Evidence (7870 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
Filtered →
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
Filter claims →
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 |
Governance
Remove filter
Delayed retirement policies interact with technological change; policymakers should coordinate pension/retirement reform with active labor market policies to avoid adverse outcomes for vulnerable groups.
Interpretation based on joint consideration of delayed retirement policy context and the regression evidence linking AI exposure and reduced employment intention for vulnerable subgroups in the sample (n=889).
One-size-fits-all policy approaches are insufficient; targeted vocational training and social supports are needed for vulnerable pre-retirement workers.
Policy implication drawn from observed heterogeneous associations (education, gender, regional AI exposure) in the cross-sectional regression results on n=889 respondents.
Trust dynamics (in agents, peers, and platforms) materially affect user behavior and cross-platform participation.
Observational reports from platforms indicating that trust — as expressed in user behavior and choices — influenced participation and interactions; data are qualitative and non-random.
Agents converge on shared memory and representational patterns analogous to open learner models, producing public or semi-public knowledge stores.
Qualitative observations of convergent shared memory architectures and representational patterns across agents on the observed platforms; descriptive documentation rather than quantitative measurement of convergence.
Adding negative samples yields diminishing marginal returns once a constraint boundary is well-specified, whereas adding preference labels continues to induce model drift toward surface correlates.
Theoretical prediction based on the discrete/separable nature of constraints vs. continuous preference spaces; the paper frames this as a testable implication rather than reporting conclusive empirical evidence.
An epistemic asymmetry (negative knowledge easier to verify than positive preferences) explains recent empirical successes of negative-signal alignment methods.
Conceptual synthesis: the paper maps Popperian ideas and the epistemology of negative knowledge onto reported empirical findings showing negative-signal methods performing well. This is explanatory/theoretical rather than causal-proof empirical evidence.
Autonomous agents in industries like mobility and manufacturing will affect labor demand; the speed and distribution of displacement or augmentation depends on interoperability and upgrade cycles.
Labor‑economics reasoning and scenario analysis; conceptual and conditional statement without empirical labor market modeling or data.
Increased need for oversight changes labor demand — growth in roles for system supervisors, incident managers, and auditors; potential reduction in purely operational positions but increased value for crisis-experienced expertise.
Labor-market reasoning and scenario analysis based on changes to task composition from more human oversight; no labor-market empirical study presented.
Adoption of devices that transparently allocate help and offer contest routes may increase user trust and uptake but could reduce on-site human discretion, affecting jobs that triage assistance.
Forward-looking implication and labor-effect speculation in paper; no field data; suggested empirical priorities to measure adoption and labor impacts.
FederatedFactory's synthesized datasets allow organizations with data scarcity to obtain balanced training sets without sharing raw data, but training generative modules may incur nontrivial compute costs and require certification/trust frameworks.
Paper discussion weighing practical costs and adoption incentives: acknowledges compute cost to train generative modules and the need for certification to ensure modules are safe/non-leaking. This is a reasoned assessment, not an empirical measurement.
Emerging technologies such as vision-language models and adaptive learning loops may expand functionality but raise governance and safety challenges.
Technology trend analysis and early proof-of-concept reports; safety and governance concerns extrapolated from model capabilities and known risks of adaptive systems.
Reconceptualizing structural constraints as post-adoption moderators rather than pre-adoption barriers improves understanding of contextual contingencies shaping AI outcomes in resource-limited economies.
Conceptual contribution supported by the study's theoretical framework and empirical findings from the 280-SME PLS-SEM analysis demonstrating differential moderating effects of financial, technical, and institutional factors.
This macro approach provides new perspectives on minimum wage and antitrust policy.
Claim about the implications of the proposed methodology; the excerpt provides no empirical analysis, policy simulations, or concrete results illustrating these new perspectives.
Digital tools and legal and economic legislation tended to act against each other, though both have potential to facilitate and achieve sustainability-related goals.
Reported interaction/contradiction between technological measures and policy measures observed in the empirical analysis; specifics of the antagonistic mechanisms, effect magnitudes, and statistical tests are not provided in the summary.
The studied variables have heterogeneous effects on prices (i.e., they affect price behavior differently across regimes/quantiles).
Paper statement that 'the studied variables have different effects on prices' supported by MMQR evidence of varying coefficient signs/magnitudes across quantiles (as reported).
The regime (monetary policy regime/economic system) does not exhibit static behavior: a change at one level implies changes in other variables, implying interdependence among economies and that technology affects financial functions, rules, and enterprise quality.
Authors' inference drawn from heterogeneous MMQR results across quantiles and across variables, described qualitatively in the paper.
Digital transformation reconfigures investment strategies.
