Study Slavery to Study the Impact of Robots on Workers

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

I started writing a post that was collecting several recent pieces about robots/technology and their impact on workers. Let me quick post those:

As I started trying to sketch out what to write, I realized that I was just barfing up a big pile of confirmation bias. I read the above articles and was able to convince myself that they supported my general “robo-optimistic” outlook. That is co-incident with my opinion that our fetish for manufacturing jobs and output is just that, a fetish.

There are, of course, lots of “robo-pessimists” out there who feel that technology is going to be very bad for labor. Low wages, or even mass unemployment are possible consequences. I’ve gone back and forth a little in the past with Richard Serlin, who generally falls into the robo-pessimist camp.

Rather than writing a post that says, essentially, “Those posts agree with my priors”, let me try to switch gears. What would constitute a good argument for robo-pessimism? In other words, what kind of argument would make me change my mind about this?

A common analogy used by robo-pessimists is the horse. As engines, and particularly internal combustion engines come into use, horses were made obsolete. The absolute population of horses has declined dramatically over the last 100 years because they became costly relative to using an engine to drive your cart around. When robots can do what humans do, the analogy goes, humans will become costly relative to using robots, and so humans will become completely unemployable.

I don’t think this is a particular good argument, mainly because horses have no ability to innovate for themselves. No horse ever looked around and said, “You know, I feel like there is more I could do.” Horses didn’t offer to become drivers of the new horseless carriages, nor did horses think to learn how to repair engines or build them so that they had something to do besides pull wagons around.

But people can innovate and invent entirely new jobs for themselves. If you tell me that people won’t possibly be able to innovate new jobs when the robots arrive, then I think you have a ridiculously low opinion of people. And it isn’t necessary that everyone innovates, just a few who invent new jobs and professions that we cannot possibly think of today. If we could, we would have invented them already.

In place of the horse analogy, let me suggest a line of reasoning to robo-pessimists that, to me, has a better chance of producing a convincing argument for that pessimism. Slavery. You want to give examples of when free people were put out of work by slaves. I’m not talking about the effect of being enslaved, that is clearly negative. I want to know the outcomes of those who remain free (small-time white farmers in the South) when workers are introduced who can exactly mimic the skills of the free workers (slaves). These are free humans being replaced by the equivalent of a living robot, and those free workers can still innovate new work for themselves. The story thus doesn’t fall into the problem that the horse analogy has.

In addition, the slavery comparison doesn’t require us to think about capital-labor complementarity, and it doesn’t rely on the introduction of a new technology that both eliminated some tasks (hand weaving) but created others (monitoring weaving machines). Slaves effectively are robots, for the purposes of the economic discussion here. They can perfectly replace free labor, but do not necessarily create any kind of other work for free labor in their wake.

I haven’t read it in a long time, but I think Gavin Wright’s Old South, New South is where I’d start. Slavery ensured that wages were kept extremely low for free workers in the South. This may have created conditions that encouraged new businesses or industries to locate there, and this is in fact what Wright suggests happened in the early 20th century after slavery was abolished. But while slavery was in place, industry did not develop in the low wage South.

One reason for this is that that slave-owners were “labor-lords”, not “land-lords” (Wright’s terms). They had no incentive to build up the value of land, as they could simply relocate further west with their major asset and start over. Thus the improvements that helped make the North more prosperous for free workers – railroads, schools, dams, ports, etc.. – were not built. Without those improvements industry could not or would not relocate to the South.

So to use slaves as an analogy for robots in their effect on free workers, the causes for robo-pessimism are not that it leaves free workers with nothing to do, but that it frees the robot-owners from any incentives to invest. Since robot-owners are free to move their capital to new locations, what incentives do they have to agree to the equivalent to railroads or ports or infrastructure?

In addition, despite having decades to come up with something else to do, the free workers of the South never created a new set of jobs or activities that allowed them to keep up with the North. They remained poor farmers, and did not (could not?) coordinate to build the infrastructure themselves. They were left to essentially scrape out a living as subsistence farmers without many connections to the broader economy. What you’d want to do is present me with evidence that free white living standards were pushed down by the introduction of slavery. Perhaps wages in areas of the South prior to the arrival of cotton versus after?

Those last few paragraphs are the result of maybe thirty minutes of thinking about robots from this perspective. If you wanted to make a really compelling argument for robo-pessimism, I think taking this analogy and running with it would be the way to go. It’s one avenue by which I think I could be convinced to switch from vaguely robo-optimistic to robo-pessmistic.

