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 |
Perception providing the material's location inside the vial was used to guide the agent.
Paper summary states perception input (material location) was provided to the agent; sensing modality and accuracy/details of perception are not specified.
Privacy-preserving accountability logs can support ex post adjudication, insurance products, and reputational dynamics, reducing moral hazard.
Conceptual claim: protocol includes privacy-minded logs; paper argues potential for post-hoc review and insurance. No empirical tests of adjudication or insurance products provided.
Observable capability and coordination-risk signals enable more granular pricing, risk-based contracts, and differentiated service tiers (e.g., primary-only vs primary+auditor).
Policy/economic implication argued conceptually in the paper; no empirical pricing experiments or market data provided.
High capability profiles for some tasks will shift delegation toward agents (automation) and reallocate human labor toward supervision, auditing, and low-win-rate tasks.
Projection based on capability profiles and economic reasoning in the paper; presented as implications rather than empirically demonstrated. No labor-market empirical data provided.
Better matching of tasks to agent competencies improves allocative efficiency across task markets.
Theoretical/economic claim derived from capability profiles enabling improved matching; no empirical market experiments or measurements reported in the summary (field experiments suggested as future work).
Task-aware signals reduce search and screening costs by acting like quality/reliability metrics in delegation markets.
Economic implication argued conceptually in the paper: task-conditioned capability and coordination-risk signals function as observable quality metrics, reducing transaction costs. This is a theoretical argument; no empirical market-level test reported.
Using CFR avoids the computational and development costs of retraining T2I models to improve color fidelity, providing a lower-cost path to better color authenticity.
Paper emphasizes CFR is training-free and applies at inference, claiming improved color authenticity without model retraining; cost implication is inferred from lack of retraining (quantitative compute savings not provided in the summary).
Once trained, these simulation-trained summary networks are fast to evaluate and can be used as amortized estimators to enable large-scale counterfactuals, sensitivity analyses, and Monte Carlo-based policy evaluation with much lower per-evaluation cost.
Practical implication claim: based on amortization principle (neural network inference is fast at evaluation time) and reported ability to replace repeated runs of iterative algorithms; the summary asserts reduced per-evaluation cost but does not provide quantitative runtime benchmarks or speedup ratios in the provided text.
Surrogate-accelerated workflows reduce energy consumption and carbon footprint per discovery because they require fewer expensive evaluations.
Stated implication in the paper linking fewer expensive quantum-chemistry/DFT evaluations to lower energy use; no measured energy/emissions data provided in the summary.
Order-of-magnitude reductions in expensive evaluations enable faster R&D cycles and higher throughput for exploration of potential-energy landscapes in materials science, catalysis, and drug design.
Policy/economic implication argued in the paper based on empirical reductions in expensive evaluations; no direct time-to-discovery experiments reported in the summary.
Organizations should consider LLM-generated feedback as a high-return, lower-cost PRF option for low-resource retrieval tasks to reduce expenses tied to corpus annotation or expensive retrieval pipelines.
Implication drawn from the paper's cost-effectiveness results (LLM-generated feedback performing well per LLM invocation cost across the evaluated BEIR tasks).
QCSC capabilities could change the economics of certain AI model classes that rely on expensive scientific simulations for training data by producing richer, cheaper training datasets.
Theoretical link between simulation output quality/cost and training-data generation for physics-informed ML and generative chemistry models; no empirical studies or cost estimates presented.
QCSC-enabled faster, higher-fidelity simulation can compress R&D cycles in chemistry and materials, lowering time-to-discovery and increasing returns to computational investment for firms.
Use-case analysis linking simulation fidelity/turnaround to R&D timelines; relies on assumed speedups and fidelity improvements but provides no measured speedup data.
The proposed approach will increase demand for edge/embedded ML expertise, GNN optimization, and HAPS integration, shifting supplier ecosystems and labor requirements.
Workforce and supply-chain implication stated in the paper's discussion of economic impacts; based on projected capabilities required to implement FL+GNN solutions, not on labor-market measurements.
FL reduces raw-data movement across jurisdictions, easing regulatory compliance for cross-border NTN services and supporting privacy-preserving business models.
Implication derived from the federated approach (local model updates vs. raw-data transfer) noted in the paper; no legal/regulatory case studies or measurements provided.
HAPS-as-aggregator creates a distributed service layer between satellites and terrestrial infrastructure, enabling new roles (HAPS operators, FL orchestration providers) and revenue streams.
Paper's market-structure implications: conceptual argument that HAPS aggregation in an FL architecture yields opportunities for new service roles and monetization; no market or revenue analysis provided.
