Evidence (8066 claims)
Adoption
5586 claims
Productivity
4857 claims
Governance
4381 claims
Human-AI Collaboration
3417 claims
Labor Markets
2685 claims
Innovation
2581 claims
Org Design
2499 claims
Skills & Training
2031 claims
Inequality
1382 claims
Evidence Matrix
Claim counts by outcome category and direction of finding.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 417 | 113 | 67 | 480 | 1091 |
| Governance & Regulation | 419 | 202 | 124 | 64 | 823 |
| Research Productivity | 261 | 100 | 34 | 303 | 703 |
| Organizational Efficiency | 406 | 96 | 71 | 40 | 616 |
| Technology Adoption Rate | 323 | 128 | 74 | 38 | 568 |
| Firm Productivity | 307 | 38 | 70 | 12 | 432 |
| Output Quality | 260 | 71 | 27 | 29 | 387 |
| AI Safety & Ethics | 118 | 179 | 45 | 24 | 368 |
| Market Structure | 107 | 128 | 85 | 14 | 339 |
| Decision Quality | 177 | 75 | 37 | 19 | 312 |
| Fiscal & Macroeconomic | 89 | 58 | 33 | 22 | 209 |
| Employment Level | 74 | 34 | 78 | 9 | 197 |
| Skill Acquisition | 98 | 36 | 40 | 9 | 183 |
| Innovation Output | 121 | 12 | 24 | 13 | 171 |
| Firm Revenue | 98 | 35 | 24 | — | 157 |
| Consumer Welfare | 73 | 31 | 37 | 7 | 148 |
| Task Allocation | 87 | 16 | 34 | 7 | 144 |
| Inequality Measures | 25 | 76 | 32 | 5 | 138 |
| Regulatory Compliance | 54 | 61 | 13 | 3 | 131 |
| Task Completion Time | 89 | 7 | 4 | 3 | 103 |
| Error Rate | 44 | 51 | 6 | — | 101 |
| Training Effectiveness | 58 | 12 | 12 | 16 | 99 |
| Worker Satisfaction | 47 | 33 | 11 | 7 | 98 |
| Wages & Compensation | 54 | 15 | 20 | 5 | 94 |
| Team Performance | 47 | 12 | 15 | 7 | 82 |
| Automation Exposure | 27 | 26 | 10 | 6 | 72 |
| Job Displacement | 6 | 39 | 13 | — | 58 |
| Hiring & Recruitment | 40 | 4 | 6 | 3 | 53 |
| Developer Productivity | 34 | 4 | 3 | 1 | 42 |
| Social Protection | 22 | 11 | 6 | 2 | 41 |
| Creative Output | 16 | 7 | 5 | 1 | 29 |
| Labor Share of Income | 12 | 6 | 9 | — | 27 |
| Skill Obsolescence | 3 | 20 | 2 | — | 25 |
| Worker Turnover | 10 | 12 | — | 3 | 25 |
We implement an auditable dictionary-and-context approach that flags agentic references and then quantifies the surrounding 'controls density' (governance and safety language) within the same local disclosure window.
Methods description: dictionary-and-context text-as-data approach and a quantified 'controls density' metric applied to filings.
We assemble a balanced panel of 2,500 firm–year observations (500 firms per year) from 2021–2025.
Stated dataset construction in the paper: balanced panel across years with 500 firm–year observations per year, total 2,500 firm–years.
Agentic artificial intelligence (AI) systems can execute actions rather than merely generate content.
Conceptual/definitional statement in the paper framing agentic AI as systems that execute actions (not an empirical test).
Transparency’s effectiveness in promoting data-sharing is amplified by, and dependent upon, user trust; fostering trust in AI may be a more vital prerequisite for data-sharing than implementing transparent designs.
Synthesis of experimental findings (N=240): transparency increased willingness only among users with pre-existing trust; null effect of transparency alone on actual sharing; authors conclude that trust moderates transparency effects and recommend focusing on trust-building.
