CoChat
Sign Up
Cited by 3,121Labor market dynamics and wage inequalityEconomic Growth and ProductivityEconomic Theory and Policy

The DOI matches its Crossref record; 64 of the 68 references it cites were found in a registry or an index; none of the 8 sources its claims cite is checked for support yet.

The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment

Open access
Acemoglu, Daron, Restrepo, Pascual·2018·American Economic Review
DOI 10.1257/aer.20160696
Open this paper in CoChat — free accountOpen PDF

Abstract

We examine the concerns that new technologies will render labor redundant in a framework in which tasks previously performed by labor can be automated and new versions of existing tasks, in which labor has a comparative advantage, can be created. In a static version where capital is fixed and technology is exogenous, automation reduces employment and the labor share, and may even reduce wages, while the creation of new tasks has the opposite effects. Our full model endogenizes capital accumulation and the direction of research toward automation and the creation of new tasks. If the long-run rental rate of capital relative to the wage is sufficiently low, the long-run equilibrium involves automation of all tasks. Otherwise, there exists a stable balanced growth path in which the two types of innovations go hand-in-hand. Stability is a consequence of the fact that automation reduces the cost of producing using labor, and thus discourages further automation and encourages the creation of new tasks. In an extension with heterogeneous skills, we show that inequality increases during transitions driven both by faster automation and the introduction of new tasks, and characterize the conditions under which inequality stabilizes in the long run. (JEL D63, E22, E23, E24, J24, O33, O41)

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

Acemoglu and Restrepo develop a task-based model in which automation replaces labor in existing tasks while new labor-comparative-advantage tasks are created, showing automation reduces employment and the labor share and may reduce wages, while new tasks have opposite effects. In the full model, unless capital is very cheap relative to labor, the economy tends toward a stable balanced growth path with both automation and new-task creation, though labor market frictions can generate excessive automation.

Key findings

  • In the static model, automation reduces employment and the labor share and may even reduce wages; the creation of new tasks raises wages, employment, and the labor share.
  • In the dynamic model, if the long-run rental rate of capital relative to the wage is sufficiently low, the long-run equilibrium involves automation of all tasks; otherwise there is a stable balanced growth path in which automation and new-task creation advance together. In the long run, automation increases the wage but reduces employment and the labor share.
  • Empirically, occupations with 10 percentage points more new job titles in 1980 experienced 0.41 percent faster employment growth between 1980 and 2015; about 60 percent of the roughly 50 million jobs added over that period are associated with additional employment growth in occupations with new job titles. In 2000, about 70 percent of computer software developers held new job titles.
  • With heterogeneous skills, both automation and new tasks increase inequality in the short run, but standardization of new tasks can stabilize inequality in the long run. With labor market frictions, automation can reduce welfare, and the decentralized equilibrium features excessive automation.

Methods

  • Theoretical: a static task-based model with fixed capital and exogenous technology, then a dynamic model with endogenous capital accumulation, and a full model that endogenizes the direction of research toward automation and the creation of new tasks via directed technological change; extensions cover heterogeneous skills, creative destruction of profits, and welfare.
  • Empirical: uses Lin (2011) data to measure the share of new job titles within occupational categories, with US Census data for 1980, 1990, and 2000 and American Community Survey data for 2015; Figure 1 controls for the demographic composition of employment in the occupation in 1980, with further details in online Appendix B.
  • Main analysis: characterizes static equilibrium and comparative statics, derives conditions for existence, uniqueness, and stability of balanced growth paths, and analyzes long-run comparative statics and the three extensions.

