In Praise of Job Destruction

Here, I explain why I support job-destroying technologies such as AI and robots.

1. The Naive Concern

Nearly any beneficial technology can be attacked for “destroying jobs.” The latest suspect: AI. Many are now worried that computers will destroy white collar jobs. I, on the other hand, am only worried that it might not be good enough to destroy many jobs. I think the current round of hand-wringing about AI job-destruction is similarly fallacious to worries about other technological advances, such as machines that harvest food or manufacture cloth.

Here is a naive concern; perhaps this is what most people worrying about “job destruction” think: Suppose some new technology makes it possible to produce some good or service, X, using half as much labor per unit of X as was previously needed. Then half of the X-workers will get laid off. Or maybe total consumption of X will go up, so maybe fewer than half of the workers will get laid off, but still a lot. Also, the remaining workers will get less pay since there is “less need” for them.

On the other hand, the price of X for consumers will go down, profits will go up, and often the quality of X’s will also go up. (If there was a hand-made car, I don’t think it would be very good.)

In deciding whether the new technology is good or bad, we subjectively weigh our feelings about the X-workers against our feelings about the consumers. (We ignore the profits of the businesses, because who cares about them?) Most people have more sympathetic feelings when thinking about the poor X-workers, which makes it seem as if the technology is bad.

2. Past Job Destruction

In the Middle Ages, over 90% of the population were farmers.

Since then, we developed harvesting machines, fertilizers, better crop breeds, etc., that vastly increase food production, so that one farmer of today can produce the output of something like 100 medieval farmers. Today, under 5% of the population are farmers, and we have a lot more and better food. Presumably, a lot of farmers were forced to stop farming along the way.

If you follow the reasoning of section 1 above, this must have been a disaster. Why aren’t 85% of the population now unemployed? Somehow, we found new things for people to do. And we keep finding things for people to do, through every wave of technological advancement. The increased mechanization of factories did not lead to mass unemployment, nor did the computer revolution that occurred during my lifetime; they just made our society richer. There is no observed trend toward unemployment as technology advances.

If we had listened to the luddites, we would have avoided most of the great advances of the last 300 years, we’d still be living as peasants, and we wouldn’t have such wonders of modern civilization as indoor plumbing, cell phones, and my philosophy books.

3. The Broken Window Story

The reasoning of section 1 suggests that increased production might be overall, economically bad. The natural extension of this is to say that destruction is good. Not many people would like to put it like that, but the argument follows the same logic.

Bastiat gives the example of a child who breaks a shopkeeper’s window. Onlookers then say that this is good for the economy, because it creates business for the glazier, who will be hired to replace the window. The glazier will then have more money, which he will use to buy something else, thus further stimulating the economy, etc.

Saying that it is good to break a window is basically equivalent to saying that it is bad to stop the breaking of windows. The latter argument follows the same logic as section 1: if you stop people from breaking windows, then some number of people who are employed in replacing windows will lose their jobs.

What is wrong with the broken window argument? One problem is that it overlooks opportunity costs: because the shop owner has to pay to replace the window, he now has less money to spend on other things. Now, he’ll be unable to afford a copy of Understanding Knowledge. As a result, I will also have less money to spend on video games, etc. So the alleged stimulatory effect of the window-breaking is counter-balanced by its inhibition of other economic activity. Overall, the economy is about as well-off as it was before, except that now one person, the shopkeeper, has less stuff overall. E.g., he could have had a window and a copy of Understanding Knowledge; now, he only has a window. (For more, see Hazlitt’s Economics in One Lesson.)

If you agree that destruction is overall bad because it results in our having less stuff overall, then you should also agree that increased production is good because it results in our having more stuff overall.

4. The Concern About AI Wealth Concentration

I have heard people worry that the new wealth produced by AI will all be concentrated in the hands of, say, five people, and the rest of us will be left to starve.

This is very confused. If Sam Altman and Elon Musk accumulate some vast wealth through super-smart robots, but somehow none of this gets shared with the rest of society, why wouldn’t the rest of society just keep doing what we’re currently doing? How would Musk and Altman stop us from trading with each other? Since the rest of us, in this scenario, don’t have the robots, we keep making stuff using normal labor, to trade it to each other. Let’s say that somehow Elon gets all the farmers and all the textile workers fired so that robots can pick the fruit and sew the clothes. Why can’t the farmers go back to picking fruit using non-robot methods, to trade it to the textile workers, who also go back to making clothes using non-robot methods, etc.?

The wealth-concentration story doesn’t make sense, because in order for the AI companies to replace the farm workers, people must be buying the products from the robot farms, which means people must be able to afford them. The same goes for all the other products in the economy.

Of course, in previous rounds of automation and productivity-increase, there was no general trend of products becoming less available to ordinary people. Almost everything has steadily become more affordable.

5. How Technology Raises Wages

One way of describing the effect of productive technology is to say that we become able to produce the same amount of stuff using fewer workers. True. But the other way of saying it is that we become able to produce more stuff using the same number of workers. If you assume that total production is fixed, then it sounds like we’re going to get unemployment. But it’s closer to the truth to assume that total employment is fixed, in which case you just get more production, i.e., society gets richer. Why does the latter happen rather than the former? Basically because roughly the same number of people want to work, and there is basically no limit to how much stuff people are willing to consume.

Do productive technologies raise or lower wages in the industry that adopts them? They might lower wages due to less demand for workers. But they could also raise wages (for the people who continue to work in that industry) because each worker is now more productive. E.g., if one worker can now produce the amount that 100 workers used to produce, then that one worker’s labor is now much more valuable than it used to be.

