Evidence (14055 claims)
Adoption
8570 claims
Productivity
7631 claims
Governance
6869 claims
Human-AI Collaboration
6491 claims
Org Design
4175 claims
Innovation
4114 claims
Labor Markets
3566 claims
Skills & Training
2966 claims
Inequality
2066 claims
Evidence Matrix
Claim counts by outcome category and direction of finding.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 758 | 199 | 100 | 900 | 2007 |
| Governance & Regulation | 826 | 400 | 191 | 122 | 1563 |
| Organizational Efficiency | 777 | 193 | 124 | 84 | 1189 |
| Technology Adoption Rate | 635 | 233 | 124 | 97 | 1098 |
| Research Productivity | 422 | 128 | 57 | 336 | 954 |
| Output Quality | 476 | 179 | 59 | 47 | 761 |
| Decision Quality | 328 | 177 | 81 | 47 | 640 |
| Firm Productivity | 435 | 57 | 88 | 20 | 606 |
| AI Safety & Ethics | 218 | 277 | 65 | 33 | 599 |
| Market Structure | 180 | 170 | 123 | 24 | 502 |
| Task Allocation | 213 | 64 | 72 | 33 | 387 |
| Skill Acquisition | 170 | 61 | 61 | 17 | 309 |
| Innovation Output | 203 | 27 | 43 | 18 | 292 |
| Employment Level | 105 | 54 | 107 | 13 | 281 |
| Fiscal & Macroeconomic | 131 | 69 | 43 | 26 | 276 |
| Consumer Welfare | 117 | 63 | 42 | 11 | 233 |
| Firm Revenue | 153 | 48 | 26 | 3 | 230 |
| Task Completion Time | 173 | 31 | 8 | 12 | 225 |
| Inequality Measures | 44 | 122 | 49 | 6 | 221 |
| Worker Satisfaction | 89 | 65 | 22 | 12 | 188 |
| Error Rate | 69 | 92 | 10 | 2 | 173 |
| Regulatory Compliance | 77 | 69 | 14 | 5 | 165 |
| Automation Exposure | 56 | 56 | 26 | 13 | 154 |
| Training Effectiveness | 94 | 21 | 13 | 19 | 149 |
| Wages & Compensation | 77 | 36 | 25 | 6 | 144 |
| Team Performance | 86 | 17 | 27 | 10 | 141 |
| Developer Productivity | 95 | 17 | 14 | 6 | 133 |
| Job Displacement | 12 | 80 | 20 | 1 | 113 |
| Hiring & Recruitment | 52 | 7 | 8 | 3 | 70 |
| Creative Output | 31 | 18 | 8 | 3 | 61 |
| Skill Obsolescence | 5 | 46 | 6 | 1 | 58 |
| Social Protection | 27 | 16 | 8 | 2 | 53 |
| Labor Share of Income | 17 | 19 | 17 | — | 53 |
| Worker Turnover | 11 | 12 | — | 3 | 26 |
| Industry | — | — | — | 1 | 1 |
'De-organized Growth' represents a structural shift toward decentralized, less formalized cultural work instead of firm-based expansion.
Synthesis of empirical findings: positive employment change without enterprise-count growth, plus evidence of increased platform-mediated gigs and procurement-driven work; derived from DID estimates and descriptive analyses of work organization patterns across cities (280 cities, 2008–2021).
Complaint-derived signals may degrade over time (concept drift) or be vulnerable to strategic manipulation (e.g., coordinated complaint campaigns), requiring ongoing retraining, monitoring, and anomaly detection.
Discussion/implications section warns about concept drift and manipulation risk and recommends model retraining and robustness checks; no formal adversarial tests reported.
Algorithmic credit scoring and AI can improve risk assessment but may encode historical biases or use proxies that disadvantage marginalized groups.
Synthesis of empirical examples and methodological literature on machine learning in credit scoring; the paper recommends audit methods but does not present new model evaluations.
The overall social outcome of FinTech adoption depends on technological capabilities, institutional quality, and regulatory design.
Analytical framing and political-economy model presented in the literature review; supported by cross-case comparisons rather than new empirical estimation.
Workforce transitions induced by AI imply distributional consequences (winners and losers), so policies should anticipate transitional unemployment and reskilling needs.
Inference from documented labor-market compositional changes (decline in routine tasks, growth in green occupations) combined with policy discussion in the paper; not a direct causal estimate of unemployment outcomes.
