The Future of Slavery

Bryan Caplan made strong, and to me incredible, claims that econ consensus predicts all ems would be fully slaves with no human personality. As he won’t explain his reasoning, but just says to read the slavery literature, I’ve done a quick lit review, which I now summarize, and then apply quickly to the future in general, and to ems in particular.

The ability to control your pain, actions, and income are distinct property rights. When someone else owns them all, you are said to be a slave, especially if they allocate these rights via something close to a “full control” package. In this package, you have little control over assets or actions. Pain is usually threatened, and often implemented, to force specific disliked but demanded actions. (Pain was used more on children than adults.) Think of rowing for a galley ship, digging up silver in a mine, picking cotton, or advancing on a simple war front line.

A second “mixed control” package allocates these rights by letting you retain control over many action details, only rarely causing pain, and letting you earn a residual income or status. This scenario was more common for domestic slaves, for slaves with better options for sabotage or escape, and for complex jobs where motivation matters more, via worker discretion, responsibility, attentiveness, pleasantness, intelligence, or creativity. By collecting a residual income, slaves might eventually buy their freedom. Free people have often sold this package of rights for short durations in traditional jobs. The main difference is your ease of changing jobs; the harder it is to change jobs, the more like this kind of slave you are.

In a third “debt” package, you must pay off a loan but are otherwise mostly free to choose your own job, location, and living arrangements. The option to impose pain is reserved for rare situations. Closely related is “share cropping” wherein the owner demands a percentage of income earned. Some combination of a fixed payment plus a percentage of income was a common scenario for slaves in southern US cities. This is also the usual way state rulers extort the locals they “own” via taxation. Many people voluntarily choose to go into debt, and sell percentages of their business income, and most legal systems reserve the right to impose pain in rare situations, a situation most people are okay with.

A fourth “ransom” approach sells these rights back to some combination of you and your associates. Often this converts these rights into debt held by someone who is better able to motivate and monitor you.

Many considerations influence the efficiency of these allocations, including costs of monitoring and restraint, losses from theft, rebellion, escape, and sabotage, individual preferences for pain, status, autonomy, and work style, effects of pain, status, and control on motivation and focus, information rents from workers being better aware of work details, complementary investments in training and capital, who knows better and has better incentives to use control rights, and signaling status, productivity, etc. to outsiders.

Historically, even when slaves were common, they were usually a minority of the population. (Beware, the term “slave” is used in different ways.) About 10% in the Roman Empire and US south. Foragers didn’t do slaves at all. About 0.3% of the world is in slavery today, mostly in forms of debt bondage.

The common existence of slavery that wasn’t converted immediately into debt or ransom does suggest that it was sometimes locally efficient as a resource allocation, ignoring larger social externalities, even given substantial costs of monitoring, enforcement, and worse motivation and allocation of skills.

Sometimes during hard times people would sell themselves or their children into slavery; better to be fed than dead. Sometimes slaves were created as collateral for loans, and freed when the loan was paid. Sometimes slavery was the contractual result of a failure to pay loans. Sometimes people sold themselves into slavery for a limited time, as with apprenticeships and indentured servitude.

But historically, slaves were mostly created in war. Drafted soldiers are slave-like. When a winning side didn’t expect to hold the territory, and feared leaving the vanquished to recover then retaliate, their remaining options were death or slavery. But slaves were only valuable when delivered to a useable location. So the worse treatment of slaves has been in transit immediately after capture.

Slave populations usually dwindled until replenished by war, probably because through most of history interest rates were too high to justify the long term investment of raising human children. Domesticated crops and animals grow much quicker. This same short term focus also often induced slave owners to work their slaves to death. A short term focus was often increased by distant ownership, as local manager’ incentives were tied more to immediate production. Workings slaves to death induced more slave revolts.

The US south was unusual in that it grew long-lived slaves from birth. Interest rates were unusually low, peace lasted long, and once US law forbad importing slaves, owners were highly motivated to preserve their big plantation industry. Slaves weren’t converted into debt perhaps because of credit market failures, or more plausibly because the full control approach was especially productive on plantations. (The sex story is overrated, as only 1-2% of slave babies were fathered by white men.)

That is, on plantations slaves plausibly produced more when threatened with pain, even if their utility was lower. The fact that humans can feel strongly disliked pain while living a long productive life and successfully reproducing does suggest that our pain signals are biologically maladaptive. But given how different is the modern world from the one where our pain signals evolved, we should expect this sort of thing sometimes.