Immediate sharing decisions were largely driven by intuitive System 1 processing rather than deliberative evaluation (System 2).
Interpretation of the pattern in experimental data (N=240): high, similar sharing rates across conditions despite differing stated willingness-to-share and measured privacy concerns; authors attribute this to dual-process dynamics (System 1 driving immediate behavior).
The positive effect of transparency on willingness to share was contingent on pre-existing user trust in AI, particularly for white-box systems.
Moderation analyses reported from the experiment (N=240): interaction between transparency (white-box vs black-box) and measured pre-existing trust in AI showed increased willingness-to-share only among users with higher trust, with the effect most pronounced for white-box systems.
We conducted a pre-registered online experiment (N=240) where participants interacted with a fictional sleep-optimization app and were randomly assigned to scenarios where data was processed by either a human expert, a transparent white-box AI, or an opaque black-box AI.
Pre-registered online experimental design described in paper; random assignment to three processing-entity conditions (human, white-box AI, black-box AI); sample size reported as N=240; measured outcomes included actual data-sharing and willingness to share, plus trust and privacy concerns.
A Metacognitive Co-Regulation Agent (in CRDAL) assists the Design Agent in metacognition to mitigate design fixation, thereby improving system performance for engineering design tasks.
Mechanistic claim supported by the paper's experimental results on the battery pack design problem showing CRDAL outperforming SRL and RWL; detailed measures of fixation reduction not provided in the excerpt.
The CRDAL system navigated through the latent design space more effectively than both SRL and RWL.
Empirical analysis on the battery pack design task comparing latent-space trajectories/exploration between CRDAL, SRL, and RWL; details on how 'more effectively' was quantified and sample size are not provided in the excerpt.
The CRDAL system achieves better design performance without significantly increasing the computational cost compared to SRL and RWL.
Empirical claim based on experiments on the battery pack design problem comparing computational cost across CRDAL, SRL, and RWL; exact computational metrics and sample size not provided in the excerpt.
In the battery pack design problem examined here, the CRDAL system generates designs with better performance compared to a plain Ralph Wiggum Loop (RWL) and the metacognitively self-assessing Self-Regulation Loop (SRL).
Empirical comparison on a battery pack design task between CRDAL, SRL, and RWL reported in the paper; exact number of test instances or runs not stated in the excerpt.
We propose a novel Co-Regulation Design Agentic Loop (CRDAL), in which a Metacognitive Co-Regulation Agent assists the Design Agent in metacognition to mitigate design fixation.
Methodological contribution presented in the paper (proposed system architecture). No empirical sample size reported for the proposal itself.
We propose a novel Self-Regulation Loop (SRL), in which the Design Agent self-regulates and explicitly monitors its own metacognition.
Methodological contribution presented in the paper (proposed system architecture). No empirical sample size reported for the proposal itself.
When models are used in research, potential threats to inference should be systematically identified alongside the steps taken to mitigate them, and specific justifications for model selection should be provided.
Prescriptive recommendation in the paper (normative guidance) based on the authors' analysis of threats to inference; no empirical testing reported in abstract.
The inferential issues that closed models present can be resolved or mitigated by certain measures.
Paper's analytic discussion of mitigation strategies and ways to resolve or reduce threats to inference; no empirical validation or quantified results provided in the abstract.
EcoThink offers a scalable path toward a sustainable, inclusive, and energy-efficient generative AI Agent.
Concluding claim in the paper asserting broader impact and scalability of the proposed method (position/interpretive claim based on reported results).
Extensive evaluations were performed across 9 diverse benchmarks.
Statement in the paper that evaluations were run on 9 benchmarks (as stated in the abstract).
EcoThink employs a lightweight, distillation-based router to dynamically assess query complexity, skipping unnecessary reasoning for factoid retrieval while reserving deep computation for complex logic.
Methodological description of the proposed framework in the paper (design/architecture claim).
EcoThink reduces inference energy by up to 81.9% for web knowledge retrieval.
