Making a task cheaper may increase total use so much that some of the efficiency gain disappears.
Jevons observed that more efficient use of coal did not necessarily mean lower total consumption. Modern rebound theory makes this more precise: a lower effective cost may induce additional demand. In AI-supported knowledge work, cheaper production may therefore create more downstream review and maintenance.
Making a task cheaper may increase total use so much that some of the efficiency gain disappears.
In 1865, William Stanley Jevons argued that more efficient steam technology could expand coal use. Rebound reduces the expected saving. At 100% rebound that saving disappears completely; only above 100% is there backfire, also called the Jevons paradox.
Rebound and full backfire must be assessed separately. Magnitude and causes vary by context; knowledge-work rebound and verification backfire remain research questions here.
Total resource use = resource use per item × number of items. First specify the unit, such as minutes, euros or energy. A lower cost may affect the number of items.
AI transfer to investigate: AI can lower the marginal production cost of some tasks. Whether this increases volume and total review burden depends on use and system boundaries and must be tested.
Proposed test, not yet conducted: Compare human production with AI-assisted production at the same required quality. Prespecify the resource unit, system boundary and observation period. Measure production, volume, verification and maintenance. Compare total use with a comparable situation without the efficiency improvement and with the expected saving at unchanged volume. Only total use above the first comparison that is caused by the efficiency improvement counts here as backfire.
Productivity per task and organisational efficiency are different outcomes.
Prespecify the same resource unit, system boundary and observation period. Rebound offsets expected savings; backfire requires total resource use above a comparable situation without the efficiency improvement, with evidence for that causal link. An increase over a historical measurement alone is insufficient.
Jevons and Gillingham, Rapson and Wagner support the historical account and the definition of rebound. The editions and passages used are identified above. The application to AI knowledge work remains a research hypothesis.
Help improve this case
We review your contribution before updating the text or sources.