Evidence (8974 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
10085 claims
Filter claims →
Productivity
8974 claims
Filtered →
Governance
8062 claims
Filter claims →
Human-AI Collaboration
7749 claims
Filter claims →
Org Design
5057 claims
Filter claims →
Innovation
4896 claims
Filter claims →
Labor Markets
4088 claims
Filter claims →
Skills & Training
3372 claims
Filter claims →
Inequality
2377 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 | 882 | 244 | 117 | 1097 | 2424 |
| Governance & Regulation | 1010 | 469 | 229 | 135 | 1875 |
| Organizational Efficiency | 977 | 235 | 149 | 90 | 1462 |
| Technology Adoption Rate | 781 | 299 | 143 | 128 | 1362 |
| Research Productivity | 506 | 155 | 74 | 363 | 1110 |
| Output Quality | 555 | 219 | 71 | 70 | 915 |
| Decision Quality | 395 | 200 | 95 | 54 | 751 |
| Firm Productivity | 523 | 67 | 101 | 27 | 724 |
| AI Safety & Ethics | 262 | 309 | 75 | 36 | 688 |
| Market Structure | 195 | 201 | 135 | 30 | 566 |
| Task Allocation | 248 | 77 | 96 | 38 | 464 |
| Innovation Output | 300 | 34 | 55 | 20 | 411 |
| Skill Acquisition | 207 | 75 | 65 | 21 | 368 |
| Employment Level | 138 | 67 | 119 | 24 | 350 |
| Fiscal & Macroeconomic | 156 | 80 | 53 | 33 | 329 |
| Task Completion Time | 211 | 38 | 13 | 16 | 280 |
| Firm Revenue | 183 | 52 | 29 | 5 | 270 |
| Consumer Welfare | 131 | 77 | 48 | 13 | 269 |
| Inequality Measures | 50 | 141 | 54 | 9 | 254 |
| Worker Satisfaction | 104 | 85 | 25 | 13 | 227 |
| Error Rate | 87 | 112 | 11 | 5 | 215 |
| Automation Exposure | 69 | 69 | 37 | 20 | 198 |
| Wages & Compensation | 102 | 49 | 31 | 11 | 193 |
| Team Performance | 115 | 30 | 30 | 11 | 187 |
| Regulatory Compliance | 88 | 74 | 17 | 7 | 186 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 116 | 21 | 15 | 8 | 161 |
| Job Displacement | 12 | 92 | 26 | 1 | 131 |
| Hiring & Recruitment | 57 | 12 | 9 | 5 | 83 |
| Skill Obsolescence | 6 | 59 | 10 | 2 | 77 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 23 | 17 | 1 | 59 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Productivity
Remove filter
This work is conceptual/theoretical and reports no original empirical dataset; it explicitly calls for mixed-methods empirical validation (case studies, field experiments, longitudinal studies), measurement development, and multi-level data collection.
Explicit methodological statement in the paper describing its nature as a theoretical synthesis and listing empirical needs; no empirical sample provided.
Four autonomous agents were benchmarked on the same fresh CTF challenge set alongside human teams.
Benchmarking experiment described in the study: four autonomous AI agents evaluated on the identical fresh challenge set used in the live onsite CTF.
Data and methods: the study used an online experiment with 861 online-retail employees performing short-duration, virtual, task-focused collaborations; analyses focused on direct effects, moderation (emotion and partner type), mediation (service empathy), and moderated-mediation.
Methods description in the paper specifying design, sample size (n = 861), task context (temporary virtual teamwork), and analytic approach (hypothesis tests including moderation and mediation analyses).
Teamwork partner type (human vs AI) has no direct, significant effect on collaboration proficiency for temporary virtual tasks.
Online experiment with employees in the online-retail industry (n = 861). Hypothesis testing showed no significant main effect of partner type on the outcome variable 'collaboration proficiency' in the reported analyses.
The paper recommends an empirical research agenda including field experiments comparing teams with and without AI mediation, structural models of labor supply and wages under reduced language frictions, microdata analysis of adopters, and measurement studies for coordination costs and mediated-action reliability.
