Evidence (3308 claims)
Search and filter individual claims pulled from the papers. Looking for a specific finding ("what's the effect on wages?"), you're in the right place. Want to compare whole outcome categories against each other instead? Use the Evidence Explorer.
The board below groups claims two ways: by broad theme (nine paper-level topics) and by outcome category (the 34 claim-level outcomes that the Explorer and Syntheses also use).
Browse by theme
Nine broad, paper-level topics. Click one to filter the claims below.
Adoption
9875 claims
Filter claims →
Productivity
8807 claims
Filter claims →
Governance
7870 claims
Filter claims →
Human-AI Collaboration
7560 claims
Filter claims →
Org Design
4892 claims
Filter claims →
Innovation
4781 claims
Filter claims →
Labor Markets
4004 claims
Filter claims →
Skills & Training
3308 claims
Filtered →
Inequality
2332 claims
Filter claims →
Claims by outcome category
Counts by direction of finding. These are the same 34 outcome categories the Explorer compares and the Syntheses are written for. A linked row has a published synthesis.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 870 | 233 | 116 | 1066 | 2363 |
| Governance & Regulation | 976 | 451 | 218 | 133 | 1809 |
| Organizational Efficiency | 949 | 224 | 144 | 88 | 1416 |
| Technology Adoption Rate | 764 | 287 | 141 | 122 | 1325 |
| Research Productivity | 501 | 152 | 74 | 362 | 1101 |
| Output Quality | 542 | 216 | 69 | 69 | 896 |
| Decision Quality | 387 | 198 | 94 | 54 | 740 |
| Firm Productivity | 513 | 67 | 101 | 27 | 714 |
| AI Safety & Ethics | 249 | 303 | 73 | 36 | 667 |
| Market Structure | 190 | 192 | 134 | 27 | 548 |
| Task Allocation | 243 | 77 | 91 | 36 | 452 |
| Innovation Output | 291 | 33 | 55 | 20 | 401 |
| Skill Acquisition | 206 | 72 | 65 | 21 | 364 |
| Employment Level | 133 | 63 | 115 | 22 | 335 |
| Fiscal & Macroeconomic | 153 | 79 | 52 | 32 | 323 |
| Task Completion Time | 206 | 37 | 12 | 15 | 272 |
| Firm Revenue | 179 | 52 | 29 | 5 | 266 |
| Consumer Welfare | 130 | 76 | 47 | 13 | 266 |
| Inequality Measures | 48 | 137 | 51 | 6 | 242 |
| Worker Satisfaction | 101 | 81 | 25 | 13 | 220 |
| Error Rate | 84 | 110 | 11 | 5 | 210 |
| Wages & Compensation | 98 | 47 | 30 | 10 | 185 |
| Regulatory Compliance | 88 | 73 | 17 | 7 | 185 |
| Automation Exposure | 66 | 64 | 33 | 16 | 182 |
| Team Performance | 105 | 29 | 30 | 11 | 176 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 114 | 21 | 14 | 8 | 158 |
| Job Displacement | 12 | 90 | 24 | 1 | 127 |
| Hiring & Recruitment | 57 | 9 | 9 | 5 | 80 |
| Skill Obsolescence | 6 | 56 | 9 | 1 | 72 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 21 | 17 | 1 | 57 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Skills Training
Remove filter
Relying on secondary literature limits the paper's ability to make causal inferences and constrains empirical generalizability to all sectors or countries.
Stated limitations in the paper's Data & Methods section acknowledging scope and inferential constraints.
Increases in K_T reduce employment levels in affected firms and industries even when aggregate productivity rises.
Panel econometric estimates at firm and industry levels relating K_T intensity to employment outcomes, controlling for demand, input prices, and firm characteristics; difference-in-differences specifications and instrumental-variable robustness checks; corroborated by sectoral case studies.
Rising technological capital (K_T) — proxied by robot/automation density, software and intangible capital accumulation, AI adoption surveys, and AI-related patenting — leads to a decline in labor’s share of output.
Firm- and industry-level panel regressions linking constructed K_T intensity measures to labor shares, supported by macro growth-accounting decompositions; robustness checks include difference-in-differences and instrumenting adoption with plausibly exogenous shocks (e.g., cross-border technology diffusion, trade shocks); validated with cross-country comparisons and case studies.
The evidence base was concentrated in system-facing applications that detect or shape inequities within recruitment, evaluation and exposure systems.
Synthesis result from the scoping review indicating thematic concentration across included studies (as reported in abstract).
The study uses LinkedIn and GitHub data to examine firms' adoption of GitHub Copilot and related SWE skills and labor outcomes.
Statement of data sources and study design reported in the paper (LinkedIn profiles/skill listings linked to GitHub repository/adoption signals).
SAFI measures LLM performance on text-based representations of skills, not full occupational execution.
Methodological caveat stated by the authors clarifying the scope and limits of SAFI.
We propose an AI Impact Matrix that positions skills into four quadrants: High Displacement Risk, Upskilling Required, AI-Augmented, and Lower Displacement Risk.
