Evidence (4857 claims)
Adoption
5586 claims
Productivity
4857 claims
Governance
4381 claims
Human-AI Collaboration
3417 claims
Labor Markets
2685 claims
Innovation
2581 claims
Org Design
2499 claims
Skills & Training
2031 claims
Inequality
1382 claims
Evidence Matrix
Claim counts by outcome category and direction of finding.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 417 | 113 | 67 | 480 | 1091 |
| Governance & Regulation | 419 | 202 | 124 | 64 | 823 |
| Research Productivity | 261 | 100 | 34 | 303 | 703 |
| Organizational Efficiency | 406 | 96 | 71 | 40 | 616 |
| Technology Adoption Rate | 323 | 128 | 74 | 38 | 568 |
| Firm Productivity | 307 | 38 | 70 | 12 | 432 |
| Output Quality | 260 | 71 | 27 | 29 | 387 |
| AI Safety & Ethics | 118 | 179 | 45 | 24 | 368 |
| Market Structure | 107 | 128 | 85 | 14 | 339 |
| Decision Quality | 177 | 75 | 37 | 19 | 312 |
| Fiscal & Macroeconomic | 89 | 58 | 33 | 22 | 209 |
| Employment Level | 74 | 34 | 78 | 9 | 197 |
| Skill Acquisition | 98 | 36 | 40 | 9 | 183 |
| Innovation Output | 121 | 12 | 24 | 13 | 171 |
| Firm Revenue | 98 | 35 | 24 | — | 157 |
| Consumer Welfare | 73 | 31 | 37 | 7 | 148 |
| Task Allocation | 87 | 16 | 34 | 7 | 144 |
| Inequality Measures | 25 | 76 | 32 | 5 | 138 |
| Regulatory Compliance | 54 | 61 | 13 | 3 | 131 |
| Task Completion Time | 89 | 7 | 4 | 3 | 103 |
| Error Rate | 44 | 51 | 6 | — | 101 |
| Training Effectiveness | 58 | 12 | 12 | 16 | 99 |
| Worker Satisfaction | 47 | 33 | 11 | 7 | 98 |
| Wages & Compensation | 54 | 15 | 20 | 5 | 94 |
| Team Performance | 47 | 12 | 15 | 7 | 82 |
| Automation Exposure | 27 | 26 | 10 | 6 | 72 |
| Job Displacement | 6 | 39 | 13 | — | 58 |
| Hiring & Recruitment | 40 | 4 | 6 | 3 | 53 |
| Developer Productivity | 34 | 4 | 3 | 1 | 42 |
| Social Protection | 22 | 11 | 6 | 2 | 41 |
| Creative Output | 16 | 7 | 5 | 1 | 29 |
| Labor Share of Income | 12 | 6 | 9 | — | 27 |
| Skill Obsolescence | 3 | 20 | 2 | — | 25 |
| Worker Turnover | 10 | 12 | — | 3 | 25 |
Productivity
Remove filter
Wage polarization is likely: middle-skill wages will be compressed while high-skill wages rise; some low-skill service roles may persist or expand.
Synthesis of skill-biased technological change literature and task substitution/complementarity arguments; paper references empirical patterns of polarization in prior studies.
Firms with better data infrastructure and higher initial IT investment will adopt AI faster, potentially widening performance gaps across firms and industries.
Theory-informed assertion and literature synthesis; no empirical heterogeneity analysis is specified in the abstract.
Complementarity between AI and skilled accountants may raise wages for analytical roles while compressing demand for routine clerical roles, contributing to wage polarization.
Prediction grounded in economic theory and prior literature; the paper does not report direct wage-change estimates in the abstract.
AI will automate routine accounting tasks, reducing demand for low-skill bookkeeping work while increasing demand for higher-skilled roles (data interpretation, advising, oversight), creating occupational reallocation and upskilling needs.
Projection based on task-based labor economics literature and the paper's synthesis; not supported by specific longitudinal labor-market estimates in the abstract.
AI can enable new revenue streams (platforms, personalized pricing, automation-as-a-service) and increase market concentration, producing 'winner-takes-most' dynamics that raise profit rates for leading adopters and compress margins for laggards.
Literature synthesis on platforms and winner-take-all effects applied to AI; conceptual argument without firm-level causal testing in the paper.
AI adoption exerts downward pressure on routine labor costs while raising capital and recurrent costs (R&D, computing infrastructure, data, cybersecurity); higher fixed and lower marginal costs favor scale and incumbents with access to data and capital.
