Evidence (7560 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
Filtered →
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 |
Human Ai Collab
Remove filter
An iterative, scenario-refined prompt engineering structure enables the LLM (ChatGPT in this study) to generate task-specific, contextualized guidance that aligns with real-time user actions and machine state.
System design and methods: authors describe developing and refining a prompt structure across multiple machine-operation scenarios and using ChatGPT as the generative engine to produce stepwise instructions and contextual overlay content. Evidence is methodological and qualitative within the paper's development process.
Participants reported lower perceived workload and improved usability when using the AR-MLLM system.
Subjective workload/usability questionnaires were administered in the CMM case study; authors report reduced reported workload under AR-MLLM guidance. (Questionnaire instrument, scales, and sample size not specified in the summary.)
Participants completed assigned CMM tasks faster when using the AR-MLLM system compared to baseline/traditional training.
Task execution time was recorded in the CMM case study; authors report statistically meaningful reductions in completion time with AR-MLLM guidance versus baseline. (Summary does not give numerical effect sizes or sample size.)
The AR-MLLM system achieved high measurement/feature-activity accuracy (participants performed correct measurements under AR-MLLM guidance).
Measurement/feature activity correctness was measured in the CMM case study; authors report high measurement accuracy under the AR-MLLM condition. (Exact rates and sample size not provided in the summary.)
The AR-MLLM system achieved high task-recognition accuracy (the system correctly identified the current task/step).
Measured task recognition accuracy in the CMM case study; authors report 'high' recognition accuracy for the system. (Exact numeric accuracy and sample size not specified in the summary.)
An AR + multimodal LLM (AR-MLLM) training system can substantially improve training and execution in complex machine operations (demonstrated on a Coordinate Measuring Machine).
Case-study experiment in the paper where human participants performed CMM measurement tasks both with and without the AR-MLLM system; metrics collected included task recognition accuracy, measurement activity correctness, task completion time, and subjective workload/usability. (Participant sample size not specified in the provided summary.)
Labor complementarities with agentic AI will shift resources toward oversight, interpretation, and coordination roles rather than routine task execution.
Economic and organizational reasoning; literature synthesis on skill complementarities; no empirical labor-market data analyzed in the paper.
Principal–agent contracting frameworks must be extended to account for evolving agent objectives and open-ended action spaces; contracts should be dynamic and include continuous renegotiation and monitoring.
Theoretical extension and recommendations based on economic reasoning; proposed formal models for future work.
Projection congruence — alignment of forecasts/plans across heterogeneous agents — becomes a central metric for assessing alignment in agentic human–AI teams.
Conceptual modeling and proposal in the paper; introduced as a new measurable construct (projection congruence indices) for future empirical work.
The DAR framework reframes human oversight as a dynamic, auditable process whose micro-level mechanics and macro-level legitimacy have direct economic consequences for productivity, contracting, regulation, and welfare.
Synthesis claim based on the conceptual framework, formal modeling, derived propositions, and policy/economics implications sections. The claim is theoretical and synthesizing rather than empirically validated.
The Reversal Register will create granular, time-stamped administrative data valuable for structural estimation of trust, error externalities, and productivity comparisons between automation and human judgment.
Design claim linking register contents to potential econometric uses; no empirical data shown—claim about potential data utility.
Reversal Register logs can enable descriptive and causal analyses of handovers and support experimental/quasi-experimental tests (e.g., randomized hysteresis thresholds, A/B override policies).
Implied empirical strategies and instrumentation described; paper outlines how register data would be used for experiments and causal inference. No empirical implementation or sample reported.
Operationalizing reversible AI leadership via DAR can preserve human accountability while enabling AI-led decisions where appropriate.
Conceptual argument supported by the combined use of authority states, Reversal Register logging, and override mechanisms; no field validation provided.
DAR incorporates stabilizing mechanisms—hysteresis bands and safe-exit timers—to reduce rapid oscillation of authority and improve stability of handovers.
Formal model components and design proposals (hysteresis and timers) with conceptual argument that these damp oscillation; no empirical validation reported.
Continuous human-in-the-loop oversight, monitoring, and retraining are required to maintain quality and prevent model drift.
Practitioner reports and conceptual literature synthesized in the review advocating monitoring and retraining; no longitudinal empirical study provided here.
Transparent disclosure to customers about AI involvement helps preserve trust.
Conceptual analyses and referenced empirical/regulatory discussions in the literature aggregated by the review; this paper presents no new experimental evidence on disclosure effects.
Hybrid designs that automate low-risk, high-volume tasks while routing complex, judgment-sensitive cases to humans produce the best operational outcomes.
Inferred best-practice from aggregated empirical studies, industry examples, and conceptual reasoning; no controlled comparative trials presented in this review.
Agent augmentation via suggested responses, summarization, and information retrieval improves agent productivity.
Aggregated evidence from prior empirical research and practitioner reports cited in the review; no new measurements or sample sizes presented here.
Generative AI enables personalization at scale through automated tailoring of messaging and recommendations.
