The Labor-From-Factories Explosion

As I’ve discussed before, including in my book, the history of humanity so far can be roughly summarized as a sequence of three exponential growth modes: foragers with culture started a few million years ago, farming started about ten thousand years ago, and industry starting a few hundred years ago. Doubling times got progressively shorter: a quarter million years, then a millennia, and now fifteen years. Each time the transition lasted less than a previously doubling time, and roughly similar numbers of humans have lived during each era.

Before humans, animal brains brains grew exponentially, but even more slowly, doubling about every thirty million years, starting about a half billion years ago. And before that, genomes seem to have doubled exponentially about every half billion years, starting about ten billion years ago.

What if the number of doublings in the current mode, and in the mode that follows it, are comparable to the number of doublings in the last few modes? What if the sharpness of the next transition is comparable to the sharpness if the last few transitions, and what if the factor by which the doubling time changes next time is comparable to the last few factors. Given these assumptions, the next transition will happen sometime in roughly the next century. Within a period of five years, the economy will be doubling every month or faster. And that new mode will only last a year or so before something else changes.

To summarize, usually in history we see relatively steady exponential growth. But five times so far, steady growth has been disturbed by a rapid transition to a much faster rate of growth. It isn’t crazy to think that this might happen again.

Plausibly, new faster exponential modes appear when a feedback loop that was previously limited and blocked becomes is unlocked and strong. And so one way to think about what might cause the next faster mode after ours is to look for plausible feedback loops. However, if there thousands of possible factors that matter for growth and progress, then there are literally millions of possible feedback loops.

For example, denser cities should innovate more, and more innovation can find better ways to make buildings taller, and thus increase city density. More better tutorial videos make it easier to learn varied skills, and some of those skills help to make more better tutorial videos. We can go all day making up stories like these.

But as we have only ever seen maybe five of these transitions in all of history, powerful feedback loops whose unlocking causes a huge growth rate jump must be extremely rare. The vast majority of feedback loops do not create such a huge jump when unlocked. So just because you can imagine a currently locked feedback loop does not make unlocking it likely to cause the next great change.

Many people lately have fixated on one particular possible feedback loop: an “intelligence explosion.”  The more intelligence a creature is, the more it is able to change creatures like itself to become more intelligent. But if you mean something more specific than “mental goodness” by “intelligence”, then this remains only one of thousands of possibilities. So you need strong additional arguments to see this feedback loop as more likely than all the others. And the mere fact that you can imagine this feedback being positive is not remotely enough.

It turns out that we already know of an upcoming transition of a magnitude similar to the previous transitions, scheduled to arrive roughly when prior trends led us to expect a new transition. This explosion is due to labor-from-factories.

Today we can grow physical capital very fast in factories, usually doubling capital on a scale ranging from a few weeks to a few months, but we grow human workers much more slowly. Since capital isn’t useful without more workers, we are forced to grow today mainly via innovation. But if in the future we find a way to make substitutes for almost all human workers in factories, the economy can grow much faster. This is called an AK model, and standard growth theory says it is plausible that this could let the economy double every month or so.

So if it is plausible that artificial intelligence as capable as humans will appear in the next century or so, then we already know what will cause the next great jump to a faster growth mode. Unless of course some other rare powerful feedback loop is unlocked before then. But if an intelligence explosion isn’t  possible until you have machines at least as smart as humans, then that scenario won’t happen until after labor-from-factories. And even then it is far from obvious that feedback can cause one of the few rare big growth rate jumps.

GD Star Rating
Tagged as: , , ,
Trackback URL:
  • Scott Young

    Interesting ideas. One quibble though:

    Where do you get the 10B year old history for life? Life on earth can’t be more than ~5B years old. Is that a mistake of extrapolation or are you assuming some kind of panspermia hypothesis and just not stating it.

    • Why not follow the link?

      • Dude Man

        Extrapolating what life forms existed before Earth was formed using only data points from Earth seems like over-extrapolation. Honestly, I’d expect a high schooler who is taking his first Statistics class to look at that conclusion with more skepticism than you have.

      • one of the dudes

        How about taking the data and seeing where you get with an open mind? There are 5 links about this in the paragraph you are uncomfortable with.

    • Evan Kostishak

      not just a quibble. i stopped reading this article when robin started talking about the evolutionary rates of lifeforms that predate the earth. any conclusions made with this as part of the basis are conjecture

  • Who do you take to be fixated on an intelligence explosion as the *NEXT* feedback loop?

