Evidence (8807 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
Filtered →
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 |
Productivity
Remove filter
Comparative analysis reveals significant institutional differences between EU and Ukrainian legal systems that are relevant to regulatory stability, the cost of innovation, data accessibility, the balance of market power, and guarantees for consumers and employees.
Qualitative comparative examination of institutional and cultural/procedural differences between EU and Ukraine as presented in the paper (method: comparative approach; no quantitative metrics provided).
Most Ukrainian laws relevant to the digital economy are based on existing legal structures and systems, and Ukraine currently lacks a unified regulatory system specifically designed for artificial intelligence.
Comparative analysis of Ukrainian and EU legal frameworks as described in the paper (method: comparative approach; legal document review referenced qualitatively).
Digitalisation is making data and algorithmic systems increasingly important economic resources, thereby changing the way markets operate, how labour is organised, how productivity is measured and how income is distributed.
Conceptual analysis and theoretical model developed via literature synthesis and comparative approach (no empirical sample reported).
Interpretability, trust calibration, and interface design matter, but they cover only part of what determines whether human-AI combination works.
Authors' argumentative claim based on their analysis and mapping of broader factors; presented as an evaluative conclusion rather than an empirical estimate.
Meta-analytic evidence shows moderate but heterogeneous effects of agentic/code-generation tools on productivity.
Reference to meta-analytic synthesis across studies reported in the paper (meta-analytic details not provided in abstract).
Some merged PRs introduce new lint or security findings while simultaneously removing existing issues (i.e., merges sometimes involve both addition and removal of issues).
Before-and-after static analysis (Pylint and Bandit) of merged PRs showing coexistence of introduced and removed findings in observed diffs.
Cross-model validation reveals architecture-level trade-offs independent of specific LLMs: Dual Process excels at numeric/temporal queries (65-90% accuracy) while RAG excels at historical retrieval (60-85% accuracy).
Empirical cross-model tests across six LLMs; reported accuracy ranges for different query types and architectures.
AI functions both as a general-purpose technology and as an innovation in the method of innovation.
Conceptual/theoretical framing presented in the paper (the authors characterize AI as both a GPT and an innovation in methods of innovation).
Clarifying-question prompts produced mean rubric scores of 6.67 out of 8, higher than raw prompts but lower than checklist-improved prompts.
Reported mean rubric scores in the abstract showing clarifying-question prompts scored 6.67, compared to 5.67 for raw and 7.50 for checklist.
Classical categories (labour, capital, firm, market, productivity, trust) remain necessary but are incomplete for describing economic action when technologies prepare decisions, coordinate workflows, support tasks, verify transactions, and reshape responsibility.
Conceptual analysis supported by diagnostic indicators showing distributed decision/action capacity across humans, AI agents, robots, protocols, compute and energy systems; argumentative/theoretical evidence rather than causal inference.
Labour projections are more consistent with task reallocation than labour disappearance.
Analysis of labour-market reallocation data and labour projections (public sources) interpreted under a task-reallocation framework rather than full employment loss, using relative growth and reallocation indicators.
Readiness and performance-related variables are associated with higher predicted success, whereas higher barrier levels are associated with lower predicted success.
Model coefficients/feature effect analyses and nonlinear diagnostics from the fitted models.
High-AIC participants realized outsized gains from GenAI access; low-AIC participants saw limited or even negative marginal returns.
Subgroup analysis of the randomized experiment comparing treatment effects by AIC level; authors report large positive treatment effects for high-AIC subgroup and small or negative effects for low-AIC subgroup.
The distribution of gains from GenAI access was highly uneven across users.
Experimental results showing heterogeneous effects across participants (variance/heterogeneity analyses reported in the paper).
The same observable behavioral signal can carry opposite meaning for different agent configurations.
Synthesis of the cross-configuration empirical findings (directional disagreements such as the error-rate example and other features).
Five other continuous features and three of seven binary patterns from prior SE literature show similar directional disagreement across configurations.
Aggregate empirical finding across the set of features and binary patterns analyzed in the 126-configuration dataset.
Error rate is the cleanest case: 47 configurations resolve more issues when their error rate is lower, while 48 resolve more when it is higher.
Empirical counts from the paper's analysis of configurations (reported 47 vs 48 configurations showing opposite sign relations between error rate and issue resolution).
On most signals, configurations disagree not merely in magnitude but in direction (i.e., the same signal correlates positively with resolution in some configurations and negatively in others).