So consider this post a bleg. Are there good historical examples of the introduction of slaves into economies where we can observe the effect on the non-slave population?

As a last aside, I think this post is instantly #1 for “Titles that tell you everything important about a post”.

15 thoughts on “Study Slavery to Study the Impact of Robots on Workers

  1. Interesting article. I think the horse example could work for the pro-robot argument as well, if we looked at the horseman and the driver of cars instead of the horse and the engine. Horses were replaced by cars with engines, however, the cars still needed drivers. Therefore, in a world of perfectly mobile labour, the horse cart drivers would become the car drivers and that would reduce or eliminate the loss of jobs.

    The slavery analogy makes me uncomfortable because you have equated slaves to robots. I understand your academic perspective on the matter, but the wounds of slavery are still too fresh for this kind of analogy in my humble opinion.

    • It’s a terrible analogy, yes, in that it dehumanizes those who were enslaved. However, I don’t think we should purposely limit our thinking. You can be at once respectful and analytical.

  2. What about looking at the effect of a ban on robots/slavery? Pretty sure you can actually find that out looking at occupational score of Southern Whites in the US Census data from 1850 to 1870 (
    More generally I have a hard time thinking of how to write a model with massive unemployment forever (or maybe that will be the true leisure society actually). Ok white southerners remained poor farmers for a period, but who can claim it would have remained this way indefinitely, would have slavery persisted? I guess they would have ultimately migrated North just like the Black did after WWI. Then what if robots/slavery had spread North too? I guess ultimately it’s a question about competition among robot owners / barriers to entry. If no one can deter people from entry, there will always be an incentive for people to work on a better type of robot, right?

    • Something like that “experiment” is the right idea empirically. From what I know of Southern economic history after the Civil War, it wasn’t particularly positive for those existing white small farmers. They got put into a pool of cheap unskilled labor with freed slaves, and the large landowners generally acted to ensure that they would not be able to raise wages – limiting education, out-migration, and labor organization. In short, their incentives were still to not invest heavily in the broader economy of the South.

  3. Fascinating post!
    “…the causes for robo-pessimism are not that it leaves free workers with nothing to do, but that it frees the robot-owners from any incentives to invest. Since robot-owners are free to move their capital to new locations, what incentives do they have to agree to the equivalent to railroads or ports or infrastructure?”
    Noah Smith had a blog post a while back that was discussing the workers in China who are assembling products that we used to do here in the US. The bottom line was that few are taking risks investing in machinery to boost productivity – they just take the path of least resistance and simply find a cheaper worker wherever they can to do it the older less productive way. Somehow this makes me wonder if there should be a labor model of some kind that simply assigns some “slave coefficient” to an occupation/location (no one is a “pure” slave or a “pure” non-slave). The tighter the labor market you are in and the less open to competition from other countries, undocumented workers, etc., then the slave coefficient is higher (lower), and there is a more urgent need to invest in machines… robots what have you, and the converse is also true. Maybe globalization with it’s hunt for the cheapest worker has depressed investment relative to savings and that’s why inflation and interest rates are so low and we are having “secular stagnation”.

    • I think people do think about this idea – without calling it the “slave coefficient”. David Autor and others have tried to characterize what skills or tasks are easily replaced or not.

  4. Oh, as far as “robo-optimism” or “robo-pessimism” goes, I guess that would depend on what your personal POV “slave coefficient” happens to be. If you are readily replaceable and out biddable by a machine or another relatively non-free human then I suppose you would be a robo-pessimist. I believe that it would be best if global labor markets should tighten fairly universally to a point where the robot replacements begin and then you have the robo designers and maintainers, etc., then we would transition to a “robo-optimist” phase.

    • I think that’s right, and similar to what Winter Tomato was saying in the comments – trying to come up with a single answer to the effect of robots/AI on the economy isn’t terribly useful. What we want is to characterize the effect on different groups (really, ideally, on each individual). So in that sense there should be a range of robe-pessimism-optimism that depends entirely on your “coefficient”.

  5. What marks this post is not robo-optimism, but let’s call it scientific optimism: the stance that believes we can get to the right or best answer by fairly considering all the arguments. In this sense the post is very much on line with Paul Romer’s recent plea – one that was well-discussed here – for a truth-seeking as opposed to position-taking approach to economics.

    In itself, I’m pretty sure that’s a good thing. What I’ve been asking myself (and not just about this post) is whether the “objective” approach to economics-as-science, after all the objections made the last 150+ years, is really complete or convincing.

