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 |
In the production stage, workers lose decision-making power.
Theoretical analysis of production relations using Marxist reproduction framework; qualitative claim without reported empirical data.
The canonical manifestation of this failure pattern is called 'Phantom Legislation' (internally consistent symbolic outputs disconnected from real business semantics).
Terminology and descriptive example provided by the authors based on their analysis of observed failure cases in the Bang-v3 project.
Incumbent workers in more robot-exposed industries are unlikely to transition outside manufacturing over 2014-2021.
Longitudinal worker-level analysis of the 2014 manufacturing cohort through 2021 showing low rates of transition out of manufacturing for workers in higher-exposure industries (administrative employer-employee data).
Incumbent manufacturing workers in more robot-exposed industries experience a reduction in cumulative workdays at their original plants between 2014 and 2021.
Worker-level (intensive-margin) panel analysis tracking a 2014 manufacturing-worker cohort through 2021 using administrative employer-employee data; comparison of cumulative workdays at original plants across workers in industries with different robot exposure.
Income inequality has the opposite effect (it works against poverty reduction).
Negative and statistically significant coefficient on an income inequality measure in the CS-ARDL estimates for BRICS (2008–2023).
In multi-brand GEO competition there is a social dilemma: when all brands adopt the same optimization strategy, individual payoff falls from +0.802 to +0.007 in our payoff proxy, and non-participating brands receive zero recommendations in our tests.
Experimental multi-brand GEO competition scenario reported in the paper with reported payoff proxy values (+0.802 to +0.007) and an observation that non-participating brands received zero recommendations under the tested conditions.
Authority-style marketing language, including fabricated clinical-evidence claims, breaks the Conditional Monopoly at a Bias Surplus Value equal to +0.17 rating points, with each model responding differently.
Experimental manipulation of marketing language (authority-style / fabricated clinical claims) and measurement of the Bias Surplus Value (reported threshold +0.17 rating points); observed heterogeneous responses across the three tested LLMs.
The dominance of well-known brands disappears with less than a +0.1-star rating advantage for a competitor.
Experimental variation in competitor rating reported in the paper showing that providing a competitor a rating advantage smaller than +0.1 stars is sufficient to remove the incumbent's dominance.
GPU-accelerated deep learning exacerbates this problem, as nondeterministic floating-point reductions can produce drift in long backtests, challenging regulatory reproducibility and auditability expectations.
Argument in paper linking known GPU-nondeterminism (floating-point reduction nondeterminism) to practical issues in long financial backtests; no empirical backtest dataset size provided in the excerpt.
For thirty years, quantitative finance has paid a costly two-language tax: models researched in Python are rewritten in C++ for production, often introducing numerical discrepancies.
Statement in paper's introduction/abstract describing historical practice; no quantitative sample size or systematic study reported in the excerpt.
Automation reduces employment-based tax revenue and increases public financial pressure.
Explicit finding reported in paper; derived from the scoping review of existing literature (method: qualitative scoping review following Arksey & O'Malley). No quantitative sample or meta-analysis size reported in the abstract.
Automation often displaces workers without adequate retraining, leading to unemployment and reduced income tax contributions, which worsens income inequality.
Statement in paper's purpose/intro; synthesized from the literature via a qualitative scoping review (framework of Arksey & O'Malley). No primary empirical sample size reported in the abstract.
Test file counts substantially overestimate verification strength.
Conclusion drawn from the high prevalence (80.2%) of test patches with weak/no oracle signals compared to mere presence-of-test-file counts.
Raw merge rates are lower for strong-oracle PRs.
Unadjusted (raw) comparisons of merge rates between PRs classified by oracle strength in the study dataset.
Applied at scale, 80.2% of test patches contain weak or no explicit oracle signals.
Automated/syntactic classification of oracle-signal categories applied to the full dataset of test-file patches (as described in methods).
Test files lacking explicit assertions execute code without verifying behavior, so quality gates based on test-file presence overestimate verification strength.
Conceptual/analytic argument supported by the paper's framing and subsequent empirical analysis of oracle signals.
The paper highlights where assumptions, such as rationality and heterogeneity, may fail in agentic markets.
Analytical critique and theoretical examples/arguments in the paper pointing out potential breakdowns of classical assumptions when agents act on users' behalf; no empirical tests provided.
