Evidence (14922 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
9047 claims
Filter claims →
Productivity
8066 claims
Filter claims →
Governance
7278 claims
Filter claims →
Human-AI Collaboration
6912 claims
Filter claims →
Org Design
4439 claims
Filter claims →
Innovation
4359 claims
Filter claims →
Labor Markets
3652 claims
Filter claims →
Skills & Training
3018 claims
Filter claims →
Inequality
2160 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 | 795 | 210 | 105 | 955 | 2131 |
| Governance & Regulation | 886 | 414 | 197 | 126 | 1654 |
| Organizational Efficiency | 826 | 204 | 129 | 87 | 1257 |
| Technology Adoption Rate | 681 | 259 | 128 | 110 | 1189 |
| Research Productivity | 464 | 138 | 65 | 349 | 1028 |
| Output Quality | 503 | 196 | 61 | 53 | 813 |
| Decision Quality | 351 | 180 | 84 | 51 | 673 |
| AI Safety & Ethics | 238 | 288 | 71 | 34 | 637 |
| Firm Productivity | 455 | 58 | 92 | 20 | 631 |
| Market Structure | 186 | 172 | 123 | 25 | 511 |
| Task Allocation | 222 | 70 | 76 | 34 | 407 |
| Innovation Output | 238 | 28 | 48 | 18 | 334 |
| Skill Acquisition | 177 | 62 | 62 | 17 | 318 |
| Employment Level | 107 | 57 | 108 | 13 | 287 |
| Fiscal & Macroeconomic | 135 | 72 | 44 | 26 | 284 |
| Firm Revenue | 172 | 50 | 28 | 5 | 256 |
| Consumer Welfare | 121 | 68 | 45 | 12 | 246 |
| Task Completion Time | 183 | 33 | 10 | 13 | 240 |
| Inequality Measures | 45 | 126 | 50 | 6 | 227 |
| Worker Satisfaction | 95 | 74 | 23 | 12 | 204 |
| Error Rate | 77 | 98 | 11 | 4 | 190 |
| Regulatory Compliance | 84 | 73 | 17 | 7 | 181 |
| Automation Exposure | 61 | 61 | 27 | 14 | 166 |
| Training Effectiveness | 98 | 21 | 14 | 19 | 154 |
| Wages & Compensation | 78 | 37 | 25 | 6 | 146 |
| Developer Productivity | 105 | 18 | 14 | 6 | 144 |
| Team Performance | 87 | 17 | 28 | 10 | 143 |
| Job Displacement | 12 | 83 | 23 | 1 | 119 |
| Hiring & Recruitment | 53 | 8 | 8 | 3 | 72 |
| Social Protection | 39 | 17 | 8 | 2 | 66 |
| Creative Output | 32 | 20 | 8 | 3 | 64 |
| Skill Obsolescence | 5 | 50 | 6 | 1 | 62 |
| Labor Share of Income | 17 | 20 | 17 | — | 54 |
| Worker Turnover | 15 | 15 | — | 3 | 33 |
| Industry | — | — | — | 1 | 1 |
For graduates of Technical and Vocational Education and Training (TVET), acquiring advanced digital skills significantly narrows the income gap with general higher education graduates.
Heterogeneity analysis on KLIPS micro-data examining interaction of educational pathway (TVET vs general higher education) with possession of advanced digital skills in extended Mincerian wage regressions; the result reported is a significant narrowing of the earnings gap (no numeric magnitude given in the excerpt).
Quantitatively, AI-adopting firms raise aggregate value-added total factor productivity by approximately 1.51% in a representative post-adoption year.
Aggregate TFP decomposition/aggregation based on estimated firm-level treatment effects and value-added weights (methodological details in paper); the 1.51% figure is the reported quantitative estimate for a representative post-adoption year.
AI functions as an innovation-enabling intangible investment that supports productivity growth.
Synthesis of empirical findings: increased patenting and patent quality, increased R&D (but not capex), improved productivity and market value; evidence derived from the firm's adoption-timing measure and stacked diff-in-diff estimates.
AI adoption enhances knowledge recombination (increased recombination across technologies).
Increases in measures such as patent originality, generality, and technological distance interpreted as evidence of enhanced knowledge recombination; estimated with the stacked diff-in-diff design.
Evidence on mechanisms indicates AI improves firm-level efficiency.
