When sensible choices add up badly
Microeconomic logic
Any individual firm facing a capable and cheap AI faces a straightforward choice: automate a task and cut costs, or pay more for a human and lose on margin to a competitor that automated. In a competitive market the first choice is nearly compulsory. Nothing about it is irrational. It may even be efficient in the narrow sense.
The aggregate problem
Now imagine every firm in the economy making the same decision at the same time. Each one cuts wages to save cost. The aggregate effect is that households have less money to spend. Less household spending means less revenue for firms. Less revenue means more cost-cutting. The loop feeds itself. Economists call this demand-led contraction; a related mechanism, debt-deflation, was described by Irving Fisher in 1933.
Why it is not automatic
A real economy has several interrupters. Central banks can cut rates and stimulate demand. Governments can tax and transfer. Wages may stop falling at some floor. New kinds of work may appear fast enough to absorb displaced workers. The contraction is a risk, not a certainty — and it is a live debate whether current AI capabilities can trigger it or not.
Why this risk is taken seriously
Three of the world's most-cited economists on this question — Daron Acemoglu, David Autor, and Pascual Restrepo — have documented labour-displacing automation running ahead of labour-augmenting innovation for four decades. The risk in their work is not that machines take all jobs, but that the productivity they generate is so lopsidedly distributed that aggregate demand erodes. Fisher's old warning, in new clothes.
Name one sector where widespread cost-cutting by firms seems to have hurt aggregate demand. What happened next?