Data on US south slave prices tells us what was valued in slaves then. For adults, age was bad, as were slaves from distant places within the US, and slaves that the owner chose to sell, as opposed to being forced to sell. New slaves imported from overseas were no more or less valued. It was good to be male, light-skinned, have artisan skills, and be guaranteed not to be sick or run away.

I didn’t find any data on slaves and docility, though I did find how docility fits into the standard five factor personality framework. Docility is lumped with “submissive, dependent, pliant” as part of “passivity”, which correlates most strongly and positively with neuroticism, but also positively with agreeableness and negatively with openness. In general only the agreeable part suggests more productivity in most jobs today; neurotic people are less productive, and the effect of openness depends more on job type.

What is there to dislike about slavery? The war and theft that cause slavery are clearly lamentable. And the possibility of slavery increases the range of possible inequality, at least if you ignore the dead. But the full control allocation package seems the main reason to dislike slavery. Other packages seem much closer to those resulting from free choices, and when they result from free choices they don’t seem strongly objectionable.

Today slavery, especially full control slavery, is discouraged not only via moral censure and political coordination, but also by stronger nation-states, few wars, better credit markets, increasing wealth, increasing vulnerability to sabotage, more automation, and more complex jobs. The only contrary factors I can think of are easier monitoring and preventing escape. If all these trends continue in the same relative proportions, we should expect a continued decline in slavery.

In the world of my book, The Age of Em, many of these trends continue. Nation-states and credit markets get stronger, and war remains rare. Automation advances, and jobs get even more complex, with motivation and sabotage mattering even more. Monitoring and preventing escape also get easier.

Individual em incomes do fall, which gives a thicker lower tail of outcomes, and in traditional societies that allowed slavery this low tail was often filled with slaves. However, ems can fall via running slower while remaining free, and this option would reduce the fraction that fall into slavery, even if slavery were allowed.

Ems are initially created via destructive scanning of high income human volunteers at the peak of their careers, in a world that forbids slavery. Soon after they are destructive scans of the most promising young children. So these volunteers do not expect to become slaves, and the world around them, being like ours, initially tries to discourage that transition.

However, since a lot changes we can’t offer much assurance that attitudes toward slavery don’t change. Also, labor supply factors matter a lot less; if even one productive em is enslaved, and slavery is allowed, then copies of it could fill a whole slave sector. What matters far more is demand, i.e., what are the more efficient ways to allocate labor? If allowed, there are probably some jobs where full control slavery is more efficient; the em world is big, with many corners. But most jobs are complex, where the full control scenario is inefficient. And the debt or mixed control allocations that are more efficient for typical jobs are probably not substantially more efficient under slavery, as slavery hurts motivation. Debt should be good enough.

So, bottom line, after a quick review of the econ of slavery literature, I still can’t find a rationale for Bryan Caplan’s claim that all ems would be fully slaves. Ancient society never got close to that state of affairs. And I see even less rationale for his claim that they would be so docile and “robot-like” as to not even have human-like personalities. Which is his main reason for saying 80% of my book is wrong. Neither the literatures on choosing employees today, nor that on choosing slaves in the past, put much emphasis on docility. And even if they did, the idea that they’d emphasize it so much as to eliminate human personality, that just sounds crazy.

So Bryan, how about actually giving an argument, instead of waving your hands in the general direction of the literature?

GD Star Rating
Tagged as: , ,
Trackback URL:
  • Eliezer Yudkowsky

    I think I might understand Bryan’s point—if you drew a line from Robin Hanson to Eliezer Yudkowsky on this issue, I suspect Bryan Caplan would be Robin + 0.05 Eliezers, or thereabouts.

    The central conceit of Age of Em is that Ems are *exactly* like humans except in two regards: they can be easily copied, and they consume many fewer resources per hour to maintain. Aside from that, Ems are exactly like humans in all other economic regards, even clock speed, so far as I can tell. This is what lets Robin analyze Ems using all the existing economic generalizations, but for these two presumptions that can both be phrased as relatively easy counterfactuals.

    If we imagine a sudden portal opening to the Dimension Where People Can Be Easily Xeroxed and the Cost of Living Is Far Lower, and that Dimension was otherwise an empty void and to be colonized entirely from Earth (such that somebody owned every atom of matter exported to it), the Dimension would behave pretty much like the Age of Em assumptions in all respects. I’m not sure you get Age of Em results, but those are the Age of Em assumptions.

    On the Yudkowskian view, this seems entirely implausible, or at least an extremely unnatural outcome. On the Yudkowskian view, 15 years before you would otherwise get scanning and emulation and computing abilities powerful enough to produce Ems (and run them cheaply at human clock rates), neuroscience decodes enough algorithms that when you run them on Google-sized supercomputers you can get neuromorphic AI smart enough to self-improve and then follows the great paperclipping.