Explicit recommendations and research agenda stated in the paper; this is a descriptive claim about the paper's content rather than an empirical finding.
The paper's primary approach is conceptual/theoretical development and agenda-setting; it does not report large-scale empirical or experimental data.
Explicit methods statement in the paper: synthesis, illustrative examples, framework development; absence of reported empirical sample or experiments.
The study's empirical base consists of 40 semi-structured interviews with cross-industry project practitioners in the UK, analyzed using thematic qualitative methods.
Stated data and methods in the paper: sample size (40), interview method, cross-industry sampling, and thematic analysis.
Limitation: Implementation heterogeneity — the costs and feasibility of the recommended HR changes vary by context and may affect generalisability.
Explicit limitation acknowledged in the paper; drawn from theoretical reasoning about contextual heterogeneity and practitioner variability.
Limitation: The framework is conceptual and requires empirical validation across sectors, firm sizes and AI‑intensity levels.
Explicit limitation acknowledged by the authors; based on the paper's method (theoretical synthesis, no original data).
The paper generates empirically testable propositions (e.g., how leader practices affect AI adoption speed, task reallocation, productivity, error rates, employee well‑being and turnover) and suggests natural‑experiment settings for evaluation.
Stated methodological output of the conceptual synthesis; the paper lists candidate empirical tests and research opportunities but contains no original empirical tests.
Typical methods used are deep learning for property prediction and representation learning, protein-structure modelling tools, generative models for de novo design, NLP for knowledge extraction, and ADME/Tox in silico models integrated with traditional computational chemistry.
Methodological survey in the paper listing these approaches and examples of their application.
Commonly used data types in AI-driven drug discovery include biochemical/binding assay data, protein structural data, HTS results, ADME/Tox and PK datasets, omics/phenotypic readouts, and scientific literature/patents.
Cataloguing of data sources used across studies and company pipelines described in the paper.
AI became widely adopted in pharmaceutical discovery during the 2010s, driven by greater compute, larger datasets, and advances in deep learning.
Historical overview and trend analysis in the paper referencing increased compute availability, growth in public and proprietary datasets, and the rise of deep-learning publications and tools over the 2010s.
The available evidence consists mainly of promising empirical studies and case studies, but there are few long-run, generalized ROI or productivity estimates; results are heterogeneous across therapeutic areas.
Self-described limitation of the narrative review: heterogeneity of study designs and outcomes precluded pooled quantitative estimates and long-run ROI assessment.
AI applications span the full drug development pipeline, including target discovery, in silico screening and de novo design, preclinical safety models, clinical trial design and patient selection/monitoring, and post-marketing surveillance.
Comprehensive literature synthesis across preclinical, clinical, and post-marketing sources in the narrative review summarizing documented uses across these stages.
Current evidence is illustrative rather than systematic; there is a lack of long-run, quantitative measures of AI’s effect on late-stage clinical outcomes in the literature reviewed.
Explicit methodological statement in the paper: study is an expert/opinion synthesis and narrative review with no new causal econometric estimates or primary experimental data.
Suggested metrics for researchers and investors to monitor include R&D cycle time, cost per IND/NDA, proportion of projects using AI, success rates at development stages, market concentration measures, and investment flows into AI-enabled biotech vs incumbents.
Recommendations made in the Implications section as metrics to watch; no empirical tracking or baseline measures provided.
Limitations of the analysis include limited empirical validation of archetypes or impacts and potential selection bias toward prominent firms and technologies.
Explicit limitations stated in the Data & Methods section of the paper.
The paper is an editorial/conceptual synthesis rather than a primary empirical study: it uses qualitative analysis and illustrative examples, and reports no new quantitative estimates.
Explicit statement in the Data & Methods section of the paper describing document type, approach, evidence base, and limitations.
Ethical oversight and governance (addressing bias, consent, downstream risks) are critical constraints that must be addressed for AI to generate sustained benefits.
Normative synthesis referencing common ethical concerns; no empirical evaluation of oversight mechanisms in the paper.