Qualitative synthesis of empirical studies and industry reports showing automated personalization use-cases; no systematic effect-size estimates or new quantitative data in this review.
Generative AI provides 24/7 availability and cost-effective scaling of routine interactions.
Industry case examples and prior empirical studies aggregated in the review; no original data or quantified sample sizes provided in this paper.
Generative AI can materially transform customer service and strategic communication by enabling continuous automation, scalable hyper-personalization, and effective agent augmentation.
Nano review: qualitative aggregation and synthesis of existing empirical studies, industry case examples, and conceptual analyses. No novel primary data or sample size; conclusion drawn from heterogeneous secondary sources and practitioner reports (not a systematic meta-analysis).
There is a need for standards around evaluation, bias mitigation, provenance, and accountability in AI-assisted ideation and design.
Policy recommendation motivated by documented biases, errors, and provenance issues in the reviewed studies; grounded in the synthesis's critique of existing practice.
There will likely be complementarity-driven increases in demand for evaluative, integrative, and domain-expert roles (curators, synthesizers, implementation experts).
Inference from task-level studies and economic reasoning about complementarities between AI generative capability and human evaluative skills; empirical labor-market evidence is limited in the reviewed literature.
Lower search and idea-generation costs enabled by LLMs may speed early-stage R&D and increase the gross flow of candidate innovations.
Theoretical economic interpretation supported by empirical findings of increased idea volumes in experimental/field studies summarized in the review; no long-run causal firm-level evidence presented.
Generative AI accelerates early-stage hypothesis and prototype development by providing scaffolded prompts and procedural suggestions.
Applied case evidence and experimental studies summarized in the review showing reduced time or increased productivity in early-stage experimental/design tasks when using LLM assistance; no pooled effect size presented.
Empirical studies document that AI-assisted tools can help break cognitive fixation and generate cross-domain analogies.
Cited experimental tasks and lab studies in the literature showing higher incidence of analogical or cross-domain suggestions from LLMs and improvements on fixation-related task metrics; heterogeneity across tasks and measures.
Generative AI provides scaffolded, structured support that aids systematic hypothesis formation, prototyping steps, and decomposition of complex problems.
Review of design/ideation studies and applied case evidence where LLMs produced stepwise plans, decomposition prompts, or hypothesis scaffolds; evidence drawn from multiple short-term experimental and applied studies, sample sizes and exact designs vary by study.
Generative models rapidly produce many candidate ideas, analogies, and associative prompts that help overcome cognitive fixation.
Synthesis of experimental ideation and design studies reporting increases in number of ideas and examples of reduced fixation when participants used LLM outputs; heterogeneous sample sizes across cited studies (not reported in review).
Generative AI can raise per-worker productivity for tasks involving brainstorming, drafting, and prototyping, but realized gains depend on downstream filtering and implementation costs.
User studies showing higher output on specific tasks (brainstorming/drafting), combined with qualitative reports of filtering/implementation effort; many studies measure immediate task output but not net realized productivity after implementation.
Generative AI can increase creative output in both lab and field tasks as judged by external raters.
Controlled experiments and field studies reporting higher judged creativity/novelty scores for AI-assisted outputs versus controls; judged creativity/novelty is typically assessed by human raters using rubric-based scoring.
AI assistance helps people overcome fixation and produces cross-domain analogies that they might not generate alone.
Experimental studies and qualitative analyses documenting reductions in fixation effects and increases in cross-domain analogical suggestions when participants use generative models.
Generative AI supports systematic problem breakdown and early-stage prototyping, accelerating hypothesis generation and prototype development.
Field case studies of AI-supported prototyping and lab/user studies reporting reduced time-to-prototype and generated hypotheses; measures include time-to-prototype and user-reported usefulness.
Generative AI boosts ideational fluency—the quantity and diversity of ideas produced in brainstorming tasks.
Controlled experiments and user studies measuring number and diversity of ideas with and without AI assistance; typical study designs compare participant idea counts/uniqueness across conditions (note: many studies use small or convenience samples).
When used as a 'cognitive co-pilot' that expands the solution space and challenges assumptions while humans curate and evaluate, generative AI generates economic value.
Inferred from experimental and field findings showing increased idea quantity/diversity and faster prototyping combined with qualitative studies showing human curation is needed; economic interpretation drawn from the review rather than direct macroeconomic measurement.
Generative AI serves a dual cognitive role: (1) a high-volume catalyst for divergent idea generation and cross-domain analogy-making, and (2) a structured assistant for deconstructing complex problems and scaffolding hypotheses and prototypes.
Synthesis of controlled experiments, lab studies, field case studies, and qualitative analyses summarized in the review; evidence includes measures of idea fluency/diversity, examples of analogy production, and observations of AI-assisted problem decomposition in prototyping tasks. (Note: underlying studies are heterogeneous and often short-term or convenience samples.)
Agent augmentation (drafting replies, summarizing histories, suggesting actions) raises frontline productivity and can improve response consistency.
Pilot deployments and internal A/B tests cited that measure time saved by agents and improvements in draft quality/consistency; mostly short-run and firm-specific reports.