    A skeptic would say that robotics, like any technological change, is a two-edged sword, benefiting some, hurting others. The skeptic might also say that which side you are on mainly reveals whose side you are on — on where your sympathies and empathies lie (or on where your “interests” lie. to cast it differently).

    I think this sort of skepticism has done a lot of harm. At the same time, it is difficult to refute.

    I think where economists fail to refute it – where they seem to fall into the trap of “objectivity masking as interest” – is where they look for or expect a single “net” answer. So, to answer, the robotics question based on its effect on per capita GDP is to fall into the trap. To look at the perhaps-differential effect of robotics on the various quintiles is better. To pose the question, as is experimentally done here, as one that might be understood by looking at the history of (here comes a loaded, double-edged word) “exploitation” – I think that is better yet. It may mean that, in the end, that we arrive at answers that include judgments as well as “truths”, but I’m OK with that.

    So this was for me a very interesting post. I really like the “deep breath” that was taken here…

    • That’s a good point regarding the idea of looking across the distribution. There is no *single* effect of robots or AI or whatever. If we simplify we might come up with some kind of average effect, but that will be useless as a description of what happens to any particular person (which is what people want to actually know).

  6. This sounds like this is talking about AI that truly can do most anything humans can do. I think automation today, which is limited to a small subset of human tasks, complements rather than substitutes for human labor overall. But for AI that is cheap and really as flexible as humans the case for technological unemployment seems overwhelming.

    The effect that pushes down wages in economic models of AI that can fully substitute for human workers is Malthusian growth of the labor supply, to push wages toward subsistence. If AI subsistence wages are lower than human subsistence then the process can continue beyond the point at which humans are no longer productive enough to buy food and such with their wages. If AI subsistence is and stays higher than human levels (which seems technologically implausible) then there would still be jobs for humans, and the human population would gradually grow to displace the higher-cost AI workers.

    It doesn’t make a difference in the models whether the AI workers are free or enslaved.

    Click to access 00b4951facaf4a1f42000000.pdf

    Enslaved humans could never reproduce at a rate comparable to the potential expansion of AI populations in a fully automated economy. It takes many years for new humans to grow up, but an automated economy could double the size of the effective labor force in less than a year, whether through construction of more power plants and robot/AI factories, improvements in computer hardware, or software improvements.

    In the United States, the kidnapping of new people to enslave was barred early, and the enslaved population grew through natural increase. But one could look to cases like the ancient Roman Republic, where there were sudden shocks to the labor supply when foreign provinces were conquered and hundreds of thousands of people enslaved. That was reported to cause severe falls in wages for Roman citizen farmers, and their displacement by large slave-operated latifundia.

    But in comparing the cases one needs to keep in mind a few things about the Malthusian economics of slavery. In a Malthusian equilibrium workers should be able to earn enough to replace themselves in the next generation. But that means that workers produce more than they need to survive themselves on average, since they also need to support their young children while they grow. So even in a Malthusian era those who could cheaply capture and enslave adults one could profit by enslaving people and taking the surplus productivity that would otherwise be spent on supporting dependents (this is why classical slave populations tended to dwindle, while American slave populations grew through natural increase: labor productivity was well above replacement-subsistence in the latter case).

    In recent times Malthusian effects have not had much bite, as agricultural land and natural resources have been a smaller part of the economy, capital has been able to grow to match the population, and a larger economy has enabled more innovation. But if humans could produce highly skilled new workers in 1 year instead of 25, then even though innovation would accelerate greatly with the enlarged labor force, it would still have a very hard time keeping up with labor growth. And if it did keep up for a time, even a short duration of such growth would run into physical limits:

    • The issue with the Malthusian case is that “subsistence wage” is endogenous. Meaning, the price of subsistence goods will be driven down as well as the wage, so it isn’t clear whether the real wage of humans would actually fall. If AI can grow food without human intervention, then food is cheap even if people’s wages are low because of the loss of demand for agricultural labor. Whether wages or prices fall more is an open question – it could go either way, but it doesn’t necessarily follow that real wages would fall.

      But thinking more about the experience of workers who were “replaced” by slaves makes me think that the net effect is negative on the real wage.

      • “Whether wages or prices fall more is an open question – it could go either way, but it doesn’t necessarily follow that real wages would fall.”

        I think we can show this for a mature robotic economy if we add in a couple of facts like the role of energy in the production function and constraints on human consumption baskets, as discussed below.