Conceptual cost-structure analysis drawing on automation and platform literature; no microdata or empirical cost estimates presented.
AI is a Schumpeterian general-purpose technology that can increase aggregate productivity potential but will do so unevenly across firms and sectors, producing heterogeneous effects on profitability.
Theoretical application of general-purpose technology and Schumpeterian literature to AI; literature-based claims without original empirical validation in the paper.
Firms' profitability and sustainability are shaped both by technological adoption (which can raise productivity and market power) and by structural pressures (trade wars, labor relations, supply constraints) that can erode margins.
Synthesis of firm-level implications from innovation and political-economy literatures; no firm-level causal estimates presented in the paper.
Contemporary crises change firms' cost structures (logistics, inputs, financing) and revenue prospects (demand shifts, market access).
Interpretive synthesis of observed firm-level impacts from pandemic, inflation episodes, and geopolitical events reported in secondary literature (no primary firm-level panel used).
Supply-chain fragilities and trade conflicts (emphasized by Mandel) mediate distributional and macroeconomic outcomes during long waves and crises.
Qualitative historical interpretation and literature references on supply-chain disruptions and trade conflicts (no systematic empirical identification in the paper).
New technological waves—most notably artificial intelligence (AI) and the green transformation—act as Schumpeterian forces that can alter productivity, competition, and profitability.
Conceptual mapping of Schumpeterian innovation-cluster theory to contemporary technologies (literature synthesis; no firm-level causal estimates reported).
Contemporary shocks (COVID-19, global inflation, geopolitical tensions) interact with long-wave mechanisms to reshape firms' cost and revenue structures.
Interpretive application of the comparative framework to recent historical episodes and macro trends; qualitative evidence from literature on pandemic and recent shocks (no primary microdata presented).
Students use GenAI as a co-designer and idea generator, which modifies workflow, decision points, and evaluative practices in their design process.
Qualitative interview data from architecture students; thematic analysis surfaced accounts of GenAI being used for ideation, variant generation, and as a collaborative partner (N unspecified).
Collaboration between architecture students and generative AI reshapes creative cognition in the architectural design process through algorithmic thinking strategies.
Semi-structured interviews with architecture students (interview sample size not specified) analyzed via inductive thematic analysis; authors synthesize recurring themes linking GenAI use to changes in cognitive strategies.
Patients classified as high‑risk by CDRG‑RSF had higher TMB, lower NK and CD8+ T cell infiltration, and model‑predicted resistance to Erlotinib and Oxaliplatin but sensitivity to 5‑fluorouracil.
CDRG‑RSF study reported immune deconvolution and TMB comparisons across risk groups and used pharmacogenomic prediction methods to infer drug sensitivity/resistance patterns for high‑risk vs low‑risk groups.
Both DNNs and LASSO correlated well at the individual‑sample level, but linear models (LASSO) struggled to recover cross‑study DEA log2FCs despite good sample‑level fits.
Same cross‑omics comparative study: reported good sample‑level prediction correlations for both model classes, but DNNs more faithfully reproduced differential expression signals across independent studies while LASSO did not recover DEA log2FCs robustly.
Compliance costs and audit requirements create regulatory barriers to entry but also incentives for standardized metadata and interoperable systems; policy can encourage open standards to reduce lock-in.
Policy analysis and recommendation in paper (theoretical); no regulatory cost quantification provided.
Algorithmic lesson planning, automated audits, and data-driven competency mapping are natural targets for AI augmentation and can reduce recurring resource burdens but require quality-labelled data, strong governance, and transparency.
Paper's discussion of AI complementarity (conceptual); no implementation trials or performance metrics presented.
The taxonomy clarifies where substitution versus complementarity are likely: AI-assisted tasks imply partial substitution of routine work; AI-augmented applications generate complementarities that increase demand for higher cognitive skills; AI-automated systems shift labor toward monitoring, exception handling, and governance.
Inference from mapping the three interaction levels to observed case features (n=4) and application of the Bolton et al. framework in cross-case synthesis.
AI-augmented systems support real-time medical tasks (e.g., decision support during procedures), amplifying human judgment and speed but raising required cognitive skills and changing training and coordination practices.
Findings from the case(s) labeled AI-augmented in the four-case qualitative sample and cross-case interpretive analysis using the service-innovation framework.