    Yes, many people seem fixated on an intelligence explosion but I don’t take them to be claiming it is the next shift to a substantially higher rate of economic growth. Rather, I take their fixation on an intelligence explosion to be based on their expectation that an intelligence explosion will lead very quickly to a world that is virtually incomprehensible to us not merely more productive.

    There seems to be something to this. While I’m sure a world with robot factory workers would be substantially different it wouldn’t be unrecognizeably so….in the same way that an ancient Roman might be surprised by much of modern technology but wouldn’t be fundamentally unable to navigate our culture. A world with quickly improving AI intelligences presents the prospect of being utterly different in a way that many other proposals don’t seem to match (though a genetic modification feedback loop might).

    • It isn’t remotely obvious to me that people get harder to understand as they get smarter. For me, it is the opposite.

      • Seems pretty obvious that a person y who is far more intelligent than x will often be incomprehensible to x.

      • The opposite is also true. In fact we can just say any two people with any set of features will often be incomprehensible to each other.

      • Dave Lindbergh

        You’re on the extreme right end of the bell curve.

        Given that almost everyone else is to your left, naturally as they get smarter they become easier for you to understand – they’re becoming more similar to you.

        For people on the left end, *you* are hard to understand.

        Perhaps you’ve noticed.

      • I don’t think that’s the explanation for Robin’s mistaken intuition. Stupid is stimulus-bound, hence easier to predict and understand. I can predict my dog’s behavior more reliably than my wife’s.

        But human society makes it advantageous in many situations to be predictable. Smart people are more able to produce predictable behavior when they need to.

      • Yes, I meant in actual society they get easier to understand. Of course they COULD get harder to understand if they wanted to, but they don’t want to.

  • Frederic Bush

    Factories produce physical things from other physical things and they all need to be moved around. Unless you can scale up the transportation system as fast as you can scale up the factories, that’s going to be a hard cap on how fast things can grow.

    • Within a local geographic region material can easily be moved between factories in much less than a month.

      • Frederic Bush

        I’m just envisioning 20+ times the number of trucks we currently have on the road to handle all the moving about of stuff. In some places that will be fine but in other places that will be a big problem. You can claim efficiencies from robot drivers but these roads have max capacities even under ideal conditions.

      • 0x0D2ERDSF

        I’m envisioning 1000x drones there are now comunitcating in real time delivering computing hardware to where its needed running. The max capacity for the atmosphere is orders beyond whats available confined to roads. Some countries have actually allowed people to use drones now in commerce, baffling eh?

    • You can scale up a transportation in space much easier due to absence of atmosphere and gravitational stress. So I think the easiest way to achieve rapid doubling rates would probably be to launch seed factory equipment into orbit (and use asteroid materials for expansion). Another option would be to land the equipment on the Moon, where you still have ultrahigh vacuum and the gravity is only a sixth of earth’s.

      • All-Out Sprint

        And then what? The consumers are on earth.

      • Doesn’t matter. We use most of our stuff remotely. Note that various entities own (and profit from) satellites, which are very expensive in a launch-only market.

  • All-Out Sprint

    What about physical inputs like energy, land and raw materials? If the economy doubled every month, they should rather quickly become the new bottlenecks.

    • You think those things can double every decade but not every month? I don’t see what the timescale has to do with it.

      • All-Out Sprint

        I don’t know if they can double at all, especially land. Though I guess it’s possible through innovation.

        My point is there are other limits to new growth modes, space fantasies nonewithstanding.

        Let’s say labor supply is a current limiting factor to growth, and then disappears. It doesn’t follow that a sustainable new growth mode results. And with doubling times of a month, the new bottleneck will become relevant rather quickly from our perspective.

      • The economy HAS been doubling every 15 years for a long time now.

      • All-Out Sprint

        Right, and it led to more and more appropriation of available land from nature toward human use, while human population grew exponentially. Now we use almost the entire planet, at least land area in moderate climates.

        Not sure how many doublings of usable land could be accomplished by application of technology now (again ignoring speculative space stuff that’s of no use to consumers on earth).

        If land becomes an input bottleneck to growth soon, labor from factories will not lead to a sustainable new growth mode.

      • free_agent

        Doubling every month has a timescale that is 120 times faster than doubling every decade. So instead of a change that runs to its limit in a century, it runs to its limit in a year.

  • Interesting thought. But what would we do with all that stuff the factories were producing? It seems to me that we have already reached the stage where production of material goods is more limited by economic demand than by potential capacity.

    • This is the key point. Industrialization didn’t double farm output, it created completely new output (of widgets) with its own demand. The notion of “doubling the economy” will have nothing to do with “moving physical goods”, it will be in the realm of ideas. Perhaps virtual worlds (which you can grow almost infinitely with the same resources).