Mechanism tests reported in the paper linking AI adoption to improved efficiency metrics (e.g., productivity measures) using the same empirical strategy; specific metrics and sample size not provided in the abstract.
The effects of AI adoption on innovation outcomes are stronger for firms with a more focused business scope.
Heterogeneity analysis by firms' business scope (more focused vs. less focused) within the stacked diff-in-diff framework; outcome assessed on innovation measures such as patenting and quality.
Post-adoption patents span more technologically distant classes (greater technological distance / broader technological scope).
Patent-class based measures of technological distance and class-spanning applied to patents from adopter firms versus nonadopters in the diff-in-diff design.
Post-adoption patents exhibit greater originality and greater generality.
Patent-level measures of originality and generality (standard patent metrics) estimated in the stacked diff-in-diff framework comparing adopters to nonadopters.
After AI adoption, firms have a higher share of 'exploitative' patents that build on the firm's existing technologies.
Classification of patents as exploitative (building on firm’s prior technologies) and comparison across adopters and nonadopters using the staggered adoption diff-in-diff design.
AI-powered developer tools (often based on large language models) aim to automate routine tasks and make secure software development more accessible and efficient.
Framing/assumption in the paper's introduction (general description of such tools' intended purpose; not directly measured in this experiment).
Organizations increasingly adopt AI-powered development tools to boost productivity and reduce reliance on limited human expertise, especially in security-critical software development.
Background/contextual claim stated in the paper to motivate the study (general trend claim; likely supported by prior literature but not by the study's experimental data described here).
AI-driven FinTech solutions function as strategic enablers of competitiveness in international markets by enhancing speed, reliability, and cost-effectiveness of trade finance operations.
Synthesis conclusion from the quantitative analysis linking AI adoption to operational gains (speed, reliability, cost-effectiveness) and competitive outcomes; competitive impact measurement and sample details not provided in the summary.
Predictive analytics and machine learning models strengthened credit evaluation and fraud monitoring, thereby reducing uncertainty and information asymmetry in global trade transactions.
Quantitative findings attributing improvements in credit evaluation accuracy and fraud monitoring effectiveness to predictive analytics/ML; the summary does not provide measures (e.g., accuracy, AUC), sample size, or statistical details.
Transaction cost reduction is a critical mediating factor linking AI-enabled FinTech innovations to improved trade outcomes.
Reported mediation relationship in the quantitative analysis indicating transaction cost reduction mediates the effect of AI adoption on trade outcomes (mediation model specifics and sample size not given).
AI minimized financial risks through enhanced risk assessment and fraud detection.
Quantitative analysis linking AI-driven mechanisms (risk assessment, fraud detection systems) to reductions in financial risk metrics; specific risk measures, effect sizes, and sample size not reported in the summary.
AI accelerated cross-border payment processes.
Reported quantitative evaluation of AI adoption effects on operational efficiency components, with cross-border payment speed cited as an improved component (measurement details and sample size not specified).
AI integration significantly improved international trade efficiency.
Quantitative analysis evaluating relationships among AI adoption, operational efficiency variables, and international trade efficiency; the paper reports a statistically significant improvement (exact tests, p-values, and sample size not provided in the summary).
Cross-talk between distributed systems and LLM-team research yields rich practical insights.
Conclusion drawn by the authors based on their mapping and findings (qualitative claim supported by the paper's arguments and examples; excerpt lacks concrete metrics).
There is recent and increasing interest in forming teams of LLMs (LLM teams).
Claim made in the paper asserting increased interest and deployment at scale; supported in the paper by literature/contextual citations and reported deployments (specific numbers or studies not provided in the excerpt).
The study contributes a conceptual architecture for next-generation accounting automation that bridges traditional compliance models and modern financial infrastructure (enabling real-time validation, automation, and transparency).
Presentation of a proposed conceptual architecture in the paper, supported by empirical evaluation and stakeholder feedback; claimed as a primary contribution. (The summary does not include architecture diagrams, implementation details, or performance benchmarks beyond the reported metrics.)
Integrating ML and blockchain represents a transformative shift that addresses limitations of traditional financial governance (static ledgers, manual reconciliation, retrospective audits).