    If you did start with Ems that were exactly like humans, somehow, and neuromorphic AI as such was somehow impossible, then on the Yudkowskian view that still wouldn’t last very long (unless some unnatural state were somehow made stable) because it would not take very long to expand or improve the human brains, and a 5x large human is something of far larger import than having 5 humans instead of 1.

    Bryan Caplan is 5% of the way to this point of view. He has correctly realized one of the (least) important non-human phenomena that would come into play, namely that Ems can be subject to much sharper selection pressures, and that this in turn would more strongly optimize away elements of human cognition that were not economically productive, and permit more extreme movement toward the edges of the possibility space. When we’re dealing with Ems, we no longer ask about the long-term effect of pain and slavery on the average human. We ask whether, if you apply pain and slavery to *all* the workers, there is *one* worker who as a slave is more productive than any non-slave workers. Perhaps there’s one person who, under threats of sufficient pain, will work every hour of every day for a week and then gets replaced by a new starting copy.

    The way in which this is unlike the modern world, again, is that in the case of Ems, we can apply much stronger selection pressures and we expect the literally optimal case that gets xeroxed all over the case to have most of its variables at more extreme settings, compared to average people who are doing pretty well.

    Personally I’d expect that slavery via pleasure-center stimulus might prove even more effective, but that kind of direct brain-hacking is 10% of the way to the Yudkowskian view and Bryan Caplan is only at 5% so he’s still imagining electric shocks. (Well, simulated electric shocks. They’re all p-zombies to him.) And that playing around with neurotransmitters might produce better workers still, shaping motivation so as to be entirely and cheerfully work-focused while eliminating all other distractions and also augmenting cognition, but now we’re at 20% on the Yudkowskian scale and the veneer of anthropomorphism is starting to wear away completely.

    Similarly, we’re no longer concerned with whether the average above-average intellectual is robotlike in some regard, but whether Paul Erdős in particular was focused entirely on his work plus taking amphetamines.

    We are no longer concerned with whether the average profitable worker will continue to be profitable if owned by someone else and enslaved by various means, but whether the *most profitable type of person on the face of the Earth* is an owned slave.

    I don’t expect I’ve exactly represented Bryan’s views here—for one thing, it makes “read about historical slaves” pretty much irrelevant—but it is my steelman of Bryan’s view.

    • I appreciate the effort. And yes give the possibility of selection, the question on each job is what particular combo of origin, training, tweaking, etc is max productive. Though if there are a thousand distinct jobs there may be a thousand distinct answers. Even so the people we’ve seen best at jobs today & in the past would seem to be our best guess for this best em. Or best thousand. Yes, you might argue that we should be quite uncertain about this best guess. Even so, we’d need some basis on which to guess that extreme-docile slaves are the max productive at pretty much all jobs. Its that basis that I’m missing.

      • Eliezer Yudkowsky

        I think you probably do need to postulate neurochemistry tweaking, direct brain stimulation, or similar things—things which, on the Yudkowskian view, you can probably learn to do pretty quickly once you have an em in hand, if that’s not prevented by some unnatural state made stable (e.g. through effectively enforced law) or obviated by other developments.

        In other words, you probably have to select on variance in mind-design-space that is visible in the modern world, but not in our world present in full width within the top thousand specialists particularly, before the most profitable accessible versions of workers are docile and happy to ship you their earnings.

        But that variance *would* be exposed if you had ems to play around with and hack, and then the docile-hacked John von Neumann is the most profitable brain to own.

      • But even if we assume a much larger space of minds to work with, that *still* doesn’t imply particular directions like more slavery or more docile. Yes we should admit our uncertainty, but you’ll need to assume something that is directional to get conclusions that are directional.

      • Eliezer Yudkowsky

        The directionality comes from maximizing profit. Entities that demand less of the gains from trade, holding effort constant, are more profitable.

      • But that effect is already part of the equilibrium today. The fact that today’s equilibrium doesn’t move all the way in that direction says that there are other contrary considerations at play. We are trying to predict the net effect of all changes given all the real constraints at play.

      • danielxvarga

        I’d so love to try out my fancy new Bayesian hyperparameter optimization algorithms on human neurochemistry. Can any of my John von Neumanns beat the baseline John von Neumann on the industry-standard benchmark task of developing game theory? Can the optimal neurochemistry settings be transferred to a different domain like art?

        Okay, this is the stuff of nightmares.

