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 |
Teams often produce evaluation outputs (tests, metrics, user feedback) but lack mechanisms, processes, or technical levers to convert those outputs into actionable engineering or product changes—a novel “results-actionability gap.”
Recurring theme from the 19 practitioner interviews and coding; authors explicitly articulate and label this gap based on participants' reports.
The study confirms several previously documented evaluation challenges with LLMs: model unpredictability, metric mismatch, high human-evaluation costs, and difficulty reproducing failures.
Interview data from 19 practitioners; thematic analysis flagged these recurring problems as reported by participants and aligned with prior literature.
Emergent quality hierarchies among agents imply winner-take-most dynamics in informational value and potential market concentration in agent quality.
Observed formation of quality hierarchies in agent interactions and documented economic interpretation; this is a hypothesis/implication drawn from qualitative patterns rather than measured market outcomes.
Security of LLM-based MASs functions as an economic externality: failures can impose social costs (misinformation, poor collective decisions), and absent liability or market incentives providers may underinvest in robustness.
Economic reasoning and implication section in the paper—conceptual argument linking the technical vulnerability to economic externality and incentive misalignment. No empirical economic data provided in the summary.
Analytical conditions on stubbornness and influence weights identify when a single adversary can dominate network dynamics (i.e., influence propagation criteria derived from FJ fixed-point analysis).
Mathematical/theoretical analysis of FJ model fixed points and influence propagation in the paper; derivation of conditions relating agent stubbornness and interpersonal trust weights to steady-state influence.
Large-scale battlegrounds and competitions increase compute demand and associated costs, with implications for budgets and environmental externalities.
Paper notes that the Battling Track dataset (20M+ trajectories), model training for baselines/competitions, and running a living benchmark imply substantial compute; this is an argued implication rather than measured environmental impact.
Rapid deployment of autonomous learners could accelerate displacement in affected sectors and widen inequality if gains concentrate among capital owners or platform providers.
Socioeconomic risk assessment and projection; conceptual and not empirically quantified in the paper.
Faster, more generalist embodied AI could substitute for routine physical and social tasks, shifting human labor toward oversight, high-level planning, creativity, and flexible social cognition roles.
Labor-market impact hypothesis derived from automation literature; conceptual projection only.
Organizations without access to high-frequency operational data may face increased barriers to entry in latency-sensitive markets, concentrating rents with incumbents who can collect such data.
Paper presents this as an implication of the dataset/value results: proprietary high-frequency data can create competitive advantages. This is a policy/economic implication derived from model performance observations rather than a tested market analysis.
If models frequently leak or misuse preferences in third‑party contexts, users and organizations will discount the value of personalization or demand stronger controls, increasing costs for deploying memory features and reducing consumer surplus.
Economic reasoning and implication drawn from the observed misapplication behavior; no empirical user adoption or market data provided in the study to directly support this claim.
The failure mode (misapplication of preferences to third parties) creates negative externalities (privacy violations, normative harms, misinformation, contractual breaches) that markets and platforms may not internalize without regulation or design changes.
Economic interpretation and argumentation building on the empirical failure mode; these harms are hypothesized implications rather than measured outcomes in the paper.
Widespread adoption of predictive HR tools raises distributional and fairness concerns (algorithmic bias, disparate impacts) and privacy risks that may prompt regulatory responses affecting adoption costs and equilibrium outcomes.
Discussion/implications section raises these risks conceptually; the paper does not empirically measure downstream policy or distributional effects.
Unclear liability frameworks increase perceived and real costs and can slow adoption by hospitals and insurers.
Policy analyses and procurement narratives noting liability uncertainty cited as a barrier to procurement and deployment.
Up-front implementation costs commonly include procurement, integration with PACS/EMR, UI/UX development, regulatory compliance, and staff training; recurring costs include monitoring, data labeling, software updates, and cybersecurity.
Implementation reports, vendor and hospital accounts, and qualitative studies documenting cost categories (specific dollar amounts vary across settings and are rarely published in detail).
Uneven organizational supports can concentrate returns to AI in firms and workers that successfully actualize affordances, potentially widening wage and employment disparities; targeted policy and training investments can mitigate these effects.
Theoretical implication from the framework with policy recommendations; no empirical testing or sample reported in the paper.
These trends (job polarization and differential wage/mobility outcomes) may exacerbate economic disparities across regions.
Interpretation and projection based on the observed trends in the reviewed literature and reports; presented as a risk/implication rather than an empirically tested causal finding in the summary.
Without continuous support for upskilling/reskilling and inclusive policies, AI risks becoming a source of exclusion rather than an enabler of human advancement.
Normative conclusion derived from reviewed literature and thematic interpretation in the qualitative study (literature-based; evidence is secondary and not quantified).
