Evidence (8974 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
10085 claims
Filter claims →
Productivity
8974 claims
Filtered →
Governance
8062 claims
Filter claims →
Human-AI Collaboration
7749 claims
Filter claims →
Org Design
5057 claims
Filter claims →
Innovation
4896 claims
Filter claims →
Labor Markets
4088 claims
Filter claims →
Skills & Training
3372 claims
Filter claims →
Inequality
2377 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 | 882 | 244 | 117 | 1097 | 2424 |
| Governance & Regulation | 1010 | 469 | 229 | 135 | 1875 |
| Organizational Efficiency | 977 | 235 | 149 | 90 | 1462 |
| Technology Adoption Rate | 781 | 299 | 143 | 128 | 1362 |
| Research Productivity | 506 | 155 | 74 | 363 | 1110 |
| Output Quality | 555 | 219 | 71 | 70 | 915 |
| Decision Quality | 395 | 200 | 95 | 54 | 751 |
| Firm Productivity | 523 | 67 | 101 | 27 | 724 |
| AI Safety & Ethics | 262 | 309 | 75 | 36 | 688 |
| Market Structure | 195 | 201 | 135 | 30 | 566 |
| Task Allocation | 248 | 77 | 96 | 38 | 464 |
| Innovation Output | 300 | 34 | 55 | 20 | 411 |
| Skill Acquisition | 207 | 75 | 65 | 21 | 368 |
| Employment Level | 138 | 67 | 119 | 24 | 350 |
| Fiscal & Macroeconomic | 156 | 80 | 53 | 33 | 329 |
| Task Completion Time | 211 | 38 | 13 | 16 | 280 |
| Firm Revenue | 183 | 52 | 29 | 5 | 270 |
| Consumer Welfare | 131 | 77 | 48 | 13 | 269 |
| Inequality Measures | 50 | 141 | 54 | 9 | 254 |
| Worker Satisfaction | 104 | 85 | 25 | 13 | 227 |
| Error Rate | 87 | 112 | 11 | 5 | 215 |
| Automation Exposure | 69 | 69 | 37 | 20 | 198 |
| Wages & Compensation | 102 | 49 | 31 | 11 | 193 |
| Team Performance | 115 | 30 | 30 | 11 | 187 |
| Regulatory Compliance | 88 | 74 | 17 | 7 | 186 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 116 | 21 | 15 | 8 | 161 |
| Job Displacement | 12 | 92 | 26 | 1 | 131 |
| Hiring & Recruitment | 57 | 12 | 9 | 5 | 83 |
| Skill Obsolescence | 6 | 59 | 10 | 2 | 77 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 23 | 17 | 1 | 59 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Productivity
Remove filter
Shared-state and adaptive forms perform better when they make context durable, inspectable, and task-contingent.
Reported result from the paper's computational/theoretical work plus simulations and trace analyses (methods listed; no sample sizes or quantified effect sizes provided in excerpt).
The article develops contextual transaction cost as the central mechanism linking these similarities and differences.
Stated theoretical contribution in the paper: introduction/development of the 'contextual transaction cost' concept as a mechanism. Method: conceptual/theoretical development.
Agentic AI is a partial organisational analogue: it resembles a human organisation because it differentiates work, coordinates interdependence, performs recurrent routines, crosses boundaries, and produces collective outcomes.
Theoretical/conceptual argument developed in the paper comparing structural and functional properties of agentic AI to human organisations (computational theorising and conceptual analysis). No quantitative sample size reported in excerpt.
These systems are entering organisational workflows under familiar labels such as teams, managers, committees, markets, and workflows.
Presented as an observed trend in the article; supported by the author's descriptive statements (methods: conceptual observation; no sample size reported in excerpt).
Agentic artificial intelligence is increasingly deployed not as a single assistant but as a collective of planners, solvers, reviewers, memory managers, tool users, and orchestrators.
Stated as an observational claim in the article's opening framing. Supported by the paper's descriptive discussion (no quantitative sample size or external dataset reported in the excerpt). Method: conceptual/observational statement.
The improvement of AI-based risk mitigation was 30%, compared with 14% for traditional methods.
Observational study combining AI modelling and questionnaires; the paper reports comparative improvement percentages for risk mitigation with AI and with traditional methods, but no sample size or detailed methodology is provided in the supplied text.
AI was proven to be 92% accurate in risk prediction.
Observational study using AI modelling (machine learning/NLP) to evaluate risk prediction accuracy; reported accuracy figure provided but no sample size, validation procedure, or performance metrics beyond the percentage in the supplied text.
The results show 40% quicker decision-making after the adoption of AI.
Observational study combining AI modelling and questionnaires comparing AI-based strategies to traditional ones; quantitative improvement in decision-making speed reported but no sample size or further methodological details provided in the supplied text.
The results show a 30% improvement in compliance accuracy after the adoption of AI.
Observational study combining AI modelling and questionnaires comparing AI-based strategies to traditional ones; quantitative improvement in compliance accuracy reported but no sample size or further methodological details provided in the supplied text.
The results show a 20% reduction in the number of operational disruptions after the adoption of AI.