Limitations

  • The paper states that the model imposes automation of tasks at the bottom, whereas in reality middle tasks may be automated; it also abstracts from industry-level technological barriers and from the need for workers to acquire new skills, which could bottleneck a rebound in labor demand. It calls for additional empirical evidence on automation’s employment and wage effects and on policy responses.
  • Evident from the text: the new-job-title empirical relationship is descriptive/associational rather than causally identified, and the theoretical results rely on specific assumptions such as balanced-growth preferences, the comparative advantage schedule, and a baseline patent structure without creative destruction.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1257/aer.20160696

Cites

64 of 68 references verified0 citation errors · 4 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 8 cited sources checked · 0 supported · 0 flagged

Show all 136 claims

  • It may state two separate claims as one. 112
  • The quote only partly states this claim. 5
  • Probably not a substantial claim. 54
148 of 148

  • The authors state this as an assumption.

    Labor has a strict comparative advantage in tasks with a higher index (and in new and more complex tasks).

    “ASSUMPTION 1: γ(i) is strictly increasing.”
    “Assumption 1 implies that labor has strict comparative advantage in tasks with a higher index”

  • The authors state this as an assumption.

    The demand for labor and capital is homothetic, ensured by either η→0 or ζ=1.

    “ASSUMPTION 2: One of the following two conditions holds: (i) η → 0 , or (ii) ζ = 1 .”

  • The authors state this as an assumption.

    The capital stock is not too large: K < K_bar, where K_bar is such that R = W/γ(N).

    “ASSUMPTION 3: We have K < _ K , where _ K is such that R = W _ γ(N) .”

  • The authors state this as an assumption.

    Skilled labor has a comparative advantage in new tasks.

    “we introduce heterogeneity in skills, and assume that skilled labor has a comparative advantage in new tasks, which we view as a natural assumption.”

  • The authors state this as an assumption.

    Capital has the same productivity in all tasks.

    “A simplifying feature of the technology described in equation (3) is that capital has the same productivity in all tasks.”

  • The authors state this as an assumption.

    Non-automated tasks can be produced with just labor.

    “Another simplifying assumption is that non-automated tasks can be produced with just labor.”

  • The authors state this as an assumption.

    New tasks created at N immediately replace tasks located at N-1.

    “we have simplified the exposition by imposing that new tasks created at N immediately replace tasks located at N -1”

  • The authors state this as an assumption.

    The utility cost of labor supply is continuously differentiable, increasing, convex, and satisfies ν″(L) + (θ -1) × (ν ′(L)) 2 /θ > 0.

    “which we assume to be continuously differentiable, increasing, and convex, and to satisfy ν″(L) + (θ -1) × (ν ′(L)) 2 /θ > 0”

  • The paper relies on this without support.

    No comprehensive framework exists incorporating automation effects and potential countervailing forces.

    “Yet, we lack a comprehensive framework incorporating such effects, as well as potential countervailing forces.”

  • The paper relies on this without support.

    Labor market frictions cause the wage rate to be above the opportunity cost of labor.

    “when the wage rate is above the opportunity cost of labor (due to labor market frictions)”

  • The authors state this as an assumption.

    The representative household's dynamic preferences are given by equation (14) with discount rate ρ > 0.

    “We assume that the representative household's dynamic preferences are given by (14) ∫ 0 ∞ e -ρt u(C(t), L(t)) dt, where u(C(t), L(t)) is as defined in equation ( 4) and ρ > 0 is the discount rate.”

  • The authors state this as an assumption.

    The comparative advantage schedule satisfies Assumption 1′: γ(i)=e^{Ai} with A > 0.

    “ASSUMPTION 1′: γ(i) satisfies (15) γ(i) = e Ai with A > 0.”

  • The authors state this as an assumption.

    The path of technology {I(t), N(t)} is exogenous.

    “The path of technology, represented by {I(t), N(t)} , is exogenous”

  • The authors state this as an assumption.

    Assumption 4: σ̂ > ζ.

    “ASSUMPTION 4: ˆ σ > ζ .”

  • The authors state this as an assumption.

    Intellectual property rights to each intermediate are held by a technology monopolist who can produce at marginal cost μψ, with μ∈(0,1), ψ>0; a competitive fringe can replicate at higher marginal cost ψ; μ is such that the unconstrained monopoly price exceeds ψ.

