Evidence (650 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
21267 claims
Filter claims →
Productivity
17978 claims
Filter claims →
Governance
17038 claims
Filter claims →
Human-AI Collaboration
16914 claims
Filter claims →
Org Design
11104 claims
Filter claims →
Innovation
11087 claims
Filter claims →
Labor Markets
6711 claims
Filter claims →
Skills & Training
5616 claims
Filter claims →
Inequality
4343 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 | 1880 | 496 | 296 | 1854 | 4721 |
| Organizational Efficiency | 2906 | 665 | 438 | 180 | 4210 |
| Governance & Regulation | 2162 | 929 | 480 | 247 | 3866 |
| Technology Adoption Rate | 1533 | 545 | 278 | 210 | 2593 |
| Decision Quality | 1391 | 534 | 321 | 173 | 2429 |
| Output Quality | 1298 | 472 | 231 | 145 | 2153 |
| AI Safety & Ethics | 682 | 821 | 230 | 90 | 1837 |
| Research Productivity | 855 | 253 | 121 | 425 | 1675 |
| Firm Productivity | 1105 | 171 | 175 | 73 | 1531 |
| Task Allocation | 735 | 229 | 361 | 99 | 1433 |
| Market Structure | 457 | 461 | 251 | 47 | 1222 |
| Innovation Output | 673 | 94 | 108 | 36 | 913 |
| Task Completion Time | 499 | 118 | 43 | 38 | 702 |
| Firm Revenue | 458 | 130 | 61 | 26 | 677 |
| Skill Acquisition | 381 | 122 | 113 | 34 | 650 |
| Consumer Welfare | 316 | 176 | 115 | 39 | 648 |
| Employment Level | 223 | 143 | 177 | 53 | 600 |
| Error Rate | 246 | 282 | 44 | 19 | 594 |
| Fiscal & Macroeconomic | 283 | 142 | 78 | 52 | 562 |
| Inequality Measures | 103 | 329 | 106 | 13 | 552 |
| Worker Satisfaction | 225 | 185 | 63 | 30 | 503 |
| Automation Exposure | 158 | 155 | 72 | 37 | 426 |
| Regulatory Compliance | 186 | 126 | 35 | 14 | 362 |
| Team Performance | 193 | 56 | 51 | 24 | 326 |
| Developer Productivity | 224 | 58 | 27 | 13 | 323 |
| Wages & Compensation | 148 | 108 | 50 | 17 | 323 |
| Training Effectiveness | 218 | 44 | 21 | 27 | 313 |
| Job Displacement | 23 | 159 | 53 | 5 | 240 |
| Hiring & Recruitment | 109 | 61 | 32 | 11 | 215 |
| Skill Obsolescence | 16 | 107 | 26 | 6 | 155 |
| Creative Output | 71 | 44 | 28 | 6 | 150 |
| Social Protection | 58 | 31 | 12 | 3 | 104 |
| Labor Share of Income | 29 | 43 | 25 | 2 | 99 |
| Worker Turnover | 45 | 29 | 6 | 4 | 84 |
| Industry | — | — | — | 1 | 1 |
Training accounted for 20.6% of broad response-frame mentions and appeared at comparable rates across party families.
Response-frame coding combined with comparisons across party families.
AI-driven educational tools may accelerate skill formation but may also produce uneven returns when access to them is unequal.
The article derives this implication from its discussion of dual effects on learning and socioeconomic differences in access; no direct quantified estimate is reported.
Early exposure to digital technologies shapes later technical, cognitive, and socioemotional skill acquisition and may affect the productivity of the future workforce.
The claim is presented as an implication for AI economics based on developmental research and the article's synthesis of evidence on childhood learning and technology exposure.
The paper reports that some studies found no impact of AI use on post-AI skill competency, while a five-day multitasking study found that AI-assistant use improved performance on tasks completed without AI support.
