timescales you mention.

  • @rick: Maybe, however I tend to think those technologies will remain human assistants rather then human replacements for quite a long time.

  • Some interesting further reading is the history of the industrial revolution:


    Clearly it made the average person better off in the long run, but there is still debate about what happened in the short term (first 50 years or so). Moreover, historians mostly agree that the distribution of income became more unequal between 1790 and 1840. Robotics might well follow a similar path.

  • Another example of huge changes that have taken place in the past:

    “For most of the 19th century, about 25% of all agriculture labor threshed grain. That job was automated in the 1860s.”

Comments are closed.