In the image of their tools | Science

In the United States, spending on AI infrastructure now dwarfs investment in the Internet and railroads combined, but spending on the institutions that would let societies absorb that technology is, by comparison, negligible. This asymmetry creates a paradox. Models post striking benchmark scores while aggregate economic output barely moves: Economist Daron Acemoglu projects a total factor productivity gain of well under one percent over a decade, and firm-level studies find that task-level gains do not necessarily add up at the firm or economy level. This paradox follows from an institutional mismatch that can be addressed by reverse alignment, the deliberate redesign of institutions, norms, and skills, rather than of models, so that societies can absorb new forms of machine cognition.