Literature review citing Karny et al. (2024), Nelson et al. (2025), and Talypova et al. (2024); the paper provides no sample sizes or quantitative effect estimate for these studies.
Freelancers use generative AI to structure their learning and explore unfamiliar skills, but generally stop short of trusting it as their primary teacher because of inconsistency, weak contextual understanding, and the need to verify its output.
The claim is based on empirical studies of freelance knowledge workers' upskilling practices [9]. No sample size or quantitative effect estimate is reported in the paper.
The impact of digital tools depends more on the epistemic function they perform than on their technical sophistication or novelty.
Thematic comparison of tool classes according to whether they support functions such as sense-making, practice, or assessment.
Aggregate effect estimates for digital technology in school mathematics are dominated by short, small-scale, researcher-implemented studies, and these conditions also show the largest reported effects.
Thematic synthesis and appraisal of studies by design, intervention duration, implementation context, and sample representativeness.
The audit profession will require greater expertise in system assurance, code review, crypto-forensics, and governance evaluation, rather than focusing primarily on transaction verification.
Conceptual assessment of how blockchain changes audit tasks and the skills required to perform the resulting assurance work.
The urban skill upgrading described for Italian cities occurred primarily through low-skill workers leaving and high-skill workers moving in, rather than through existing workers acquiring new skills.
The paper's interpretation of the Capello and Lenzi evidence on urban labor-market adjustment.
Cross-domain transfer is strongest between structurally similar interaction domains, while structurally isolated games show no transfer.
Full cross-environment transfer matrix involving policies trained separately in each of six domains.
Students given unrestricted access to a generative AI tool perform better when using the tool but worse without it than peers who never used such a tool.
The paper cites early empirical evidence comparing students with unrestricted generative-tool access against peers without prior tool use; the cited study's sample size is not reported in the supplied text.
Kazakhstan's digital readiness is mixed: internet penetration and mobile connectivity are relatively high by regional standards, but advanced AI skills, AI research capacity, and rural and regional connectivity remain limited.
Descriptive assessment of infrastructure, skills, research capacity, and regional disparities.
Transferred experience across tasks affected model performance differently: it increased DeepSeek-V4-Pro's avg@3 by 0.093 but decreased Gemini-3.1-Pro's avg@3 by 0.017.
Controlled inter-task comparisons with and without transferred experience.
Accumulated experience within a task usually improves the next solution, but can also propagate misleading conclusions or anchor agents to local optima.
Controlled comparisons of subsequent decisions with and without accumulated experience in the intra-task setting.
Automating junior tasks can generate short-run productivity gains while weakening the longer-run pipeline through which workers develop accountable expertise.
The paper's expertise-formation argument that junior tasks simultaneously produce current output and train future judgment.
The paper characterizes the emerging evidence on generative AI and independent performance as mixed rather than uniformly beneficial or harmful.
Narrative review citing studies on mathematics tutoring, misinformation, dialogue interventions, persistence, programming, and critical thinking.
As firms value algorithm-friendly credentials and platform-based signals, incentives may shift toward short-term credentialing and digital upskilling rather than broader human-capital investments.
The paper identifies a possible change in skill-formation incentives, but provides no longitudinal education or training data.
Intensive AI use produces strong performance on the main task even with low task engagement, but follow-up performance is substantially higher only when intensive AI use is combined with sustained engagement.
Analysis dividing treated participants into four groups based on AI-assistance intensity and task engagement; the analysis uses behavioral measures from the task and chat logs.
The review finds that AI reorganizes human-resource competencies in banking toward digital, analytical, operational, strategic, and ethical capabilities.
Systematic literature review of 68 peer-reviewed articles drawn from banking, HRM, and sustainability literature using the PRISMA 2020 protocol.
Few-shot evolution tends to overfit its source task group and can harm performance on other groups, whereas reflection-based evolution transfers more robustly across groups.
Cross-task-group transfer analysis comparing few-shot and reflect supervision.
