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
Die Befunde unterstreichen die Bedeutung kontextspezifischer Einführung, rollenbezogener Qualifizierung und Governance für eine nachhaltige Akzeptanz generativer KI in Organisationen.
Interpretation/Schlussfolgerung der Autoren basierend auf den survey-Ergebnissen und beobachteten Unterschieden zwischen Rollen sowie zeitlichen Entwicklungen (im Abstract formuliert).
Der größte Mehrwert von Copilot liegt bei klar strukturierten, textbasierten Aufgaben.
Befragungsergebnisse zur Nutzenabschätzung für typische Tätigkeiten der Wissensarbeit, wie im Abstract zusammengefasst (präferierte Aufgabenarten: strukturierte, textbasierte Aufgaben).
Microsoft 365 Copilot wird überwiegend als benutzerfreundlich und technisch zuverlässig wahrgenommen.
Selbstberichtete Beurteilungen zu Benutzerfreundlichkeit und technischer Zuverlässigkeit in der wiederholten Querschnittsbefragung (Angabe im Abstract).
Wissenschaftliche Mitarbeitende entwickeln im Zeitverlauf positivere Einschätzungen, insbesondere hinsichtlich Produktivität und Arbeitserleichterung durch Copilot.
Längsschnittähnliche Beobachtung über die wiederholten Querschnittserhebungen; zeitliche Veränderung der Selbsteinschätzungen wissenschaftlicher Mitarbeitender im Abstract beschrieben.
Verwaltungsmitarbeitende bewerten die Nützlichkeit und die Zuverlässigkeit von Microsoft 365 Copilot höher als wissenschaftliche Mitarbeitende.
Selbstberichtete Bewertungen in der wiederholten Querschnittsbefragung; Vergleich zwischen Berufsrollen (Verwaltung vs. Wissenschaft) angegeben im Abstract.
The framework shifts manual harness engineering into automated harness engineering, and takes one step further — automating the design of the automation itself.
Conceptual claim about the scope/implication of the proposed framework stated in the paper; the excerpt contains no empirical measures, experiments, or sample sizes to verify the claim.
The Meta-Evolution Loop optimizes the evolution protocol Λ across diverse tasks, learning a protocol Λ^(best) that enables rapid harness convergence on any new task — so that adapting an agent to a novel domain requires no human harness engineering at all.
Strong methodological claim and intended outcome stated in the paper (formalization and algorithms promised); no empirical validation, benchmarks, or sample sizes given in the excerpt to substantiate the universality or 'no human' guarantee.
The Harness Evolution Loop optimizes a worker agent's harness H for a single task: a Worker Agent W_H executes the task, an Evaluator Agent V adversarially diagnoses failures and scores performance, and an Evolution Agent E modifies the harness based on the full history of prior attempts.
Description of the proposed algorithmic component/architecture in the paper (conceptual specification); no empirical results or sample size provided in the excerpt.
We present a two-level framework that automates this process.
Methodological claim: the paper proposes a two-level framework (Harness Evolution Loop and Meta-Evolution Loop) and states it in the text; no experimental validation or sample size reported in the excerpt.
Traditional software engineering artifacts can serve as effective control mechanisms in AI-assisted development.
Concluding claim in the abstract synthesizing the preliminary evaluation results; presented as the paper's implication/recommendation (based on the exploratory study noted).
Embedding machine-readable requirements and architectural artifacts reduces implementation drift.
Reported as a preliminary finding from the exploratory evaluation; the abstract claims a reduction in implementation drift when using Shift-Up artifacts versus unstructured approaches (no quantification provided).
This paper proposes Shift-Up, a framework that reinterprets established software engineering practices (executable requirements / BDD, C4 architectural modeling, and architecture decision records / ADRs) as structural guardrails for GenAI-native development.
Design-science research (DSR) artifact: the Shift-Up framework is presented as the paper's primary design contribution (description/proposal in the paper; no broad empirical validation in the abstract).
Generative AI (GenAI) is reshaping software engineering by shifting development from manual coding toward agent-driven implementation.
Stated as a high-level premise in the paper's introduction/abstract; presented as an observed trend motivating the research (no empirical sample or quantified measurement reported in the abstract).
The results demonstrate a 'less is more' pattern: simpler combination (memory + reflection) yields better performance than adding architectural complexity.
Authors' interpretation of the ablation study results showing that adding multiple extra mechanisms degraded performance compared to the memory+reflection configuration.
A nine-variant ablation reveals that memory and reflection together produce a 58% cumulative improvement over the stateless baseline.
Ablation study with nine variants on the sequential portfolio benchmark; authors report a 58% cumulative improvement when combining memory and reflection versus the stateless baseline.
AEL outperforms five published self-improving methods and all non-LLM baselines while maintaining the lowest variance among all LLM-based approaches on the benchmark.
