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
These infrastructural and access constraints create unequal starting points that can amplify later disparities in labor-market preparedness.
Inference drawn from observed survey disparities in access, hands-on training, and preparedness; the study did not directly measure labor-market outcomes but links preparedness to potential labor-market effects in discussion.
Top-down AI guidance from institutions is common, while grassroots input from educators and students is often missing, which reduces policy relevance and uptake.
Survey items and thematic coding indicating the origin and participatory nature of institutional AI guidelines; comparative prevalence reported in open and closed responses.
Overreliance on GenAI CDS may lead to deskilling of clinicians, eroding judgment over time and increasing systemic vulnerability.
The paper cites theoretical risk and references limited longitudinal concerns; empirical longitudinal studies demonstrating deskilling are scarce per the paper’s stated evidence gaps.
Commercial structural biology services for routine solved folds may be commoditized, pushing firms toward complex validation, novel targets, or high‑value contract research.
Paper suggests this in 'Disruption of service markets' as a projected industry response; it is a strategic implication rather than an empirically demonstrated trend in the text.
Organizational compliance, governance, and transaction costs shape which AI uses are feasible, producing heterogeneity in adoption across firms; trust and accountability frictions can slow adoption even when productivity gains exist.
Workshop participants (n=15) reported compliance and governance considerations; authors infer broader organizational heterogeneity and friction effects from these qualitative data.
Designers’ expressed concerns about skill development suggest potential long-term effects on human capital accumulation; adoption that reduces learning opportunities could lower future wages or employability.
Participants' concerns captured in qualitative workshops (n=15); claim is an extrapolation to labor-market outcomes rather than direct measurement in the study.
Private governance and firm-level solutions (internal standards, bargaining with unions) may proliferate, but these can entrench firm-specific norms and increase market power asymmetries.
Conceptual argument drawing on governance and industrial organization literature; no empirical measurement of prevalence or market-power effects included.
Inadequate protections reduce public trust in mobile-AI services, which can slow diffusion and undercut the growth trajectories that policy narratives anticipate.
Inferred from stakeholder commentary and policy discourse combined with communication-rights theory; the paper does not present survey or adoption-rate data.
Low-wage and platform workers are particularly exposed to algorithmic management and surveillance, with potential downward pressure on wages, bargaining power, and job quality.
The paper's qualitative analysis of stakeholder comments and policy omissions, combined with literature-based inference about platform labor dynamics; no primary labor-market survey or quantitative wage data provided.
Soft‑law governance and growth-first narratives risk concentrating benefits (investment, productivity gains) while externalizing costs (privacy harms, biased decisioning) onto vulnerable populations, exacerbating inequality and reducing inclusive economic development.
Analytic inference from qualitative review of governance instruments and policy narratives combined with communications-ecology and political-economy reasoning; not based on quantitative economic measurement in the paper.
Uncertainty about long-run agentic behavior increases option value and downside risk of investing in agentic systems, which may raise discount rates and required returns.
Economic argument applying risk/return logic to agentic uncertainty; no quantitative empirical evidence provided.
Economic rents and advantages may accrue to agents who control large datasets, computing resources, and organizational processes that effectively integrate AI as a co-pilot, potentially increasing market concentration among AI providers.
Economic theory on scale economies and platform effects combined with observed industry patterns; reviewed literature provides conceptual arguments and case examples rather than broad empirical market-structure measurement.
Generative AI poses substitution risk for entry-level or routine cognitive work focused on generation or drafting without evaluative responsibility.
Task-based analyses and case studies indicating automation potential for routine generation tasks; empirical demonstrations of AI-produced drafts/outputs that could replace such work, but longer-run displacement evidence is limited.
Upfront integration and recurring governance costs mean smaller firms may face higher relative costs — potentially increasing scale advantages for larger incumbents.
Deployment case studies and cost reports indicating significant fixed integration and governance costs; inference to market structure is speculative.
There is a risk of deskilling through excessive reliance on AI, implying a need for continuous training and certification to preserve human judgment.
Qualitative interview evidence and observed concerns about overreliance; authors recommend training/governance based on identified risks; no direct longitudinal measurement of deskilling provided in summary.
Recommendation algorithms and widespread automated advice can induce herding or increase common exposures across retail investor portfolios, with potential macroprudential implications.
Theoretical discussion supported by examples from retail trading episodes and algorithmic amplification literature referenced in the review (conceptual and anecdotal evidence; limited systematic empirical quantification).
Insurance markets may price AI-specific fraud risk, raising premiums or creating new products (AI-fraud insurance).
Speculative economic implication suggested by the authors; no market data or insurer statements cited.