Optimization variance admits a necessary lower bound on its decay rate, implying fundamental constraints on how quickly uncertainty dissipates regardless of the predictor used.
Mathematical proof/theoretical result presented in the paper (formal lower-bound derivation; no empirical sample size in excerpt).
The high cost of fine-tuning LLMs poses a significant economic barrier.
Stated in the paper's motivation/intro as a high-level observational claim (no sample size or numeric cost figures provided in the text excerpt).
Disclosing AI involvement in visual content creation is associated with a weaker funding penalty than disclosing AI involvement in textual content creation.
Subgroup/moderation analysis within the same dataset (41,073 Kickstarter projects) comparing projects that disclosed AI-use in visual modalities versus textual modalities, using LLM-assisted classification to determine modality and entropy balancing for covariate adjustment.
AI-use disclosure is associated with a significant decline in funding performance for Kickstarter projects.
Observational analysis of 41,073 Kickstarter projects using LLM-assisted text classification to identify AI-use disclosure and entropy balancing to adjust for covariate differences; statistical tests reported as significant in the paper.
The AI-investment paradox persists because firms govern AI as a broad technology program rather than as a set of discrete, investable decision opportunities embedded within workflows.
Argument/theoretical claim developed by the authors as the central explanatory hypothesis of the paper; presented conceptually rather than tested empirically within this work.
Despite enterprises continuing to invest heavily in AI, many initiatives fail to scale or generate sustained business value (the 'AI-investment paradox').
Background claim stated in the paper's introduction/abstract and presented as motivating fact; supported implicitly by citations to prior literature and industry reports (no original empirical sample or quantitative analysis reported in this paper).
Urbanization reduces environmental carrying capacity (LCF), with reductions more pronounced under higher ecological pressure.
Panel ARDL estimates for the G-7 (1990–2019) finding a statistically significant negative relationship between urbanization (World Bank WDI data) and LCF; authors emphasize stronger negative effects under greater ecological pressure.
Globalization reduces environmental carrying capacity (LCF), particularly at higher levels of ecological pressure.
Panel ARDL estimates on the G-7 panel (1990–2019) report a statistically significant negative effect of the KOF globalization index on LCF, with stronger negative effects at higher ecological pressure (authors note the impact is pronounced 'particularly at higher levels of ecological pressure').
Developing a process twin is costly because it requires accurately modelling the entire production process (process steps, equipment and product-specific settings, process variations) and binding the model to live operational data.
Authoritative/technical description of development requirements and costs in paper (methodological claim), not quantified in abstract.
Emerging evidence suggests that general disclosures of AI involvement may unintentionally undermine consumer trust and reduce purchase intentions.
Paper cites 'emerging evidence' as motivating prior work; no specific studies, sample sizes, or effect estimates given in the provided text.
These findings highlight the risk of bias propagation in AI-assisted writing, calling for fairness-aware design in educational AI tools.
Authors' conclusion and recommendation based on the experimental results described above (N = 123 study showing bias transfer).
The project of commons-governed AI faces tensions and constraints including openwashing, the compute bottleneck, free-riding, and a tension between scale and sustainability.
Analytical discussion in the paper identifying and describing challenges and constraints observed in the literature and practice.
Agent-caused losses are still not clearly assigned, priced, or transferred; providers often disclaim consequential damages, users are left with uncompensated losses, and default human review limits the efficiency gains of automation.
Statement in paper introduction describing the current state of practice (no empirical sample size reported).
In a two-type heterogeneous-agent economy, high-cognitive-capital agents adopt AI more intensively and may eventually erode their unaided cognitive capital below that of initially lower-skilled agents.
Heterogeneous-agent extension of the analytical model; stated as a derived proposition. No empirical validation.
The decentralised equilibrium over-adopts substitutive AI relative to the social optimum because of systemic risk, cognitive public goods, and arms-race externalities.
Equilibrium analysis and welfare comparison in the formal model (decentralised vs socially optimal allocation). No empirical sample.
Expected crisis losses are convex in aggregate leverage.
Analytical result/proposition derived within the model showing convex relationship between expected losses and aggregate leverage. No empirical sample.
Temporary accommodation has become a major fiscal and administrative pressure for English local authorities, particularly in London, where demand and costs have risen sharply.