AI-enabled macro and fiscal models can improve policy testing and contingency planning but require transparency, validation, and safeguards against overreliance.
Conceptual argument and illustrative examples; no empirical trials or model performance metrics reported.
AI shifts the locus of economic governance from static rules to living systems that anticipate shocks and adapt in real time.
Policy-analytic framing and scenario-based reasoning within the book; supported by illustrative examples rather than empirical measurement.
International spillovers of AI-driven productivity depend on trade linkages and cross-border data flows; they are weaker when such linkages are limited.
Cross-country comparisons using trade flow data and measures of cross-border data policy/infrastructure; heterogeneous treatment effects in firm-level panels and country aggregates conditional on trade openness and data flow indices.
Emerging and low- and middle-income economies show smaller productivity gains (roughly 2–6%) and larger short-run job losses in routine occupations after AI adoption.
Estimates from worker-level microdata and firm panels in emerging economy samples, event studies of employment by occupation, and occupational task classification (ISCO/ISCO-08) to identify routine jobs.
White‑box mandates can constrain some high‑performance black‑box models and thereby incentivize research into explainable AI and new feature-engineering approaches compatible with rights protections.
Argument in "Innovation vs. compliance tradeoffs" linking regulatory constraints to R&D incentives; theoretical reasoning without empirical validation.
Enforced non‑discrimination and explainability requirements may change model design (fewer opaque proxies, constrained feature use), altering risk assessment and possibly increasing measured lending costs in the short run.
Theoretical modeling of model-design incentives and pricing effects in the compendium; no empirical estimation provided.
Strict upfront compliance may slow deployment but also reduce long‑run liabilities and reputational externalities, affecting venture timelines and expected returns.
Policy trade‑off analysis in the compendium; theoretical and normative argumentation without empirical longitudinal study.
Enforced explainability and non‑discrimination tests may change the design and variable use in credit models, affecting risk assessment, interest spreads and access for historically excluded groups.
Technical and policy analysis synthesizing literature on model design and fairness trade‑offs; normative projections rather than empirical demonstration.
Attribution and measurement innovations affect how value is credited across channels, altering budget allocation across publishers and influencing platform revenues.
Conceptual and policy analysis, supported by literature on attribution effects on budgets; no new empirical allocation dataset presented.
AI-driven bid optimization can increase short-term allocative efficiency (better matching) but may generate welfare-reducing externalities like privacy loss and attention capture.
Auction-market theory and empirical studies cited in literature on bid optimization; the paper synthesizes these findings rather than presenting a new randomized experiment.
Model performance, fairness, robustness, and sustainability are co-produced by technical choices plus contracts, platform policies, and regulation (co-production claim).
Conceptual synthesis combining technical evaluation literature with institutional analysis; no controlled empirical partitioning of effects provided.
Automated market and model optimization create economic efficiencies but reduce transparency for buyers, sellers, and regulators (Efficiency vs opacity trade-off).
Auction and market analysis literature and theoretical arguments; examples from RTB market structure and opaque bid optimization policies discussed; no new controlled experiment provided.
More targeted messaging can improve relevance and conversion but increases risks of nudging and informational harms (Relevance vs manipulation trade-off).
Conceptual trade-off illustrated via causal inference and targeting literature; supported by empirical studies in cited literature (not reproduced here) showing higher conversion with targeting and separate literature on persuasion risks.
The economic performance, social impacts, and durability of AI-driven advertising are determined as much by institutional arrangements (platform design, governance, regulation, market structure) as by model accuracy.
Theoretical and institutional analysis, case-study style arguments and literature references; paper does not present new randomized or large-sample empirical results quantifying the relative contribution.
Federated systems can lower barriers for advertisers and publishers who previously lacked aggregated data, but they also create coordination and infrastructure costs that may favor organizations able to invest in shared infrastructures or consortium governance.
Economic analysis and policy discussion outlining effects on entry, competition, and coordination costs. Evidence is conceptual; no empirical market-entry case studies provided.
Land-transfer effects on AGTFP are positive but constrained: institutional frictions limit the contribution of land transfer to green transformation.
Mediation results indicating a positive but limited indirect effect via land transfer/scale expansion, supplemented by discussion of institutional barriers in the paper.
Widening cross-country divergence in labor costs implies heterogeneous pathways for AI adoption and labor-market impacts across the region (high-cost countries may see faster automation and different skill-demand shifts than lower-cost ones).