    “we deviate from our earlier assumption of a perfectly competitive market for intermediates, and assume that ( intellectual) property rights to each intermediate, q(i) , are held by a technology monopolist who can produce it at the marginal cost μψ in terms of the final good, where μ ∈ (0, 1) and ψ > 0 .We also assume that this technology can be copied by a fringe of competitive firms, which can Re”

  • The authors state this as an assumption.

    Developing a new intermediate that automates or replaces an existing task is viewed as infringement; the firm must compensate the existing patent holder, with compensation via a take-it-or-leave-it offer.

    “We assume that developing a new intermediate that automates or replaces an existing task is viewed as an infringement of the patent of the technology previously used to produce that task.For that reason, a firm must compensate the technology monopolist who owns the property rights over the production of the intermediate that it is replacing.We also assume that this compensation takes place with”

  • The authors state this as an assumption.

    There is a fixed supply of S scientists allocated to automation or creation of new tasks.

    “There is a fixed supply of S scientists, which will be allocated to automation ( S I (t) ≥ 0 ) or the creation of new tasks ( S N (t) ≥ 0 ), so that S I (t) + S N (t) ≤ S.”

  • The authors state this as an assumption.

    When a scientist is employed in automation, she automates κ_I tasks per unit time; in creation, she creates κ_N new tasks per unit time.

    “When a scientist is employed in automation, she automates κ I tasks per unit of time and receives a wage W I S (t) .When she is employed in the creation of new tasks, she creates κ N new tasks per unit of time and receives a wage W N S (t) .”

  • The authors state this as an assumption.

    Automation and creation of new tasks proceed in the order of the task index i.

    “We assume that automation and the creation of new tasks proceed in the order of the task index i .”

  • The authors state this as an assumption.

    Scientists differ in cost of effort: working in automation costs χ_I j Y(t), creation costs χ_N j Y(t); the distribution of χ_I j - χ_N j is smooth and increasing G over [-υ,υ], with υ small enough that all scientists work.

    “we assume that scientists differ in the cost of effort: when working in automation, scientist j incurs a cost of χ I j Y(t) , and when working in the creation of new tasks, she incurs a cost of χ N j Y(t) .26 Consequently, scientist j will work in automation if”

  • The authors state this as an assumption.

    Normalization G(0)=κ_N/(κ_I+κ_N).

    “For notational convenience, we also adopt the normalization G(0) = κ N _ κ I + κ N .”

  • The authors state this as an assumption.

    In equilibrium n(t) > max{n̅(ρ), ~n(ρ)}, so that I*(t)=I(t) and newly-automated tasks start being produced with capital immediately.

    “To simplify the exposition, let us assume that in this equilibrium n(t) > max { n ̅ (ρ), ~ n (ρ)} , so that I * (t) = I(t) and newly-automated tasks start being produced with capital immediately.”

  • The authors state this as an assumption.

    High-skill workers have a comparative advantage in new tasks.

    “The assumption that high-skill workers have a comparative advantage in new tasks receives support from the data.”

  • The authors state this as an assumption.

    There are two types of workers: low-skill workers with time-varying productivity γL(i,t) in task i, and high-skill workers with productivity γH(i).

    “we assume that there are two types of workers: low-skill workers with time-varying productivity γ L (i, t) in task i , and high-skill workers with productivity γ H (i) .”

  • The authors state this as an assumption.

    Assumption 1″: productivities are given by γH(i)=e^{A_H i}, γL(i,t)=e^{ξ A_H i} Γ(t-T(i)), where Γ is increasing with lim Γ=1, ξ∈(0,1], and T(i) is the time task i was first introduced.

    “The productivities of high-skill and low-skill workers are given by γ H (i ) = e A H i , γ L (i, t ) = e ξ A H i Γ(t -T(i)), where Γ is increasing with lim x→∞ Γ(x) = 1 , ξ ∈ (0, 1] , and T(i) denotes the time when task i was first introduced.”