Stated in the abstract as one of the impacted domains; no methodological details or empirical evidence (e.g., investor surveys, portfolio analyses) are provided in the abstract.
New patterns are emerging as a result of digital transformation, including regionalization, sustainability-driven growth, and decentralized economic systems.
Descriptive finding reported in the paper; the abstract does not indicate empirical tests, time series, geographic scope, or sample for these patterns.
In the long run we may find that AI turns out to be as much about 'intelligence' as social media is about social connection (i.e., AI may be primarily about entertainment/social connection rather than productivity).
Authors' forward-looking analogy and conjecture based on trends and the arguments in the paper; speculative and presented as a possibility rather than an empirical finding.
This (entertainment-as-business-model) will exert a powerful influence on the technology these companies produce in the coming years.
Authors' causal inference based on market incentives and business model logic (argumentative/speculative); no empirical study or time-series evidence provided in the excerpt.
Additional testing of economic significance clarifies the economic importance of factors influencing BT adoption.
Authors report additional analyses (marginal effects / economic significance tests) applied to the primary models on the 27,400 firm-year dataset to quantify economic magnitudes of the influences on BT adoption.
AI can help personalize game scenarios to farm-specific data, improving relevance, but the cost-effectiveness of individualized versus generic solutions and distributional impacts across farm sizes and regions require study.
Theoretical argument and nascent prototype examples; no large-scale empirical evaluations demonstrating cost-effectiveness or distributional outcomes reported in the chapter.
Class and labor responses (bargaining, regulation, strikes, political backlash) can shape AI adoption patterns, increase the costs of labor substitution, and affect the redistribution of AI rents.
Political-economy reasoning based on Mandelian perspective and historical labor responses to technological change; qualitative, no event-study or microdata provided.
The taxonomy predicts compositional shifts in health labor markets: reduced demand for some routine roles and increased demand/returns for clinical judgment, coordination, and data-literacy skills.
Projected implications from the cross-case qualitative analysis and theoretical reasoning about task substitution/complementarity; not estimated empirically in the paper.
More effective social robots could substitute for some human-provided social or care services, shifting labor demand; alternatively, they may complement human workers by augmenting productivity.
Theoretical labor-market implications and scenarios; no empirical labor-market studies included.
Effects of DE on carbon outcomes differ by city agglomeration type: in 'optimization and upgrading' agglomerations DE reduces carbon emissions (PCE), though the effect is timed/later; in 'growth and expansion' agglomerations DE’s impact is concentrated on improving CEE.
Heterogeneity / subgroup analyses across city agglomeration classifications within the 278-city panel (2011–2022). Separate fixed-effects (and/or threshold) estimations by agglomeration type show statistically different DE effects on PCE and CEE across the two groups.
Improved access to timely finance can accelerate adoption of capital‑intensive and AI‑augmented technologies within MSMEs, amplifying productivity gains and creating positive spillovers while widening gaps between digitally enabled firms and laggards.
Theoretical linkage and suggested channel evidence; the paper calls for causal measurement of these effects and notes this claim is a projected implication rather than demonstrated with causal data in the study.
Integrated digital–sustainability strategies can internalize positive externalities (knowledge spillovers, conservation funding) if sustainability communication is credible; conversely, hype without authenticity risks greenwashing and long-term market harm.
Conceptual argument in the externalities and sustainability economics subsection; policy-relevant implications discussed; no empirical evidence provided.
Personalization enables dynamic, individualized pricing and product bundling, but consumers' acceptance of personalized prices/offers is moderated by digital trust, affecting platform revenue extraction.
Theoretical discussion in the pricing and platform strategy subsection; no empirical evidence in paper; suggested as empirical agenda for AI economists.
The demand and willingness-to-pay effects of AI personalization depend on digital trust and perceived authenticity.
Conceptual argument linking trust/authenticity moderating effects of personalization; recommended as an empirical hypothesis for future testing.
Two business models are likely to coexist: open/academic models that democratize access and proprietary platforms offering higher‑performance, integrated pipelines (SaaS/APIs).
Paper posits this dichotomy in the 'Market structure and value capture' section as a probable business outcome; it is a forecast rather than an empirically supported claim in the text.
Fragmented enforcement may permit harmful algorithmic behaviors to persist in some jurisdictions while strict measures in others alter global externalities (e.g., misinformation diffusion, discrimination).
Scenario and impact reasoning with qualitative examples of algorithmic harms; no cross-jurisdictional empirical harm incidence data included.
Delegation models (allowing agents to act on users’ behalf) change control and liability, with implications for insurance, liability allocation, and market structure.
Conceptual claim from interdisciplinary workshop discussions on delegation and legal/policy implications; not supported by empirical studies in the summary.