      • This is the most likely explanation of the Fermi paradox too, I think.

      • Joseph Miller


        is the most likely solution to the Fermi Paradox 🙂

      • All-Out Sprint

        Almost infinitely isn’t the same as infinitely. Even with super efficient black hole computing, it’s still implausible they would leave resources unused, at least in the long run.

    • Demand isn’t inelastic. If you can buy something cheaper you are more likely to buy it. You currently don’t own any giant Mecha or a Large Hadron Colliders because those would cost more than you can afford, not because they wouldn’t be cool to have. The thing about self replicating factories it’s that you can own arbitrarily large projects for the cost of the uniform unit plus some time and rental of (currently very abundant) natural resources.

      • Matthew Light

        I could own an almost infinite amount of nearly-free things by shopping at the thrift store, but I have zero interest in that.

        Some things I used to buy and own I avoid now like the plague (physical books, DVDs, CDs etc.) because they are a hassle.

        There are people who like to own lots of physical possessions – they make TV shows about them (hoarders).

        I think my attitude is increasingly common these days. Do we own our possessions or do they own us?

      • What you’re noticing is that the junk one fills one’s house with carries hidden costs to ownership; a house full of junk actually has less use value than a house that contains only the bare essentials. If you had a robotic system to organize and store all of your possessions seamlessly and invisibly, your threshold for how much is too much would likely be higher.

        What I’m talking about is goods which are mostly used remotely (or would be, given the premise). Computer servers, communications satellites, particle accelerators, laboratory spaces. By the time it reaches the customer, the product is mostly just data, but the physical steps in the middle still involve lots of matter bring moved around. By using the data, or products made using the data, you still create demand for the physical stuff.

      • All-Out Sprint

        There is a large subset of consumption that is not reducible to just data on the consumer end. People will want their food, hot showers, swimming pools, sports cars etc., especially in an economy that grows so fast. You are assuming that everybody will be priced out of physical consumption and into virtual reality, but experience shows that the richer the world gets, the more physical throughput is used per capita despite growth in virtualized goods.

      • I don’t expect that sector not to grow, but I don’t think it will grow as fast as the space based productive sector. I assume that physical throughput in this context includes what is used remotely, and agree that increased throughout and energy consumption is likely to closely correlate with increased wealth. The earth based productive sector will benefit synergistically from computational and data resources from space, initially. At some point, there would also be energy, and even material resources from space.

    • Joseph Miller

      “It seems to me that we have already reached the stage where production
      of material goods is more limited by economic demand than by potential

      Are you saying that everyone is producing the wrong things? To get money to buy something, you have to first produce something, which implies that there is something someone wants.

      Out of curiosity, what do you think would happen if everybody subjectively decided that they want everything half as much? Would anything change?

      • one of the dudes

        debt would be defaulted on

      • Joseph Miller

        Why do you think that?

      • one of the dudes

        ex-post over-investment of fixed capital fixed capital. Debt-deflation.

      • Joseph Miller

        When I said people “want everything half as much” I meant everything, which includes the desire to hold cash.

        Until you can reverse engineer brains, you’ll never even be able to distinguish between one person who “wants everything half as much” as another person.

        My point was that in equilibrium, relative demand is the only thing that matters.

      • Really what I’m saying is that we are currently limited more by our economic and political system than we are by technology in terms of what and how much we are producing. We live in a world where paradoxically the big economic problem is a “demand shortfall” while at the same time there is a huge amount of incredibly useful stuff going undone. The easy solution of having the state play the role of ‘demander of last resort’ and putting the unused capacity to use for something is ideologically opposed by our ruling elite, and hence our current pickle. I don’t know if much more efficient consumer good production would really change that equation much.

      • Joseph Miller

        “putting the unused capacity to use for something is ideologically opposed by our ruling elite”

        This is a strange theory. You think that politicians can and should give all of the unemployed people jobs but don’t because they are ideological?

        Remember the JOBS act? Politicians are always talking about getting people employed by various projects.

        “We live in a world where paradoxically the big economic problem is a
        “demand shortfall” while at the same time there is a huge amount of
        incredibly useful stuff going undone.”

        In my earlier comment I tried to address this. It doesn’t seem right that people wanting things too little can be an explanation.

  • Sharknado

    If you think that human labor is the current bottleneck in the expansion of physical capital, then how do you explain things like the explosion of service jobs, high rates of unemployment, and high rates of permanent exit from the labor force?

    • The fewer people who work, the more that limited labor is a problem.

      • Sharknado

        But in developed countries, it’s the *demand* for labor that’s limited, not the supply. Manufacturing jobs are at an all-time low, and people who work with their hands are leaving the workforce because they can’t find any jobs. You almost never hear about a factory that shuts down because it can’t find enough workers.

