I’ve just read Race Against The Machine, a new Kindle Single by Erik Brynjolfsson and Andrew McAfee, which argues contra Tyler Cowen’s The Great Stagnation that we are witnessing not a slowdown, but a positive acceleration of technological change. Brynjolfsson and McAfee argue that the fast pace of innovation is creating mismatches between humans and new technology, which has resulted in a lot of technological unemployment. The jargon is skill-biased technical change (SBTC). All recessions bring unemployment, but recent recessions have resulted in “jobless recoveries” that are the result not of cyclical forces but of deep structural change in the economy.
Brynjolfsson and McAfee are not wrong, but I think a better picture emerges if we attempt to reconcile their argument with Cowen’s rather than viewing them as contradictory. As Tyler argues, we have not had the kind of growth we might have expected 40 years ago if we had extrapolated based on the prior 40 years. See this chart on total factor productivity by David Beckworth. I think McJolfsson’s view and Cowen’s view are complementary if viewed from a sufficiently “big picture” perspective; the slowdown in TFP and the speedup in SBTC are, after all, decades-long trends.
Here’s my model. First we need to differentiate between two kinds of innovation and think about their effects. The first kind of innovation is geared toward brute maximization of production. It is typically centralized and makes use of economies of scale. Examples might include an assembly line factory or a big, coal-fired power plant. Because these innovations tend to be centralized, they introduce points of control. The capital is typically fixed and therefore easy to tax and regulate. It’s well known in the development literature that it’s really hard for governments to control rural peasants who live off the grid. Once they move to the cities and plug into centralized services, it is easier to require them to send their children to school, for instance. Because these innovations introduce points of control, I will call them technologies of control.
On the other hand, not all innovations are about brute maximization of production. Some are about producing things that we already know how to produce in ways that have ancillary benefits. An important ancillary benefit is evading control. Examples of these innovations include 3D printers and solar power. The evasion of control that is possible with 3D printers is the subject of Cory Doctorow’s short story Printcrime. And portable solar power cells can make people harder to control by supplying electricity without the need to register an address, have a bank account, stay put, and so on. These are obvious examples, but control can be evaded through more subtle innovations as well. I will call innovations that circumvent points of control that can be used by governments or monopolies to exploit, tax, or regulate technologies of resistance.
Now, postulate some background rate of innovation. How many resources will be devoted to technologies of control and how many to technologies or resistance? The answer is that it depends on how invasive the state (or other monopolies) are. When the state is invasive, at the margin the incentive is to find ways to circumvent the points of control; a greater proportion of resources will go into technologies of resistance. When the state is non-invasive, at the margin the incentive is a purer maximization of production; a greater proportion of resources will go into technologies of control, which results in higher growth.
What determines how invasive the state will be? Call me a cynic, but I think it correlates strongly with the availability of points of control. When factors of production are fixed, when demand for government supplied public goods is inelastic, when there are lots of points of control, the government will exercise more control. When the opposite is true, when there are few points of control, the government is unable to act invasively.
As you can see, there is a system of feedback. But the countervailing forces need not push outcomes to a stationary equilibrium. As we all know, time-to-build can result in cycles. Since technologies take time to change direction and develop, and since politics is slow to adapt, we should expect a non-stationary equilibrium. I think this is consistent with the broad facts. A hundred years ago, at least as it concerns white males living in the US, the government was relatively non-invasive. As a result, they developed centralized technologies that created a lot of growth, technologies of control. As new points of control were introduced, the government became more invasive. The modern state was born. At some point, innovation gradually increased toward technologies of resistance. The low-hanging fruit from the prior era eventually petered out, and sometime around 1974 we began to see lower TFP growth. As technologies of resistance improve relative to technologies of control, I can’t say exactly what will happen. A lot depends on whether government becomes gradually less invasive as points of control disappear or whether it continues to overreach; if the latter, we could observe some kind of interesting political turmoil.
So far, I’ve been pretty general about technologies of resistance, but I want to tie it back into McJolfsson’s story about rapid skill-biased technical change. The key point is that labor is extremely regulated; firms that use labor are subject to intense government control. In part this is because policies that give labor a “bigger piece of the pie” are popular with voters, and in part it is because labor can complain and enforce its rights in a way that machines cannot. If you own a business and you are subject to intense government control, you are going to invest resources in circumventing the points of control. In our economy, that means getting rid of lots of labor as cheaply as possible, which means skill-biased technical change. As Arnold Kling has said, “if a job can be defined, it can be automated or outsourced.” But it’s because there is so much control exercised in the labor market that the incentive to automate and outsource is so high.
On the other side of the labor market, I wonder if post-materialism is not also part of an attempt to evade control. A lot of talented people are scaling back their labor efforts, and while surely not all of this is due to taxes and regulations, some of it may be. And other innovations which seem truly new, such as the development of autonomous vehicles, are the result of control of which we may not even be aware; for instance, how profitable would it be to develop autonomous vehicles if Pareto-improving trade with immigrant drivers were not made impossible by immigration and labor restrictions?
The Internet has been somewhat insulated from the kind of political control that I am claiming leads to the cycle of control and resistance. As a consequence, I think we observe an epicycle there. Internet technologies can be centralized at the company level or standardized at the protocol level. Email is an example of a technology that is standardized at the protocol level, and it was developed in the early days of the Internet, when market power was a serious concern. Today, there are so many competitors in the online messaging field that market power is not a real problem. Consequently, we observe services like Facebook and Twitter, which are centralized and can provide “higher production” by reducing spam, for instance. If Facebook and Twitter ever abuse their market power too much, that is when distributed, protocol-based substitutes such as Diaspora and Status.net will take over. And when the government starts exerting more control over the Internet, we’ll observe the adoption of new technologies to circumvent that control, such as encryption and mesh networking.
In a strange way, this theory is a partial vindication of Ayn Rand; the only problem is that she was too literal. The productive people do not go on strike when they are over-controlled. Instead, they innovate around the points of control. They go on strike at the margin. And it doesn’t take a big, dramatic exit. A little bit cumulatively over decades is sufficient to both be noticeable in the data and to reduce the amount of control that can be exercised.
At the risk of being accused of now-more-than-everism, I’ll point out that the problems associated with a greater focus on technologies of resistance and with skill-biased technical change could be much ameliorated by a government that dramatically reduced its control over its citizens. Stick to supplying public goods and providing a small safety net. It won’t fix everything overnight—technology has momentum—but it will make things better than it otherwise would be. However, I think there is little chance of this happening. It requires out-of-equilibrium political play. Instead, if my theory is correct, we will find out what happens when large, invasive governments overextend and are forced to shrink.