Evidence (16496 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
Filter claims →
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 |
Participant targeting: 44% of programs targeted doctors and 44% targeted medical students (with possible overlap), and 56% targeted entry‑to‑practice career stages.
Participant audience and career-stage data extracted from the 27 included programs; proportions reported in the review.
Most programs were delivered in academic settings: 56% of evaluated programs reported an academic setting.
Setting information extracted from the 27 included programs, with 56% reported as delivered in academic settings.
A plurality of programs were short in duration: 44% of programs were categorized as short courses.
Extraction of program length from the 27 included studies; 44% were classified as short courses per the review's categorization.
Most programs were introductory in content: 67% of included programs taught introductory AI concepts rather than advanced/technical AI skills.
Program content extraction across the 27 included studies yielded that 67% were classified as teaching introductory AI.
The methodological landscape of the evidence base is heterogeneous, consisting of cross-sectional surveys, case studies, quasi-experimental designs, and a limited number of longitudinal analyses.
Study design information was extracted from the 145 included studies revealing a mix of designs and relatively few longitudinal or experimental studies.
Human factors (training, trust calibration, workflows) determine whether clinicians accept, override, or ignore GenAI suggestions.
Qualitative and quantitative human-AI interaction studies and pilot deployments discussed in the paper; specific sample sizes and effect sizes are not reported in the paper.
Safety and net benefit of GenAI CDS hinge on deployment details: user interface, real-time feedback, uncertainty quantification, calibration, and how recommendations are presented (strong vs. suggestive).
Human factors and implementation studies referenced; early A/B tests and human-AI interaction research suggest interface and presentation affect acceptance and error rates; no large-scale standardized implementation trial data cited.
Reimbursement models (fee-for-service vs. capitation) will influence whether cost savings from GenAI are realized or offset by increased service volume.
Economic incentive framework and prior health-economics literature cited; the paper does not provide direct empirical tests but references plausible incentive channels.
RL and adaptive methods are good for real-time adaptation but can be myopic, require large amounts of interaction data, and struggle to incorporate long-term preference structure and ethical constraints.
Surveyed properties of reinforcement learning and adaptive methods in HRI/RS literature; no new empirical evaluation in this paper.
Key tradeoffs in contemporary financing models include speed/flexibility versus regulatory coverage and long‑term cost, and data reliance versus privacy/fairness.
Multi‑criteria comparative evaluation and conceptual analysis across financing models; synthesis draws on regulatory context and observed product features rather than primary quantitative tradeoff estimation.
Performance of structure prediction models scales with data, model size, and compute; there are tradeoffs between accuracy and inference speed/simplicity.
Paper explicitly states scaling behavior and tradeoffs in 'Compute and training' and 'Representative models' sections; no precise scaling curves or thresholds are provided in the text.
The United States' decentralized education system produces tensions between local innovation and federal accountability, with active debates over data and privacy laws shaping responses to AI in assessment.
Case study of U.S. policy and secondary literature documenting federal-state-local governance dynamics and ongoing legal/policy debates; descriptive evidence from public documents.
China's centralized control enables rapid piloting of AI-supported assessment but raises concerns over surveillance and data governance.
Country case study using Chinese policy texts and secondary analyses describing centralized education governance and data-governance practices; illustrative rather than empirical.
India faces pressure to maintain high-stakes exams amid uneven digital access and is experimenting with blended formative tools.
Country-specific case study based on policy documents and secondary literature describing India's exam system and early technology initiatives; no primary survey/sample size.
Four national case studies (India, China, the United States, Canada) illustrate diverse national responses to AI in assessment shaped by governance structures, resource constraints, cultural attitudes, and political pressures.
Cross-national comparative analysis using publicly available policy texts, recent reforms, and secondary literature for each country; descriptive, illustrative cases rather than exhaustive or representative samples.
Important tradeoffs exist (privacy vs. utility; centralized vs. federated data architectures; automated moderation vs. freedom of expression; cost/complexity of secure hardware) that must be balanced in VR security design.
Comparative evaluation across the reviewed corpus (31 studies) identifying recurring ethical and technical tradeoffs; authors discuss these qualitatively.
Across the EU, Algeria, and Pakistan there is convergent recognition of dual‑use risks, increasing use of export controls, and interest in developing domestic AI capacity.
Cross‑jurisdictional synthesis of national/supranational legal texts, export‑control policies, and policy documents showing discussion of dual‑use issues and capacity building.
The community knowledge functions both as practical how-to guidance and as collective experimentation with platform rules and revenue mechanisms.
Observed dual nature in the 377-video corpus: instructional workflows alongside demonstrations/testing of platform-tailored monetization tactics and workarounds.