Transparency and auditability for model behavior, provenance, and decisions are essential for trustworthy deployment and regulatory acceptance.
Policy and governance synthesis drawing on regulatory dynamics; no empirical study of regulatory outcomes included.
Rigorous model validation and reproducibility across datasets and settings are necessary constraints for successful AI deployment.
Normative claim in the editorial based on reproducibility concerns in ML and biomedical research; no reported validation trials within the paper.
Recommendation (research): Future research should link AI adoption to objective performance metrics (profitability, default rates, processing times) and use longitudinal or quasi-experimental designs to identify causal effects.
Authors' suggested research directions noted in the summary, motivated by limitations of cross-sectional, self-reported data.
The summary omits important reporting details: p-values, standard errors, model control variables, and exact variable operationalizations are not provided.
Explicit reporting gap noted in the paper summary (absence of p-values, SEs, controls, and operationalization details).
Because the data are cross-sectional and self-reported, the design limits causal inference about AI adoption causing the observed outcomes.
Study design (cross-sectional survey, self-reported measures) and explicit limitation noted in the paper summary.
Key measures are self-reported Likert scales for AI adoption/usage and the dependent outcomes (financial decision-making efficiency, operational efficiency, financial resilience, and AI-based analytics effectiveness).
Measurement description in Methods: independent and dependent variables reported as self-reported Likert measures collected in the cross-sectional survey.
The study is a cross-sectional quantitative survey of 312 professionals in banks, fintechs, and financial service firms.
Study design and sample description reported in Data & Methods; sample size explicitly given as N = 312 and composition described as professionals across financial institutions, fintech organizations, and financial service companies.
The SKILL.md used in the with-skill condition encodes workflow logic, API patterns, and business rules as portable domain guidance for agents.
Paper description of the with-skill intervention specifying the content and intended role of SKILL.md.
We evaluated open-weight models under two conditions: baseline (generic agent with tool access but no domain guidance) and with-skill (agent augmented with a portable SKILL.md document encoding workflow logic, API patterns, and business rules).
Experimental design in paper describing the two agent conditions; SKILL.md described as the injected domain guidance artifact.
Each scenario is grounded in live mock API servers with seeded production-representative data, MCP tool interfaces, and deterministic evaluation rubrics combining response content checks, tool-call verification, and database state assertions.
Methods/benchmark design described in paper specifying environment: live mock APIs, seeded data, MCP tool interfaces, and deterministic evaluation combining content checks, tool-call verification, and DB assertions.
SKILLS comprises 37 telecom operations scenarios spanning 8 TM Forum Open API domains (TMF620, TMF621, TMF622, TMF628, TMF629, TMF637, TMF639, TMF724).
Framework specification in the paper; explicit statement of scenario count (37) and list of 8 TMF Open API domains.
We introduce SKILLS (Structured Knowledge Injection for LLM-driven Service Lifecycle operations), a benchmark framework for telecom operations.
Paper describes the design and release of the SKILLS benchmark framework as the contribution; methods section outlines framework components and usage.
The paper identifies three core mechanisms underlying calibrated trust and complementarity: (1) calibrated trust balancing reliance and oversight, (2) complementarity–trust interaction for optimal performance, and (3) dynamic feedback loops producing reinforcing learning cycles.
Explicit identification of mechanisms claimed in the paper's synthesis; this is a descriptive claim about the paper's content rather than an empirical finding—no sample or empirical test reported in the abstract.
AI-adopting firms do not increase capital expenditures following adoption.
Firm-level capex analysis showing no significant change in capital expenditures for adopters versus nonadopters post-adoption in the paper's empirical framework.
SWE-Skills-Bench is available at https://github.com/GeniusHTX/SWE-Skills-Bench.
Repository URL provided in the paper for the benchmark's code/data.
SWE-Skills-Bench provides a testbed for evaluating the design, selection, and deployment of skills in software engineering agents.
Benchmark design pairs skills, repositories, and deterministic verification tests; intended use stated by authors as a testbed for evaluation of skills.
