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 |
Games can act as front-ends to underlying models and datasets or bridge multiple DSTs, improving interoperability and workflow fit for farmers.
Examples of prototypes and deployed tools that connected game interfaces to models/datasets or multiple DSTs; evidence is case-based and demonstrates feasibility rather than large-scale adoption.
Serious games can explicitly model economic outcomes alongside environmental metrics, showing how mitigation/adaptation actions affect enterprise resilience and income.
Prototype demonstrations and case studies that combined economic models with environmental outputs in game interfaces; economic outcome data in these examples are limited and typically short-term or simulated rather than long-term observed incomes.
Dynamic, scenario-based visual outputs in serious games help users understand trade-offs over time (for example, carbon sequestration versus yields).
Comparative demonstrations and workshop observations where scenario visualization was used to communicate temporal trade-offs; evaluation mostly via self-reported comprehension and qualitative feedback from participants.
Interactive, transparent simulations in games reduce skepticism by letting users explore assumptions and model behavior, thereby building trust in DST recommendations.
Qualitative interviews, user testing in workshops, comparative demonstrations where participants explored model assumptions and reported increased confidence; evidence primarily anecdotal and from small pilots.
Co-design through serious games facilitates participatory design with farmers and stakeholders, producing tools that better match on-farm decision contexts and preferences.
Reports from participatory workshops and co-design sessions, case studies of prototype development with farmer groups; evidence largely qualitative (user feedback, design iterations) and based on small-group engagements.
Serious games—interactive, simulation-based decision support tools—can materially increase farmer uptake of land-use decision support tools (DSTs) needed to meet global net zero targets by enabling co-design, building trust, visualizing outcomes, demonstrating profitability–environment links, and integrating with other tools.
Synthesis of literature and practice examples including case studies and deployed game prototypes used with farmer groups, participatory workshops, and qualitative interviews/surveys. Evidence is primarily from small-scale pilots and demonstrations rather than large randomized trials; sample sizes are heterogeneous and often small or not reported.
Cost–benefit analyses in AI economics should internalize long-term, hard-to-quantify harms (autonomy loss, social trust erosion) rather than rely solely on market price signals.
Normative critique of standard welfare analysis with literature support from ethics and political philosophy; no empirical recalculation of cost–benefit models provided.
Investing in privacy-preserving AI methods (differential privacy, federated learning, synthetic data) and governance institutions is warranted as an alternative to atomized data markets.
Policy and technical recommendation based on literature on privacy-preserving techniques and governance models; paper does not present original technical evaluations or cost–benefit analyses.
Economists modeling AI markets should incorporate non-pecuniary harms, externalities, and moral constraints when assessing welfare, innovation trade-offs, and optimal policy.
Normative recommendation grounded in philosophical argument and critique of standard welfare frameworks; not supported by empirical methodological comparison in the paper.
The paper's conceptual contribution challenges macro-centric crisis narratives by centering social mechanisms (support systems, peer benchmarking, institutional trust) as critical determinants of small-firm adaptation.
Theoretical framing (novel socially embedded analytical lens) combined with empirical results showing the importance of networks, identities, and normative motivations in explaining adaptation outcomes relative to macro-structural explanations.
AI governance for training should require content validation, transparency of model use, data minimisation, human accountability, and auditable logs to prevent hidden biases and exclusion.
Policy recommendation from conceptual risk analysis and best-practice governance principles; no field implementation or audit data provided.
Skills recognition should emphasize functional, employer‑usable verification and portability (e.g., micro‑credentials, QA/transparency instruments), not formal legal harmonisation.
Policy recommendation coming from conceptual analysis and review of transferable instrument layers (drawing from EU tools); no empirical comparison provided.
Mandatory pre-departure training in South–South labour corridors (examined via the Myanmar–Malaysia corridor) is a highly implementable, cross-level lever for improving regularity and rights-protecting mobility in contexts with limited enforcement and coordination capacity.
Conceptual analysis anchored in the Myanmar–Malaysia corridor using a structured desk review of policy/program materials, corridor process mapping, and governance gap analysis. No new causal field experiments or quantitative impact estimates reported.
Practical recommendation: include policy historians and political‑economy scholars in AI advisory bodies and require replication/open data for influential results to limit covert ideological influence.
Normative and institutional recommendations based on the historical case study showing interdisciplinary gaps and channels of influence; proposed remedies in the paper.
Practical recommendation: increase transparency and disclosure of funding, affiliations, and normative assumptions in AI economics research to make potential persuasion effects visible.
Policy recommendation derived from the case study's findings about how funding and institutional contexts shaped intellectual influence; prescriptive inference rather than empirical demonstration.
These anti‑democracy/anti‑market ideas gained legitimacy and wider influence through elite channels (notably Nobel laureates and canonical publications), increasing their influence on policy and public discourse.
Tracing dissemination pathways via publication venues, prestige of authors (including Nobel laureates), citation and institutional channels, and archival records indicating engagement with policy circles; qualitative inference from prominence of authors and outlets.
AI adoption raises executives' human capital/market value, which contributes to higher compensation.
