Jevons’ Paradox seems alive and well

While many people fear that AI will eliminate millions of white-collar jobs, many have argued that Jevons’ Paradox indicates that the opposite may happen. The proliferation of AI may increase employment and productivity simultaneously.

For those who don’t know Jevons’ Paradox, let me quickly recap. During the Industrial Revolution, when steam engines became ubiquitous and increased the productivity of manufacturing, many feared that increased productivity would lead to lower employment. After all, if a man with a machine can make twice as many shoes as a man making them by hand, you can theoretically reduce employment by half without reducing output. In practice, the opposite happened. As steam engines and machines proliferated, employment in the manufacturing sector soared.

The relationship that higher productivity leads to increased employment can be observed time and again. The introduction of the cash machine (ATM) did not reduce the number of bank tellers. Rather, their number grew, but instead of simply handing out cash, they now offered advice to customers.

And when it comes to AI, Microsoft CEO Satya Nadella argues that Jevons’ Paradox holds as well and will lead to higher employment in software engineering and other areas penetrated by AI.

The basic idea is that as productivity increases, coding becomes easier to do and thus people will do more of it, which ultimately creates more jobs.

That’s as far as theory goes, but what does the data say? It is early days, but in recent years, we have seen surveys that indicate that the proliferation of AI has so far not cost any jobs. But recently, I have come across another study that looks at actual employment data in the US between 2017 and 2024. It then combines this employment data with output and productivity data from the GDP data releases and wage data from the American Community Survey.

Hence, it is a top-down estimate of the macroeconomic impact of AI proliferation, starting with industrial enterprise AI in 2021 to the end of 2024. And this study indicates that Jevons’ Paradox may really hold in the case of AI.

First, the proliferation of AI in industrial applications starting in 2021 and later the proliferation of generative AI, starting with the launch of ChatGPT in 2023, has likely increased productivity. The study finds that a one standard deviation increase in AI exposure in any business increased output by 7% between 2020 and 2024.

To measure job exposure to AI, they used the index created by OpenAI and the Centre for the Governance of AI at Oxford University, which is really just a number that is hard to explain. For what it’s worth, a one standard deviation increase is roughly the equivalent of the difference in AI exposure between a worker at the US Postal Service and a worker at an online retailer.

Second, the study finds that industries across the US that had higher AI exposure also saw employment and real wages rise. A one standard deviation increase in AI exposure increased employment by 3.9% and the wage bills of affected employers by 4.8%. These numbers are smaller than the growth in output, indicating that not all productivity gains drop through to higher employment or higher real wages, but at the very least, there is statistically significant job growth.

Again, it is early days, and new data is becoming available all the time, so we need to take these results with some caution, but they are encouraging nevertheless.

Impact of AI exposure on worker earnings (top) and employment (bottom)

Source: Johnston and Makridis (2026).