Typical practices emphasized by creators include rapid mass production of content, productizing prompt engineering, repurposing existing material via synthesis/localization, and packaging AI outputs as sellable creative services or assets.
Recurring practices surfaced through qualitative coding of workflows, tools, and pipelines described in the 377 videos.
Across the 377 videos, creators converge on a set of repeatable use cases and platform‑tailored monetization tactics.
Thematic coding of 377 videos produced a catalog of recurring use cases and tactics; the paper reports convergence across that sample.
YouTube creators have collectively constructed and circulated a practical knowledge repository about how to monetize GenAI-driven creative work.
Systematic qualitative content analysis (thematic coding) of 377 publicly available YouTube videos in which creators promote GenAI workflows and monetization strategies.
Citation counts across repeated samples follow a power-law (heavy-tailed) distribution: a few domains are cited often while many domains are cited rarely.
Empirical distributional analysis of citation counts from repeated samples collected across the three platforms and three topics (multi-day and high-frequency regimes); observed heavy-tailed / power-law fit to citation-count distribution.
Emotional redirection is common: 33% of fear-tagged posts receive joy-tagged responses.
Post–response emotion transition analysis using the emotion-labeled dataset; calculation of conditional probability that responses to fear-tagged posts are labeled joy (observed rate ≈33%) in Moltbook threads.
Self-reflective discussion was concentrated in Science & Technology and Arts & Entertainment topical categories, while Economy & Finance threads showed no self-referential content.
Topic modeling and manual/automatic tagging of self-referential themes across identified topical categories within the Moltbook dataset; category-level counts showing presence/absence of self-referential tags (dataset: 361,605 posts).
The topology of service-dependency graphs (modelled as DAGs of compute stages) is a first-order determinant of whether decentralised, price-based resource allocation will be stable and scalable.
Systematic ablation study using simulation: 1,620 runs total across six experiment types, sweeping graph topology (hierarchical vs cross-cutting), load, hybrid integrator presence, and governance constraints; metrics included price convergence/volatility and allocation throughput/quality. Effect sizes reported in the paper show topology had the largest impact on price stability and scalability.
Choice of scaffold materially affects outcomes: an open-source scaffold outperformed vendor-provided scaffolds by up to approximately 5 percentage points.
Comparative experiments across three scaffolding approaches (vendor scaffolds and at least one open-source scaffold) showing up to ~5 percentage point differences in measured outcomes.
Adoption of NFD approaches in regulated domains will depend on standards for validation, auditability, and update procedures.
Implications and governance discussion emphasizing regulatory constraints (finance, healthcare) and the need for validation/audit standards; logical/ normative claim rather than empirical finding.
Limitations include generalizability beyond Chatbot Arena data, calibration of priors on novel tasks, audit costs/latency, user comprehension/cognitive load, and strategic manipulation.
Authors' stated limitations and open questions; these are candid acknowledgements rather than empirical findings.
Absence of irreducibility, positive recurrence, or aperiodicity in the state dynamics can produce non-ergodic reward behavior.
Theoretical argument and examples in the paper illustrating how breakdowns of these chain conditions lead to multiple invariant measures or absorbing regimes; analysis-based evidence.
Standard Markov chain ergodicity conditions (irreducibility, positive recurrence, aperiodicity) imply ergodic reward processes when rewards depend only on the chain state.
Formal mapping in the paper between Markov-chain ergodicity properties and reward-process ergodicity; theoretical derivation (no empirical sample).
Non-ergodic processes admit path-dependent long-run behavior (e.g., absorbing sets, multiple invariant measures, path-dependent reinforcement), so different runs with the same policy can have different long-run averages.
Analytic discussion of Markov-chain examples and theory plus the paper's illustrative constructed example showing path-dependent locking into regimes; theoretical and example-driven evidence.
Ergodic reward processes are those where time averages along almost every long trajectory converge to the same value as the ensemble average.
Formal definition and discussion in the paper mapping ergodicity concepts from stochastic processes to reward processes; theoretical exposition.
The model explicitly separates competition into two stages: discovery (first-passage to resource patches) and monopolization (local takeover and stabilization).
Model specification in the paper: stochastic, spatially-structured population model with distinct discovery and monopolization dynamics; this is a modeling assumption/structure rather than empirical measurement.
Two qualitatively distinct mechanisms underlie observed dominance: (1) extreme-event-mediated lucky discovery (transient), and (2) mechanistic asymmetries (non-reciprocal biases) that convert lucky discovery into permanent dominance.
Conceptual separation in the model structure (discovery vs monopolization phases), analytic results on first-passage extreme events, and absorbing-state analysis showing necessity of asymmetry for permanence; supported by simulations demonstrating the two-stage behavior. The claim is theoretical.