    • As a correction, I definitely don’t assume ems are all human speed. And I talk quite a lot about what changes we should expect selection to produce. I may have gotten many of those wrong of course, but I definitely don’t assume ems are just like the typical human.

  • Recalling back to when I read Sowell’s “The Real History of Slavery”, most of the African slave trade actually went east to Ottoman territory. These slaves were engaged in what Adam Smith would consider “unproductive labor” as servants rather than producers, and represented what Veblen would call “conspicuous consumption”. In stark contrast to the New World, slaves were castrated and replaced through repurchase rather than natural growth. It seems what we would normally consider “inefficient” practices lasted a long time.

    You’ve earlier discussed serfdom, with the conclusion that when the market rate of labor is low enough, there’s not much point in imposing slavery because the distributive result is the same.

    • Your description of Ottoman slaves sounds right to me. I don’t think it disagrees with anything I’ve said.

      • It doesn’t disagree directly, but it should caution one before saying that if some practice existed, it’s probably because it was efficient. Sometimes a “thick” understanding of the context of historical societies is necessary instead of the “thin” generalizations of economic logic.

      • But using slaves as servants to gain status is locally efficient in that world.

  • Ari

    More betting!

  • Workings slaves to death induced more slave revolts.

    How much did the fear of revolts moderate the treatment of slaves? Ems can’t revolt. Does that provide unprecedented advantages for enslavement?

    • Who says ems can’t revolt?

      • Eliezer Yudkowsky

        You *really* don’t expect *any* brain hacking? Not even overseers who can read out your auditory cortex? And why doesn’t the plantation owner just control the hardware and the off switch?

      • Saying it will get easier to detect and repress revolt is not at all the same as saying revolt is impossible.

      • The last, control of the hardware, was what I had in mind. This would make revolt impossible, no?

  • Bryan argues on Twitter that the success of the Soviet nuclear program argues for the efficiency of slavery. Did he take the old cold-war propaganda about “Communist slavery” too seriously?

    [Since Bryan thinks slavery is efficient, one better sees why he disparages efficiency as the ultimate criterion.]

  • Lord

    All I can think of is workers in a company town, each day deeper in debt, with replacement threatened for those unable or unwilling to deliver, but rather unnecessary.

    • Once a horror scenario like that has been lodged in your subconscious, you literally can’t imagine using analysis to figure out which scenarios might actually happen?

      • Lord

        So unlikely to be pointless. If there was a reason for replacement, money likely wouldn’t matter.

  • Brett

    In a summary of the literature, it would be valuable to have references to the literature – at least so we have mutual understanding of what constitutes “the slavery literature.”

  • free_agent

    It’s an interesting question: If two societies are in contact (i.e., competition), and one allows EMs full citizenship and one makes them all slaves, which society out-competes the other?

  • Wei Dai

    It doesn’t seem to me that who owns what rights matters. If pain or threat of pain is good for productivity, then competition will favor ems willing to apply pain or threat of pain to themselves (e.g., by using an app or hiring a third party service to monitor them and apply pain accordingly) even if they own the right to control their own pain. That we don’t see this happening today can be explained by lack of sufficient selection pressure, so it’s not predictive of a future where much stronger selection pressure applies.

    Pain or threat of pain must be good for productivity in some situations, but not others, since we see evolution use it only some of the time, not constantly. Or was pain actually a constant threat in our ancestral environment, where nearly everyone was at risk of injury, hunger, and exposure almost every day, and we are only relatively free from it today due to being out of evolutionary equilibrium? Biological death is naturally almost always painful and highly motivating as a result, and perhaps it’s reasonable to expect that in high selection pressure scenarios, ems will at least be faced with a similar threat for failing to maintain a competitive productivity, rather than expecting being painlessly shut down or going into retirement, regardless of the distribution of property rights?

    • Even if selection has been week the last 300 years, 10,000 years of farming had a lot of selection pressure, with slavery possible the whole time. Most wild animals are not constantly in pain.

      • Wei Dai

        Deliberately using pain as motivation during farming era would have been much more costly (in terms of monitoring productivity and applying pain) than it would be during the em era. It’s true that most wild animals are not constantly in pain, but they are under constant threat of pain (from being eaten, or failing to find enough food to stave off hunger, or suffering serious injuries), certainly more than modern humans. One might naively expect ems to have less pain and threat of pain than modern humans living in industrialized societies if they own the right to control their own pain, and I’m arguing that (in high selection pressure scenarios) they would have more, comparable to wild animals. (Not sure if you already made this point in your book, but I didn’t find it from a quick scan.)

