This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papersSummary Main Finding Notes QJE
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
[Claude classification]: This paper makes a foundational methodological contribution by developing parallel measures to distinguish augmentation (labor-complementary) from automation (labor-substituting) innovations using patent texts. The IV strategy using breakthrough patents from 20 years prior is innovative. The paper finds that automation effects have intensified in recent decades (1980-2018) while augmentation effects have not kept pace. The measure of 'new work' using Census Alphabetical Index titles is validated against Google Ngram usage patterns. The paper studies historical automation broadly rather than contemporary AI specifically, though it discusses implications for AI in the conclusion.
New Frontiers: The Origins and Content of New Work, 1940-2018
Autor, Chin, Salomons, Seegmiller
2022NBER Working Paper Series65 citations
Observational labor marketCausalTheoretical model
Automation / RobotsAI (General)Augmentation vs. substitutionGeneral automationRoutine task change
Autor et al. use a novel database of new job titles from Census Alphabetical Indexes (1940-2018) and patent-based measures of augmentation and automation innovations to study the origins, content, and labor market effects of new work creation in the United States.
The majority of current US employment (60% as of 2018) is found in job titles that did not exist in 1940; augmentation innovations causally increase new work creation and occupational employment, while automation innovations causally reduce employment but do not generate new work; the locus of new work shifted from middle-paid occupations (1940-1980) to high-paid professional and low-paid service occupations (1980-2018).
- Key Methods
- Natural language processing (word embeddings) to link patent texts to occupational outputs (augmentation) and tasks (automation); instrumental variables using breakthrough patents from 20 years prior to identify causal effects; long-difference regressions at occupation and occupation-by-industry levels.
- Sample Period
- 1940-2018
- Geographic Coverage
- US
- Sample Size
- Approximately 30,000 occupational micro-titles and 20,000 industry micro-titles per decade; ~7 million patents (1920-2018); 166 consistent occupations (1940-1980), 306 consistent occupations (1980-2018); 33,977 occupation-by-industry cells
- Level of Analysis
- Occupation, Industry
- Occupation Classification
- Census occupation codes
- Industry Classification
- Census industry codes
- Replication Package
- Yes