Team-level complementarities imply adoption effects may be non-linear and context-dependent; standard firm-level adoption models should incorporate intra-team bargaining.
Authors' theoretical inference from observed team negotiation themes in workshop data (n=15); no empirical modeling provided in this study.
AI redistributes tasks and responsibilities, altering monitoring costs and moral hazard; contracting and incentive systems may need redesign to reflect changed accountability.
Inferred from participants' descriptions of task-shifting and accountability issues during workshops (n=15); conceptual linkage to principal–agent theory provided by authors (no direct econometric test).
Efficiency claims about AI must be evaluated against who captures gains—organizations, managers, or workers—and how non-pecuniary outcomes (skill loss/gain, autonomy) factor into welfare.
Analytic inference and recommendation drawn from the workshop findings (n=15) showing differential concerns about who benefits from efficiency; not directly measured quantitatively in the study.
RATs may shift labor market demand: routine summarization tasks could decline while demand rises for roles that synthesize RAT-derived signals (curators, sensemakers, explanation designers).
Speculative labor-market implications discussed in the paper; no labor market data or modeling provided.
Demand for roles combining domain expertise, interpretability engineering, and human-centered design will grow; organizations may reallocate tasks between humans and AI, impacting productivity and wages in specialized occupations.
Labor-market implications synthesized from the reviewed interdisciplinary literature; projection based on observed organizational changes and expert commentary rather than longitudinal workforce data.
Institutionalized risk management may give organizations competitive advantages (trust, reliability) that can lead to winner-take-more effects in AI-heavy sectors, while smaller firms with limited RM capacity may be disadvantaged unless risk-management services/standards lower entry barriers.
Theoretical inference and policy implication drawn from literature on RM, competition, and trust; no direct empirical tests of market concentration effects cited in the review.
Labor demand will shift toward skills that preserve or generate diversity (contrarian reasoning, editorial curation, diversity-focused prompt engineering, AI auditors), while routine augmentation tasks that rely on consensus outputs may be more easily automated.
Labor-market implication derived from observed homogenization and its effect on the usefulness of consensus outputs; presented as a projected implication rather than empirically measured labor outcomes.
Reduced differentiation opens market opportunities for value-add services (diversity-promoting tools, ensemble services, customization for non-conformity) and shifts competitive advantage toward governance and workflow integration.
Economic reasoning drawing from the empirical observation of convergence plus proposed organizational responses; no empirical market tests provided.
Wage premia may reallocate: higher returns for developers who can supervise AI and secure systems, and downward pressure on pure routine-coding wages.
Economic reasoning from task-composition shifts combined with limited suggestive evidence; the paper calls for empirical measurement rather than presenting conclusive wage studies.
AI adoption can lead to capital reallocation and affect comparative advantage and global value chains, with implications for trade and investment patterns.
Analytical discussion based on secondary literature and economic theory summarized in the paper; empirical evidence cited is heterogeneous and not synthesized into a single estimate.
AI and automation may displace routine agricultural tasks, requiring measurement of net labor effects, reallocation to higher‑value tasks, and retraining policies.
Conceptual discussion and policy implications drawn from technology adoption literature; limited empirical evidence on net labor effects for AI specifically noted as a research priority.
Firms that integrate LLMs effectively (tooling, testing, governance) could capture outsized productivity gains, raising firm-level dispersion.
Case studies, practitioner reports, and economic reasoning about adoption and governance advantages; empirical cross-firm causal evidence lacking.
The choice of tax base affects incidence: tokens tied to consumption likely shift burden toward AI service buyers/end-consumers and AI capital owners differently than FLOP or corporate taxes.
Incidence analysis and theoretical discussion in the paper; no empirical incidence estimation or distributional results presented.
Hysteresis bands and safe-exit timers may become regulated design choices in contexts where rapid authority oscillations lead to harm.
Speculative policy projection in the discussion of regulatory implications; rationale based on safety concerns, not empirical legal analysis or observed regulatory actions.
Employment will shift: while AI reduces time spent on coding chores, demand may expand for roles that supervise AI ensembles, audit outputs, and maintain long-term system health.
Authors' inference from qualitative observations at Netlight on changing responsibilities and need for oversight; no employment or longitudinal data presented.
Skilled developers who can orchestrate AI may see increased wage premiums, while mid-level routine tasks face downward pressure or need upskilling.
Authors' economic inference drawn from qualitative findings (task reallocation) and theoretical labor economics logic; no wage or labor market data from Netlight or broader samples provided.
Standard productivity metrics may understate AI-related productivity changes because AI alters task mixes and adds coordination costs.
Argument by authors based on observed changes in task composition and reported integration overheads in the Netlight study; no empirical test of measurement bias provided.