Experimental result reported in the paper (maximum observed reduction for the web knowledge retrieval benchmark, as stated in the abstract).
EcoThink reduces inference energy by 40.4% on average across 9 diverse benchmarks.
Experimental evaluations reported in the paper across 9 benchmarks comparing inference energy of EcoThink versus baseline (as stated in the abstract).
Policy efficacy varies significantly across corporate profiles, with the strongest effects observed in non-state-owned enterprises, high-tech firms, and firms located in eastern regions.
Heterogeneity analyses reported in the study (subgroup analysis by ownership, technology intensity, and geographic region).
The estimated positive effect of the pilot zones on corporate NQPF is robust across a comprehensive battery of robustness and endogeneity tests.
Paper reports multiple robustness and endogeneity checks (details not provided in abstract) that reportedly do not overturn main findings.
Mechanism analysis identifies three systemic transmission pathways for the policy: optimizing factor allocation, deepening digital technology empowerment, and promoting green innovation and sustainability.
Mechanism analysis reported in the study (methods not detailed in abstract) attributing the policy effect to three pathways.
The pilot zones create an optimized 'digital environment' that underlies the positive impact on corporate NQPF.
Empirical analysis in the paper attributes improved corporate NQPF to an optimized digital environment created by the policy intervention; mechanism analysis referenced.
The DML approach flexibly controls for high-dimensional confounding variables and functional form misspecification, enabling highly rigorous causal inference compared with traditional linear models.
Methodological claim based on use of Double Machine Learning in the study (described as addressing high-dimensional confounders and misspecification).
Establishment of China’s National Digital Economy Innovation and Development Pilot Zones significantly enhances corporate New Quality Productive Forces (NQPF).
Quasi-natural experiment using Double Machine Learning (DML) framework applied to A-share listed companies over 2015–2023; empirical results reported as statistically significant.
The proposed system and findings have policy-relevant implications for policymakers and fiscal institutions, improving their ability to name (identify) and react to potential instabilities.
Paper discussion claims implications for policymakers and fiscal institutions based on the proposed framework and synthesized empirical findings; specific policy-impact evaluations are not provided in the excerpt.
This paper proposes a novel framework that uses machine learning and news data to create a regulatory early-warning mechanism for predicting and mitigating fiscal risk.
Paper text describes a proposed framework combining machine learning with news streams; described as a methodological contribution (conceptual design/architecture). No numeric evaluation or sample size reported in the provided excerpt.
Integrating AI into financial ecosystems can strengthen both economic and climate resilience, provided that regulatory frameworks, ethical AI practices, and capacity-building measures are simultaneously addressed.
Paper's concluding recommendation based on combined qualitative and quantitative findings from the three case studies and the 1,500 interviews; framed as conditional policy guidance in the abstract.
Predictive AI models can facilitate climate-resilient decision-making in agriculture.
Reported as a finding from the Thailand AI-supported smart agriculture finance case study, supported by qualitative evidence and (implied) predictive-model-driven finance decisions noted in the abstract.
Women exhibit higher adoption and savings patterns on AI-enabled financial platforms.
Abstract reports gendered impacts derived from 1,500 semi-structured customer interviews plus account-activity data across the three case studies, noting higher adoption and savings for women.
AI-enabled platforms reduce vulnerability to climate-related income shocks.
Abstract claims findings that AI-enabled platforms reduce vulnerability to climate-related income shocks based on case studies (including smart agriculture finance in Thailand), interviews and transaction/loan data analysis.
AI-enabled platforms promote savings behavior among customers.
Abstract reports findings based on mixed-methods: qualitative interviews (1,500) and quantitative account-activity analysis indicating increased savings behavior on AI-enabled platforms.
AI-enabled platforms significantly improve credit access for low-income and rural customers in the case-study contexts.
Quantitative analysis of transaction records and loan repayment histories combined with qualitative insights from 1,500 interviews across three case studies (M-KOPA, TymeBank, and smart agriculture finance in Thailand) as described in the abstract.