Observed increased divergence in the 2013–2023 comparison across the 19-country sample plus theoretical mapping from cost levels to likely automation incentives; no direct panel evidence linking country-level cost divergence to differential AI adoption rates is provided.
The note provides 2025 projections that incorporate recent legal reforms in six countries, changing future cost estimates.
Projection exercise using the 19-country baseline (2023) and explicitly incorporating known legislative/reform changes enacted in six countries to update NWC, MCSL and CFIL projections to 2025.
Applying VIS to the electric generation sector (2014–2023) reveals significant discrepancies between conventional productivity measures and VIS-derived measures, implying conventional measures can under- or over-estimate true labor productivity once upstream labor is included.
Case study on electric generation using 2014–2023 BEA/BLS/IMPLAN data and VIS computation; comparison of VIS labor productivity against traditional direct-only productivity measures reported as materially different (magnitude/details in paper).
A subsystem methodology using Vertically Integrated Sectors (VIS) built from public BEA, BLS, and IMPLAN data (2014–2023) produces materially different labor productivity estimates for U.S. industrial and electric power sectors than conventional direct-only measures.
Empirical application of the VIS method to U.S. sectors using public BEA, BLS, and IMPLAN annual data covering 2014–2023; direct comparison reported between VIS-derived output-per-labor and conventional direct-only output-per-labor measures (case study: electric generation).
Automation reshapes job tasks — reducing demand for some routine manual roles while increasing demand for technical, supervisory, logistics-planning, and service roles — implying substantial reskilling needs rather than outright net job collapse.
Labor-market analysis using occupational employment and job-posting data (task content), supplemented by qualitative interviews and surveys tracing task changes and reskilling needs; scenario sensitivity checks on net employment under alternative adoption paths.
Broader conclusion: AI has the potential to raise productivity and create value, but without proactive policy the benefits risk being concentrated among skilled workers and firms, exacerbating inequality and regional disparities.
Integrative interpretation drawing on productivity and distributional findings from the 17 studies and theoretical considerations about differential complementarities and adoption patterns.
Whether AI is net job‑creating depends on context (sector, country, policy environment, and workforce skill composition).
Observed heterogeneity across the 17 studies by sectoral setting, country context, and policy environment; studies report differing net employment outcomes depending on these factors.
AI contributes to labor‑market polarization: growth in high‑skill opportunities alongside contraction in many middle- and low‑skill roles.
Comparative synthesis of occupational and wage-composition findings across the 17 studies shows recurring patterns of expansion at the high-skill end and reductions in middle/low-skill employment.
Cross-country variation in demand versus supply of new skills is large, and this variation is captured by a Skill Imbalance Index.
Construction of a Skill Imbalance Index at the country level that compares skill demand (vacancies requesting new skills) to proxies for skill supply (worker skill endowments or related measures); country-level comparisons show wide variation in the index.
Labor-market polarization intensifies: gains are concentrated among high-skilled workers.
Occupation-level analyses of employment and wage changes showing larger positive effects for high-skilled occupations following adoption of new skills.
Overall employment and wages rise where new skills are adopted, but these gains are uneven across workers and occupations.
Cross-sectional and panel analyses relating diffusion of new skills (measured from vacancies) to changes in employment and wages across occupations and demographic groups.
Expected differential wage pressure: wages are likely to fall for routine/low‑skill occupations and rise or remain stable for high‑skill workers who possess complementary AI skills.
Econometric studies summarized in the review (cross‑sectional and panel regressions) and theoretical consistency with SBTC; the review highlights heterogeneity in findings and limited long‑run causal certainty.
AI contributes to skills polarization: demand rises for advanced cognitive, digital, and socio‑emotional skills while routine cognitive and manual task demand declines.
Theoretical integration (SBTC), task decomposition studies showing shifts in task demand by skill content, and labour‑market analyses reporting changes in occupational skill mixes; evidence comes from cross‑sectional and panel studies summarized in the review.
AI/ML has a dual, sector- and skill-dependent effect on labor: widespread displacement of routine and lower-skilled tasks coexists with augmentation of professional and cognitive work and the creation of new labor forms (gig, platform-mediated, and human–AI hybrid roles).
Systematic synthesis of peer‑reviewed empirical studies, industry and policy reports, task‑based analyses, and firm/establishment case studies across cross‑country and sectoral analyses; empirical approaches include econometric (cross‑sectional and panel) studies linking automation/AI adoption to employment and wages, task decomposition analyses, and surveys of firm adoption and restructuring. The review notes heterogeneity across studies and limited long‑run causal evidence.
