Jevons Paradox of Intelligence

The Jevons Paradox of Intelligence extends William Stanley Jevons's 1865 observation about coal to computational cognition. Jevons demonstrated that as steam engines became more efficient—extracting more work per ton of coal—total coal consumption increased because efficiency reduced cost, expanding the range of economically viable applications. Smil applies this structural pattern to AI: as tools make knowledge workers more productive, they do not reduce total computational demand but expand it. Workers take on more tasks, tackle adjacent domains, fill pauses with prompts—the Berkeley study documented this empirically. The twenty-fold productivity multiplier Segal celebrates translates, under sustained behavioral patterns, into twenty-fold growth in inference queries, token generation, GPU-hours, electricity consumption, heat dissipation, and cooling water evaporation. Efficiency gains per operation are real and substantial; they are outpaced by demand growth, producing net increases in aggregate resource consumption. The paradox operates at individual and systemic scales simultaneously, making it one of the most reliable structural predictors of AI's physical footprint.