If you’ve been following technology news recently, you’ve probably noticed a sudden increase in references to a 19th-century economics theory called the Jevons Paradox, which is named for the neoclassical economist William Stanley Jevons, and captures the observation that increasing the efficiency of a resource can lead to greater consumption.
Jevons first articulated this idea in an 1865 book, pithily titled, The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. He argued that building more efficient steam engines – ones that required less fuel to generate the same power – would not solve the problem of England’s diminishing coal supplies. If you made the engines more efficient, Jevons predicted, people would find more applications for steam power, and even more coal would be burned overall.
This is indeed what happened. (At least, the part about increased coal consumption. The feared coal shortage was averted through new mining techniques.)
The Jevons Paradox is popular again because it provides a useful frame for understanding the potential impact of AI on jobs. Many fear that this technology will make workers so efficient that the labor market will shrink. If one programmer can now do the work of five, then companies will fire 80% of their programmers!
The Jevons Paradox implies the opposite might occur. If you make workers more efficient, their output will become cheaper, and the demand for their services might grow. If one programmer can now do the work of five, the effective cost of creating software will become so cheap that many more individuals and organizations will now pay to develop their own tools and applications.
This is a fascinating prediction that’s worth keeping an eye on. (For a deeper dive into the counterintuitive economics of AI, I recommend Derek Thompson’s recent interview with Alex Imas.) But there’s also a darker side to the Jevons Paradox that hasn’t been discussed as much recently: suddenly increasing demand for a resource can create unexpected negative side effects.
More efficient steam engines, for example, led to soot-stained buildings and the smoky start to the era of human-driven climate change. More recently, in the context of knowledge work, the arrival of digital communication tools such as email and Slack created similar unanticipated problems. By making communication significantly more efficient, the demand for fast interaction exploded, leading to our current moment in which the average knowledge worker is now interrupted once every two minutes. (For more on how this descent into communication madness occurred, check out my 2021 bestseller, A World Without Email.)
If AI ends up making certain types of workers more efficient, I hope the Jevons Paradox holds, as it’s better than the alternative of labor market contraction. But we need to remain vigilant about its side effects. It’s tempting to assume that increasing efficiency, in any context, can only make things better, but economic history has often told a more complicated tale.
This reminds me about the Pareto principle, that 20% of inputs yield 80% of (desired) outputs. Often, the ratio is starker than this.
Has anyone done the statistical work to see if the opposite is true? That is, if there is a Dark Pareto?
Do 20% of inputs also yield 80% of *undesirable* outcomes? It’s not intuitive for obvious reasons—nobody (until recently) thought it was a good idea to hit himself in the face with a hammer—but I think it’s real.
Smoking is one case. It’s a very small, simple behavior that leads to some of the worst health outcomes possible.
Does the phone foyer method remove Dark Pareto inputs?
indeed, sounds like a lot of people are contemplating Jevon these days, me included. if only he knew how popular he became. 🙂
https://open.substack.com/pub/theventureout/p/habits-specifically-for-2026?r=d2dl2&utm_medium=ios
Like steam barges, steam trains got replaced by diesel and then electric… and over short-haul flight distances these replaced flying (in civilised countries)… not all progress makes things worse. BUT that should be the default expectation… Post Silent Spring, ‘we’ clearly have not learned much of a precautionary principle – still. : (((((((
And no, no notifications on messages or email, etc., they can and will wait.
Really interesting perspective! Thanks for bringing this up for reflection, Cal.
Absolutely agreed on the increase in expected output here… I’ve noticed how many projects and ways I’ve been encouraged to use AI in, and I can already see the depth and rigour to which I can ‘keep the human in the loop is going to be increasingly limited…
I’m Catholic, I admire your work, and I see that you’ve treated our religion with respect. Thanks, Cal. Greetings from Argentina.