Robot Macroeconomics: What can theory and several centuries of economic history teach us?

John Lewis.

Advances in machine learning and mobile robotics mean that robots could do your job better than you.  That’s led to some radical predictions of mass unemployment, much more leisure or a work free future.   But labour saving innovations and the debates around them aren’t really anything new.  Queen Elizabeth I denied a patent for a knitting machine over fears it would create unemployment, Ricardo thought technology would lower wages and Keynes famously predicted a 15 hour working week by 2030.   Understanding why these beliefs proved to be wrong gives us important insights into why similar claims about robotisation might be incorrect.  But automation could nevertheless have sizeable distributional implications and ramifications well beyond the industries in which it’s deployed.

Technological progress won’t create mass unemployment…

Technology can lead to workers being displaced in one particular industry, but this doesn’t hold for the economy as a whole.  In Krugman’s celebrated example, imagine there are two goods, sausages and bread rolls, which are then combined one for one to make hot dogs.  120 million workers are divided equally between the two industries:  60 million producing sausages, the other 60 million producing rolls, and both taking two days to produce one unit of output.  Now suppose technology doubles productivity in bakeries.  Fewer workers are required to make rolls, but this increased productivity will mean that consumers get 33% more hot dogs.  Eventually the economy has 40 million workers making rolls, and 80 million making sausages.  In the interim, the transition might lead to unemployment, particularly if skills are very specific to the baking industry. But in the long run, a change in relative productivity reallocates rather than destroys employment, even if the distributional impacts of that reallocation can be complicated and significant.

…and it probably won’t make your working week much shorter…

As productivity rises, people could just work fewer hours and enjoy the same level of consumption. But equally, they could work the same hours and devote the productivity boost entirely to raising consumption or, more likely, enjoy a bit more of both.  This so-called “income effect” means working hours should fall, but by less than one for one with the rise in productivity.

But that’s not the only thing going on – rising productivity tilts the relative prices of leisure and consumption in favour of the latter – what economists call the “substitution effect”.   Gregory Clarke’s fascinating dataset suggests that in 1700 a craftsman needed to work for almost 10 hours to earn the 2 old pence required to purchase a kilo of beef.  But by 2014 a median UK worker can earn the ten pounds or so need to buy that kilo of beef in less than hour.  And so measured in beef, or goods in general, the reward for working that extra hour is much bigger.

The overall effect on hours depends on the balance of the two.  Angus Maddison’s 2001 magnum opus estimates that between 1820 and 1998, real GDP per capita in Western Europe increased 15-fold.  Over the same period hours declined by about a half. So the productivity dividend was split about 7:1 in favour of consumption.  On that basis, unless automation leads to vast productivity gains, any fall in hours would be modest and slow. It would take a 75% rise in productivity to deliver a 10% fall in hours.  Or a 150% rise to knock a day off the working week.

…and it’ll probably push up average wages

The consensus view amongst economists is that technological change is labour augmenting- i.e. it acts to increase output in the same way as an increase in labour input.  If that sounds counterintuitive it is merely the flip side of assuming that innovations are labour-saving.  History provides a good testbed for this hypothesis – if technical change is labour augmenting, then technological change should lead to a rise in the wage rate, but leave the interest rate unchanged.   The Bank’s own historical datasets suggests long rates have (periods of high inflation excepted) hovered around 4% since the 1500s.  And the chart below shows that  since 1800, the return to labour – i.e. the real wage rate – has  grown by a factor of around 15.  Of course these aggregate figures might well mask substantial variation across industry and worker groups.  Certain workers may be hit very hard, especially if their human capital is rendered obsolete. Robotisation may not be good news for all workers and may pose important distributional challenges.

Figure 1: Wages and Interest rates over time

figure 1

Is robotisation different?

So to argue that robotisation will benefit capital at the expense of labour you have to believe there is something intrinsically different about it compared to innovations that went before.  One thing that might, and I stress might, be different is the substitutability between labour and capital. Under labour augmenting technological change, if this parameter is less than one, then rising capital to output ratios over time increases the labour share, if it’s equal to one, the labour share is unchanged.

Looking at back data suggesting a constant or rising labour share, for the bulk of the post-industrial era many economists concluded the elasticity was less than or equal to one.  So more capital meant its relative price had to fall by more than its quantity increased – hence a lower share of income goes to capital.  How might technology change this?

Imagine a taxi and its driver – there is in essence no substitutability between the two.  They have to be combined in fixed proportions, and so having a taxi with two drivers, or a driver with two taxis creates no extra output.  Earlier technological progress, faster cars, satnav, Uber, didn’t change much on that score.  But perhaps robots will make labour and capital much more interchangeable – so the driver can be substituted by a computer, and the passenger rides round in a driverless car.  If this pushes substitutability above one, growth in the capital stock over time leads to a higher share of income for capital.  Indeed Piketty and Zuchmann argue this applies to a much broader range of technological change than just robotisation and has driven up inequality in the past two decades.