Demand is expected to shift toward AI and data skills and digital literacy, while complementarities between human capabilities and generative AI may influence wages and firm staffing strategies.
Qualitative literature synthesis on labor and skill composition in AI-enabled digital entrepreneurship; no wage or employment estimates are reported.
The aggregate labor-market evidence on AI-related adaptation is mixed, including evidence of adaptive capacity, substantial heterogeneity in retrainability, and null effects in some settings.
The paper summarizes labor-market findings from Manning and Aguirre (2026), Hyman et al. (2025), and other cited studies, but does not report their sample sizes or effect estimates in the supplied text.
The distinction between assisted performance and independent capability is empirically established when studies measure unaided performance after AI exposure.
The paper characterizes this claim as supported by studies that directly test unaided performance after AI exposure, citing Budzyń et al. (2025) and Macnamara et al. (2024).
AI assistance can improve performance during an assisted task without improving subsequent unaided performance.
The paper cites experimental evidence from Wiles et al. (2024), reporting that participants performed significantly better while AI assistance was available, but that the advantage did not transfer to later unassisted performance.
Within the programming-learning experiments, students who used the model to generate solutions covered more material while understanding less, whereas students who used it to request explanations deepened their understanding.
Usage-pattern analysis within two pre-registered experiments by Lehmann et al. (2025).
The literature reports rising skill demands alongside widespread skill mismatches.
Qualitative synthesis of studies examining worker skills, education, task content, and labor-market adjustment.
Accountants increasingly focus on interpreting, validating and governing AI outputs rather than generating predictions themselves, signalling a shift in professional roles and required expertise.
Theoretical argument and synthesis of literature on changing professional tasks and required skills in an AI-enabled accounting environment; no empirical sample size specified.
Participants anticipate workforce recomposition rather than immediate displacement, emphasizing upskilling in AI literacy, analytics/forecasting, cybersecurity awareness, and AI governance.
Interview responses from 45 practitioners expressing expectations about workforce changes and training priorities.
We identify six distinct AI interaction patterns, three of which involve cognitive engagement and preserve learning outcomes even when participants receive AI assistance.
Qualitative and quantitative analysis of participant–AI interaction logs and behavior from the randomized experiments, producing a taxonomy of six interaction patterns and noting which patterns associated with preserved learning.
The AI-related burden dynamics unfold across the interrelated dimensions of learning, compliance, and psychological costs (aligned with Moynihan, Herd and Harvey's 2015 framework).
Observational mapping from the 6-month organizational ethnography linking observed AI effects to the three dimensions in the referenced conceptual framework.
The dual revolution is creating new skill demands for accounting professionals.
Statement in the review asserting that technological and reporting changes generate new skill requirements (qualitative argument; no empirical sample or quantified skill gap reported).
Modeling mastery as learning a reliability map via Gaussian process regression yields a learning-rate bound driven by information gain, clarifying when discovering 'where the model works' is slow.
Theoretical learning analysis: the paper models user mastery as GP regression and derives a bound on learning rate that depends on information gain (mathematical result).
Digital transformation has affected skill demands in Hungary (impacts on skills and the need for reskilling/upskilling).
Synthesis of literature and policy reports focusing on skills and education responses; no specific sample size reported in the summary.
AI integration shifts the entrepreneur’s role from direct execution toward the orchestration of distributed Human–AI systems under conditions of uncertainty.
Framed as a theoretical/analytic conclusion in the thesis excerpt; the excerpt contains no empirical details (methods, sample size) supporting the statement.
China’s adaptive capacity reflects not only post-2019 policy responses, but a substantial stock of human capital and systems-level knowledge accumulated before comprehensive controls.
Comparative historical analysis and industry data indicating pre-existing human capital and systems knowledge in China that supported adaptation post-2019 (qualitative, case-based).
Exposure to contrasting expert narratives shifts stated beliefs about the likely impact of AI on the labor market; there is no evidence of a significant impact on behavioral outcomes in the current sample (small effects possible with more data).
