David Autor on the Future of Work and Polanyi's Paradox

Episode Summary

Executive Summary: Russ Roberts and David Autor discuss Polanyi’s Paradox: humans perform many tasks they can’t fully explain, making them hard to automate directly. Autor argues computerization reshapes jobs by substituting for routine tasks while complementing high-skill work and expanding low-skill services, producing job and wage polarization. He urges better education, more adaptable skills, and policies like a broader EITC.

Main Topics: Polanyi’s Paradox and Tacit Knowledge (Priority: 5/5): Autor explains the central idea that people routinely perform tasks they cannot explicitly describe, from bicycle riding to empathy and language interpretation, which makes these tasks difficult to codify into software. Why Some Tasks Get Automated and Others Don’t (Priority: 5/5): The conversation contrasts routine, explicit procedures that computers handle well with non-routine tasks requiring judgment, flexibility, and intuition. The key issue is not intelligence alone but whether a procedure can be written down. Environmental Control and Machine Design (Priority: 5/5): Autor argues that many successful technologies work by simplifying the environment rather than building fully humanlike machines. Driverless cars and Kiva warehouses illustrate how machines can operate when the world is structured for them. Employment Polarization (Priority: 5/5): Computerization has reduced middle-skill clerical and production jobs while increasing high-skill professional work and low-skill in-person services. This changes the occupational structure even if total employment does not collapse. Wage Effects and Labor Market Dynamics (Priority: 4/5): Autor explains that wages depend on substitution/complementarity, demand elasticity, and labor supply elasticity. High-skill workers gain more because technology complements them and supply is less elastic, while low-skill services face abundant labor supply. Education, Skills, and Adaptation (Priority: 4/5): Both speakers stress that the labor market increasingly rewards math, communication, problem-solving, and flexibility. Autor argues U.S. education is not producing enough people with the foundational skills needed for technical and adaptable work. Policy Responses and Social Adjustment (Priority: 4/5): Autor supports improving education pathways, vocational routes, and expanding the Earned Income Tax Credit, especially to childless men. He emphasizes that the main challenge is distributional, not a lack of wealth creation.

Key Arguments: Polanyi’s paradox shows that many human abilities rely on tacit knowledge, so they are hard to automate directly because they cannot be fully specified as procedures. Automation often works best by simplifying the environment so machines can perform narrow tasks reliably rather than by creating fully general human substitutes. Driverless cars are not truly humanlike; they depend on pre-mapped, curated environments and still need to stop when conditions deviate too much. Machine learning can outperform humans at narrow recognition tasks, but it often succeeds by brute-force pattern matching and can fail in obvious edge cases because it lacks deeper reasoning. Computerization substitutes for routine, codifiable middle-skill tasks while complementing high-skill work that uses judgment, creativity, and fluid intelligence. The labor market has polarized: growth at the top and bottom, decline in the middle, with many routine office and production jobs disappearing or shrinking. Wage outcomes depend on whether technology substitutes for or complements a worker, how elastic final demand is, and how easy it is for new workers to enter the occupation. Low-skill service jobs grow because they are hard to automate, but wages remain depressed because labor supply is highly elastic and many displaced workers can enter them quickly. College and STEM are not a universal answer; many jobs require broader adaptability, writing, communication, and problem-solving more than narrow technical specialization. Policy should focus on better education, stronger pathways into skilled trades and technical occupations, and labor subsidies like the EITC to raise participation and support lower-wage workers.

Data Points: Podcast date: September 18, 2014 - Opening introduction by Russ Roberts Year of Polanyi reference: 1961 - Autor cites Michael Polanyi’s The Tacit Dimension U.S. employment in agriculture around 1900: 38% - Used to illustrate how technological change can transform employment over a century U.S. employment in agriculture around 2000: 2% - Used to show the long-run decline in farm employment Google cat-recognition system processors: 16,000 processors - Example of brute-force machine learning in image recognition Google cat-recognition training data: millions of pictures - Training set used to identify cats versus non-cats U.S. area covered by impermeable surfaces: an area the size of Ohio - Illustration of how roads and built infrastructure adapt the environment for cars Labor market response timing: 1980 to present - Autor notes the return to education in the U.S. has risen almost continuously since 1980 High school completion increase: first time in decades in the 2000s - Autor notes improvement in educational attainment after a long plateau Amazon/Kiva warehouse labor: 2 physical handling points - People only load goods onto shelves and unload them; robots handle transport and organization EITC subsidy for women with dependent children: up to about $5,000 - Autor contrasts current policy generosity for this group with others EITC subsidy for men without dependent children: about $400 a year - Autor argues this group should receive a larger subsidy to increase labor force participation

Pivotal Quotes: "we know more than we can tell" — David Autor: Used to define Polanyi’s paradox and the limits of explicit procedural knowledge "I don't think the answer will be through a chemical analysis" — David Autor: Autor argues that understanding cognition will require information-processing models, not just chemistry "the human skill that is most complimented, you know, at a fundamental level by automation is flexibility" — David Autor: Autor summarizes what machines most strongly reward in human workers

Implications: Automation will likely keep reshaping work rather than eliminating it, but the gains will be uneven. The big issues are wage polarization, education gaps, and how to help workers adapt through broader skills, vocational pathways, and more supportive labor policies.