Of course we cannot say that a round of automation or other technological improvement will always be good for everyone. There’s no guarantee that someone won’t be worse off in any given case. But the overall effect of productive technology in general is to make almost everyone vastly better off than people were in earlier centuries.

This is particularly true when the improvements are widespread, i.e., affecting many industries. When you have an improvement narrowly focused on one industry, then it is more likely that specific people in that industry will wind up worse off—certainly in the short term, and possibly even in the long term.

6. Say’s Law

Why don’t widespread productivity improvements produce the sort of unemployment, demand shortages, and impoverishment that people are worrying about? The answer lies in

Say’s Law: Aggregate supply = aggregate demand.

Aggregate supply is the total supply of all goods and services in the economy, i.e., the total value (as valued by the market) of the stuff that people are willing to sell. Aggregate demand is the total demand for all goods and services, i.e., the total amount that people are willing to pay for stuff. The reason these are said to be equal is that the supply creates the demand.

Note: This is not the idea that if you create some product, then people will start wanting that product just because you created it. It is not claiming that production alters people’s desires. (That might be true in some cases, but it’s not what Say’s Law is about.)

To see the basic point, it’s helpful to look past money, which is a distraction. Fundamentally, people don’t want money. Fundamentally, people trade valuable goods and services (“stuff”) for other valuable stuff, and money is just an intermediary that facilitates exchanges. When you “demand” blueberries in the economic sense, this just means that you offer something else of value in exchange for blueberries. Your total demand for stuff is the total amount of stuff that you are willing to trade for other stuff. And that is generally going to just be the total amount of stuff (including services) that you yourself provide. Your productivity creates your demand for other stuff, in the sense that you are only able to buy stuff because you have created some value that you can exchange for that stuff.

By the same token, the total demand of everyone for everything in the economy is determined by the total productivity of everyone in the economy. Note that “demand” in this context does not refer to people’s purely subjective desires; it refers to people’s ability and willingness to actually offer value in trade for the things that they are said to demand. Demand in that sense does not create supply; it is created by supply. This is why, if you want an economy to be stronger, you should not be trying to “stimulate demand”; you should be trying to increase productivity, and the rest will more or less take care of itself.

Btw, more recent economists have criticized Say’s Law; they think you can sometimes have an excess of supply and shortage of demand. This might happen if people increase the amount of money they want to hold and therefore refrain from consumption. If people are slow to adjust prices (including wages), then there can be an excess supply and some involuntary unemployment. However, I think these are relatively small and short-term effects, and they are not what the “technological unemployment” pessimists are worrying about. The latter people appear to be worrying that we’ll have massive unemployment because we “won’t need” workers anymore.

Say’s Law is the basic explanation for why productivity boosts over the last 200 years have not led to 90% unemployment. The productivity boosts themselves created the increased demand for labor that prevented the unemployment. People can always think of something useful for people to do. The more productive businesses are, the more they can afford to pay for people to do those things, whatever they are.

7. What Will the New Jobs Be?

Sometimes, techno-optimists try to respond to the unemployment worry by listing new jobs that will be created by the new technology. E.g., someone worries that combine harvesters are going to cause massive unemployment for farmers, and the optimist says, “Oh, don’t worry. There will be new jobs for, um, building combine harvesters. And oiling them. And repairing them, I guess.” (This almost makes it sound like we should be hoping for them to break down a lot.)

These defenses often sound lame. The new “jobs created” by the technology usually sound as if they’d be far less numerous than the jobs destroyed. If AI causes massive white collar layoffs, just how many “prompt engineer” positions do we expect to pop up to take all these disemployed workers?

That is the completely wrong approach to defending a new technology. It isn’t good because it will create jobs in that way. Insofar as it creates that kind of job—e.g., jobs fixing harvesters, or jobs checking AI outputs for hallucinations—the technology is flawed. The ideal situation is that new technologies don’t create any jobs of that kind.

The desired way for new technologies to create jobs is that they increase total productivity, which raises the total demand for all labor that is still being done by people.

Objection: “But given how amazing AI is, what if it literally replaces all human jobs?”

That would be amazing. If only that would happen! That would mean the end of scarcity. The fact that there was no demand for labor would mean that there was no work anyone could do that would add to total productivity, which would basically mean that we have satisfied all human desires (that labor previously contributed to satisfying) with no work. There would be no need for anyone to charge money for any of these things, since they’d be unlimited.

That won’t happen, though. AI will not do literally everything. I don’t know which occupations will expand and which will shrink. But in all previous rounds of productivity advances, we’ve found that there were still useful things for people to do. Whatever those things are, they wind up being worth more. As machines approach being able to do everything, the value of an hour of labor approaches infinity.

8. Trends

The value of an hour of labor has dramatically risen over the long term. Here’s a graph of worker wages in England since 1200 A.D. (source):

You can see the dramatic takeoff with the industrial revolution.

I know that I’m now going to hear that wages have stagnated in the U.S. recently, so here’s a graph of median wages in the U.S. since 1979 (source):

You can see that wages are on the increase.

There is also no trend toward unemployment resulting from technology; here is the U.S. unemployment rate since 1948 (source):

In sum, there is no sign of the economic problems the alarmists are worried about actually happening.

By the way, there are some financial markets as well as economic experts that predict economic growth for the next few years. They generally expect moderate growth; there’s no sign of an imminent growth explosion due to AI or whatever else. (1, 2) Unfortunately, AI isn’t ready to take our jobs yet.