Hyper-personalization at scale can increase relevance of responses and customer engagement when fed high-quality signals.
Case studies and pilot deployments that applied personalization signals (customer history, behavioral data) and reported improved relevance/engagement metrics; evidence conditional on availability and quality of signals and largely non-randomized.
24/7 automation reduces routine handling time and operational costs for simple, repetitive queries.
Operational deployments and pilot studies reporting reduced handling times and cost-per-interaction for routine queries; some vendor-supplied before/after or A/B comparisons, but heterogeneous measurements and limited randomized evidence.
Perceptions—specifically trust and perceived accuracy—are central frictions in AI adoption within finance; interventions that raise perceived and demonstrable accuracy (e.g., explainability, transparent validation) will increase uptake and productivity gains.
Study finds correlations between perceptions and adoption/productivity proxies from questionnaire and performance data; authors combine these empirical associations with qualitative insights to recommend explainability/validation as interventions. Evidence is correlational and inferential (causal impact of interventions not estimated in summary).
Higher perceived accuracy of AI outputs is associated with increased perceived utility of AI for forecasting and risk-management tasks.
Survey items measuring perceived accuracy and perceived utility for specific tasks (forecasting, risk management) and quantitative association analysis; supported by interview excerpts illustrating task-specific utility; exact effect sizes and sample counts not provided in summary.
Greater trust in AI correlates with greater willingness to adopt AI tools and to incorporate AI recommendations into decisions.
Correlational findings from structured questionnaires linking measures of trust with adoption intentions and self-reported incorporation of AI recommendations; supported by qualitative interview evidence; sample across multinational financial institutions (size not specified).
When trust and accuracy are high, human–AI collaboration improves organizational agility, enabling faster, data-driven strategic pivots and better risk management.
Quantitative analysis estimating relationships between perceived trust/accuracy and organizational agility indicators (speed of strategic pivots, risk-management metrics) augmented by interview accounts describing faster responses; sample: finance professionals across multinational financial institutions (sample size and exact agility metrics not specified).
Perceived accuracy of AI-generated insights increases decision confidence and perceived utility for forecasting and risk management.
Quantitative questionnaire measures of perceived accuracy correlated with self-reported decision confidence and perceived utility for forecasting/risk management, with qualitative interviews used to explain mechanisms; sample: finance professionals across multinational financial institutions (sample size not specified).
Perceived trust in AI tools is a key driver of finance professionals' willingness to use AI and their confidence in AI-assisted decisions.
Mixed-methods: quantitative analysis of structured questionnaires measuring perceived trust together with measures of willingness to use AI and decision confidence, supplemented by semi-structured interview evidence; sample described as finance professionals across multinational financial institutions (sample size not specified in summary).
The Adaptive Agent Routing and Coordination (AARC) module performs intent recognition with confidence scoring, triggers proactive clarification dialogues on low confidence, and provides a planning feedback loop to refine plans during execution.
System design description: AARC includes intent classifier confidence thresholds, clarification dialogue behavior, and a feedback loop. Its role is supported by routing/coordination performance improvements and ablation experiments, but the summary lacks quantitative measures of clarification frequency or confidence calibration.
The Multi-Modal Contextual Memory (MMCM) stores multi-modal (visual, linguistic, temporal) contextual memory units in a relational graph and uses an advanced retrieval mechanism with temporal decay weighting to support multi-hop reasoning.
System design and implementation description: MMCM encodes modality, timestamp, and relational links; retrieval uses similarity plus temporal decay. Its effectiveness for multi-hop QA is supported by the reported improvement in Knowledge Base Response Validity and ablation results, though quantitative retrieval performance metrics are not provided in the summary.
The Semantic-Enhanced Task Planning (SETP) module enriches LLM-generated plans with object-relationship graphs, hierarchical task decomposition, and implicit physical/affordance constraints to improve plan plausibility.
System design description: SETP augments LLM plans with semantic object graphs and hierarchy enforcement. Its contribution is supported indirectly by ablation results showing performance drop when SETP is removed; direct quantitative attribution to specific SETP mechanisms not detailed in the summary.
An ablation study shows that removing any of the three core modules (SETP, MMCM, AARC) degrades CRAEA's performance; each module contributes meaningfully to overall gains.
Ablation experiments reported in the paper where SETP, MMCM, and AARC were each removed in turn and performance degradation was observed across metrics. The summary describes the qualitative outcome but omits numerical ablation results and sample sizes.
Human evaluators rate CRAEA higher on perceived coherence, naturalness, and user satisfaction compared to baselines.
Subjective human evaluation studies reported in the paper—comparative ratings on coherence, naturalness, and satisfaction. The summary does not specify number of human raters, rating scales, or statistical significance.
CRAEA improves Agent Routing and Coordination success relative to baseline agents.
Objective metric 'Agent Routing Success Rate' measured in simulation; CRAEA compared to baseline LLM-driven agents (e.g., memoryless or statically routed controllers) with reported higher routing success. Exact task counts and effect sizes not included in the summary.