Statement in paper introduction/background; contextual claim based on administrative observations and cited motivation for building DOMUS (no specific sample size or numerical data reported in the provided text).
New accusations function as social gatekeeping of perceived authenticity without actually screening for AI.
Synthesis of findings: large-scale vocabulary trends, speech-act coding, and matched-control results indicate accusations are used for social signaling/gatekeeping rather than accurate AI detection.
Many existing approaches (interaction metrics, behavioral coding schemes, activity traces) often struggle to capture higher-order interaction dynamics, including how collaborative processes reorganize, stabilize, regulate, and evolve through time.
Theoretical critique and literature-level argument presented in the paper (conceptual analysis); no empirical sample or quantitative evaluation reported.
The increase in ICD risk at higher levels of AI investment is weaker among firms with above-normal external audit attention.
Moderator (heterogeneity) tests reported in paper showing the ICD-risk increase with AI investment is attenuated for firms receiving above-normal external audit attention.
The increase in ICD risk at higher levels of AI investment is weaker among firms with CIO presence.
Moderator (heterogeneity) tests reported in paper showing attenuated ICD-risk increase for firms that have a Chief Information Officer (CIO).
The increase in ICD risk at higher levels of AI investment is weaker among IT-industry firms.
Moderator (heterogeneity) tests reported in paper showing smaller upward slope of ICD risk with AI investment for firms in the IT industry.
At lower levels of AI exposure, AI investment is associated with lower ICD risk.
Substantive component of the reported U-shaped relationship estimated in regressions on the 41,725 firm-year sample.
High price strongly decreases the probability an assistant recommends a hotel: a high price lowers selection by 30.0 percentage points.
Randomized conjoint experiment estimating AMCEs; reported point estimate of −30.0 percentage points for high price.
Directly evaluating agents on physical high-precision instruments is impractical due to high cost, safety risks, limited accessibility, and difficulty in ensuring reproducible evaluation.
Argument presented by authors as motivation for creating a simulated testbed; no empirical cost/safety/accessibility metrics provided in the excerpt.
Comparing a six-group score spread against a two-run noise difference overstates disparity by approximately 2.4X through statistic arity alone.
Methodological comparison reported in pilot analysis showing that using a six-group spread versus a two-run noise baseline produces ~2.4X larger apparent disparity attributable to arity; based on pilot dataset and statistical comparisons.
Claude Haiku 4.5 exhibits an idle-drift failure mode, repeatedly choosing inaction despite producing coherent assessments and plans.
Qualitative observation from agent trajectories showing that Claude Haiku 4.5 repeatedly selects no-action decisions over time while still generating coherent internal assessments and plans.
Secure attachment further moderated the indirect effect of organizational AI adoption on employees' turnover intentions via identity threat (i.e., it attenuated the mediated effect).
Moderated mediation (conditional indirect effect) analysis reported on three-wave survey data of 312 employees; secure attachment reported to weaken the indirect AI adoption → identity threat → turnover intentions pathway.
Secure attachment negatively moderated the relationship between organizational AI adoption and identity threat (i.e., higher secure attachment reduced the AI adoption → identity threat effect).
Moderation analysis (interaction effect) reported in the three-wave survey data (N=312); secure attachment reported to negatively moderate the AI adoption to identity threat path.
Signal-planting and directed-versus-random experiments show that novelty alone is insufficient: random orthogonal jumps expand coverage but do not improve yield without predictive alignment.
Empirical claim supported by experimental comparisons (signal-planting experiments and directed vs. random proposal experiments); specific sample sizes and quantitative results not provided in the abstract.
Scientific discovery saturates when new hypotheses cease to provide independent information, even if the nominal hypothesis space remains large.
Conceptual/theoretical claim stated in the paper's framing; argued via the notion of informational independence of hypotheses (no specific sample size reported).
We should not train AIs to share those specific value systems (i.e., we should not align AI to aggregated or particular human value sets that may be oppressive or unhealthy).
Normative recommendation offered by the authors as part of their argumentation; presented without empirical quantification in the abstract.
Aligning AI to aggregated human preferences is the wrong target.
Normative/argumentative claim stated in the paper; no empirical sample, presented as the authors' thesis.