Conceptual/interpretive framework introduced by the authors; described in text as proposed by the paper.
Using a strictly algorithmic baseline (mathematical bottleneck aggregation), we calculate Relative Occupational Automation Indices (OAI) for the U.S. labor market based on the DWA-level scores.
Method and calculation claim: algorithmic baseline aggregation applied across the 923 occupations / 2,087 DWAs to produce OAIs mapped to the U.S. labor market. Specific aggregation formula referenced but not numerically detailed in the excerpt.
We deconstructed 923 occupations into 2,087 Detailed Work Activities (DWAs).
Explicit data processing claim in the paper: mapping of 923 occupations to 2,087 DWAs for analysis.
Through a thematic review of existing research, the authors identified recurring themes about incentive schemes: their components, how researchers manipulate them, and their impact on research outcomes.
Authors' stated method and findings: thematic review (the scope/number of reviewed papers not specified in excerpt).
A critical aspect of conducting human–AI decision-making studies is the role of participants, often recruited through crowdsourcing platforms.
Claim based on the authors' thematic literature review noting participant sourcing practices (specific studies and counts not given in excerpt).
Researchers conduct empirical studies investigating how humans use AI assistance for decision-making and how this collaboration impacts results.
Statement summarizing the research landscape; supported implicitly by the authors' thematic review of existing empirical studies (number of studies not specified in excerpt).
Returns to AI are heterogeneous across firms; estimating treatment effects requires attention to selection, complementarities, and dynamic adoption pipelines.
Methodological argument referencing treatment-effect literature and observed firm heterogeneity; supported by conceptual examples rather than a single empirical treatment-effect estimate.
Using aggregate data, the study provides no evidence that AI benefits any particular group of workers — neither highly educated nor less-educated ones.
Authors' interaction analysis between AI adoption and human capital using aggregate panel data; reported null finding for differential benefits across education/skill groups (1995–2017, 35 OECD countries).
Survey data were collected from firms located in major Chinese cities (Beijing, Shenzhen, Xi’an, and Zhengzhou), resulting in 750 valid responses for analysis.
Reported survey sampling and data collection in the paper; explicit statement of cities sampled and number of valid responses (750).
The study relies on secondary evidence from the U.S. Census Bureau, U.S. Bureau of Labor Statistics, OECD, IMF, Stanford AI Index, McKinsey Global Institute, NBER, and recent experimental research published from 2020 onward.
Explicit methodological statement in the paper.
The paper synthesizes evidence drawing on reports from the World Economic Forum, PwC, McKinsey Global Institute, Gartner, and the International Monetary Fund.
Literature/report synthesis explicitly described in the paper (citation list to those organizations).
These pilot findings motivate a pre-registered replication that is now in preparation.
Statement in the paper reporting intention to run a pre-registered replication study following the pilot.
The results are preliminary but statistically robust.
Authors' characterization of the pilot findings (explicit statement in the paper indicating preliminary status and statistical robustness).
Raw cognitive ability or model benchmark metrics did not distinguish who engaged in complementary reasoning.
Pilot study reports lack of predictive power from cognitive ability measures and model benchmark scores for identifying participants who achieved complementary, high-performing collaboration.
Most participants deferred to the model, producing forecasts that matched the model's predictions.
Statement in the paper summarizing distribution of individual behaviors in the pilot (majority reported as deferring/matching the model).
The study used a real-money prediction market (Polymarket) as an objective, externally resolved benchmark.
Pilot study described in the paper explicitly states use of Polymarket as an external, real-money benchmark for forecast resolution.
The study employed a simplified multiple criteria assessment methodology based on global and regional expert evaluations of education quality determining knowledge and innovation development.
Methodological statement in the paper describing the approach used to analyze the Visegrad and Baltic states over 2022–2025; implies use of expert evaluations and multiple-criteria assessment.
ATHENA is not presented as a validated measurement instrument; rather, it is a conceptual and methodological scaffold for empirical validation and responsible organizational experimentation.
Explicit qualification in the paper that ATHENA is a conceptual scaffold and has not been validated as a measurement instrument (stated limitation).
A limitation of the study is that using occupation-level data prevents capturing within-occupation wage heterogeneity.
Paper statement of research limitations indicating occupation-level data as a constraint to measuring within-occupation variation.
The AI exposure index reflects potential exposure to AI technologies rather than actual firm-level AI adoption.
Paper explicit limitation: the AI exposure index measures potential exposure and is not a direct measure of firm-level adoption.
The study uses occupation-level data for 671 occupations combining wage information with an AI exposure index.
Methods description: occupation-level dataset of 671 occupations and an AI exposure index constructed/used by the authors.
The same model rebuilt to withhold answers erased the harm (i.e., removed the negative effect on unaided exam performance).
Reported as a finding from the same causal evidence referenced above (details of the study design and sample size not provided in the excerpt).