Returns to AI and digital investments are heterogeneous across firms and industries, implying adoption barriers and varied productivity impacts.
Across the 145 studies, reported effect sizes and qualitative findings vary by firm characteristics, industry sector, and technology readiness, as summarized in the review.
Impacts of digital transformation on productivity vary substantially by moderators such as digital competencies, organizational culture, leadership, and technology readiness.
Multiple included studies identified these factors as moderators/mediators in their empirical analyses; moderator effects were synthesized in the review.
DeFi components could enable automated milestone disbursement instruments but face regulatory and counterparty risk barriers.
Paper mentions DeFi as a potential disbursement automation mechanism and notes regulatory/counterparty risk; this is a conditional, context-dependent claim without pilot evidence for large-scale DeFi use.
Task-based labor effects: GenAI will substitute routine tasks (documentation, triage) and complement complex decision-making; net employment effects are ambiguous and vary by role.
Task-based model of labor and early observational/pilot studies; the paper highlights heterogeneity by specialty and role, but presents no comprehensive empirical employment-impact studies.
GenAI can reduce clinician time per case (productivity gains) but may increase utilization (more tests/treatments) if it lowers thresholds for intervention or aligns with revenue incentives.
Economic reasoning supported by early empirical and simulation work; the paper notes the possibility based on task substitution and induced demand literature; direct causal empirical evidence from large-scale deployments is limited.
Heterogeneous returns: returns to AI will vary across SMEs due to differences in managerial capabilities and local institutional contexts; targeting complementary capabilities may be more cost‑effective than uniform subsidies for hardware/software.
Theoretical conclusion drawn from integrating RBV, dynamic capabilities, and institutional theory across reviewed studies; supported by cited heterogeneity in the literature.
Improved personalization via RS techniques can increase consumer surplus by better matching robot behaviors to user needs, but it also creates the potential for finer-grained price or content discrimination if monetized.
Economic reasoning and implications section; conceptual analysis without empirical measurement.
Evidence of labour reallocation within rural economies following AI-driven productivity changes was observed in the reviewed literature.
Reported findings across several reviewed studies noting shifts in labour allocation and task composition on farms and in related value-chain activities.
Paper‑based regulatory environments slow DT diffusion; digitised compliance and standardised data schemas can accelerate adoption and enable AI‑driven oversight.
Findings in the review noting regulatory friction and proposed solutions; supported by case evidence where digitisation of compliance facilitated digital workflows.
DT adoption is a socio‑technical transformation that requires governance, standards, collaborative delivery models, and workforce capability building — not just technology deployment.
Conceptual synthesis and cross‑study recommendations in the reviewed literature emphasizing organizational, contractual, and governance changes alongside technology.
Both initial trust and inertia have statistically significant effects on GAICS adoption decisions.
Inferential statistical tests reported in the quantitative phase indicating significant pathways from initial trust and from inertia to adoption outcome (exact effect sizes and sample size not provided in the abstract).
Organizations’ adoption of Generative AI–enabled CRM systems (GAICS) is driven by initial trust and inertia.
Quantitative inferential analysis in the study's second phase testing the conceptual model (paper reports statistically significant relationships between initial trust, inertia, and GAICS adoption). Sample size and sector/country scope not reported in the abstract.
Better predictive models can shrink asymmetric‑information wedges and potentially reduce interest spreads for high‑quality but thin‑file borrowers; however, model errors or biased features can systematically exclude certain groups.
Conceptual analysis of model performance, bias risk, and implications for pricing; supported by literature on algorithmic bias and selective case evidence but not empirical causal tests within the paper.
Blockchain applications (tokenization, smart contracts) have potential for transparent, programmable financing and lower transaction costs but remain nascent and face legal and market adoption barriers.
Qualitative synthesis of emerging blockchain use cases and legal/regulatory analysis; characterization is forward‑looking and based on current maturity levels rather than empirical measurement of outcomes.
Crowdfunding is useful for market validation and early‑stage capital but has limited ticket sizes and is not scalable for growth capital needs.
Comparative assessment of financing models and illustrative examples; conclusion based on typical crowdfunding ticket sizes and market practice rather than new representative data.
Revenue‑based financing offers flexible repayments tied to cash flow and suits startups with recurring revenues, but can be more expensive over time and is less regulated.
Qualitative evaluation of product features in the comparative framework and literature synthesis; based on product design characteristics rather than primary quantitative pricing analysis in the paper.
