Jevons' Paradox is the observation that making something more efficient tends to increase total consumption of it, not decrease it. The name comes from Victorian economist William Stanley Jevons, who noticed in 1865 that improvements to the steam engine — which made it burn far less coal per unit of work — did not reduce coal consumption in Britain. It caused it to explode. Cheaper, more efficient engines made it economical to run more of them, in more places, doing more things. The efficiency gain unlocked new demand that had previously been too expensive to satisfy.
The paradox reappears reliably across economic history. Fuel-efficient cars lead to more driving, not less, because the lower cost per mile makes longer trips worthwhile. More efficient lighting leads to more brightly lit buildings, not a reduction in electricity use. In every case, the efficiency improvement does not shrink the footprint of the technology — it expands it, because it lowers the threshold at which using more becomes rational. The expected saving gets absorbed by new activity rather than pocketed as reduction.
In the context of AI, the paradox matters enormously for thinking about labour and productivity. The natural assumption is that if AI makes cognitive work faster and cheaper, less cognitive work will be needed from humans. Jevons' insight suggests the opposite is more likely: cheaper and faster thinking will make it economical to tackle far more ambitious problems than we currently attempt. Drug discovery, personalised education, climate modelling, and mental healthcare are all domains where the bottleneck is not imagination but the sheer cost of the cognitive work required. Remove that bottleneck, and demand for the work — and for the humans who can direct and refine it — tends to grow.
The practical implication is not that efficiency improvements are bad, or that we should resist them. It is that the effects of efficiency rarely run in the direction common sense predicts. When someone argues that AI will reduce the need for human intelligence, they are implicitly assuming that the total demand for intelligence is fixed. Jevons spent his career demonstrating that this assumption is almost always wrong. It has not become more correct since 1865.