Policymakers should pursue integrated policies linking energy transition, macroeconomic stability, and digital innovation to preserve the United States' technical supremacy in AI.
Normative recommendation based on the paper's empirical findings (WQR/WQC on 2013Q1–2024Q4 US data) showing links between energy policy, macro determinants, and AI investment.
Stable energy policy, continuous economic growth, and improved global integration are significant for promoting AI development in the United States.
Policy implication drawn from empirical associations found using WQR/WQC on US quarterly data (2013Q1–2024Q4), where renewable energy, growth, trade openness, and globalisation positively associate with AI investment and energy policy uncertainty exhibits nonlinear effects.
Wavelet Quantile Regression (WQR) and Wavelet Quantile Correlation (WQC) effectively capture distributional asymmetries and time–frequency dynamics in the relationships between macro/policy determinants and AI investment.
Methodological claim supported by the paper's use of WQR and WQC on the 2013Q1–2024Q4 US quarterly dataset; results are reported across quantiles and scales (as stated).
Globalisation positively influences AI investment in the United States.
Empirical analysis using WQR and WQC on US quarterly data from 2013Q1 to 2024Q4 (48 quarters).
Trade openness positively influences AI investment in the United States.
Empirical analysis using WQR and WQC on US quarterly data from 2013Q1 to 2024Q4 (48 quarters).
Economic growth positively influences AI investment in the United States.
Empirical analysis using WQR and WQC on US quarterly data from 2013Q1 to 2024Q4 (48 quarters).
Renewable energy consumption positively influences AI investment in the United States.
Empirical analysis using Wavelet Quantile Regression (WQR) and Wavelet Quantile Correlation (WQC) on US quarterly data from 2013Q1 to 2024Q4 (48 quarters).
AlphaFold represents an 'oracle' breakthrough in AI for scientific discovery.
Cited as an example of an algorithmic breakthrough that changed a specific scientific subtask (protein structure prediction). The paper frames AlphaFold as a milestone in the history reviewed; no new experimental data presented.
The resulting policy matrix includes R&D funding, regulatory sandboxes, public procurement incentives, and tax relief, tailored to each stage of technological evolution.
Paper presents a policy matrix produced by the study listing these instruments mapped to maturity stages; no quantitative evaluation of impact reported in text provided.
To validate and prioritise policy instruments, Delphi rounds with domain experts and Analytic Hierarchy Process (AHP) weighting are employed.
Paper reports use of Delphi method and AHP for validation and prioritization; methodological description without reported number of experts or rounds.
A technology maturity classification categorises innovations into emerging, developing, and mature stages, forming the basis for strategic policy matching.
Paper defines a maturity classification (emerging/developing/mature) and indicates it is used to match policy instruments; categorical description provided, no quantitative validation details in text provided.
Temporal mapping and citation networks reveal distinct technology maturity patterns, which are visualised using S-curve and hype cycle models.
Paper describes use of temporal mapping and citation network analysis and visualization via S-curve and hype cycle models; methodological description without quantitative sample-size details.
Technologies such as AI-driven healthcare, quantum communication, hydrogen energy, and smart educational AI are identified as key domains of convergence.
Paper reports these domains were identified via the applied analytic framework and multi-source data triangulation; no numeric counts/sample sizes provided.
The study applies advanced techniques such as LDA topic modelling, BERT-based clustering, and co-citation analysis to detect innovation trajectories.
Paper states these specific analytic techniques were applied (method description).
The research leverages large AI models and multi-source data—including global patent databases (WIPO, USPTO, Lens.org), scientific literature corpora, and industry intelligence platforms (CB Insights, Qichacha).
Paper statement of data sources and use of large AI models; methodological description (no sample sizes reported).
Recommended regulatory responses include algorithmic transparency mandates, mandatory mental health risk audits, participatory co-design, human review of deactivations, and minimum wage protections aligned with ILO principles.
Authors' policy recommendations derived from the review's synthesis and identified psychological risks.