  • The authors state this as an assumption.

    New technologies do not infringe the patents of the products that they replace.

    “new technologies do not infringe the patents of the products that they replace”

  • The authors state this as an assumption.

    The function ι(n(t)) is nondecreasing.

    “the function ι(n(t )) is included and assumed to be nondecreasing”

  • The authors state this as an assumption.

    There exists an upward-sloping quasi-labor supply schedule L_qs(ω) which constrains employment L ≤ L_qs(ω).

    “suppose that there exists an upward-sloping quasi-labor supply schedule, L qs (ω) , which constrains the level of employment, so that L ≤ L qs (ω)”

  • The authors state this as an assumption.

    Assumption 2′: one of three conditions holds: (i) η→0; (ii) ζ=1; or (iii) |σ-ζ| < ...

    “ASSUMPTION 2′: One of the following three conditions holds:”

  • The authors state this as an assumption.

    Labor has a comparative advantage in new tasks.

    “it is assumed that labor has a comparative advantage in these new tasks”

  • The authors state this as an assumption.

    Automation is modeled as the endogenous expansion of the set of tasks that can be performed by capital, replacing labor in tasks that it previously produced.

    “Automation is modeled as the (endogenous) expansion of the set of tasks that can be performed by capital, replacing labor in tasks that it previously produced.”

  • The authors state this as an assumption.

    The new type of technological change complementing labor takes the form of the introduction of new, more complex versions of existing tasks.

    “this takes the form of the introduction of new, more complex versions of existing tasks”

  • The authors state this as an assumption.

    The supply of high-skill labor is fixed at H and the supply of low-skill labor is fixed at L.

    “assume that the supply of high-skill labor is fixed at H and the supply of low-skill labor is fixed at L”

  • The authors state this as an assumption.

    Technology evolves exogenously with Ndot=Idot=Δ and n(t)=n > max{nbar(ρ), ntilde(ρ)} and A_H(1-θ)Δ < ρ.

    “Suppose also that technology evolves exogenously with N ˙ = I ˙ = Δ and n(t) = n > max{ _ n ( ρ), ~ n ( ρ)} (and A H (1 -θ )Δ < ρ ).”

  • The authors state this as an assumption.

    ρ > ρbar and κI/κN > κbar.

    “Suppose that ρ > _ ρ and κ I / κ N > _ κ .”

  • The authors state this as an assumption.

    Assumptions 1′, 2, and 4 hold, and there is creative destruction of profits.

    “Suppose that ρ > _ ρ , Assumptions 1′ , 2, and 4 hold, and there is creative destruction of profits.”

  • The authors state this as an assumption.

    The quasi-labor supply schedule L_qs(ω) is increasing in ω, has elasticity εL > 0, and is binding in the sense that W/C > ν'(L).

    “Suppose also that the quasi-labor supply schedule L qs (ω) is increasing in ω , has an elasticity ~ ε L > 0 , and is binding in the sense that W/C > ν ′ (L) .”

  • The authors state this as an assumption.

    Labor market frictions also constrain the social planner's choices.

    “assuming that labor market frictions also constrain the social planner's choices”

  • The authors state this as an assumption.

    Assumption 2′ holds.

    “We prove this proposition under the more general Assumption 2′ .”

  • The authors state this as an assumption.

    Assumption 2 holds.

    “Under Assumption 2, this last expression further simplifies to A K = B as claimed in the text.”

  • The authors state this as an assumption.

    Assumptions 1′, 2′, and 4 hold.

    “Suppose that Assumptions 1′ , 2′ , and 4 hold.”

  • The authors state this as an assumption.

    Assumption 3 holds.

    “This is ensured by Assumption 3, and avoids the need for additional notation at this point.”

  • The authors state this as an assumption.

    θ=0.