High-level argument supported by the study's empirical improvements (fraud detection, reconciliation time, transaction accuracy) and conceptual analysis mapping system capabilities to shortcomings of traditional models. (This is a synthesis/interpretation rather than a single measured outcome.)
Stakeholder validation confirms the system's operational feasibility with 95% approval.
Stakeholder validation (presumably via survey or consultation) reporting 95% approval for operational feasibility. (The summary does not specify the number of stakeholders, selection criteria, or survey instrument.)
The study validates theoretical frameworks such as triple-entry accounting (Grigg, 2024) and X-Accounting (Faccia et al., 2020).
Conceptual/theoretical alignment demonstrated by mapping the hybrid ML-blockchain architecture and empirical findings to the premises of the cited frameworks. (Summary does not specify formal validation method or criteria.)
The system maintains 99.8% transaction accuracy.
Reported transaction accuracy measured on the same empirical datasets (public-sector financial records and private-sector supply chains) used to evaluate the hybrid system. (The summary does not provide sample size, timeframe, or definition of 'transaction accuracy'.)
The hybrid system produces a 60% reduction in reconciliation time.
Empirical measurement of reconciliation time on datasets from public-sector financial records and private-sector supply chains comparing hybrid ML-blockchain workflows to traditional reconciliation processes. (No sample size or absolute times provided in the summary.)
A hybrid ML-blockchain system achieves a 9.8% improvement in fraud detection accuracy (F1-score).
Quantitative evaluation using empirical data drawn from public-sector financial records and private-sector supply chains; improvement reported as change in F1-score between the hybrid system and baseline (traditional) oversight approaches. (Paper does not report sample sizes or exact baseline metrics in the summary.)
Both stable individual differences and moment-to-moment fluctuations in perspective-taking influence AI response quality.
Analyses reported in the paper linking both trait-level (stable) and state-level (moment-to-moment) measures of perspective-taking to variation in AI response quality across the benchmark dataset; assessed via the Bayesian IRT model and supplementary within-subject analyses.
Theory of Mind (the capacity to infer and adapt to others' mental states) emerges as a key predictor of synergy.
Statistical association reported between participants' Theory of Mind measures and the estimated synergy (improvement in performance with AI), based on analysis of the benchmark dataset (n = 667) within the Bayesian IRT framework.
These AI formulation models reduced experimental workload by 30–50%.
Reported in the review as estimated reductions in experimental workload when using AI-driven formulation optimization. The excerpt lacks details on how workload was measured, which experiments were replaced or reduced, and sample sizes.
In formulation optimization, artificial neural networks, neuro-fuzzy systems, and hybrid model-based AI models have been able to predict dissolution profiles and critical quality attributes with accuracy rates of over 90%.
Reported model performance in formulation optimization studies summarized by the review. The excerpt does not include which specific studies, datasets, cross-validation protocols, or sample sizes produced >90% accuracy.
AI has reduced clinical trial duration by up to 59%.
Reported in the review as an observed maximum reduction in trial duration associated with AI-driven approaches. The excerpt omits details on which trials, therapeutic areas, trial phases, or sample sizes produced this figure.
AI has sped up compound screening by 1–2 years.
Presented in the review as a comparative reduction in time-to-screening attributed to AI methods. The excerpt does not provide the underlying studies, screening scope, or sample sizes.
AI-enabled platforms have cut the drug discovery pipeline timelines (compared with the traditional 4–6 years) down to 46 days.
Reported as an outcome of AI-enabled platforms in the review. The excerpt does not list the specific platform(s), individual study design(s), or sample sizes underlying the 46-day figure.
Artificial intelligence (AI) is transforming pharmaceutical research and development (R and D), and making measurable improvements in efficiency, precision, and cost-effectiveness in drug research and development.
Stated as a summary conclusion in the review based on cross-domain literature synthesis. Specific studies or quantitative meta-analytic methods and sample sizes are not provided in the excerpt.
Experiments on simulated and real-world data show that humans assisted by the adaptive AI ensemble achieve significantly higher performance than humans assisted by single AI models trained either for independent AI performance or for human-AI team performance.
Empirical experiments reported in the paper on both simulated datasets and real-world data; the abstract states results are statistically significant but does not provide sample sizes, datasets, or statistical details in the excerpt.
An adaptive AI ensemble that toggles between two specialist models (an aligned model and a complementary model) using a Rational Routing Shortcut mechanism overcomes the complementarity–alignment limitation of single-model approaches.