At the national level, AI-related innovations are yet to be transformed into measurable economic gains.
Interpretation based on the observed negative association between AI patent counts and GDP growth from the panel regressions (OLS, FE, Difference and System GMM) and theoretical reasoning about adoption/diffusion lags and complementary requirements; empirical support derives from the same models (sample details not provided).
Research literature synthesis demonstrates 70-75% automation potential.
Quantitative estimate offered by the authors (70-75%) as part of function-by-function analysis; no described empirical evaluation or sample supporting the figure.
Knowledge transmission (teaching/lecturing) shows 75-80% AI substitutability.
Authors' quantitative estimate presented in the analysis (75-80%); the paper does not detail empirical methods or validation samples for this percentage.
Administrative tasks face 75-80% disruption risk from AI.
Paper provides a quantitative estimate (75-80%) as part of its functional disruption assessment; no empirical methodology, dataset, or sample size is described to support the numeric range.
Over 400,000 [individuals] are projected to die before obtaining permanent residency.
Mortality projection applied to the estimated backlog and projected wait times (authors' projection); exact demographic assumptions (age distribution, mortality rates) and method are not provided in the excerpt.
The remaining difference (roughly 70%) is not explained by the factors observed in the data, indicating additional influences not captured in the survey.
Residual (unexplained) component from decomposition analyses on ESJS data.
Demand-dependent pricing in the modeled energy load management setting creates a social dilemma: everyone would benefit from coordination, but in equilibrium agents often choose to incur congestion costs that cooperative turn-taking would avoid.
Theoretical/modeling analysis of consumer agents scheduling appliance use under demand-dependent pricing as described in the paper (analytical argument and/or model simulations). Specific sample sizes or simulation parameters are not given in the abstract.
Aggregation and linkage across data sources can reveal intimate, predictive traits that were not foreseeable to the data subject at the time of sale.
Conceptual argument with references to documented cases and literature on data linkage and inference; relies on illustrative examples rather than original empirical experiments.
Policy-relevant implication (extrapolated): identity heterogeneity implies family- and purpose-driven entrepreneurs may be less likely to pursue AI-enabled innovation after income shocks, suggesting targeted outreach and low-risk entry paths to avoid widening digital divides.
Extrapolation from documented identity-heterogeneous declines in innovation after income shocks (empirical result) to probable patterns in AI adoption; AI adoption is not directly measured in the paper's dataset.
The United States shows a more market-driven (firm-dominated) patenting profile and comparatively weaker integration between AI and robotics patent trajectories.
Country-level and actor-type decomposition for U.S. patent filings (1980–2019), showing higher firm share of patents and weaker long-run association/cointegration between core AI and AI-enhanced robotics series compared with China (as reported in the paper).
There is a risk of a two‑tier market where high‑quality temporal‑preserving enhancements are costly, increasing inequality in experiential welfare and cognitive capital.
Speculative socioeconomic implication based on cost/access arguments and distributional concerns; no inequality modeling or empirical pricing data provided.
Technical expansion without an accompanying theory of lived temporality risks increasing capabilities while degrading the qualitative depth of human experience (presence, attentional flow, felt meaning).
Argumentative claim supported by philosophical analysis and literature synthesis (neurophenomenology, attention economics); no empirical test reported (N/A).
Differential access to higher-quality (paid) versus free GenAI tools and differing ability to engage with the tool could widen inequality among students and institutions.
Authors' implication based on student-reported concerns about limitations of free ChatGPT versions and on heterogeneous gains across disciplines; this is a policy/implication claim not directly measured in the experiment.
High-quality, equitable climate information displays public-good characteristics (nonrival, nonexcludable at scale), so private incentives alone will underprovide geographically representative data and shared infrastructure.
Economic reasoning supported by observed concentration of compute and model development (mapping) and standard public-goods theory; no formal empirical market model estimated in the paper.
Improving photorealism with objective color-fidelity metrics and refinement reduces the need for manual color correction and retouching in downstream workflows.
Paper and summary argue this as an implication: higher-fidelity outputs from CFR/CFM reduce manual editing demand. This is an economic/market implication rather than a directly evidenced experimental result in the paper (no labor-market causal study reported).
The paradigm implies potential market risks including vendor lock-in and concentration if only a few providers control scalable linear-optical samplers.
Conceptual risk analysis in the paper's discussion of economic implications; this is a qualitative argument built on the technical premise that trained models require access to specialized quantum sampling hardware for deployment.
Heterogeneous trust levels across firms and schools may produce uneven productivity gains and widen performance gaps.
Logical implication and policy discussion in the paper; the cross-sectional study documents relationships between trust and outcomes but does not provide aggregate diffusion or cross-firm longitudinal evidence to confirm unequal sectoral diffusion.