Lightweight GNNs enable more intelligence on-board or at HAPS without requiring major hardware upgrades, potentially deferring capital expenditures (CapEx).
Economic/operational implication in the paper based on the stated compactness of the GNN model and its suitability for edge/on-board deployment; no quantified hardware or CapEx comparison provided.
Improved predictive beam selection (from the proposed GNN/FL approach) reduces link outages and retransmissions, cutting operational costs and improving user experience.
Economic implication stated in the paper linking better beam prediction/stability (experimentally observed) to reduced outages and retransmissions; no direct measurement of outages/retransmissions or operational cost savings reported in the summary.
Adopting DPS-like efficiencies reduces the marginal compute cost of online prompt-selection workflows (dominated by rollouts), thereby shortening finetuning cycles and increasing developer productivity.
Paper's implications section: logical inference from reported reduction in rollouts and rollout compute; not an empirical market study—no dollar or industry-scale numbers provided.
There is a strong complementarity between AI investments and organizational change: firms with better leadership, cross-functional processes, and data practices capture disproportionate benefits, implying increasing returns to scale and potential winner-take-most dynamics.
Authors' theoretical inference from cross-case patterns and economic reasoning; supported qualitatively by cases showing disproportionate gains in better-managed firms.
Firms that can credibly supply explainability and governance may capture a premium—explainability can be a competitive differentiator and a signal of quality and lower regulatory risk.
Conceptual synthesis and market-structure arguments from the reviewed literature; reviewed studies provide theoretical and some qualitative support but not systematic market-price estimates.
Policy should incentivize transparency, auditability, standards for human–AI interfaces, workforce development, certification of teaming practices, and liability frameworks to ensure accountability and equitable outcomes.
Normative recommendation based on ethical and governance considerations synthesized in the paper; not supported by policy evaluation evidence within the paper.
Orchestrating attention and interrogation through interface and workflow design helps manage what humans and AI focus on and how they challenge/verify each other, thereby reducing errors and misuse.
Prescriptive claim grounded in human factors and HCI literature synthesized by the authors; the paper suggests these mechanisms but does not report empirical trials demonstrating effects.
Design principles (define goals/constraints, partition roles, orchestrate attention/interrogation, build knowledge infrastructures, continuous training/evaluation) are necessary design levers to build high-performing, transparent, trustworthy, and equitable Human–AI teams.
Prescriptive synthesis from reviewed literatures and conceptual modeling; these principles are proposed heuristics rather than empirically validated interventions in the paper.
Embedding AI produces operational gains: automation of routine tasks, fewer errors, faster decision cycles, and continuous model learning/refinement.
Operational claim articulated conceptually with suggested evaluation metrics (forecast accuracy, latency, false positive/negative rates); the paper does not present empirical measurement, sample sizes, or deployment results.
Risk management can accelerate AI adoption by lowering uncertainty for managers and investors, thereby affecting diffusion and productivity gains from AI.
Conceptual implication derived from the review's synthesis and discussion (policy/implication section); not supported by primary empirical testing within the reviewed literature.
Firms that adopt structured risk management for AI projects can reduce model failure, operational losses, and reputational costs—improving risk-adjusted returns on AI investment.
Theoretical and practical extrapolation from general RM frameworks and thematic findings in the literature; no AI-specific primary empirical studies included in the review.
Structured risk management can produce potential cost savings via reduced loss events and more efficient capital allocation.
Reported as a benefit across some reviewed studies and practitioner reports; the review notes lack of primary empirical quantification of effect sizes.
Firms that design processes to preserve human diversity and elicit diverse AI outputs may capture greater productivity gains, increasing returns to organizational capability rather than to raw model access.
Theoretical implication and prescriptive recommendation based on observed homogenization; no direct causal firm-level evidence presented, inference based on economic reasoning.
Investments to build trust in AI (transparency, reliability, training) are likely to have positive returns via higher adoption rates and realized AI benefits.
This is presented as an implication derived from observed positive associations between trust and outcomes; the study did not conduct cost–benefit or longitudinal causal tests of such investments in the reported analyses.
Practical levers to increase AI trust include transparency of AI models, demonstrated reliability, and manager-focused AI literacy/training.
Paper proposes these levers based on study findings and discussion (recommendations), but they were not tested experimentally in the reported cross-sectional survey.
A stronger data-driven decision culture that stems from AI trust yields better operational and academic outcomes.
Study reports positive associations between AI trust → data-driven culture → operational and academic outcomes in survey-based analyses; however, the summary does not specify which operational/academic metrics were measured or sample size.