Methodological contribution described in the paper; includes the design of the ensemble and the Rational Routing Shortcut; theoretical guarantees of near-optimality are claimed in the paper (proofs referenced but not shown in the excerpt).
EASP offers a practical tradeoff between reasoning quality and latency by avoiding iterative LLM tool-calls at inference time while still producing grounded plans.
Methodological claim in the paper: Probe-then-Plan uses a lightweight probe to avoid heavy iterative LLM tool calls during serving; supported by design rationale and performance-focused evaluations (offline and online).
EASP has been successfully deployed in JD.com's AI-Search system.
Statement in the paper that EASP was deployed in JD.com's AI-Search system; presumably validated by internal deployment logs and online A/B testing reported.
Online A/B testing on JD.com demonstrates that EASP achieves substantial lifts in UCVR (user conversion rate) and GMV (gross merchandise volume).
Reported results from online A/B testing on JD.com referenced in the paper indicating lifts in UCVR and GMV (no numerical magnitudes provided in the abstract).
Extensive offline evaluations and online A/B testing on JD.com show that EASP significantly improves relevant recall.
Empirical claims in the paper citing extensive offline evaluations and online A/B testing on JD.com as the basis for observed improvements in relevant recall (specific datasets/sizes not reported in the abstract).
Environment-Aware Search Planning (EASP) resolves the blindness-latency dilemma in LLM-based e-commerce search by grounding planning in the real retrieval environment via a Probe-then-Plan mechanism.
Conceptual design and empirical evaluation described in the paper: introduces a lightweight Retrieval Probe to expose a retrieval snapshot and a Planner that diagnoses execution gaps and generates grounded search plans; supported by offline evaluations and online A/B testing on JD.com (section describing method and experiments).
The findings provide valuable insights for entrepreneurs, policymakers, and academic institutions to implement adaptive strategies for sustainable and inclusive entrepreneurial growth in the era of artificial intelligence.
Authors' implications/conclusions based on the study results (n=350; statistical analyses) recommending adaptive strategies targeted at stakeholders.
AI functions as a strategic enabler that reshapes entrepreneurial practices, labour dynamics, and innovation strategies.
Conclusion drawn from the study's quantitative findings (survey of 350, regression/SEM results) that linked AI adoption to changes in opportunity recognition, labour substitution, and innovation processes.
AI-driven innovation processes accelerated product development, improved operational efficiency, and supported experimentation, thereby strengthening entrepreneurial performance.
Survey data from 350 AI-adopting SMEs analyzed with regression and SEM showing positive associations between AI adoption and measures of product development speed, operational efficiency, experimentation, and overall entrepreneurial performance.
AI facilitated labour substitution by automating repetitive tasks, allowing human resources to focus on creative and analytical roles.
Responses from the same sample (n=350) of AI-adopting SME entrepreneurs/managers; descriptive statistics and inferential analyses (regression/SEM) linking AI adoption to increased automation and role reallocation.
AI adoption significantly enhanced opportunity recognition by enabling entrepreneurs to identify emerging market trends, assess risks, and make informed strategic decisions.
Quantitative survey of 350 entrepreneurs and managers of SMEs who had adopted AI; relationships tested using regression analysis and structural equation modelling (SEM) reported a significant positive effect of AI adoption on opportunity recognition.
Sustainable human capital development requires coordinated interaction between education systems, employers, and public institutions.
Normative recommendation derived from the paper's systemic analysis and comparative review of institutional responses; no empirical policy evaluation or quantified cross-country causal analysis reported.
Alignment of educational strategies with labor market dynamics is necessary to support effective reskilling and upskilling.
Supported by comparative assessment of international practices and systemic analysis linking education strategies to labor market requirements; evidence is analytical rather than experimental or longitudinally quantified in the paper.
Effective reskilling and upskilling depend on the development of continuous learning ecosystems.
Analytical conclusion drawn from organizational learning models and international practice comparison; no controlled trials or quantitative evaluation of specific ecosystems reported.
As technological change accelerates, the ability of individuals and organizations to adapt becomes a central condition of economic resilience and long-term competitiveness.
Analytical generalization from organizational learning models and systemic analysis of labor-market dynamics; supported by comparative observations but not by a reported empirical causal study.