Complementarity and Substitution
The complementarity/substitution distinction is the analytical engine that connects Autor's task-based framework to wage outcomes. A technology substitutes for labor when it performs tasks workers previously performed, reducing demand for those workers. A technology complements labor when it increases the productivity of tasks workers continue to perform, raising their marginal product and therefore their wages. Most technologies do both simultaneously, and the net effect on any worker depends on the balance: whether the substitution of AI for some of her tasks is offset by the complementarity of AI with her remaining tasks. This balance is not fixed by the technology; it depends on how work is designed, which skills workers bring, and which tasks AI is deployed to perform. The distinction reframes the AI debate from a binary 'will AI take jobs?' into the more precise question: which workers will find AI substituting for them and which will find it complementing them?
In the AI Story
The distinction has deep roots in labor economics. Capital-skill complementarity — the empirical finding that physical capital is more complementary with skilled than unskilled labor — was documented by Griliches in the 1960s and extended by Krusell, Ohanian, Ríos-Rull, and Violante in the 1990s. Autor's contribution was to move the analysis down from factor-level aggregation to the task level, showing that the same technology could substitute for some tasks within an occupation while complementing others.
Segal's amplifier metaphor in The Orange Pill captures the complementarity pole of this distinction: AI amplifies the signal the human brings, making good thinking better, good writing better, good code better. But Autor's framework insists on holding both poles simultaneously. The same AI that amplifies the senior engineer's architectural judgment substitutes for the junior engineer's syntactic knowledge. The amplification and the substitution are not different technologies; they are different consequences of the same technology operating on different workers and different tasks.
The policy implications are substantial. If AI were purely substitutive, the response would be redistribution — tax the machines, support the displaced. If AI were purely complementary, the response would be acceleration — train everyone to use it. Because AI is both, the response must be more nuanced: support workers through transitions while designing work in ways that maximize complementarity. This is what Acemoglu and Restrepo call the 'race between automation and reinstatement' — whether new complementary tasks are created faster than old substituted tasks are destroyed.
Origin
The framework was formalized in Acemoglu and Autor's 2011 Handbook of Labor Economics chapter 'Skills, Tasks and Technologies: Implications for Employment and Earnings,' which remains the canonical technical treatment. The AI-specific application has been developed in Autor's work since 2020, particularly his 2024 NBER paper on rebuilding middle-class jobs through AI.
Key Ideas
Both forces operate simultaneously. Every significant technology both substitutes for and complements human labor; the question is not which, but what is the balance in a given context.
The balance is designed, not determined. How work is organized, how AI is deployed, and what skills workers bring jointly determine whether AI substitutes more than it complements — this is an institutional and managerial choice.
Complementarity requires skill. AI complements workers who bring judgment, context, and non-routine capability; it substitutes for workers whose contribution was only the routine portion now automatable.
The net effect is heterogeneous. The same AI tool produces different wage outcomes for different workers depending on their task mix and skill bundle, which is why AI's aggregate labor-market effects are so difficult to forecast.
Appears in the Orange Pill Cycle
Further reading
- Acemoglu, Daron, and David Autor. Skills, Tasks and Technologies: Implications for Employment and Earnings. Handbook of Labor Economics, 2011.
- Acemoglu, Daron, and Pascual Restrepo. Automation and New Tasks: How Technology Displaces and Reinstates Labor. Journal of Economic Perspectives, 2019.
- Brynjolfsson, Erik, Tom Mitchell, and Daniel Rock. What Can Machines Learn and What Does It Mean for Occupations and the Economy? AEA Papers and Proceedings, 2018.