FinTech lending platforms provide high accessibility and speed through alternative data and automated underwriting, with variable costs and scalability but raise regulatory and data‑privacy concerns.
Comparative qualitative assessment and illustrative case studies demonstrating faster approvals and broader reach for thin‑file borrowers; evidence is descriptive and not nationally representative or causally identified.
Traditional sources (bank loans, government schemes) offer lower nominal cost for creditworthy borrowers and regulatory protections, but suffer from collateral requirements, slow processes, and limited outreach to informal/small firms.
Comparative framework evaluation across five variables and institutional/regulatory synthesis; findings are qualitative and built on established banking characteristics rather than new representative quantitative data in the paper.
AI‑driven protein structure prediction will reallocate economic value across the biotech R&D stack—compressing early discovery costs, increasing returns to downstream validation/optimization, and favoring actors combining data, compute, and domain expertise.
Paper summarizes this as an overarching implication in the 'Overall' paragraph, integrating prior methodological and economic arguments; no quantitative economic model or empirical measurement is provided.
Labor demand will shift away from low‑throughput experimental structure determination toward ML model engineers, computational biologists, and integrative experimentalists, requiring retraining in experimental groups.
Paper states this in 'Labor and skill shifts'; it is an inferred labor market consequence without workforce surveys or models in the text.
Single‑sequence protein language models (e.g., ESMFold) trade some accuracy for much higher speed and scalability compared with MSA/template‑based models.
Paper describes single‑sequence approaches that remove MSA dependence and rely on very large pretrained language models, stating they trade accuracy for speed/scalability; no head‑to‑head metrics are presented in the text.
Effectiveness of ChatGPT varied by discipline; not all course contexts showed significant gains from allowing its use.
Heterogeneous treatment effects observed across the six courses; GLM and non-parametric tests indicated variation in effect sizes and statistical significance by course/discipline.
Analytical inequalities derived in the model delineate parameter regions (functions of AI capability growth rate, diffusion speed, and reinstatement elasticity) that separate stable/convergent adjustments from explosive demand-driven crises.
Closed-form analytical derivations presented in the model section of the paper, supplemented by numerical exploration of parameter space (phase diagrams).
Heterogeneous and changing users (skill, mental models, incentives) produce heterogeneous and time-varying treatment effects, complicating inference from average uplift estimates.
Practitioner descriptions from 16 interviews highlighting user heterogeneity and learning/adaptation over time; authors' implication that averages may be insufficient.
Human uplift studies (typically RCTs measuring how AI changes human performance relative to a status quo) are a useful tool for informing deployment and policy decisions but face systematic validity challenges when applied to frontier AI systems.
Qualitative thematic synthesis of semi-structured interviews with 16 experienced practitioners across biosecurity, cybersecurity, education, and labor; authors' analytic mapping of interview themes to research lifecycle stages.
Governance constraints induce measurable trade-offs between efficiency and compliance; the magnitude of these trade-offs depends on topology and system load.
Simulation experiments in the ablation study varied governance constraint parameters and load, measuring compliance rates and efficiency (value/throughput). Results show systematic reductions in efficiency as compliance constraints tighten, with the effect size modulated by graph topology and load levels.
AI agents are useful as breadth tools and for pre-deployment checks but lack the protocol-specific and adversarial reasoning required to replace human auditors; human-in-the-loop workflows are the best use.
Study observations: agents reliably flag well-known patterns and respond to human-provided context, but fail to perform robust end-to-end exploit generation and are sensitive to scaffolding and configuration.
NFD can raise productivity in expert-heavy tasks by capturing tacit process knowledge and reducing repetitive cognitive effort, but the effect on employment is nuanced—routine parts may be automated while humans remain central to oversight and knowledge contribution.
Claims drawn from implications and the case study where analyst effort per task decreased and practitioners reported value; employment impact discussion is conceptual and speculative.
Highly personalized agents developed via NFD create stronger switching costs because crystallized knowledge assets are sticky, and economies of scale depend on the transferability of those assets across users or firms.
Conceptual reasoning in the paper's market structure and returns sections; supported by qualitative observations from the case study about personalization and reuse limits. No large-scale market data.
NFD shifts the economic tradeoff from large up-front engineering investment to ongoing human-in-the-loop investment; marginal cost of improving an agent becomes tied to practitioner time and crystallization efficiency rather than purely engineering labor.
Implications for AI economics section—conceptual analysis drawing on the NFD model and case study observations. No large-scale economic data provided.