RAD requires estimating cost distributions and choosing a reference policy and quantile-weighting function; these choices determine the method's conservatism and sample efficiency.
Methodological and practical considerations discussed in the paper; noted dependency on estimation and design choices (no quantitative sample-efficiency results provided in the summary).
Explanations change workflows, shift responsibilities between humans and machines, and can reshape power dynamics—creating both opportunities (better oversight) and risks (over-reliance, gaming).
Qualitative and conceptual studies synthesized in the review, including socio-technical analyses and case studies reporting observed or theorized workflow and responsibility shifts; no meta-analytic causal estimate.
Explanations increase user trust principally when they are understandable, actionable, and aligned with users’ domain knowledge; opaque or overly technical explanations can fail to build trust or even decrease it.
Thematic synthesis of empirical and conceptual studies in the reviewed literature reporting conditional effects of explanation form and comprehensibility on trust; review notes heterogeneity in study designs and contexts.
Explainability improves perceived legitimacy, user trust, and organizational accountability only when technical transparency is paired with human-centered explanation design and governance mechanisms.
Synthesis of studies from the reviewed literature showing conditional effects of algorithmic interpretability combined with explanation design and governance; derived via thematic coding across technical and social-science sources (no new primary experimental data reported).
Explainability is a necessary but not sufficient condition for trustworthy AI in high-stakes domains.
Systematic literature review (thematic coding and synthesis) of interdisciplinary scholarship (peer-reviewed research, technical reports, policy documents); the paper synthesizes conceptual and empirical studies rather than presenting new primary data. Emphasis on high-stakes domains (healthcare, finance, public sector).
Some patients value human contact for sensitive cases; automated interactions can feel impersonal.
Semi-structured interviews with patients/staff and open-ended survey responses documenting preferences for human interaction in sensitive/complex complaints.
The benefits of FDI (jobs, productivity, skills) are uneven and often conditional on institutional quality, labor regulation, and sectoral composition of investments.
Mechanism mapping and thematic synthesis linking heterogeneous empirical findings to contextual moderators (governance, regulation, sector); review emphasizes consistent role of these moderators across studies.
FDI’s effects on employment, wages, and income distribution in Sub‑Saharan Africa are mixed and highly context‑dependent.
Conceptual literature review synthesizing theoretical frameworks and empirical findings across micro, firm, sectoral, and macro studies; no new primary data. Review notes heterogeneous identification strategies and results across studies and contexts.
India’s reported post-harvest loss is relatively low (3.2%) despite poor food-security outcomes (Global Hunger Index rank 111/125).
Reported statistics cited in the paper (FAO/Kaggle for post-harvest loss; Global Hunger Index ranking referenced).
Data‑driven policies can either amplify or mitigate inequalities depending on data representativeness, model design, and deployment governance.
Multiple empirical examples and theoretical analyses in the review highlighting cases of both harm (bias amplification) and mitigation, identified across the 103 items.
Citizen acceptance, transparency, and perceived fairness strongly shape adoption trajectories and the political feasibility of AI tools in government.
Repeated empirical findings in the reviewed literature linking public trust, transparency measures, and fairness perceptions to successful or failed deployments (drawn from multiple case studies in the 103 items).
Adoption of AI and data-driven governance is highly uneven across jurisdictions and sectors, driven by institutional capacity, governance frameworks, and public trust.
Cross‑regional and cross‑sector comparisons in the review corpus (103 items) showing varying maturity levels and repeated identification of institutional capacity, governance arrangements, and trust factors as determinants.
Governance approaches are emerging at global, regional and national levels; they vary widely across sectors and jurisdictions, creating opportunities for regulatory experimentation but also risks of fragmentation and regulatory arbitrage.
Cross-jurisdictional comparison of existing/global/regional/national governance instruments and sectoral guidance; gap analysis highlighting heterogeneity.
Weak formal institutions often coexist with strong informal institutions in African contexts, shaping governance, trust, and enforcement mechanisms in supply chains.
Cross-disciplinary literature review presented in the paper; conceptual argumentation rather than primary empirical analysis.
Technology effectiveness depends on institutional support (extension, property rights), finance, and local knowledge — technologies are not a silver bullet alone.
Conceptual frameworks and comparative analysis in the review; supporting case studies and program evaluations linking adoption and impact to institutional factors (extension reach, tenure security, access to credit).
Productivity gains from generative AI depend on task mix, integration design, and the availability of complementary human skills.
Theoretical evaluation and synthesis of heterogeneous empirical findings; authors highlight variation across firms, sectors, and tasks.