Observational study combining AI modelling and questionnaires comparing AI-based strategies to traditional ones; quantitative analysis reported but no sample size or further methodological details provided in the supplied text.
An occupation one standard deviation higher in interaction-and-communication content has 0.36-standard-deviation higher market-implied AI premium.
Quantitative occupational-skill regression linking standardized interaction-and-communication content to standardized market-implied AI premium; reported coefficient of 0.36 (SD units).
The AI premium reaches beyond technology firms into consumer-facing and capital-heavy parts of the economy.
Cross-sectional analysis across sectors showing positive AI beta–return relationships in consumer-facing and capital-intensive industries, not limited to technology sector.
The AI premium is large for loadings on the intensive, frontier-oriented margin of AI consumption—closed-source models, paying and seasoned users, and long prompts.
Decomposition of AI factor by consumption margins (model openness, user payment/tenure, prompt length) and analysis of how loadings on these components relate to the AI premium.
A value-weighted long-short strategy (long high-AI-beta firms, short low-AI-beta firms) earns 64.1 basis points per week.
Backtest/portfolio analysis using firm-level AI betas to form a value-weighted long-short strategy; reported weekly return statistic.
Firms whose returns covary more positively with the AI factor (high AI beta firms) earn higher subsequent returns; the AI premium is large and heterogeneous.
Empirical asset-pricing analysis: firm-level AI betas estimated from stock return comovement; subsequent returns compared across firms with differing AI betas (methodology described in paper).
Education–skills alignment, active labour force programmes and fair transition mechanisms can support growth by reducing the social costs of transformation.
Policy recommendation in the paper, presented as complementary measures to accompany technological and green transitions; rationale based on observed negative growth effects from unemployment and the need to mitigate social costs.
Countries need to strengthen R&D, digital transformation, renewable infrastructure, industrial policies and inclusive employment strategies in a coordinated manner for long-term stability.
Policy recommendation derived from the paper's empirical findings linking technological capacity, renewables, industrialisation and employment to growth; presented as a suggested policy package.
Industrialisation is an important driver of growth via economies of scale and added value growth.
Paper's empirical findings showing a positive association between the level of industrialisation and economic growth across the 27-country panel (2008–2020), with discussion of economies of scale and value-added as mechanisms.
The shift towards green (renewable) energy contributes to growth by reducing production costs and encouraging investment consistent with energy security and emission reduction goals.
Empirical analysis in the paper relating renewable energy use to economic growth for the 27-country panel (2008–2020); authors report a positive contribution of renewable energy adoption to growth and discuss mechanisms (costs, investment).
Increases in technological capacity and artificial intelligence significantly support growth.
Empirical estimation using the paper's panel data methods on 27 top-GDP countries (2008–2020); authors report a statistically significant positive relationship between measures of technological capacity/AI and economic growth.
AI supports economic growth.
Aggregate synthesis of literature (194 articles) reported in the abstract indicating links between AI and economic growth.
AI fosters innovation.
Synthesis from the systematic review of 194 peer-reviewed articles; the abstract lists innovation as one of the dimensions showing positive effects of AI.
AI functions as a general-purpose technology capable of enhancing productivity.
Synthesis of findings from the systematic review of 194 peer-reviewed articles across dimensions including productivity and innovation (as stated in the abstract).
Company-maintained repositories show a higher percentage of code and comments detected as likely to be generated by LLMs compared to community-maintained repositories.
Comparative detector-based analysis across active company- and community-maintained repositories from 2021–2025.
Code detected as likely to be generated by LLMs appeared frequently in test cases.
Detector-based analysis of repository contents (2021–2025) identifying location/context of code flagged as likely LLM-generated (test code vs. other).
Reports from large technology companies showed that around 20% to 30% of their code are generated by LLMs.
Cited reports from large technology companies (no specific report names or sample sizes provided in the text).
Personality prompting shapes how large language models communicate.
Authors manipulated personality prompts across multiple frontier LLMs and observed systematic shifts in communication behavior across experiments spanning three task domains (structured coding, open-ended research collaboration, competitive bargaining). Specific sample sizes and model names are not provided in the abstract.
About half of the variation in integration friction stays with the repository after the contribution, its author, its size, and its agent are accounted for.
Variance decomposition / multilevel modeling of integration friction across the dataset (controls for contribution, author, size, and agent are reported).
Workforce development should be grounded in systems design principles, constraint reduction, and continuous evaluation (i.e., key design principles for workforce development are proposed grounded in systems design).
Prescriptive recommendation emerging from the paper's systems-oriented analysis and synthesis of adult learning theory and organizational design (no empirical evaluation reported).
Artificial intelligence (AI) comprises not only models, but full socio-technical systems involving data pipelines, instrumentation, human-machine interfaces, deployment architectures, and organizational processes for design, monitoring, and evaluation.
Conceptual/definitional claim presented via a systems-oriented analytical framework and literature synthesis in the paper (no empirical sample reported).
We assemble an instrument, the Gini-Adjusted GDP per Capita Index (GAGI): a reproducible, publicly computable formulation that rescales each country's GDP per capita by its inequality-adjustment factor (1-G) and its price level, normalised to a 2010 baseline.