        If labor were really the bottleneck, I’d expect to observe things like: employers competing for apprentices, new streamlined schools forming all the time to train people faster, real wages rapidly increasing, fierce pressure to scale back child labor laws, and large subsidies for having many children. Wouldn’t you expect to see those features in industries where labor is the primary bottleneck?

        I see some of those features in the software engineering industry, but I see none of them in manufacturing.

      • All-Out Sprint

        The growth mode is about cheap substitutes for labor. Imagine an algorithm that can do any task a human brain can do, learns faster and can be copied (with specialized and durable robot bodies being built at a fast pace). Once invented, what’s the marginal cost for substituting one hour of labor?

        As for factories not shutting down, consider that many factories may not be built if they can’t be expected to be profitable, which they would be if the above mentioned substitute existed.

      • It isn’t remotely obvious to me that the problem is with demand. Heard of ZMP workers?

      • Sharknado

        Let’s frame the question as, “Why don’t investors build more factories?”

        Theory 1 is “We can’t find enough useful workers to operate the factories.” (ZMP workers, ala Tyler Cowen)

        Theory 2 is “Consumers can’t afford to buy more manufactured goods.” (structural deficit in aggregate demand, ala Keynes)

        Theory 3 is “Consumers would get zero utility from higher quantities of current manufactured goods.” (ala Matthew Light below)

        Theory 1 could be correct; people aren’t born knowing how to be useful in a factory. But if Theory 1 were correct, then at a minimum, I would expect to see for-profit schools that teach people how to be a useful factory worker, because there are millions of unemployed, able-bodied workers with IQs of at least 100. I don’t see those schools. Is the claim that without an above-average IQ, you can’t be profitably taught how to tend to a semi-robotic assembly line? If so — and if the true bottleneck is smart factory workers — then why don’t we see some investors build new factories using older, medium-tech designs so that they can take advantage of all the desperate (and thus bargain-priced) 100-IQ labor?

  • Ken Arromdee

    I find it odd that Robin asks “What if the…?” near the top of the article, and then the entire remainder of this article is about what would take place if his arbitrarily chosen stipulations were true. I could just as easily write an article that asks what if lizards were taking over the world.

    If he is trying to imply, by asking “what if X?”, that X is likely to be true, then he should justify his belief that it is likely to be true, not use the sentence structure “what if X?” to omit this important facet of his argument.

    • sflicht

      Did you read the fourth paragraph?

      Also, the remainder of the article is actually talking about what might be the explicit mechanism via which X happens.

  • Walter

    Programs as ‘smart’ as humans are likely to do a lot more than kick start the labor force, yeah? Wouldn’t they take over and, depending on what we programmed their values to be, ruin everything?

    • I didn’t say this post told about all of the consequences. That would be hard to fit into one post.

      • Walter

        Right, I get that…it’s just…it feels a bit disingenuous without it. Like, if we were talking about diet methods, and I was like “Set yourself on fire, the fat literally melts away.” Then you’d be like “Well, ok, but there are other consequences to that that make considering it as a diet strategy not useful”, and I’d be like “I didn’t say this post told about all…”

        I feel like enough would change if we get programs that are people that considering them in an example where all else is held static (or, enough is held static that there is still an economy) is a bit unfair. Like, they trivially take over the gov if they vote (all 999 billion BobGuids on this server vote for Bob1.0 as president), and they are slaves if they don’t, either case (we are slaves or we are slave owners) seems to change our notion of “the economy” to one where comparisons to the current are unfair, dig me?

      • If you want more detail, read my book

  • arch1

    It’s not original but it does seem true that each mode transition has involved qualitatively better information processing. Human-plus AI is in this category but labor-from factories doesn’t seem to be.

    • What better info processing do you associate with the farming revolution?

      • one of the dudes

        Coordination over long time (harvest cycle) and sometimes distance (for trade).

      • arch1

        Farming produced goods which needed
        tracking. It also enabled high density
        population centers which required better record keeping and administration (driving
        technologies such as writing) and also enabled more intensive communication and
        faster cultural evolution.

  • David Condon

    There are several emerging technologies which could substantially substitute for current labor force participation: quickly reprogrammable and cheap robot arms, autonomous vehicles, automated and online checkout, and 3d printing. Looking at BLS industry data (, I’d say those technologies could substantially replace construction, mining, manufacturing, wholesale and retail trade, and transportation and warehousing. Together those account for about a third of the labor force. There is also likely to be some if less innovation in finance, professional services, and leisure. Education and health care seem to be a current bottleneck due to government regulation. Finance appears to be a bottleneck due to market irrationality. Actively managed funds still represent about 75% of the market ( although index funds and ETFs are slowly making gains. We might hit a market growth acceleration well before AGI arrives. I think an innovation in how information is exchanged over the internet could also happen at any moment, and could present a major leap forward.