Vendors offering integrated governed hyperautomation stacks may capture premium pricing and increase switching costs, potentially widening adoption gaps between large incumbents and SMEs.
Market-structure and competitive dynamics discussed theoretically in the Implications section; no market-share or pricing data provided.
Higher compliance and liability costs may be passed to districts, potentially affecting the affordability of EdTech for underfunded schools unless federal guidance or subsidies offset costs — a distributional concern.
Economic distributional reasoning (theoretical), not supported by empirical pricing or budget impact data in the Article.
Regulators and standard-setters who value transparency and auditability will need to account for the gap between evaluation results and actionable fixes; firms may require incentives or rules to ensure evaluation leads to remediation, not just documentation.
Authors' policy implication derived from the study's finding of a results-actionability gap and discussion of auditability concerns; speculative recommendation rather than empirical finding.
This study represents the first attempt to conduct a comprehensive evaluation of artificial intelligence (AI) and its influence on job displacement based on the existing body of literature.
Author assertion in the paper; the excerpt provides no external verification (no citation of prior reviews/meta-analyses to justify the 'first attempt' claim).
We currently lack an understanding of how political parties perceive the potential impact AI has on employment, the role of regulations in protecting workers from AI-related job losses, and the importance of AI educational and training programs.
Statement of a literature/knowledge gap motivating the study (assertion by the authors; no empirical basis provided in the excerpt).
Signal legitimacy was validated through negative control experiments.
Experimentation claim: the paper asserts that negative control experiments were run to validate that signals are not due to memorized ticker associations. The excerpt does not specify the design, number, or results of these negative controls.
Pidgin should not be treated as 'broken English' but as necessary linguistic infrastructure for repaired, sustainable development; failures often reflect language-sovereignty crises requiring political solutions.
Normative claim supported by mixed-methods findings on comprehension, adoption, and legitimacy, and Critical Discourse Analysis of institutional language hierarchies.
The paper advances a new conceptual framework called 'Developmental Sociolinguistics' and formalizes Three Laws of Linguistic Justice (Epistemic Access, Discursive Parity, Sovereignty), operationalized via a proposed 'Pidgin Protocol' for decolonized development practice.
Conceptual/theoretical contribution based on synthesis of field results and literature; proposal of framework and laws as normative prescriptions rather than empirically tested policy interventions.
Standards for provenance, labeling of AI-generated content, and interoperable evidence formats would lower verification costs and create beneficial network effects.
Policy recommendation derived from identified verification frictions and the study's analysis of data/model governance needs.
There is growing market demand for AI-assisted fact-checking tools, creating opportunities for software, monitoring services, and labeled datasets.
Analytic implication drawn from findings about increasing AI use and needs for automation/labeling; based on interviews and market inference in the study.
Regulators should consider guidelines on AI monitoring, algorithmic fairness in performance evaluations, and protections to prevent hybrid‑induced career penalties.
Policy recommendation based on conceptual assessment of risks identified in literature synthesis; not an empirical claim—no policy evaluation data provided.
Hybrid agency implies complementarity between GenAI and managerial/knowledge‑worker skills (curation, evaluation, coordination), potentially increasing returns to those skills while automating routine cognitive tasks—consistent with skill‑biased technological change.
Synthesis of recurring themes linking GenAI capabilities with managerial skill topics in the thematic clusters; positioned as an implication for labour demand and skill composition rather than an empirically tested effect.
Policy prescriptions for developing countries to mitigate these vulnerabilities include: diversify supply sources, invest in local human capital and mid-stream capabilities, create legal/regulatory flexibility to navigate competing standards, and pursue regional cooperation to build bargaining leverage.
Policy analysis and recommendations grounded in the mechanisms identified via process tracing and comparative cases; intended as prescriptive synthesis rather than empirically demonstrated interventions in the paper. (Based on inferred best-practice interventions; no empirical evaluation/sample size provided.)
There is demand for tooling that bridges evaluation outputs to actionable fixes (e.g., failure-mode libraries, standardized remediation templates, evaluation-to-priority mapping), signaling economic opportunities for third-party tools and consulting services.
Authors' inference based on the documented results-actionability gap and participants' descriptions of pain points; presented as a market implication rather than direct market measurement.
Firms that invest in instrumentation, cross-functional processes, and remediation levers capture more value from LLMs; organizations with better evaluation-to-action pipelines will obtain higher productivity gains and market edge.
Authors' inference from observed heterogeneity among teams in the interviews and comparison of practices in teams that reported more success converting evaluations into changes.
Public investments in standards, verification infrastructure, and public-interest datasets can correct market failures and support trustworthy AI.
Policy recommendation informed by governance and public-good theory and examples from the literature; the claim is prescriptive and not validated by new empirical evidence within the paper.