Methodological development reported in the paper (authors present a formula and construction procedure; no external validation reported here).
The automated fill rate of core attributes during item listing exceeds 80%.
Reported metric in the paper/abstract for automated attribute population during listing; no methodological details or labeled evaluation set described in the abstract.
Item-information quality issues drop by 37%.
Reported reduction in item-information quality issues attributed to Oxygen AIIC in the paper/abstract; the abstract does not specify the evaluation period, baseline, or labeling methodology.
Search-traffic coverage reaches 80.4%.
Specific metric reported in the paper/abstract describing the fraction of search traffic covered by Oxygen AIIC data/services; no measurement protocol or baseline specified in the abstract.
Deployed across core business scenarios (search, recommendation, operations, category planning), Oxygen AIIC has delivered measurable gains at scale.
General deployment and impact statement in the abstract; no specific experimental design or quantified causal attribution provided in the abstract (though specific metrics are reported elsewhere in the abstract).
Oxygen AIIC has accumulated hundreds of billions of item-knowledge assets.
Cumulative asset-count claim presented in the paper/abstract; no detailed schema or breakdown of 'item-knowledge assets' provided in the abstract.
Oxygen AIIC processes hundreds of millions of item updates per day on Huawei Ascend NPUs.
Production throughput metric reported in the paper/abstract (per-day update rate and hardware used); no raw logs or benchmarking methodology included in the abstract.
Oxygen AIIC now covers tens of thousands of JD categories.
Deployment scope statistic reported in the paper/abstract; no supporting breakdown or evaluation details provided in the abstract.
Self-evolving item-understanding LLMs/VLMs improve in a stable and controllable manner, enabling knowledge production with 94.2% precision and 82.8% recall.
Reported evaluation metrics (precision and recall) for the knowledge-production component referenced in the paper; specific evaluation dataset, labeling protocol, and sample size are not provided in the abstract.
A 'Semantic Search then Discrimination' (S2D) knowledge identification architecture, combined with throughput improvement strategies, enables scalable, extensible, and high-throughput AI Item Library production for tens of billions of SKUs.
Architectural claim supported implicitly by reported production throughput and scope elsewhere in the abstract (processing and SKU scale); no full experimental methodology included in the abstract.
Ontology engineering driven by efficient human-AI collaboration supports the dynamic evolution and agile expansion of an ontology with millions of entries.
Design claim in the paper stating an ontology engineering approach with human-AI collaboration and scale (millions of entries); no quantitative evaluation details provided in the abstract.
Oxygen AI Item Center (Oxygen AIIC) is an industrial-scale platform built on LLMs/VLMs for item-knowledge production and service.
System/architectural description in the paper asserting the platform's use of large language and vision-language models; no detailed empirical validation in the abstract.
JD.com serves over 700 million active users and millions of merchants, with a catalog of tens of billions of SKUs.
Background platform statistics stated in the paper/abstract (company-reported usage and catalog size). No supporting external dataset or evaluation described in the abstract.
The situated analytics in ARTOO-DARTU, when paired with the ODM to prevent real-world obstructions, can significantly enhance efficiency and user experience in AR-HRC warehouse scenarios.
Summary claim supported by the 34-participant user study results (reported 46% overall efficiency increase and 61% faster visibility-sensitive subtasks) and qualitative/UX measures reported in the paper.
The ODM pipeline detects and mitigates AR content obstructions of important real-world elements, preserving visibility and thereby reducing safety/usability risks in mobile-robot warehouse HRC scenarios.
Description of ODM pipeline in paper plus empirical evaluation via user study (improvements on visibility-sensitive subtasks and overall efficiency when ODM active).
ARTOO-DARTU is an AR system tailored for warehouse HRC that enables real-time robot situated analytics and control while preserving visibility of the real world through an obstruction detection and mitigation pipeline (ODM).
System design and implementation described in the paper; presentation of ARTOO-DARTU architecture and the ODM pipeline as part of the contribution (methodological/software artifact).
Participants with the ODM active were 61% faster on subtasks requiring visibility of the real world.
Same 34-participant user study (Pocket MonstARs); measured subtask completion times for subtasks that required seeing real-world elements, comparing ODM-active condition to other condition(s).
In a 34-participant user study, our AR situated analytics yielded a 46% increase in efficiency on the overall HRC task, but only when the ODM was active.
Controlled user study reported in the paper (Pocket MonstARs gamified abstraction), N=34 participants; comparison of overall HRC task performance with AR situated analytics + ODM active versus relevant baseline(s).
The positive effect of industrial robots on firm-level TFP is statistically significant among firms with high R&D investment.
Heterogeneity/subsample or interaction analysis using the paper's matched dataset (2006–2019) that conditions the robots→TFP effect on firms' R&D investment levels.
The positive effect of industrial robots on firm-level TFP is statistically significant among firms that experience low financial misallocation.
Heterogeneity/subsample or interaction analysis using the panel data (industrial robots and Chinese listed firms, 2006–2019) examining financial misallocation as a conditioning factor.