39 of 49 skills yield zero pass-rate improvement.
Empirical evaluation over 49 skills and ~565 task instances reporting that 39 skills produced no improvement in test pass rate when injected.
The authors introduce a deterministic verification framework that maps each task's acceptance criteria to execution-based tests, enabling controlled paired evaluation with and without the skill.
Method: creation of a deterministic verification framework that converts acceptance criteria into executable tests; used to perform paired evaluations (with skill vs. without skill).
SWE-Skills-Bench pairs 49 public SWE skills with authentic GitHub repositories pinned at fixed commits and requirement documents with explicit acceptance criteria, yielding approximately 565 task instances across six SWE subdomains.
Benchmark construction: 49 public skills, repositories pinned to fixed commits, requirement documents with acceptance criteria, producing ~565 task instances spanning six SWE subdomains (as reported by the paper).
The article introduces a novel Bayesian Item Response Theory framework that quantifies human–AI synergy by separately estimating individual ability, collaborative ability, and AI model capability while controlling for task difficulty.
Methodological contribution described in the paper: development and application of a Bayesian Item Response Theory model that includes separate parameters for individual ability, collaborative ability, AI model capability, and task difficulty (method section of the paper).
The Planner is trained via Supervised Fine-Tuning (SFT) to internalize diagnostic capabilities and then aligned with business outcomes (conversion rate) via Reinforcement Learning (RL).
Method description in the paper specifying SFT initialization followed by RL alignment targeting conversion rate (UCVR) as reward signal.
EASP's Offline Data Synthesis stage: a Teacher Agent synthesizes diverse, execution-validated plans by diagnosing the probed environment.
Method description in the paper detailing the Teacher Agent's role in synthesizing execution-validated plans during offline data synthesis.
The Probe-then-Plan mechanism uses a lightweight Retrieval Probe to expose the retrieval snapshot, enabling the Planner to diagnose execution gaps and generate grounded search plans.
Methodological description in the paper: design and implementation of Retrieval Probe and Planner; validated through synthesized data and downstream evaluations (offline and online).
A quantitative methodology was employed, utilizing a structured questionnaire administered to 400 small business owners.
Explicit methodological statement in the paper: structured questionnaire survey with sample size N=400 small business owners.
Foi realizada etnografia organizacional orientada ao SCF, com roteiro e triangulação de evidências.
Método qualitativo divulgado no resumo: etnografia organizacional com roteiro e triangulação; o resumo não fornece número de organizações, duração ou amostragem.
Foi construído e validado um instrumento psicométrico (escala SCF-30) e calculado um índice 0–100, com modelagem por Equações Estruturais (SEM) e testes de confiabilidade/validade.
Descrição metodológica explícita no resumo: construção e validação da escala SCF-30, uso de SEM e testes de confiabilidade e validade. O resumo não detalha estatísticas, amostra ou resultados numéricos.
O SCF é operacionalizado por três vetores centrais: Percepção de Complexidade (PC), Aversão ao Risco Institucional (AR) e Inércia Cultural (IC).
Estrutura conceitual e operacional apresentada no artigo; especificação explícita dos três vetores como componentes do construto SCF.
This research conducts a critical analysis of the ethical implications of artificial intelligence in terms of job displacement during the fifth industrial revolution.
Author-declared methodology: a literature-based critical analysis drawing on novel studies and the existing body of literature; no further methodological details (e.g., inclusion criteria, databases searched) provided in the excerpt.
This study uses panel data on agricultural firms listed on the Shanghai and Shenzhen A-share markets from 2007 to 2023 and applies a multidimensional fixed-effects model to estimate the impact of AI on firms’ total factor productivity (TFP).
Methodological statement in the paper: dataset = panel of listed agricultural firms (Shanghai and Shenzhen A-share markets), time period 2007–2023; empirical approach = multidimensional fixed-effects model.
The paper explores risk frameworks, ethical constraints, and policy imperatives related to AI.
Descriptive claim about the paper's analytic content (thematic/policy analysis); no empirical details or measurement approach are given in the abstract.