Invocations of the Lump of Labor Fallacy have recently been superseded by appeals to Jevons paradox in claiming that we shouldn’t worry about what AI progress will do to workers. As with the case of calling something a fallacy, a paradox also suggests that those who understand it are smarter and more sophisticated than those who don’t. Case in point is Einstein’s famous Twin paradox, which people love to throw around in discussions of space travel but is quite difficult to actually understand (I highly recommend this fantastic explanation).
So what is Jevons Paradox? It goes back to English economist William Stanley Jevons who in the mid 1800s observed that as technological progress allowed coal to be used more efficiently, the aggregate demand for coal went up, instead of down as one might at first expect. What gives? If output stayed constant then using less coal per unit of output would indeed reduce total consumption. This is what one would call “ceteris paribus” (all else being equal). But Jevons observed that output was going up. Why? Because by using less coal, companies were able to make their products cheaper which in turn resulted in an increase in demand. In fact in some industries, demand increased by a higher percentage than the percentage reduction in the per unit use of coal thus resulting in a net increase in coal consumption.
In economics these types of counterintuitive outcomes are surprisingly common: second order effects often outweigh first order ones. A classic case is the increased fuel efficiency of cars which resulted in an increase in overall fuel consumption: people started driving a lot more as driving became cheaper. Now we can do some simple math to help us understand the effect better. Say cars become 10% more fuel efficient per mile, but people wind up driving 25% more miles, well then the total consumption of fuel will increase: 1.25
Invocations of the Lump of Labor Fallacy have recently been superseded by appeals to Jevons paradox in claiming that we shouldn’t worry about what AI progress will do to workers. As with the case of calling something a fallacy, a paradox also suggests that those who understand it are smarter and more sophisticated than those who don’t. Case in point is Einstein’s famous Twin paradox, which people love to throw around in discussions of space travel but is quite difficult to actually understand (I highly recommend this fantastic explanation).
So what is Jevons Paradox? It goes back to English economist William Stanley Jevons who in the mid 1800s observed that as technological progress allowed coal to be used more efficiently, the aggregate demand for coal went up, instead of down as one might at first expect. What gives? If output stayed constant then using less coal per unit of output would indeed reduce total consumption. This is what one would call “ceteris paribus” (all else being equal). But Jevons observed that output was going up. Why? Because by using less coal, companies were able to make their products cheaper which in turn resulted in an increase in demand. In fact in some industries, demand increased by a higher percentage than the percentage reduction in the per unit use of coal thus resulting in a net increase in coal consumption.
In economics these types of counterintuitive outcomes are surprisingly common: second order effects often outweigh first order ones. A classic case is the increased fuel efficiency of cars which resulted in an increase in overall fuel consumption: people started driving a lot more as driving became cheaper. Now we can do some simple math to help us understand the effect better. Say cars become 10% more fuel efficient per mile, but people wind up driving 25% more miles, well then the total consumption of fuel will increase: 1.25
Let’s now consider another example to hone our intuition here. Take food as an example and consider making farm production more efficient through increased automation, such as advanced combines.. Will the demand for food be highly elastic? Well not really because there is only so much people can eat (of course we are holding population constant here, something that Malthus had worried about a lot). So if people can’t eat a lot more then what will happen? Employment in farming will go down as labor is replaced by the more efficient machines.
So for Jevons paradox to result in an increase in labor as AI technologies are broadly deployed we need elastic demand for goods. But we actually need something else first which we have so far implicitly assumed. We need competitive markets. Suppose I am a monopolist. From the analysis of monopoly we know that monopolists charge higher prices and produce less quantity than would occur under competition. Here more efficient machines translate into higher profits for the monopolist. Today’s economy has many highly concentrated markets so the assumption of lower prices is not a given. And we also don’t know yet whether the use of AI will make markets more or less competitive.
For now let’s assume that prices will in fact go down as AI unlocks efficiencies. Can demand be expected to grow so much that it will offset the labor savings? The short answer is: we don’t know! The longer answer is: it’s really complicated. Why? Well first of all it should be quite clear that much of it depends on just how much more labor efficient AI becomes over time. Obviously in the extreme where AI and robots are so radically more efficient that you can produce output without humans at all there is no amount of demand increase that can offset this. It is quite useful to write out a formula:
Units produced * human labor / unit = Total labor
Consider a 90% labor savings and now ask how much would units produced have to grow to at least offset this?
X units produced 0.1 human labor / unit = 1 Total labor
X = 1 / 0.1 = 10!
So at a 90% labor savings from AI you need a 10x demand growth to just break even on labor demand. For many products that’s clearly not going to happen. On this ground alone we should reject a simplistic invocation of Jevons paradox.