Could robots help combat secular stagnation?

On the plus side, if you are worried about secular stagnation then robots offer you a couple of reasons to be cheerful.   First up, if robotisation does constitute a major productivity gain that raises the marginal productivity of capital, then this should push up on long run-equilibrium real rates, and hence ease fears of secular stagnation.  Second, whilst economic theory usually assumes that technological growth means capital is just costlessly melted down and made into newer, more productive machines, in practice, some innovations might require scrapping of old capital, and hence a wave of new investment.  If you believe in the loanable funds model (and not everyone does) this increased investment demand should push up on equilibrium rates.  This is somewhat speculative but it does make the point that it’s hard to believe in a dystopian world of robot induced productivity growth where secular stagnation also occurs.   You have to at least pick where you are going to be pessimistic…

Beware unexpected consequences

But perhaps the most important historical lesson of all for economists is to remember that often the biggest implications of an innovation occur far away from that good’s own industry.  Take the humble shipping container.  Transporting goods in pre-packed locked containers, which can be lifted straight onto a lorry or train, yielded enormous savings relative to having cargo transported in crates which needed loading and packing individually at each port.   Their inventor estimated that the combined savings on labour costs, time at the dockside and insurance for breakage and theft reduced the price of a tonne of cargo 39-fold.  Bernhofen et al calculate this led to an eight fold increase in bilateral trade between countries with container ports. Whilst employment fell, productivity of labour increased nearly 20 fold. For the shipping industry this wasn’t a massively disruptive technology- though trade patterns changed, the industry became more concentrated and ironically less profitable.

But by reducing the cost of trading, containerisation opened up the possibility of new supply chains and trading arrangements that were previously too expensive to undertake. And, inso doing, the resultant trade flows led to a substantial spatial reallocation of economic activity. The real macro impact of containerisation didn’t occur at sea or on the dock side.  Perhaps the biggest effect of robotisation might occur far away from the industries which adopt the robots, and in ways which today’s macroeconomists could never imagine.

Robots are (probably) our friends

So in the long run, labour-saving technological change means we can make more stuff.  And that is a generally a good thing. In the long run it doesn’t create unemployment and might even help avoid secular stagnation.  But it might alter how that output is divided up.  Working out the theory behind that, and unpicking the effects of particular innovations will probably keep economists in human or robot form occupied for years to come.

John Lewis works in the Bank’s Research Hub.

If you want to get in touch, please email us at You are also welcome to leave a comment below. Comments are moderated and will not appear until they have been approved.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

14 thoughts on “Robot Macroeconomics: What can theory and several centuries of economic history teach us?

  1. The comparison of inflation adjusted wages with a nominal interest rate series brings the article into question ( as a BOE article); especialy when there is no indication of public funding pressure . The former is especialy bad as the reader form a conscious or sub-conscious impression of the trend (s) that may be quite wrong. Please re-issue with a more usable and trustworthy chart.

  2. My reading of the research is that economic history contains episodes where complementarity dominates but there are times when technological advances lead to disruption in the labor market and for wages. The economic historian Joel Mokyr, for instance, noted that real wages in the UK did not improve much between 1750-1850. The question amounts to this: to what extent is the empirical evidence from the last century a good guide to the effects stemming from digitalization? I think there are two reasons to be more concerned and less sanguine that historical estimates will hold:
    – digitalization is a multi-sector shock, affecting most areas of the economy at the same time (similar to effect of introducing electricity)
    – the pace of change can be faster than human ability to adapt, at least for a period of transition.

    These are issues that I explore in a published report available from my web page and in a forthcoming book to be published by Edward Elgar in Jan 2017 (“Digitalization, Immigration and the Welfare State”).

  3. I don’t understand why consumption would go up, just because production goes up. I’m not an economist, but I see people living with less overall — and happier for it.

  4. This post reminds me of the adage that the market can remain irrational longer than you can remain solvent. Unemployed masses can remain “irrational” longer than it takes to see the sunny side of this dislocation. The question is just how irrational everyone gets.

  5. “Make more stuff” is a good outcome if matched by demand. The supply side thread of this sanguine narrative skirts the demand problem of the downward trend of the wage share of GDP being masked by the upward debt trend. We don’t have many years to come for the orderly resolution of this unsustainable model which requires cooperation between economists and philosophers to inform politicians. Rendering human capital obsolete may well be adopted as a goal if capital assets are viewed as being at the service of humanity rather than the prevailing opposite assumption.