  • Pingback: unblocking Feedback Loops! | Buzzography()

  • J Storrs Hall

    This is something like trying to foresee the Internet in 1916 by thinking about pneumatic message tubes (which quite a few SF writers actually did). Robin is one of the best at avoiding Failure of Nerve, but no one can avoid Failure of the Imagination. Remember that latter 21st century nanotech could most likely produce fully-formed, fully educated, adult humans in factories.

    • I don’t understand what you are saying. Are you suggesting that we should not try to foresee the next big change? That none of today’s concepts arerelevant? You seem to think the concept of nanofactories is relevant; why is that concept more useful than the others I invoke?

      • J Storrs Hall

        No; no; and it isn’t. I’m just trying to give an intuitive feeling for the extent to which we can grok the future in its fullness, so to speak.

      • I’m not sure why we should trust such an intuition much. We don’t know how much we’ll be able to understand anything until we get into the details and try.

      • J Storrs Hall

        There are trends (e.g. Productivity of manufacturing) and scaling laws that let us readily imagine a major takeoff in the coming century. Something a lot like the computing takeoff when the corporate computer center gave way to PCs. I don’t see AI as providing a sudden foom in the process, though; something more like your log normal analysis is likely.

      • Seems you are saying that some kinds of abstractions, eg scaling laws in manufacturing, are trustworthy even into strange future where many things change, but that other kinds of abstractions, eg growth theory that says growth booms when labor substitutes can be made in factories, are less trustworthy, and can’t be relied on in strange futures. This seems close to simple favoritism of the disciplines you were trained in, relative to others.

      • J Storrs Hall

        Think again, Hammurabi. I would have come closer to agreeing with your OP 15 years ago, when I was more of a pure AI type. Productivity in manufacturing is already slightly hyperexponential

      • There are (at least) two routes to AI: 1) accumulating better code and 2) porting brain code via emulation. Lognormal post applies only to 1, this post applies to both, but only after reach near full displacement of human workers.

  • Pascal Aschwanden

    I would like to think that this could someday come true. But, realistically, a much larger limiting factor looms ahead: Government. It is the ultimate road block which prevents progress in almost every meaningful area of advancement: from self driving cars, to flying cars, to building height restrictions to innovations in building techniques to medical advancements, etc. Although government is necessary – it’s also far too slow. Before anyone starts making the exponential jumps this article hopes for, you’ll have to reform government – and that takes a long, long time.

    • Doesn’t that support the contention that the nearest rapid material increases will be space based? There aren’t many rules governing space based manufacturing/mining. There eventually will be, but this is an area where I imagine governments moving slow works in favor of growth.

    • free_agent

      OTOH, behind many technological innovations, the US government paid for the basic research. Microelectronics, the Internet and the current boom in biotech come to mind. Not to mention that government provides the infrastructure that defines and enforced property rights. And certainly there are places on earth where the hand of government rests very lightly, such as Afghanistan and the northwest province of Pakistan, but they’re not noted for high levels of prosperity.

  • Pingback: Tracking Evolution In Real Time – Just a Thought()

  • free_agent

    Something like this is already happening: Consider data-center-based information services like Google and Facebook. They resemble manufacturing companies, but the “workers” are the computers in the data centers. Their software engineers are analogous to industrial design engineers. The “workers” for these companies are essentially produced in factories, so the population of “workers” can expand very quickly.

    What you don’t get is exponential growth. But it seems to me that’s a general limitation — you can only get pure exponential growth of New Things if *every* component of the New Thing can be produced by New Things — but real New Things are created from many inputs, and the rate of creating New Things ultimately becomes limited by whichever input has the slowest exponential growth rate. E.g., if you had a factory that could produce clones of itself whenever a human pushed a button, their growth rate would ultimately becomes limited by the population of humans available to push the button.

    I note that GDP growth in industrialized society grows faster than the population. (Which is equivalent to saying humans escaped the Malthusian trap.) That seems to be achieved by the fact that GDP is not a thing, there is no particular resource whose production is growing with a doubling time of 15 years. GDP is a number formed by a complicated weighted sum of the amounts of various resources. Which points out that if there is a Next Wave with a much shorter doubling time, that, too, will be of an abstract quantity, not a specific thing, and it might be calculated in a substantially different way than GDP.