Rebuilding the Middle Class

In The You On AI Encyclopedia

The thesis represents a significant evolution in Autor's own thinking. His earlier work documented the destructive pattern of polarization; his recent work asks whether the pattern is structurally necessary or whether technological and institutional choices could produce different outcomes. The shift is from diagnosis to prescription, and it depends on a specific claim about AI's character: that it reduces the cost of performing high-skill tasks by workers who lack the formal credentials previously required.

The argument resonates deeply with Segal's democratization narrative in You On AI. The engineers in Trivandrum are not junior workers doing expert work because they have become experts; they are doing expert work because AI has lowered the threshold of expertise required. The backend engineer building frontend features is not a frontend engineer; she is a backend engineer whose tool now enables her to produce frontend features. If this pattern generalizes — if AI turns out to lower thresholds rather than eliminate them — Autor's hopeful scenario becomes plausible.

But the thesis is explicitly conditional. Whether AI rebuilds the middle class or hollows the top depends on how the technology is deployed, which is an institutional question. Firms can deploy AI to enable broader participation (the nurse-as-diagnostician scenario) or to concentrate expertise (the senior engineer managing AI agents scenario). Autor's implicit argument is that the choice is not yet made, and the direction will be determined by the regulatory, educational, and organizational structures society builds over the next decade.

Origin

The thesis was developed in Autor's 2024 NBER paper 'Applying AI to Rebuild Middle Class Jobs' and elaborated in his MIT working paper 'AI Could Actually Help Rebuild the Middle Class,' published in Noema magazine. The argument represents the culmination of three decades of empirical work on technology and labor markets.

Key Ideas

AI lowers expertise thresholds. Unlike prior automation, which eliminated routine middle-skill tasks, AI can enable moderately trained workers to perform tasks previously restricted to experts with formal credentials.

The outcome is institutionally determined. Whether AI rebuilds the middle class or hollows the top depends on how firms, regulators, and educators deploy the technology — not on inherent technological properties.

Licensing and credentials matter. Occupations protected by credentialing barriers (medicine, law, architecture) are particularly consequential test cases: if AI enables paraprofessionals to do credentialed work, the institutional response determines the distributional outcome.

The window is narrow. The decade 2025-2035 is likely decisive — institutions built during this window will shape AI's labor-market effects for a generation, just as post-1980 deregulation shaped globalization's effects.