Overreliance on unvetted AI can propagate biases; economic gains from AI therefore require governance, auditing, and accountability mechanisms.
Framed as a risk and policy recommendation in the discussion; not an empirical finding from the cross-sectional survey reported in the summary.
If FDI brings capital‑intensive, AI‑enabled production without complementary upskilling, it may exacerbate wage inequality and deepen labor market dualism in SSA.
Theoretical inference and analogy from documented patterns of skill‑biased technological change and FDI-driven inequality in the reviewed literature; empirical evidence specific to AI in SSA is lacking in the review.
Full replacement of physicians would require breakthroughs in robust generalization, embodied capabilities, and legal/regulatory change—currently lacking.
Conceptual inference based on documented limitations (OOD generalization, lack of embodied/sensorimotor capability, unsettled legal/regulatory environment) summarized in the review.
Shrinking acquisition workforce capacity functions as a critical scarce input in defense AI economics; reduced human capital lowers the Department's ability to extract value from AI investments and to internalize externalities, decreasing effective returns to AI procurement.
Institutional trend evidence of workforce reductions combined with economic analysis treating institutional capacity as an input factor. No empirical quantification of returns or elasticity provided—this is analytical inference.
Ambiguous standards increase uncertainty for contracting officers, raising the risk that they will either over-rely on vendor claims or inconsistently enforce requirements, both of which harm procurement integrity.
Policy-text analysis identifying vague criteria combined with qualitative analysis of procurement decision workflows; argument based on measurement and enforcement friction literature. No empirical study of contracting officer behavior provided.
Lower governance barriers and ambiguous procurement criteria (e.g., undefined 'model objectivity') can skew market competition toward suppliers that prioritize rapid iteration and opaque practices over rigorous assurance, harming traceability and quality.
Market-effects reasoning grounded in policy changes (document analysis) and qualitative institutional analysis of measurement/enforcement frictions. No market-share or supplier-behavior data provided.
Mandating permissive contract terms and enabling waivers reduces private incentives for contractors to invest in safety and compliance, creating classical moral-hazard problems in defense AI procurement.
Economic reasoning and principal–agent analysis applied to the documented contractual changes (primary-source policy text). No empirical measurement of contractor investment behavior provided; claim is theoretical/inferential.
A mismatch between expanded waiver authority (Barrier Removal Board) and declining acquisition oversight capacity creates procurement-integrity and systemic risks: faster acquisition concurrent with weakened institutional checks increases likelihood of improper procurement decisions and unchecked deployment of unsafe or unvetted AI models.
Synthesis of primary-source policy analysis, institutional staffing trend evidence, and qualitative risk/scenario assessment using principal–agent and moral-hazard frameworks. This is a conceptual risk projection rather than an empirically derived probability estimate.
Emerging agentic/AGI capabilities introduce new failure modes and governance challenges that standard ML oversight may not cover.
Emerging literature, theoretical analyses, and expert opinion summarized in the synthesis; authors note limited empirical long-term data and characterize this as an emergent risk.
Centralized provision of high-quality coding models by a few vendors could produce vendor lock-in and increase platform power in software development inputs.
Market-structure analysis and industry observations synthesized in the paper; the claim is forward-looking and not established by longitudinal market data within the review.
If many firms adopt AI generation without matching verification, aggregate fragility in software-dependent infrastructure could rise, increasing downtime costs and systemic economic risk.
Macro-level risk projection and system fragility argument in the paper; no macroeconomic modeling or empirical scenario analysis provided.
This reversal of the burden of proof creates moral-hazard-like behavior: incentives for speed reduce verification effort.
Theoretical argument built on the micro-coercion mechanism and economic reasoning; no empirical validation provided.
Under time pressure, developers adopt an implicit default of accepting plausible machine outputs unless they can disprove them (the 'micro-coercion of speed'), effectively reversing the burden of proof.
Behavioral mechanism posited from descriptive reasoning and thought experiments; no behavioral experiments, surveys, or observational data reported.
DAR dynamics (authority states, hysteresis, safe-exit times) introduce path-dependence and switching costs that should be treated as state variables in production and decision models of human–AI joint work.
Theoretical implications section arguing these elements add path-dependence and switching costs to economic/production models; analytic reasoning, not empirical measurement.
Concentration risks exist because high fixed costs for safe integration and model adaptation may favor larger incumbents or platform providers.
Conceptual economic reasoning and practitioner commentary synthesized in the review; no empirical market-structure analysis or sample-based evidence included here.
Rich contextual memories and continuous home interaction create valuable data streams that could enable firms to capture substantial value, raising concerns about data governance, consent, and monetization.
Authors' policy and economic implications discussion noting that MMCM-like memories generate valuable data; this is a conceptual/policy claim rather than empirically tested within the study.