Mediation tests linking AI application to measures of executive human capital (skills/market value) and linking those measures to higher pay in the reported analyses.
AI adoption increases firm total factor productivity (TFP), and higher TFP is associated with higher executive compensation.
Mechanism analysis reporting that firms with higher AI application have higher estimated TFP, and TFP is positively related to executive pay (mediation tests on the sample).
AI adoption alleviates financing constraints, and this channel contributes to higher executive compensation.
Mediation/mechanism tests in the paper showing AI adoption is associated with reduced financing constraints, and reduced financing constraints are associated with higher executive pay (mediation analysis on the A-share firm panel).
The rapid rise of AI-enhanced robotics since the 2010s signals a shift toward increased embedding of AI into hardware systems, accelerating cross-sector spillovers.
Interpretation based on observed acceleration in AI-enhanced robotics patents (patent filings 1980–2019) and the convergence patterns reported in the paper. This is an inference drawn from patenting trends rather than a directly measured measure of cross-sector spillovers.
Nonlinear adoption/diffusion models that allow for thresholds, complementarities, and endogenous firm investment responses will better capture tipping points and adoption dynamics than linear models.
Modeling proposal arguing theoretical need for nonlinear specifications and endogenous adoption; no empirical fit comparisons or simulated sample evidence are presented in the paper.
Estimating micro-level gross flows at occupation × industry × geography × demographic granularity (and at higher frequency) will better capture transitions such as reemployment paths, upskilling, and churn.
Proposal to use CPS, LEHD/LODES, JOLTS, administrative unemployment records and firm panels to estimate high-resolution flows. No empirical estimates or sample-size specifics provided.
Nowcasting and real-time analytics (including LLM re-scoring and streaming signals like job postings/platform activity) can update OAIES and short-term projections to improve monitoring.
Proposal to ingest real-time/near-real-time inputs (job-posting APIs, platform data, administrative records) and re-score tasks via LLM embeddings. No implemented nowcast results or sample-based evaluation are presented.
Incorporating causal identification methods (DiD, event-study, synthetic controls, IV) with task-based exposure will yield more credible causal estimates of AI’s effects on employment, wages, and mobility than correlational risk scores.
Methodological claim supported by standard econometric approaches proposed for use with the OAIES and staggered adoption/panel data. No empirical demonstration is provided; evidence is methodological rationale.
Crises (pandemics, supply shocks) tend to accelerate digital and AI adoption, potentially shortening adjustment time to new technological regimes.
Interpretation of recent historical episodes (e.g., COVID-19) and diffusion literature; qualitative assertion without presented microeconometric identification.
AI and the green transformation function as modern long-wave drivers by improving operational efficiency, enabling new products and services, and reorganizing competitive hierarchies.
Conceptual argument linking general-purpose technology literature to observed/anticipated capabilities of AI and green tech; literature synthesis without original empirical tests.
Schumpeterian cycles are driven by clusters of technological innovations and entrepreneurial activity; AI and green technologies represent contemporary innovation clusters with strong potential for productive disruption.
Application of Schumpeterian theory to contemporary technology trends via literature synthesis and conceptual argument (no empirical quantification provided).
The paper's qualitative framework can be operationalized for economists into measurable constructs such as task-level time use, output quality metrics, billable hours, client satisfaction, wages, and employment composition.
Authors propose next steps and measurement opportunities; suggestion comes from translating interview-derived categories into empirical variables for future work.
Architectural education should integrate AI tool training and algorithmic thinking to align workforce skills with evolving task demands.
Authors' recommendation grounded in interview evidence that students are adopting algorithmic strategies and in the constructed conceptual framework; presented as pedagogical implication.
Algorithmic thinking strategies—procedural, iterative, and prompt-based reasoning—are central to how students engage with GenAI during co-design.
Inductive thematic analysis of student interviews identified recurring descriptions of procedural/iterative prompting and tool orchestration as core practices.
Integrating lived temporality into design and evaluation is necessary to preserve and enhance the qualitative aspects of human life in transhumanist transformation.
Normative/philosophical argument supported by literature synthesis and conceptual reasoning; no empirical demonstration (N/A).
AI/ML methods can reduce reliance on animal models by simulating biology, optimizing experiments, and prioritizing candidate drugs—supporting the 3Rs (Replacement, Reduction, Refinement)—but this is contingent on rigorous validation and ethical oversight.
Conceptual and methodological arguments (Manju V et al.) and cited examples of validated in silico alternatives and experiment‑optimization workflows; no single trial or sample size—recommendation based on synthesis of studies and caveats about validation and regulation.
CDRG‑RSF identified five prognostic genes including UBASH3B, which is associated with reduced NK activation and may mediate drug resistance—making it a candidate therapeutic target.
Feature selection within the CDRG‑RSF model yielded five prognostic genes; UBASH3B shown to correlate with immune suppression (reduced NK activation) and inferred links to drug resistance (associational analyses; functional validation not specified in summary).
PIGRS prognostic model (LASSO + Gradient Boosting Machine ensemble using 15 programmed‑cell‑death immune genes) outperformed most published LUAD prognostic models.