      • But in the full control slavery mode, pain was in fact used a lot as a motivation. That doesn’t seem to have been greatly expensive.
        Legal systems have long placed most everyone under threats of pain, but threats that are rarely implemented.

  • Brad Hansen

    I’m afraid that your quick review of the literature was a bit too quick. Some of the statements are simply wrong others can reasonably be contested.

    You state that
    “Historically, even when slaves were common, they were usually a minority of the population. (Beware, the term “slave” is used in different ways.) About 10% in the Roman Empire and US south. ”

    This statement is simply incorrect. Slaves accounted for substantially more than 10 percent of the population of the South. Slaves were as much as 57 percent of the population (South Caroline) and at least 25 percent (Tennessee). See, for instance, Jenny Wahl at EH.Net. Or you can check at the Historical Census Browser at UVA

    You state that
    “(The sex story is overrated, as only 1-2% of slave babies were fathered by white men.)”

    The first thing to note is that, unlike population, the number of children born to slave mothers and white fathers is difficult to estimate. Some estimates put it as low as 1-2 percent, but Stephen Crawford found that in ex-slave interviews, by the WPA and Fisk University, as many 10 percent of slaves reported that their father was white. The 10 percent figure was when the interviews were done by African –American interviewers. In other words we don’t know. It may be possible use genetic studies to produce a more accurate estimate, but I don’t know of such a study. There is also the question of “How large is large?” Stating that “the sex story is overrated” suggests that 1-2 percent is somehow not important. Given that the vast majority of enslaved people in the South lived on plantations of 15 or more people, even 1 or 2 percent could be consistent with a relatively large percentage of slave owners fathering an enslaved child (or a smaller percentage fathering numerous children). It is not obvious to me that 1 or 2 percent is small in this case.

    You state that
    “Sometimes people sold themselves into slavery for a limited time, as with indentured servitude.”

    This is just kind of odd. It may be that he is using a definition of slavery that makes this make sense, but I don’t know of any historian who regards slavery and indentured servitude as equivalent.

    You also state that
    “Slaves weren’t converted into debt perhaps because of credit market failures, or more plausible because the full control approach was especially productive on plantations.”

    I’m not entirely clear about what this means, but it does not sound consistent with current understanding of slavery in the United States. Historians have devoted considerable attention to the well developed credit markets that facilitated slave markets. See, for instance, recent work by Bonnie Martin, John Clegg, Calvin Schemerhorn, Kathryn Boodry, and Gonzalez, Marshall, and Naidu.

    • The 10% figure was for the whole US; yes it wasn’t evenly distributed.
      The 1-2% figure isn’t trivial; it is just smaller than what many people think.
      Wrt indentured servitude, I warned that people use the word “slave” in different ways.
      Well-developed markets can still have failures; that is in fact the usual claim about credit markets even today.

  • The next big jump in “slaves” in the US will be sex robots, with or without a small (or large?) AI. Just as games drive PC hardware, desire for attractive sex-bots will be driving, with cash, lots of “willing” android bodies increasingly with a human feeling skin.

    Sex-bots will start out as mindless, then nearly mindless — but as property. And such sex slaves will be what the vast majority of buyers will want.

    • Lumen

      I doubt they’ll have much emotional depth. They’ll be programmed to fake it, but it will be quite narrow.

      It’s actually not efficient to program all the mental depth of a real person into a sex bot even if they are designed to mimick some complexity of behavoir, because mimicking is easier and serves the purpose just as well.

  • Pingback: Recomendaciones | intelib()

  • lump1

    Here is my argument for why it makes no sense of ems to be slaves: Slvae owners have to house and feed their slaves, which is expensive. By analogy, em slave owners would have to cover the slave’s cpu time, which they would only do if that was the most lucrative way to use it, in terms of returns per cycle. But the owner could also sell those cycles to an incredibly driven, competent and brilliant em who would use her own initiative and hypercreativity to find a way to make money. Whoever sells the cpu cycles could pocket almost the em’s earnings as rent, because the alternative for the em is to not run at all – i.e. death. So this em is not a slave in any of the senses you mention. Yet the sword of Damocles does hang over her head, moreso than for any historical slave, because if her rate of earnings fall below another em’s, she literally perishes, because a copy of that other em will outbid her for the cycles she’s using. In these circumstances, it’s hard to imagine that the dread of oblivion is not a source of motivation for the ems that happen to be running. They are not slaves, but they spend every living moment running on a very fast treadmill, barely staying ahead of the rotating knives.

  • Pingback: Overcoming Bias : Oarsman Pay Parable()

  • Pingback: Overcoming Bias : Seduced by Tech()