Across-configuration comparison of behavior–outcome correlations for many signals in the dataset of 126 configurations / 64,380 runs.
Swapping the framework while the LLM is held fixed produces large behavioral differences in every action feature.
Comparative analysis across configurations holding LLM fixed; reported observation across action features.
The system is generically bistable, with a stable partial adoption equilibrium coexisting alongside full genuine adoption.
Analytical results from the evolutionary game-theoretic model demonstrating multiple stable equilibria (bistability). No empirical sample (theoretical proof / model analysis).
Doctors choose among three strategies: genuine adoption, partial adoption, and rejection, where genuine adoption is required for systemic benefits to materialise above a population threshold.
Model specification in an evolutionary game-theoretic framework; analytical description of strategy set and threshold condition. No empirical sample (theoretical model).
Rising density from rack- and pod-scale AI systems shapes these outcomes (deployable capacity, capex, performance) — we quantify how density changes these outcomes.
Modeling/simulation results reported in the paper quantifying the impact of rising rack/pod-scale density on deployable capacity, capex, and performance; specific numeric quantification not included in the abstract.
The negative quadratic term confirms a concave (inverted-U) relationship between AI and economic growth (diminishing marginal returns of AI).
Panel data for 19 G20 countries (2005–2023) estimated with a quadratic specification in GMM; reported negative and statistically significant coefficient on the AI-squared term.
Anhand von Fallstudien aus den G7-Ländern werden verschiedene Einsatzmöglichkeiten veranschaulicht und die wichtigsten Erfolgsfaktoren benannt – Netzanbindung, KI-Inputs, Kompetenzen und Finanzierung.
Evidence comes from G7 country case studies reported in the paper; method = qualitative case studies identifying key success factors (no number of case studies or sample size provided in excerpt).
The results vary across the 10 selected countries: the magnitude and significance of AI’s effects differ due to varying technological readiness and differing industrial structures.
Paper statement that results vary across the 10 selected countries and that nuances differ across countries due to varying industrial structures and technological readiness. Implied heterogeneity analysis across countries using the firm-level dataset and regression approaches; no country-level sample counts provided in the excerpt.
Differences in human intervention effectiveness across escalation types are partly explained by variation in workers' post-escalation intervention effort.
Observed correlations (and subgroup comparisons) in the randomized experiment showing that measures of post-escalation effort (e.g., message counts, share of chat rounds, proactivity) vary across escalation types and relate to outcome differences.
There is a fundamental reward-coverage tradeoff: concentrating probability mass on high-reward actions reduces variance but risks missing signal on actions the target policy may take.
Explicit characterization in abstract; claimed theoretical analysis/derivation of the tradeoff between variance reduction and coverage when designing logging policies.
These AIECI benefits were contingent on complementary conditions—particularly data quality, governance, managerial interpretation, and integration of intelligence outputs into operating decisions.
Cross-case pattern-matching across five analytical dimensions (intelligence source, AI mechanism, decision domain, economic implication, boundary condition) identifying recurring contingencies in the four firms' archival evidence.
Accounting for heterogeneity in AI literacy (agents' ability to identify and adapt to inaccurate AI outputs) can produce skill polarization in the long-run steady state.
Analytical/theoretical steady-state distribution analysis of agent skill dynamics with heterogeneous AI literacy parameters; paper reports conditions under which polarization emerges (theoretical, no empirical sample).
The dominant explanation for the gap locates it in model capability; instead, software-engineering capability emerges from a model-harness-environment system where a runtime substrate (the harness) mediates how an agent observes a project, acts on it, receives feedback, and establishes that a change is complete.
Conceptual argument and reframing presented in the paper (abstract). The paper formalizes this perspective rather than reporting a large-scale empirical test in the abstract.
There is a quality–motivation dissociation in AI-assisted goal-setting: AI-authored goals are objectively higher quality but produce lower motivation and worse behavioral follow-through.
Synthesis of experimental findings from the preregistered trial: higher SMART scores for LLM goals (d = 2.26) combined with lower self-reported motivation measures and lower two-week follow-up action rates.
The research challenges for this vision stem from a broader flexibility–robustness tension that requires moving beyond the on-the-fly paradigm to navigate effectively.
Analytical claim in paper identifying a design trade-off (flexibility vs. robustness) as the core challenge motivating the proposed shift; no empirical demonstration provided.
Current LLM agents are proficient at calling isolated APIs but struggle with the "last mile" of commercial software automation.