    “Because θ = 0 , we also have R = ρ + δ , and capital adjusts immediately and its equilibrium stock only depends on n , which becomes the unique state variable of the model.”

  • The authors state this as an assumption.

    S < ~S.

    “Suppose that S < ~ S so that Lemma A3 applies.”

  • The authors state this as an assumption.

    S < S˘.

    “By continuity there exists a threshold S ˘ such that, for S < S ˘ , the locus for v ̇ = 0 crosses zero from below at a unique point n B”

  • The authors state this as an assumption.

    S < ˇS.

    “Online Appendix B shows that there exists a threshold ˇ S such that the BGP in this case is locally stable for S < ˇ S”

  • The authors state this as an assumption.

    ρ+(θ-1)AΔ>0.

    “Finally, the transversality condition, (19), is satisfied given the condition ρ + (θ -1 ) AΔ > 0 in this part of the proposition.”

  • The authors state this as an assumption.

    ρ+(θ-1)Δ>0.

    “Thus, to generate a constant wage, aggregate output and capital growth, we must have N ˙ (t) = Δ , with ρ + (θ -1)Δ > 0”

  • The authors state this as an assumption.

    g=0.

    “When g = 0 , the locus for v ˙ = 0 crosses zero from below at a unique point”

  • The authors state this as an assumption.

    n(t)≤_n(ρ).

    “Suppose n(t) ≤ _ n (ρ) .”

  • The authors state this as an assumption.

    n(t)>_n(ρ).

    “Suppose, instead, that n(t) > _ n ( ρ ) .”

  • The authors state this as an assumption.

    n<~n(ρ).

    “Suppose that n < ~ n ( ρ) .”

  • The authors state this as an assumption.

    n≥max{_n,~n}.

    “• If n ≥ max { _ n , ~ n } , both v N (n) and v I (n) are positive and increasing in n ;”

  • The authors state this as an assumption.

    n<~n(ρ) and ρ<_ρ.

    “• If n < ~ n (ρ) (and ρ < _ ρ ), we have κ I v I (n) > 0 > κ N v N (n).”
54 of 54

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • An increase in I cannot increase n when n = N − I

    Minor

    Holding N fixed, an increase in I lowers n by definition. Proposition 5 describes an increase in n caused by an increase in I.

  • Eventual-automation BGP places the equilibrium threshold above the technology threshold

    Minor

    The paper defines the equilibrium threshold as I* = min{I, ~I}, so I* cannot exceed I. Proposition 4's eventual-automation case states the opposite.

2 of 2

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Nelson Richard R, 1966, American Economic Review 56 (½), 69

Cited by

Sorted by citations

See all 3,121

Related

See all

Metadata

Authors
Acemoglu, D., & Restrepo, P.
Affiliations
  • Department of Economics, MIT, 77 Massachusetts Avenue, Cambridge, MA 02142 (email: )
  • Boston University (US)
Venue
American Economic Review
Publisher
American Economic Association
Published
2018-06-01
Volume
108
Issue
6
Pages
1488-1542
ISSN
0002-8282 (ISSN-L)
DOI
10.1257/aer.20160696
OpenAlex
W3125498346
Semantic Scholar
60eaae8784648ca44c1b6100760ccafad03162ee
Cited by
3,121
References
66
Citations by year
2026: 698 · 2025: 667 · 2024: 570 · 2023: 371 · 2022: 295 · 2021: 218 · 2020: 141 · 2019: 99 · 2018: 32
FWCI
185.65
Type
journal article
Language
en
Open access
Locations
  • American Economic Review (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc)
Topics
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Economic Growth and Productivity
  • Social Sciences › Economics, Econometrics and Finance › General Economics, Econometrics and Finance › Economic Theory and Policy
Keywords
automation, task creation, labor share, employment, economic growth, capital accumulation, endogenous innovation, balanced growth path, wage effects, skill heterogeneity, income inequality, technological change

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review