Analyses were conducted using Ordinary Least Squares (OLS) regression and Random Forest models to examine the AI–inequality relationship.
Methods statement in the paper specifying the use of OLS regression and Random Forest analysis.
The study's dataset was compiled from World Bank and OECD indicators covering a cross-section of countries.
Data description in the paper stating World Bank and OECD as primary sources for the compiled cross-country dataset.
The study uses panel data covering 30 Chinese provinces from 2010 to 2022 to analyze AI's impact on value chain upgrading.
Stated dataset description in the paper (30 provinces, 2010–2022); used for the econometric analyses reported.
The study uses panel data for Shanghai and Shenzhen A-share listed companies to examine the relationship between talent introduction and corporate AI development.
Data description in the paper specifying the sample frame as Shanghai and Shenzhen A-share listed firms in a panel setup.
The study uses microdata from the China Labor-force Dynamics Survey (CLDS) 2014–2018 combined with city-level indicators of AI diffusion and a cohort-based measure of educational mismatch, estimated with extensive fixed-effects models.
Study design and data description reported in the paper abstract/introduction.
General life satisfaction remains unaffected by AI diffusion.
Auxiliary/sensitivity analyses using life-satisfaction measures available in CLDS; reported null effect of city-level AI diffusion on general life satisfaction.
AI diffusion is not significantly associated with individual wages.
City-level indicators of AI diffusion linked to CLDS microdata; estimated main effects of AI diffusion on individual wages in fixed-effects models found no significant association.
The analysis is based on a panel of 18 European countries (2017–2024) using two-way fixed-effects models with contemporaneous, one-year, and two-year lag specifications and Driscoll–Kraay robustness checks.
Methods description in the paper's abstract (explicit methodological statement).
The total AI Vibrancy Score is not a statistically significant predictor of participation in education and training.
Two-way fixed-effects panel models (country and year effects) on a panel of 18 European countries, 2017–2024; contemporaneous specification. Reported contemporaneous coefficients: 0.4822 (ages 18-74), 0.1054 (ages 45-54), 0.5006 (ages 50-74).
Artificial intelligence is taking on advising functions and automating both the production of student work and employer-side candidate screening.
Statement in the essay (perspective/argumentative piece). The claim is supported as a conceptual observation drawing on literature on AI adoption; no empirical sample or quantified measurement reported.
There is no evidence of a sizeable effect on wages following return from cross-border employment.
Chapter 4: wage outcomes examined in linked Belgian administrative registers comparing returnees and stayers; reported null/insignificant effects on wages.
Automation AI has no significant effect on aggregate Bachelor graduations.
Chapter 3: aggregate graduation margin analyzed for automation AI exposure using IV (lagged CS research intensity) on U.S. data 2010–2022; reported null/insignificant estimates.
LLM guidance did not increase the total number of victims saved (no increase in total victims saved relative to baseline).
Same experimental comparison (two LLM-guided conditions vs no-LLM) in the simulated SAR environment; behavioral measure of total victims saved reported.
South Korea exemplifies national-scale under-augmentation: high human capital (H), substantial AI (A), but low convergence capacity (C) produce phi = 0.
Case/example presented in the paper as an illustrative national example (descriptive/case-study evidence).
This systematic literature review (SLR) synthesizes empirical studies concerning AI and the implications of these changes on labor skills across all sectors between 2017 and 2025.
Methodological claim in the paper describing its scope and timeframe (SLR of empirical studies, 2017–2025). No numeric count of included studies provided in the excerpt.
During preparation of the dissertation, I used generative AI tool ChatGPT for limited language assistance (grammar correction, stylistic refinement, and improving clarity); the intellectual content is entirely the author's own.
Author statement in the dissertation (declaration of use of ChatGPT for language assistance).
The study uses a pragmatic research philosophy and conducts a qualitative scoping review following the framework of Arksey and O’Malley.
Methodological statement in the paper (explicit).
Devil's Advocate (DA) is an AI assistant that critiques the human's initial ideas, whereas Dialectical Inquiry (DI) provides alternatives and synthesizes a resolution.
Conceptual/definitional claim in the paper describing the operationalization of DA and DI for the experiments.
This research empirically compares DA and DI in AI contexts.
Paper reports experimental comparison between AI behaviors implementing Devil's Advocate (DA) and Dialectical Inquiry (DI) across the studies.
Both studies examine benefit (information elaboration) and cost (cognitive load) pathways when AI supports SDM.
Paper explicitly frames both studies to measure information elaboration as a benefit pathway and cognitive load as a cost pathway; stated measurement plan in methods.
Study 2 tests mind-shaping interventions through user strategy training.
Study design described in the paper: a second experiment (Study 2) manipulating user strategy training (mind-shaping) to evaluate effects on SDM processes and outcomes.
Study 1 tests tool-shaping interventions by comparing three AI bot prototype conditions (Information-only, DA, DI) against a control treatment.
Study design described in the paper: randomized/controlled experiment (Study 1) with four conditions (three AI prototype conditions plus control).