Efficiency Can Backfire

Make something cheaper or more efficient and we often use far MORE of it, not less — which flips a lot of confident forecasts on their head.

The Future of Technology · Lesson 17 · 7 min read

A city, sick of traffic, spends billions widening its busiest highway to twice the lanes. For a few months it is glorious — then it is just as jammed as before, now with twice the cars. The engineers did not make a mistake; they ran into one of the most reliable surprises in all of systems thinking. What is it? Hold that question; it overturns a lot of confident “just make it efficient” forecasts.

Cheaper or more efficient often means we use MORE, not less

There is a classic, counterintuitive pattern — the Jevons paradox (also called the rebound effect, or, for roads, induced demand): when the cost of using something drops, demand for it often rises enough that total usage increases.

Widen a road → driving is easier → more people drive → traffic returns. Make engines efficient → using them is cheaper → we run far more of them. The intuition “efficiency reduces usage” quietly assumes demand stays fixed — but demand responds to price.

Why it happens: lower cost releases suppressed demand

A high cost was holding demand back: people who would have driven, computed, or lit things did not, because it cost too much. Drop the cost and you cross thresholds (Lesson 7) — all that suppressed demand shows up, plus brand-new uses that were never worth it before.

It is the same “falling cost unlocks new uses” engine from the compute lesson, now viewed from the resource side: the new uses are exactly what drives total consumption up.

Computer chips got vastly more efficient — far more calculation per unit of energy and money. The naive prediction: we would do the same computing for much less energy. What actually happened: because computing got cheap, we did astronomically more of it — streaming video, smartphones for billions, storing everything, training huge AI models — so total compute and the energy it draws exploded, even as each calculation got cheaper.

Use it to forecast — with nuance

The practical model: when you hear “this efficiency gain will reduce our use of X,” be suspicious — ask whether cheaper X will instead increase total demand for X. Efficient AI may mean we run vastly more AI (so energy use can rise even as each task gets efficient); a labor-saving tool may expand the work rather than shrink it (Lesson 16).

The honest nuance: rebound is not always total. Sometimes efficiency does net-reduce usage — especially when demand is already near-saturated (almost everyone has all they want). So the rule is not “efficiency always raises usage”; it is “never assume efficiency lowers total usage — check what happens to demand.”

Run the Jevons check on a claim: “More efficient air conditioners will cut our cooling energy.”
1. First-order: each unit of cooling costs less energy. (True.)
2. Rebound: because cooling is cheaper, more people install AC, cool more rooms, and keep them colder for longer — uses the old cost suppressed.
3. Net: total cooling energy can rise even though each unit is more efficient.
Now the nuance: if nearly everyone who wants AC already has all the cooling they want (near-saturated demand), the rebound is small and efficiency can net-reduce energy. So the forecast is not automatic — it depends on how much demand the cost was holding back.