Policy instruments (law and markets) should be designed to remain institutionally and procedurally responsive to ethical claims that resist full codification (e.g., through participatory governance, oversight mechanisms, equitable redress, care-centered procurement standards).
Normative policy prescriptions derived from the Levinasian diagnosis and case illustrations; proposed measures are normative and not empirically evaluated within the paper.
Integrating Object-Oriented Ontology (OOO) and the material turn enables attention to nonhuman actors and assemblages without collapsing them into human-centered instrumentalism.
Theoretical synthesis of OOO/material-turn literature and argument that this synthesis offers analytic resources for socio-technical assemblages; illustrated conceptually in domains.
Humans who configure and teach agents gain understanding and skills themselves — learning-by-teaching generates human capital accumulation endogenous to agent deployment (bidirectional scaffolding).
Qualitative, naturalistic observations and comparative documentation of users configuring/teaching agents during the one-month study; no randomized assignment or pre/post quantitative skill testing reported.
Models trained primarily on negative constraints will generalize constraint adherence more robustly under distribution shift than models trained primarily on preference rankings.
Presented as a central, experimentally falsifiable prediction derived from the paper's theoretical account; the paper does not present large-scale empirical confirmation and recommends controlled experiments to test this.
Negative examples function as counterfactual eliminators that rule out regions of behavior space, allowing a model to settle on robust acceptable behavior, whereas positive preference signals require continual calibration in a high-dimensional, context-sensitive space.
Informal/structural theoretical argument and analogy to falsification presented in the paper; no direct empirical test reported there demonstrating this exact mechanism.
Regulators may prefer systems that support contestability and audit trails and could mandate argumentation-style explainability in certain sectors.
Speculative policy prediction; no regulatory statements or empirical policy adoption evidence cited.
Better contestability may reduce litigation and regulatory frictions if decisions are transparently defensible.
Speculative legal-economic claim; no case studies or empirical legal analysis provided.
New service layers may emerge (argumentation-as-a-service, audit firms, explanation certification, human-in-the-loop orchestration platforms).
Speculative market/industry evolution claim based on analogous tech-service cretions; no empirical evidence.
New metrics are needed to value resilience (robustness to out-of-distribution events, graceful degradation) in procurement and contracting; performance-based contracts and regulated minimums for oversight mode selection can help align incentives.
Prescriptive recommendation based on gaps identified in procurement and contracting practice; conceptual proposal without empirical testing.
Demand will grow for tools and services that enable oversight (auditability, explainability, safe fallbacks), creating markets for verification, certification, safety middleware, and human-in-the-loop platforms.
Market-structure and demand-side reasoning based on the proposed governance needs; forecast-style projection without empirical market-data analysis.
Allocation decisions should be explicit, auditable, and adaptive — with provisions for overriding, fallbacks, and graceful degradation during unanticipated conditions.
Normative recommendation based on safety and accountability principles combined with crisis-management practices; argued via conceptual analysis and illustrative design features.
Investment in multimodal continual learning, scalable and reliable knowledge-editing methods, and retrieval architectures that guarantee cross-modal consistency is economically justified.
Research/prioritization recommendations based on empirical benchmark findings showing current gaps; argumentation for R&D focus areas.
The findings argue for policies requiring disclosure of training-data timeframes and robust monitoring for time-sensitive factual accuracy in deployed systems.
Policy recommendations in the paper drawing on benchmark results and identified failure modes; prescriptive argumentation rather than empirical policy evaluation.
Models and platforms that offer transparent update mechanisms (frequent data updates, reliable RAG pipelines, clear training snapshot metadata) will have competitive advantages in the market.
Economic and market analysis in implications section recommending transparency and update mechanisms as differentiators; speculative/business-analytical evidence rather than experimental.
Embedding culturally aligned moderation and multi-layer safety orchestration can reduce regulatory frictions and increase adoption in conservative or tightly regulated markets.
Paper claims regulatory and safety economics implications from their safety/moderation architecture; this is an asserted implication rather than an empirically validated outcome in the summary.
The methods used (data quality focus, continual pre-training, model merging, modular product stacks) are potentially transferable to other underrepresented/low-resource languages, lowering barriers to regional AI competitiveness.
Paper posits this policy/transferability implication as an argument in the 'Implications for AI Economics' section; no cross-language experimental evidence provided in the summary.
Fanar 2.0 demonstrates that targeted data curation, continual pre-training, and model-merging can be a viable alternative to the raw-scale pre-training arms race for language-specific competitiveness.
Paper argues this implication based on achieving benchmark gains on Arabic and English using curated data (120B tokens), continual pre-training, model-merging, and a 256 H100 GPU training budget rather than massively larger-scale pre-training.