Prognostic modeling using LASSO feature selection followed by GBM ensemble on a 15‑gene panel; comparative benchmarking against published LUAD prognostic models reported superior performance (metrics and external cohort testing referenced).
Multi‑omics integration and consensus clustering (10 methods) in lung adenocarcinoma (LUAD) identified three molecular subtypes (CS1–CS3) with distinct prognoses.
PIGRS study integrated transcriptome, DNA methylation, and somatic mutation data and applied ten clustering algorithms to define molecular subtypes; reported three subtypes with differing survival outcomes (external validation cohorts used).
Data augmentation with Gaussian noise improved DNN performance for small sample cross‑omics training sets.
Cross‑omics study applied Gaussian noise augmentation during DNN training on small paired viral datasets and observed improved model performance and DEA recovery relative to non‑augmented training.
Dynamic Ensemble Selection‑Performance (DES‑P) produced parsimonious, high‑accuracy classifiers within the EPheClass pipeline.
Use of DES‑P for model selection in EPheClass reportedly yielded small, high‑performing ensembles (example: periodontal disease AUC = 0.973 with 13 features).
Applying centred log‑ratio (CLR) transformation and RFE to compositional microbiome data improves model parsimony and supports reproducibility in diagnostic classifiers.
EPheClass preprocessing: CLR to handle compositional 16S data and RFE to reduce feature sets; resulted in small feature panels (e.g., 13 features) with high performance and emphasis on rigorous validation to avoid prior overfitting issues.
The same EPheClass approach produced successful parsimonious classifiers for IBD (26 features) and antibiotic exposure (22 features).
EPheClass applied to additional microbiome outcomes (IBD and antibiotic exposure) with RFE selecting 26 and 22 features respectively; performance described as 'successful' (exact AUCs not provided in summary).
Ethical and policy considerations require disclosure of synthetic participant use, protection against contamination of human-data pools, and attention to consent and representation issues.
Authors' ethical recommendations based on harms and risks identified across the reviewed studies (contamination, misrepresentation, labor-market effects for participants).
There is a need for standardized benchmarks for economic behaviors (e.g., strategic interaction, intertemporal choice, risk, social preferences) to enable cross-study comparisons and rigorous validation of synthetic participants.
Authors' synthesis and recommendations motivated by heterogeneity in metrics and methods across the reviewed literature.
LLM-generated synthetic participants are a promising low-cost, flexible adjunct for research and data-collection tasks (useful for pilots, prototyping, hypothesis generation, stress-testing, and augmenting small human samples).
Synthesis of reviewed literature identifying applied use-cases and reported benefits across multiple studies; authors' recommendations based on aggregated findings.
The value-of-deception metric can be used to monetize the benefit of deception technologies relative to non-deceptive alternatives, supporting investment and cost–benefit comparisons.
Conceptual/analytical proposal in the implications section: metric defined in utility units and argued to be interpretable for economic valuation (no empirical monetary valuation provided).
There exist parameter regimes where simple allocation heuristics nearly match optimal allocations (heuristics are practically sufficient in some regimes).
Combination of analytical approximation guarantees and simulation results comparing heuristic performance to computed optima (analytical proofs plus simulated evaluations; sizes/instances not specified).
Computational experiments across heterogeneous simulated scenarios produce consistent cross-setting comparability of the proposed metrics (value of deception and price of transparency).
Simulated computational experiments sweeping parameters (decoy realism, budget, attacker rationality, observability); results reported showing comparability across scenarios (simulations; sample sizes/number of scenarios not specified).
Included studies (n=27) reported improvements in learner outcomes mapped to Kirkpatrick‑Barr levels 1–3 (learner reaction/satisfaction; attitudes/perceptions; knowledge/skills; behavior change).
Outcome extraction and mapping reported in the review: evaluations in included studies used learner surveys, knowledge/skill tests, and self-reported behavior-change measures to classify outcomes into Kirkpatrick‑Barr levels 1–3 across the 27 programs.
AI-enabled upskilling and AI-guided procedures weaken the negative effect of workplace stress on employee retention (AI moderates the stress→retention link).
Moderation test in PLS-SEM on N = 350. Reported moderator effect (AI × Stress → Retention): β = 0.078, p < 0.005 (interpreted as a buffering/weakening effect of AI interventions on the stress→retention relationship).
The framework’s emphasis on traceability and IT integration creates rich datasets suitable for econometric evaluation (causal impact on earnings, placement) and for training ML models (curriculum recommendation, skill-gap prediction).
Argument in paper about secondary uses of integrated data (conceptual); no datasets or empirical model training described.
Modelling artefacts (flowcharts/logigrams and algorigrams) can encode repeatable lesson-planning, assessment and audit algorithms.
Paper's modelling artefacts description (conceptual/tools).
Firms and hospitals need differentiated investment and governance strategies by interaction level: integration and workflow redesign for AI-assisted; training and decision-support protocols for AI-augmented; process redesign, liability allocation, and oversight for AI-automated systems.
Prescriptive recommendations derived from cross-case findings (n=4) and the conceptual mapping to innovation management implications.