But there is another problem here. In a case like coal, we can maybe get away with what economists call a partial equilibrium analysis where we consider only one market (here the demand for coal). But AI will be everywhere and so what we really need is a general equilibrium analysis where we consider how all these changes ripple through the entire economy. This is especially true because there is a deep linkage between labor and demand for goods. If the demand for labor goes down and wages decrease, then incomes start to fall. And if people have less money they will consume less, not more. Of course here too we don’t know exactly what will happen because what if prices fall even faster than incomes? Well then maybe we can get demand growth after all!
It turns out to be easy for accomplished economists to write down models that will give you almost any result you want. You just need to fiddle a bit with the structure of the model or with the assumptions. And as we just saw nobody knows what assumptions to use because we have no idea just how powerful AI will turn out to be, how competitive markets will be, whether incomes drop faster or prices, and so on.
But that’s not even the worst part. Even if we could get the assumptions and model roughly right, economics is fundamentally descriptive. What we really need though is normative. We need a vision for the kind of society we want to live in with our growing technological capabilities. There is a coda to the Jevons paradox. Jevons ultimately turned out to be wrong in his predictions about the total demand for coal because he failed to take substitutes (such as oil and natural gas) into account. We are making much the same mistake today by assuming that there is no substitute for humans to have purpose other than work. I for one am firmly advocating for a Star Trek vision where we do in fact shrink labor. We should do to all work what we have done to farm work – automate most of it away. And instead enjoy lives of leisure, contemplation, community, exploration, etc. Plenty of ways for us humans to have purpose!
Let’s now consider another example to hone our intuition here. Take food as an example and consider making farm production more efficient through increased automation, such as advanced combines.. Will the demand for food be highly elastic? Well not really because there is only so much people can eat (of course we are holding population constant here, something that Malthus had worried about a lot). So if people can’t eat a lot more then what will happen? Employment in farming will go down as labor is replaced by the more efficient machines.
So for Jevons paradox to result in an increase in labor as AI technologies are broadly deployed we need elastic demand for goods. But we actually need something else first which we have so far implicitly assumed. We need competitive markets. Suppose I am a monopolist. From the analysis of monopoly we know that monopolists charge higher prices and produce less quantity than would occur under competition. Here more efficient machines translate into higher profits for the monopolist. Today’s economy has many highly concentrated markets so the assumption of lower prices is not a given. And we also don’t know yet whether the use of AI will make markets more or less competitive.
For now let’s assume that prices will in fact go down as AI unlocks efficiencies. Can demand be expected to grow so much that it will offset the labor savings? The short answer is: we don’t know! The longer answer is: it’s really complicated. Why? Well first of all it should be quite clear that much of it depends on just how much more labor efficient AI becomes over time. Obviously in the extreme where AI and robots are so radically more efficient that you can produce output without humans at all there is no amount of demand increase that can offset this. It is quite useful to write out a formula:
Units produced * human labor / unit = Total labor
Consider a 90% labor savings and now ask how much would units produced have to grow to at least offset this?
X units produced 0.1 human labor / unit = 1 Total labor
X = 1 / 0.1 = 10!
So at a 90% labor savings from AI you need a 10x demand growth to just break even on labor demand. For many products that’s clearly not going to happen. On this ground alone we should reject a simplistic invocation of Jevons paradox.
But there is another problem here. In a case like coal, we can maybe get away with what economists call a partial equilibrium analysis where we consider only one market (here the demand for coal). But AI will be everywhere and so what we really need is a general equilibrium analysis where we consider how all these changes ripple through the entire economy. This is especially true because there is a deep linkage between labor and demand for goods. If the demand for labor goes down and wages decrease, then incomes start to fall. And if people have less money they will consume less, not more. Of course here too we don’t know exactly what will happen because what if prices fall even faster than incomes? Well then maybe we can get demand growth after all!
It turns out to be easy for accomplished economists to write down models that will give you almost any result you want. You just need to fiddle a bit with the structure of the model or with the assumptions. And as we just saw nobody knows what assumptions to use because we have no idea just how powerful AI will turn out to be, how competitive markets will be, whether incomes drop faster or prices, and so on.
But that’s not even the worst part. Even if we could get the assumptions and model roughly right, economics is fundamentally descriptive. What we really need though is normative. We need a vision for the kind of society we want to live in with our growing technological capabilities. There is a coda to the Jevons paradox. Jevons ultimately turned out to be wrong in his predictions about the total demand for coal because he failed to take substitutes (such as oil and natural gas) into account. We are making much the same mistake today by assuming that there is no substitute for humans to have purpose other than work. I for one am firmly advocating for a Star Trek vision where we do in fact shrink labor. We should do to all work what we have done to farm work – automate most of it away. And instead enjoy lives of leisure, contemplation, community, exploration, etc. Plenty of ways for us humans to have purpose!
>300 subscribers
>300 subscribers
Share Dialog
Albert Wenger
Share Dialog
Albert Wenger
Spot on, as usual. In some fields, we will see 1/10 people needed.
Blog post for your weekend reading: More Lazy Employment Thinking (Jevons Paradox Edition) https://continuations.com/more-lazy-employment-thinking-jevons-paradox-edition