        Which parts of the Hanson model do you disagree with?

        “If AI can grow food without human intervention, then food is cheap even if people’s wages are low because of the loss of demand for agricultural labor.”

        Is this going back to automation in only part of the range of human tasks (namely agricultural)? If not, it sounds like the situation I discussed in the last paragraph, where the growing population and economy counteract scarcity of natural resource inputs with innovation and economies of scale for a while (if those keep ahead of population growth, and bounded by physical limits).

        You can have that for a time, but the faster the payback on producing new robots, the more implausible it becomes with time. Solar cells, mining and manufacturing equipment, and computers can be built with common elements and free energy. One could illustrate the principle with a self-replicating factory complex that produces all of those products without human labor, although in reality it would be more efficient to integrate with the larger economy:

        If such factory complexes can reproduce every 6 months (doubling energy and industrial output) then it will not be many years before they are limited by access to ‘land’ (broadly construed to include rights to solar energy and minerals on and off earth) rather than the automated labor and capital they embody. Prices of robot labor will reflect mainly the natural resource cost (energy and materials) of construction and operation.

        Now let’s consider the price of food in this mature robo-economy. To grow food you need inputs including labor, machinery, water, nutrients, and free energy (normally provided by the sun). The free energy that goes into the food could also be used to manufacture and power computers, robots, and equipment, replacing food farms with solar farms.

        So a landowner must decide how much of her land to allocate to producing wheat and how much to solar electricity, in order to buy goods and services. How much will she earn from producing each of these alternative crops?

        If she produces solar energy (and perhaps sells some elements from the ground), she will be able to buy an amount of robot labor close to the quantity that can be built and sustained with that amount of energy and material.

        For it to be worth allocating her land to wheat production humans must be able to offer more than the value of the robot labor that could be sustained with the solar use.

        If there are humans with wealth from government subsidies (paid for by taxes on land or machines) or land or capital ownership, then that will provide some market for wheat. For example if the landowner is human herself, she would be able to pay for wheat.

        But if we assume robot labor can match or exceed human labor in every task, and the levelized resource costs of robot labor are lower than the levelized resource costs of growing , training, and maintaining human labor, then human workers who lack such non-wage income won’t be able to pay enough to favor wheat over solar, since the same energy cost produces more real goods and services when complementing robots than humans.

        So landowners don’t allocate enough land to produce food for human workers with no non-wage income (including food subsidies from charity or government) and the latter would starve. Of course if humans are still playing a disproportionate role in politics or landownership charity or subsidies might easily ensure all humans had some non-wage income since total GDP would be so great, but the wage effect stands.

        In this world many prices would fall dramatically, e.g. the cost of movies and information goods would be very low, measured in hours of labor or even kilowatt-hours. But energy prices in hours of human labor would skyrocket, and any consumption basket that humans can survive on is going to involve enough energy to power our brains and metabolism.

  7. I’ll just make two quick points, one about displacing low-skilled workers with slaves, and another about skill levels of the displaced. The first is that we need not necessarily look as far back as slavery to find evidence for what happened to low-skilled workers displaced by slavery; we have a contemporary example of displacement when unskilled immigrants displace the low-skilled native-born (indeed, this sets the evidence bar a little higher, since immigrants are not a one-off fixed cost but are merely a consistently low-cost factor substitute, at least as long as there remains an unskilled labor inflow). And while the empirical evidence is not definitive (as is the norm in economics), it does suggest that while some segments of the native-born are rendered worse-off, many end up moving up the skills ladder (exploiting, say, their command over the native language) to secure a higher value-added job (overall net societal gains are often estimated to be positive, too, as one might expect from standard trade theory). Which, naturally, brings me to my second point: what happens if we run with the thought experiment, and allow displacement continuously up the skills ladder? I think that is where the robo-pessimists are ultimately getting at. For the sake of argument, imagine if all capital (robots) were owned just by one capitalist, and these robots possess skills all the way up the skills ladder (and, with flexible mechanical limbs and AI, can easily learn within a short time any new job that may be imagined by displaced workers). Then, we become completely dependent on the altruism of that capital owner to redistribute all the private gains from robot ownership to the rest of society, since nobody would have the ability to earn income to accumulate saving and hence capital of their own. Of course, this is a thought experiment, and in practice we would expect both the existence of publicly-owned robots as well as taxation that would limit robot-capital accumulation. But I think you can see the thread of truth among the robo-alarmists.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s