  6. While we have always found new things to make with the surplus of labor from disruptions in the past, the coming robotics/A.I. revolution could be very different. The industrial revolution was a shift from human/animal labor to machines, but the displaced laborers could get a job working the machines. We are currently eliminating the least skilled jobs and creating highly skilled jobs. An out of work truck driver won’t get a job at Google building self-driving car algorithms. I could see a future where the bottom 10-20% of people in terms of intelligence and education are unemployable, because we have robots that can do anything they can. No one will think of employing a human to do a job a robot could do, just like no one thinks about how to get a horse to power a device today.

  7. Come on, this is unthinking. In the past, there wasn’t mass unemployment, except for horses. What was the difference between horses and humans? With horses, the machines could do so much the horses could do, and far better and/or cheaper, that it was not possible to find enough alternative uses of high enough value for their market wage not to drop below subsistence without a 99+% drop in the supply of horses.

    With humans, there have always been too many alternative uses because the machines just have always still fell very short in many important things humans could still do, capabilities they have. But now, take low skilled humans. These AI’s and robots are very different. What is it that low skilled humans can do that these AI’s and robots may not be able to do in 10-50 years? There may be very very little, and this could easily plummet the market wage necessary to employ all unskilled humans, or even half of them, to well below the poverty, or even subsistence, level.

  8. “Eventually the economy has 40 million workers making rolls, and 80 million making sausages. In the interim, the transition might lead to unemployment, particularly if skills are very specific to the baking industry.”

    The problem is this: Substitute roll makers and sausage makers with high-skilled workers and low-skilled workers. Suppose the economy starts at 4 billion low-skilled workers and 400 million high-skilled workers, and produces $80 trillion in goods.

    Now suppose that advances in AI and robotics result in there being, for the purposes of producing pecuniary goods and services, 12 billion low-skilled workers, but only 500 million high-skilled workers. With the old production processes, you had a ratio of 1 high-skilled worker to 10 low-skilled workers. Keeping these processes, you would need just 5 billion low-skilled workers. So what happens to the other 7 billion low-skilled workers, machine and human?

    Well, you could go to other production processes that use less high-skilled workers to low-skilled workers (again where low-skilled workers now include the machine kind), but the problem may be, and I think is, in the real world, that production processes that have a low proportion of high-skilled workers produce a lot less per worker.

    And if they produce a lot less, then it does not make sense in the market to do them, unless they cost a lot less; thus, cost must drop. The high-skilled workers can be utilized for the old high-skilled production process, so the market must pay them at least that old wage.

    But now to get businesses to employ a very heavy low-skilled production process, they can only do it if the wage of the low-skilled workers drops, and maybe very dramatically, perhaps to poverty level, or below. And a human subsistence wage that need not be a floor, as the subsistence wage for a machine may be well below that, with the cheap solar power of the future; and with how smart and advanced these machines are, they may have very low maintenance costs; they may mostly maintain themselves, and each other.

    Essentially, the problem is, what if the roll makers don’t have the skills to make the sausages, then you can’t just shift to this new higher production economy with more hot dogs produced, but with a lower proportion of roll workers and a higher proportion of hot dog workers. So what do you do? You train roll makers to now make sausages? What if sausage making requires far more education and skill? This may be very costly, and the benefit may be mostly hard to recoup for the payer of this training. Meanwhile, governments may be unwilling or unable to pay. Many workers may be too old to practically learn, and so on.

    What then? You use a different production process? This may be a lot less productive. You end up with three rolls per sausage. It generates a lot less GDP per worker. It won’t be done unless the workers’ wages go down. But the sausage workers’ wages won’t go down; if they did, they would be hired away into old style 1-1 facilities. The roll workers’ wages will have to go down – and as far as it takes, to employ even the less productive roll makers, if they aren’t to be unemployed.

    And this is just the strong trend we’ve seen over the last two generations. See:

  9. Most of the discussions on automation and jobs reference the history of technological advancement BUT ignores the context.
    1. Previous periods of progress happened at a time of rapid urbanization growth (which boosts economic activity by itself). Urbanization likely continues but urbanization growth is much lower due to base effects
    2. The world population almost tripled since 1946 with the added demographics, supporting growth
    3. During the last 70 years consumers built up debt slowly to all time highs which boosted growth. One would argue that debt growth has to slow or even reverse
    4. Productivity growth has slowed (amongst other reasons) as economies are over-indebted and an extra dollar of debt generates a fraction of the growth it once did (funding consumption or social spending compared to GDP enhancing infrastructure). Debt to GDP globally is at an all time high – higher than 2007 pre-crisis
    5. Multiple technologies are accellerating change at the same time so adjustment is required at a much higher speed
    Summary – I suspect the contextual cross currents make the current adjustment far harder than history will predict. We are entering it at higher speed but without the tail winds of urbanization and demographics and with the headwind of high debt

Comments are closed.