Reinstatement Effect
The reinstatement effect names the process by which technological change creates new tasks that expand rather than contract human employment. Historically, reinstatement roughly matched or exceeded displacement: the same industrial revolution that destroyed handloom weaving created factory management, machine operation, and maintenance tasks. The empirical claim central to Acemoglu's post-2018 work is that since approximately 1980, the reinstatement rate has fallen while displacement has accelerated, producing the stagnant wages and falling labor share that have characterized the US and much of Europe. AI threatens to widen the gap further by automating precisely the task categories that previous reinstatement waves created.
In the AI Story
The empirical methodology Acemoglu and Restrepo developed uses occupational task data from the Department of Labor's O*NET database to track which tasks exist, when they emerge, and how they spread across occupations. The finding: new task creation slowed dramatically after 1980, while task automation continued. The slowdown was not driven by slower technological progress — technology in this period accelerated — but by a change in the character of technological progress, toward replacement rather than complementation.
The mechanism behind the slowdown is contested but Acemoglu's proposed explanation centers on institutional changes: weakening of labor bargaining power removed pressure for firms to invest in worker-complementing technology; tax changes favored capital over labor; research incentives favored glamorous replacement projects over mundane augmentation; and the shareholder value movement created CEO incentives that reward headcount reduction regardless of productivity effect.
Applied to AI, the reinstatement question becomes urgent. Large language models target the cognitive task categories — writing, analysis, coding, design — that previous reinstatement waves created. If the replacement happens faster than new tasks emerge, the transition resembles the early industrial revolution's first century more than it resembles the mid-twentieth century's postwar expansion. The software death cross is the most visible current manifestation.
The prescriptive implication is that reinstatement is not automatic — it must be cultivated. Public investment in complementary capabilities, tax treatment that rewards hiring alongside automation, machine usefulness research agendas, and labor organization all contribute to conditions under which reinstatement can keep pace with displacement. None of these happens by accident. The postwar balance was a political achievement, and restoring it requires political mobilization adequate to the technological pressure.
Origin
The concept was formalized in Acemoglu and Restrepo's 2018 American Economic Review paper 'The Race between Man and Machine' and extended in their 2019 JEP piece and 2020 JPE paper. It provides the theoretical frame for Acemoglu's 2021 'Harms of AI' working paper and the 2024 'Simple Macroeconomics of AI' piece.
Key Ideas
Reinstatement is the historical counterweight to displacement. Without it, automation produces wage stagnation rather than shared gains.
The balance has shifted since 1980. Displacement has continued while reinstatement has slowed, producing labor's declining share of income.
The shift was institutional, not technological. Changes in bargaining power, tax structure, and research incentives, not changes in underlying science, drove the divergence.
AI threatens to widen the gap. Current frontier AI systems automate the task categories that earlier reinstatement created, without generating visible replacement tasks at comparable rates.
Appears in the Orange Pill Cycle
Further reading
- Acemoglu and Restrepo, 'The Race between Man and Machine,' AER (2018)
- Acemoglu and Restrepo, 'Automation and New Tasks,' JEP (2019)
- David Autor, 'Why Are There Still So Many Jobs?' JEP (2015)
- Acemoglu, 'The Simple Macroeconomics of AI,' Economic Policy (2024)