There has been a lot of discussion recently about America’s Great Stagnation and its potential causes. But one cause is obvious: The moment we get close to a leap forward in an important field of innovation, artificial intelligence, everyone starts freaking out.
The “AI doomers” are going full apocalyptic, projecting that “AI will kill us all.” (James Cameron has much to answer for.) But for those whose hair is a little less on fire, Jim Pethokoukis spots the beginnings of a new wave of panic—that AI may not destroy us, but it is going to take away our jobs.
The idea is that all the money from automated systems run by AI will enrich a small, elite group of tech bros, leaving everyone else out of work. Advocates of a “universal basic income” then try to make lemonade out of this lemon by saying that AI should be taxed so the rest of us can get paid to do nothing. Since “artificial intelligence is coming for all our jobs,” writes Annie Lowrey in The Atlantic, “the government could capture some of the enormous bounty generated by this new technology and redistribute it.”
This fear of “technological unemployment” is as old as the Luddites and persists to this day. (In Kenya, modern-day Luddites are smashing tea-leaf-picking machines.) Yet the feared result has never really happened, and it’s worthwhile to understand why that is—and why this time will not be different.
‘Gilligan’s Island’ Economics
The key to grasping this is a little-understood principle of economics known as Say’s Law, after the 19th-century French economist Jean-Baptiste Say. The law is sometimes stated as “supply creates its own demand,” which may not seem plausible. After all, only the king of the salesmen could sell refrigerators to Eskimos. But the actual point is that supply and demand are really the same thing, just viewed from different perspectives. What one person produces, his supply, is also what he has to trade for everything else, which is his demand. As Say put it, “As each of us can only purchase the productions of others with his own productions—as the value we can buy is equal to the value we can produce, the more men can produce, the more they will purchase.”
To illustrate this, let’s start with one of those simplified thought experiments economists like to use: Imagine two men washed ashore on a remote island. Let’s say that one decides to specialize in hunting wild pigs, while the other specializes in picking the island’s wild-growing fruits and vegetables. What would the two men trade? The first man would trade meat for vegetables, and the second man would trade vegetables for meat. Each man’s supply, the amount he produces, is also the demand he has to offer for trade. The hunter’s supply of meat is, from another perspective, his demand for vegetables, and the planter’s supply of vegetables is his demand for meat. What each man produces (over and above what he consumes himself) is what he has to trade for the other man’s product.
The key point is that the more each man produces, the greater the benefit to the other person. One man’s prosperity doesn’t take away from the other’s—nor, as Say concluded, is one country’s prosperity earned at the expense of its neighbors. Instead, each person’s prosperity adds to that of his neighbors.
How does this apply to automation? Well, suppose we have another island where two men wash ashore. Let’s call one of them “the professor.” He’s a scientific genius and prolific inventor who devotes his time to creating gadgets that do various bits of useful work and provide goods that would not normally be obtainable on an undeveloped island. Let’s call the second man Gilligan. (If you are over a certain age, you will start to get the joke now. Look it up, kids.) Not being quite as bright as the professor, but still industrious and eager to help, Gilligan does most of the manual labor: pedaling a bicycle to run a generator for electricity or gathering fruit from the island’s trees. In this two-man economy, whatever is produced by the professor’s machines constitutes his demand for Gilligan’s time and labor.
So what will happen to Gilligan as the professor builds more machines and makes them better and more sophisticated? Will he be better off or worse off? Will the professor’s inventions increase or decrease the demand for Gilligan’s labor? Well, remember that the total amount produced by machines constitutes the demand for Gilligan’s labor—so as the machines get better, that demand increases, and does so sharply. The more the professor is able to make and do with his gadgets, the more goods he will have to trade for Gilligan’s work.
This is the application of Say’s Law to automation. Call it Say’s Law of Robots: The sum total of goods produced by automation constitutes the demand for everything that is not automated.
Ned Ludd Was Wrong
This analogy assumes, of course, that not everything can be automated. But not everything is easy to automate, the more complex tasks will resist being reduced to formulas, and everything can’t be automated all at once. The effort of automation goes first to the areas where it is most profitable, either because it is cheapest to implement (for example, an online chatbot) or because it produces the highest value (for example, self-driving cars—when we finally get them). And it always happens more slowly than we expect.
Moreover, even if the professor and Gilligan could automate everything they need for basic subsistence, they would then want to go beyond subsistence to achieve greater abundance and live in comfort, and then in luxury. Or, you know, maybe they would want to use some of that technology to get off the island. So there is always going to be something more that they want to do, and Gilligan will never be out of a job.
None of this is speculation. The simplified thought experiment just helps us explain the real-world results of 200 years of innovation and automation beginning with the Industrial Revolution. Over the past two centuries, there have been many changes in how people work, and there’s no reason to think that will stop. But “technological unemployment” has constantly been forecasted and has never actually happened. What actually happens is that as mechanization and automation increase production, they increase the total demand for labor, and the average worker gets paid more. This is how, one after another, developing countries have risen from poverty to prosperity.
Cost Disease, but in a Good Way
It’s worth noting that industrialization and automation do not increase the wages only of people directly involved in the industries that are automated first. Its benefits tend to spread widely across the economy. The explanation for this beneficial phenomenon has a deceptively unpleasant name: Baumol’s Cost Disease.
The purpose of William Baumol’s theory was to explain why increased productivity from automation tends to raise the wages of workers whose production is not automated. A summary in the Chicago Booth Review explains the reasoning:
The example Baumol and the late William G. Bowen made famous is that of the string quartet. The number of musicians and the amount of time needed to play a Beethoven string quartet for a live audience hasn’t changed in centuries, yet today’s musicians make more than Beethoven-era wages. They argued that because the quartet needs its four musicians as much as a semiconductor company needs assembly workers, the group must raise wages to keep talent—to keep its cellist from chucking a career in music and going into a better-paying job instead.
To be sure, many of us are willing to take jobs as musicians or, ahem, writers at a discount in pay because we enjoy the work and don’t want to toil on an assembly line—or, worse, become lawyers. But the pay gap can’t get too large.
More fundamentally, if we bear in mind Say’s Law of Robots, we will remember that the pay for these nonautomated jobs increases precisely because so many new goods are being produced by everybody else. The supply of goods produced by automation is the demand for everything not produced by automation.
This is not to say that every specific job that exists now will continue to exist. Some will be automated away, and people will have to seek out new work that requires a human. But history shows that the number of jobs that disappear altogether is very small. Usually, people do the same job but in a different way, using the new technology. There may not be many typesetters left in the old-fashioned sense of people who physically put together printing plates with moveable type—a rare novelty skill today. But there are an awful lot of people doing the same thing on a computer. It’s still called typesetting, but the workers who do it today usually call themselves “graphic designers.”
AI chatbots, for example, will undoubtably replace some very routine kinds of writing and in fact have already begun to do so. Or to be more exact, they will shift that writing work to people whose job is to feed prompts to the chatbot and check its results.
If we want to warn people about the effect of technological change on their jobs, we should at least warn them about the right things. We should advise them to look for how AI can enhance their work rather than replace it, and we should challenge them to focus on the creative and conceptual-level thinking that machines can’t imitate.
But the historical evidence from past technological change is on the side of Say’s Law of Robots. The supply created by machines is the demand for everything not created by machines. Once again, we should expect not to be put out of work but to get richer. It is only the doomers who will have to look for new jobs—though I expect that they, too, will find a way to adapt, seeking out all the downsides in the next new technological leap to come along.