On-Premise RAG provides a viable path for SMEs sensitive to security and cost to adopt advanced language capabilities without perpetual vendor fees or data exposure.
Synthesis of technology, organizational, and environment/security analyses (TOE framework) and implications section arguing SMEs can adopt on-prem RAG; presented as an implication rather than proven adoption data.
Procurement contracts for AI systems can require staged validation (pilot, local fine-tuning) and performance-linked payments to align incentives and reduce adoption risk.
Policy recommendation drawn from procurement and incentive-design literature synthesized in the review; not an empirical claim about observed outcomes but a proposed intervention to mitigate identified risks.
Clear regulatory standards for synthetic data quality, provenance, and acceptable validation pipelines will lower transaction costs, reduce liability risk, and stimulate private-sector offerings (synthetic-data services, marketplaces).
Policy and governance analyses in the review arguing that regulatory clarity reduces uncertainty and promotes market activity; this is a policy inference supported by comparative regulatory studies rather than direct causal empirical proof specific to African markets.
The dissertation implies policy interventions (subsidies, tax incentives, training and integration assistance) can accelerate welfare-improving AI adoption by helping firms overcome the early negative part of the U-shaped profit profile.
Policy implication derived from the theoretical U-shaped profit relationship and model interpretation; not supported by randomized or quasi-experimental policy evaluation in the provided summary.
Vendors that embed robust cognitive interlocks into development platforms can command premium pricing by reducing downstream risk; verification features may become a competitive moat.
Market-structure and product-differentiation reasoning in the paper; no market data, pricing studies, or competitive analyses presented.
Human verification (and automated verification infrastructure) becomes the limiting factor and a scarce complement to AI generation, raising demand and wages for verification expertise and tooling.
Theoretical labor-market analysis and complementarity argument in the paper; no labor market data or econometric estimates provided.
AI contributes to flatter, more networked and modular organizational forms, with increased cross-functional coordination enabled by shared data platforms and real-time analytics.
Conceptual reasoning supported by cross-sector illustrative examples; no standardized cross-firm comparative empirical study reported in the book.
Valuation of AI services should account for initiation assistance (fixed-cost reduction to starting tasks); monetizable value extends beyond direct task automation and could affect pricing/willingness-to-pay models.
Economic argument and implication drawn from the conceptual model; the paper does not provide empirical willingness-to-pay or pricing studies.
Conversational initiation assistance could complement human labor by increasing worker throughput and engagement, rather than directly substituting for skilled tasks.
Economic/managerial speculation in the paper; no empirical workforce or productivity studies presented.
Designing interfaces and metrics that focus only on task completion or execution misses value derived from initiation assistance.
Analytic recommendation based on the proposed model; no empirical metric-validation or A/B test results presented.
Conversational AI provides a distinct, non-executive mode of value — acting as an action-initiation interface in addition to being a task-execution tool.
Conceptual/economic argumentation in the paper; no empirical valuation or willingness-to-pay estimates provided.
Iterative conversation with AI surfaces sub-tasks and structures problems (structuring), creating clearer action plans and reducing initiation barriers.
Conceptual argument and illustrative example; paper does not present systematic coding, task analyses, or empirical tests.
Externalization (expressing frustration/stress to an external interlocutor) reduces affective load and decision paralysis, facilitating task start.
Theoretical reasoning supported by an illustrative anecdote; no empirical measurements or sample-based evidence provided.
Verbalization (talking through a problem with the AI) helps users organize thoughts and identify next steps, thereby lowering barriers to action.
Mechanistic argument in the paper; no experimental or observational data reported to validate the mechanism.
The 'Peripheral Approach' — beginning with casual, low-stakes dialogue (complaints, describing where one is stuck) rather than immediately requesting task execution — gradually reduces initiation friction.
Theoretical argument and illustrative anecdote from the author. No controlled studies or quantitative measures presented.
Casual, conversation-style interactions with AI can reduce psychological barriers that prevent people from starting tasks.
Conceptual/theoretical argumentation in the paper; illustrated by an anecdote (author's use of casual AI conversation to begin drafting the paper). No systematic empirical data, no experiments or observational samples reported.
Model and platform providers may capture significant rents through APIs and integrated developer tooling.
Market-structure analysis and observations of current platform monetization strategies; speculative projection based on platform economics.
Skill premiums may shift toward workers who can effectively collaborate with AI (prompting, verification, security auditing).
Theoretical and early observational studies suggesting complementary skills add value; limited empirical wage/earnings evidence to date.