AI technical capability in the U.S. labor market is substantially larger and far more geographically diffuse than visible adoption suggests.
Agent-based simulation that maps thousands of AI tools to a skills taxonomy and a synthetic population representing the U.S. workforce (151 million agents), covering 32,000+ skills and ~3,000 counties; comparison of the Iceberg Index (skills-based exposure) to a visible-adoption wage-share metric.
The paper presents hypothesis tests assessing whether university status (and Alliance ranking) and the presence of specialized AI programs affect graduate employment effectiveness, and reports identification of key/high-performing universities.
Statement of empirical approach: hypothesis testing on effects of university status/Alliance ranking and specialized programs using the monitoring dataset; results and significance levels are reported in the full article.
Heterogeneity across universities implies that targeting high-performing institutions and diffusing their practices could be more effective than uniform expansion of AI training.
Observed variation in employment effectiveness, placement outcomes, and wages across the 191 universities; policy implication drawn from comparative performance patterns.
Labor market institutions (unions, collective bargaining), education and training systems, social safety nets, and regulations substantially mediate distributional and aggregate outcomes of AI adoption.
Comparative institutional analysis and equilibrium models linking institutional settings to wage-setting and reallocation dynamics, supported by empirical cross-jurisdiction comparisons where available.
Developing economies face different trade-offs from AI adoption than advanced economies, due to different occupational structures and complementarities.
Comparative analyses and sectoral studies drawing on cross-country microdata and institutional comparisons; theoretical models highlighting differences in task composition and absorptive capacity.
Occupational reallocation occurs: declines in some routine occupations alongside growth in AI-complementary roles (e.g., AI maintenance, oversight, and creative tasks).
Administrative and household employment data analyzed with occupational breakdowns, supplemented by task-mapping methods and panel/event-study approaches documenting shifting occupational shares over time.
Lower-skill roles experience mixed outcomes: some see adverse effects from automation while others benefit where AI is complementary to their tasks.
Microdata analyses and case studies showing heterogeneous effects by task complementarity; task-based exposure measures that differentiate which low-skill tasks are automatable versus augmentable.
AI contributes to wage polarization: earnings grow at the top of the distribution and stagnate or fall for middle occupations.
Wage distribution decompositions and panel regression studies that examine percentile-level wage changes, combined with task-based exposure measures linking AI adoption to differential impacts across the wage distribution.
The employment impact of automation depends crucially on labour-market structure (formal vs informal), availability of alternative employment, and social protections.
Theoretical framing supported by secondary literature comparing institutional contexts and their mediating effects on automation outcomes; no primary causal estimates in this paper.
Standard policy responses focused on retraining and active labor-market programs are necessary but insufficient to fully offset structural job losses where K_T substitutes broadly for tasks.
Model simulations and policy experiments in the calibrated dynamic model comparing scenarios with aggressive retraining versus structural fiscal/interventionist reforms; discussion of empirical limits from case studies and historical reskilling outcomes.
Routine automation of routine drafting tasks by GLAI may reduce demand for junior drafting labor while increasing demand for skilled reviewers, auditors, and legal technologists.
Labor-market reasoning based on task automation literature and illustrative vignettes; no labor-force survey or longitudinal employment data provided.
Improved efficiency of data centres significantly reduces capacity needs and system peaks.
Counterfactual/efficiency-improvement scenarios within the optimisation model showing lower capacity requirements and peak loads.
Low entry costs into agentic markets encourage excessive and often low-value entry, and together with behavioral frictions and signal dilution can trap agentic markets in inefficient equilibria.
Theoretical synthesis presented in abstract combining low entry cost argument with prior points about agent behavior and signal dilution; appears to be an analytical conclusion rather than reported experimental result. No sample size given.
AI-generated content contributes to signal dilution, reducing the informativeness of offers by AI agents and weakening effective product differentiation.
Claim stated in abstract as an observed/argued effect of AI-generated content on signals and differentiation; no empirical quantification or sample size given in abstract.
Empirical studies reveal persistent deviations of AI agents from optimal search behavior that function as behavioral search costs even when technical search costs approach zero.
Abstract references empirical studies (unspecified in the abstract) documenting deviations from optimal search behavior by AI agents; no sample sizes or study details provided in abstract.