Randomized experiment embedding exposure to contrasting expert narratives within the worker survey; reported results are from preliminary pilot data and may change with further data collection.
The European Union’s AI Act has extraterritorial application that reshapes Indian domestic employment by embedding regulatory compliance into the definition of relevant AI skills.
Paper's theoretical/policy analysis and argumentation linking extraterritorial scope of the EU AI Act to Indian IT sector skill definitions; no empirical sample size or quantitative estimates reported in the provided text.
The apparent democratisation of technical capability therefore depends on new dependencies and new literacies.
Argumentative inference from the paper's analysis of infrastructural and protocol-mediated shifts; conceptual claim about conditions for democratisation rather than supported by empirical measurement.
Generation Z simultaneously embraces AI while seeking careers in less automatable fields that promise stability, dignity, and work-life balance.
Summarised as a generational response in the paper; the provided excerpt does not include the study's sampling or methodology for this claim.
Employee productivity and performance tools entail both job displacement and job creation and therefore require workforce reskilling or upskilling for talent attraction, retention, progression, and promotion across structural labor market transformation.
Policy implication summarized in the systematic review of 2024–2025 literature emphasizing the need for reskilling/upskilling.
Workers pursue platform labor as a path to prestige and mobility but sustain themselves through resourceful, situated learning (renting cyber-cafe computers, copying gig templates, following tutorials in unfamiliar languages, and relying on peer networks).
Ethnographic observations and reported worker practices collected during the eight-month fieldwork (specific participant counts not provided in abstract).
The SCAN framework has implications for the future of work, such as upskilling and deskilling.
Argumentative discussion in the paper about how using SCAN could influence worker skills (upskilling/deskilling); presented as implications rather than results from empirical study.
Low-skill roles may experience mixed outcomes: some will benefit from technology-enabled upskilling while others will face downward competitive pressure.
Synthesis from sectoral case studies and task analysis indicating heterogeneous effects for low-skill occupations.
Theoretical research suggests that when corporates adopt AI, demand for high-skilled labor will increase while some low-skilled positions will be replaced.
Paper cites theoretical literature and/or presents theoretical argument about labor demand shifts with AI adoption (framing/background of the study).
Geopolitical shifts have reshaped global trade dynamics, resulting in complex consequences for skill development.
Stated in the book's summary as a high-level claim; no specific studies, methods, or sample sizes provided in the excerpt.
There is a curvilinear (inverted-U) relationship between job complexity and AI self-efficacy: employees in both low- and high-complexity roles exhibit a low level of AI self-efficacy compared to those in moderately complex roles.
Reported empirical finding from two survey studies analyzed using structural equation modeling (as stated in the abstract). Specific sample sizes and measurement details are not provided in the snippet.
The HAT model formally encodes an economic asymmetry between human skill acquisition (costly and slow) and AI capability scaling (different cost/risk profile), and this asymmetry is central to the model's predictions about substitution.
Model specification and stated assumptions in the paper (formal asymmetry assumption); conceptual/theoretical argument rather than empirical measurement.
If AI complements effort, then improvements in AI quality make high-skill students learn faster but low-skill students learn slower.
Comparative-static/analytical result from the theoretical model under the assumption that AI complements effort. No empirical data; conclusion derived from model behavior across different initial skill levels.
The study reveals an 'AI Competency Paradox'—AI raises technical skills while increasing demand for meta-competencies that established frameworks fail to assess.
Synthesis of empirical findings reported in the paper linking measured increases in technical skills with unmet assessment needs for meta-competencies.
Some skills generalize broadly across tasks and models, whereas others become specialized to role-specific workflows and lose effectiveness under transfer.
Analyses reported in the paper showing heterogeneous transfer behavior across the 22 procedural skills in the AFTER benchmark, with some skills showing broad cross-task and cross-model generalization and others showing role-specific specialization and reduced transfer performance.