The Race Between Machine and Man: Implications of Technology for Growth, Factor Shares, and Employment
Source: https://www.aeaweb.org/articles?id=10.1257/aer.20160696 ↗
Full text: open-access via Unpaywall ↗
A rigorous theoretical framework on the competition between automation (which displaces labour) and the creation of new tasks (which generates employment).
Acemoglu and Restrepo offer the analytical counterweight to Brynjolfsson's more optimistic view: not every productivity improvement ends up benefiting labour, and it depends heavily on whether the technology is augmenting or substituting.
Essential reading to avoid both naive optimism and fatalism when discussing AI's effects on employment — the outcome is a function of specific design choices, not a force of nature.
Central argument
Acemoglu and Restrepo argue that automation alone does not doom labor, because the economy generates a counterforce: the creation of entirely new tasks in which human labor holds a comparative advantage. Using a task-based model, they show that automation reduces employment and the labor share, but the introduction of new tasks raises wages, employment, and the labor share — and that in a stable long-run equilibrium, both forces advance in tandem. Crucially, stability is self-correcting: when automation outpaces task creation, market incentives slow further automation and accelerate the creation of new tasks, preventing a 'horse equilibrium' where labor becomes fully redundant.
Critique
The model's stabilizing mechanism — that market forces will endogenously rebalance automation and new task creation — rests on the assumption that humans retain a meaningful comparative advantage in emerging tasks. This is precisely what is contested in the current AI moment: large language models and generalist AI systems are encroaching on cognitive and creative tasks simultaneously, potentially compressing the window in which new comparative advantages can emerge before the next wave of automation arrives. The paper, published in 2018, does not adequately stress-test this assumption against AI capable of rapid cross-domain generalization, which may break the historical pattern its empirical evidence relies on.
Why it matters for product
For a CPO, the paper's core distinction between automation (replacing existing tasks) and task creation (inventing new roles where humans lead) is a direct lens for workforce and team design: investing in AI to automate current product ops is strategically different from investing in capabilities that open new product surfaces humans must invent and lead. The empirical finding that occupations with more new job titles saw 60% of US job growth from 1980–2015 suggests that product organizations should actively create new specializations — in AI oversight, model behavior design, or human-AI interaction — rather than purely optimizing headcount out, as the latter destroys the organizational substrate needed to define and own new task categories before competitors do.