Authors' comparative characterization based on literature context and their benchmark motivation; stated in introduction rather than a quantified experiment in the excerpt.
Fine-tuning and reinforcement learning improve in-distribution performance, but generalization to unseen part families remains limited.
Experiments reported in the paper/abstract applying fine-tuning and reinforcement learning to models evaluated on BenchCAD; observed improvements on in-distribution data and limited generalization to unseen families.
Across 10+ frontier models, current systems often recover coarse outer geometry but fail to produce faithful parametric CAD programs.
Empirical evaluation reported in the paper/abstract across more than ten contemporary multimodal / large language models on the BenchCAD dataset; observed pattern that coarse outer geometry is often recovered while faithful parametric program synthesis fails.
AI exhibits a significant U-shaped spatial effect on Lae.
Spatial econometric analysis (spatial Durbin model) on panel data for 30 Chinese provincial regions (2012–2022); kernel density estimation used for distributional analysis.
AI has a significant inverted U-shaped impact on the low-altitude economy (Lae), with diminishing marginal returns after a certain turning point.
Panel data from 2012–2022 for 30 Chinese provincial regions; composite AI and Lae indices constructed via the entropy method; estimated using spatial Durbin models and non-linear specification to detect inverted U-shape.
Evidence suggests both top-down and bottom-up diffusion: worker use can occur without firm adoption, and vice versa.
Cross-tabulation of firm-level adoption indicators and reports of worker-level use in the BTOS AI supplement (Nov 2025–Jan 2026) indicating non-perfect overlap between firm-declared adoption and reported worker use; analytic approach descriptive (no sample size in excerpt).
The study reframes VTech adoption as legitimacy-seeking rather than efficiency-driven.
Thematic analysis using Rogers' diffusion of innovations and institutional theory, resulting in the institutionally mediated diffusion of innovations (IDOI) framework which emphasizes legitimacy concerns.
Practitioners stress that human judgement remains indispensable, positioning technology as an aid rather than a replacement.
Interview responses from valuers and firm leaders emphasizing the continued role of human judgement; thematic analysis framed by the IDOI model.
Public discussion of generative AI in accounting swings between the allure of full automation and job-displacement anxiety, yet the most immediate reality in organizations is human + AI work.
Paper's background/intro synthesizing recent research and practitioner commentary (2023–2025); conceptual observation rather than empirical test.
Integrating Generative AI into agile development processes has potential benefits and limitations for planning efficiency.
High-level conclusion based on the controlled experiment with GitLab Duo and qualitative participant feedback discussed in the paper.
Larger models do not consistently outperform smaller ones on tool-use tasks.
Empirical observations from the paper's evaluations across the five function-calling benchmarks.
Model routing can mitigate the cost of agentic tool use, but existing routers are designed for chat completion rather than tool use.
Argument/positioning in the paper and literature discussion (no specific empirical test reported for existing routers in this statement).
The finding that recurrence and neighborhood statistics are stronger predictors than complaint volume has direct implications for complaint routing given the demographic correlates of those features.
Interpretive implication drawn by the authors from the SHAP results; presented as a logical consequence rather than a separately tested empirical result in the excerpt.
Successful AI implementation in logistics requires not only technological capability but also organizational readiness and effective data governance.
Conclusion drawn from the structured qualitative review of 31 scholarly sources synthesizing reported success factors and preconditions for AI adoption.
There are factor-share consequences from agent adoption (i.e., implications for the shares of income accruing to factors such as labor and capital).
Model-based discussion and comparative-static analysis in the paper deriving implications for factor shares as agents/compute capital alter production technology. The excerpt indicates qualitative/theoretical analysis rather than empirical measurement.
The CAW result generalizes through CES aggregation and, when tasks are separated into substitutable versus complementary, yields a directional inversion of skill-biased technical change.
Theoretical extension of the core model using CES (constant elasticity of substitution) aggregation and task decomposition in the paper; the claim arises from model generalization and comparative-static reasoning. No empirical validation provided in the excerpt.
Agents are not labor; they are a production technology that converts compute capital K_c into effective units of cognitive labor L_A.
Theoretical argument and definitional framing in the paper: the authors recast agents as a technology that transforms compute capital into effective cognitive labor units within an analytical model (textual/theoretical exposition). No empirical sample or experimental data reported in the excerpt.
The trajectory of AI systems is shaped not only by model design, but by the dynamics of human-AI co-evolution.
Conclusion drawn from the minimal model, analytical regimes, and simulation experiments presented in the paper.