Comparative empirical evaluation on the same sequential portfolio benchmark, comparing AEL to five published self-improving methods and multiple non-LLM and LLM baselines (reported relative ranking and variance).
On a sequential portfolio benchmark (10 sector-diverse tickers, 208 episodes, 5 random seeds), AEL achieves a Sharpe ratio of 2.13 ± 0.47.
Empirical experiment on the sequential portfolio benchmark with 10 tickers, 208 episodes, evaluated across 5 random seeds (reported Sharpe ratio and standard deviation).
We introduce Agent Evolving Learning (AEL), a two-timescale framework in which a Thompson Sampling bandit at the fast timescale learns which memory retrieval policy to apply each episode, while LLM-driven reflection at the slow timescale diagnoses failure patterns and injects causal insights into the agent's decision prompt.
Methodological description and proposed algorithmic design in the paper (no additional experimental sample size—design/algorithmic claim).
The sustainability of the algorithmic state rests on a movement from technocratic secrecy to value-based transparency to ensure AI- and human collaboration is founded on institutional accountability and algorithmic justice.
Authorial conclusion from the systematic review synthesis (2018-2026) advocating a policy/practice shift; presented as normative policy recommendation rather than quantified empirical finding.
Empirical evidence shows great gains in efficiency in fiscal forecasting.
Empirical studies included in the PRISMA-guided review (2018-2026) reporting improved fiscal forecasting outcomes; no quantitative effect sizes provided in abstract.
Empirical evidence shows great gains in efficiency at routinised administrative tasks.
Empirical studies reported in the systematic review (2018-2026); the abstract claims empirical evidence of efficiency gains but does not report specific study counts, sample sizes, or effect magnitudes.
Digital infrastructure is a primary determinant of both the pace of AI diffusion and its resulting economic returns.
Synthesis of descriptive patterns, difference-in-differences causal estimates, and instrumental-variable results using Turkish administrative and survey data (2021-2024).
Infrastructure-driven AI adoption shifts labor composition toward ICT-related roles.
Instrumental-variable estimates showing changes in occupational composition (increase in ICT-related roles) associated with infrastructure-driven AI adoption; based on administrative employment data and enterprise survey (Turkey, 2021-2024).
Infrastructure-driven AI adoption raises export intensity.
Instrumental-variable estimates linking infrastructure-driven adoption to firm export intensity using administrative and survey data (Turkey, 2021-2024).
Infrastructure-driven AI adoption raises labor productivity.
Instrumental-variable estimates where infrastructure-driven adoption is instrumented (IV) and linked to firm-level labor productivity measures; data from administrative records and enterprise survey in Turkey (2021-2024).
Improved connectivity (due to pipeline-driven fiber deployment) significantly increases AI adoption, particularly for software-intensive technologies and among small and medium-sized enterprises.
Causal inference using difference-in-differences estimates exploiting staggered pipeline expansion as variation in connectivity; sample drawn from administrative records and nationally representative enterprise survey (Turkey, 2021-2024).
AI adoption is concentrated among large firms and in regions with high-speed broadband and proximity to data centers, particularly for software-intensive and cloud-based applications.
Descriptive analysis using administrative data and a nationally representative enterprise survey from Turkey (2021-2024).
HAF-DS provides a scalable and adaptable solution for modern textile and PPE supply chains.
Author claim in conclusions indicating scalability and adaptability as properties of the proposed framework; supported implicitly by application to multiple datasets.
Coupling predictive forecasting with prescriptive optimization enhances both accuracy and efficiency in textile and PPE supply chains.
Summary conclusion drawn from the reported experimental improvements in forecast errors and operational metrics on textile and PPE datasets.
Service level rose from 95.5% to 97.8%.
Reported experimental operational metric (service level) improvement values under HAF-DS versus baseline (95.5% -> 97.8%).
Stockouts decreased by 27.5%.
Reported experimental operational metric indicating a 27.5% reduction in stockouts under HAF-DS compared to baseline.
Inventory cost decreased by 5.4%.
Reported experimental operational metric (inventory cost) showing 5.4% reduction under HAF-DS relative to baseline.
On the combined dataset, HAF-DS reduced Mean Absolute Percentage Error (MAPE) from 9.5% to 8.1%.
Reported experimental result on the combined dataset comparing MAPE of HAF-DS vs baseline (values given: 9.5% -> 8.1%).
On the combined dataset, HAF-DS reduced Root Mean Squared Error (RMSE) from 19.53 to 17.11 (12.4%).
Reported experimental result on the combined dataset comparing RMSE of HAF-DS vs baseline (values given: 19.53 -> 17.11, with percent reduction 12.4%).
On the combined dataset, HAF-DS reduced Mean Absolute Error (MAE) from 15.04 to 12.83 (14.7%).
Reported experimental result on the combined dataset comparing MAE of HAF-DS vs baseline (values given: 15.04 -> 12.83, with percent reduction 14.7%).
Experiments on textile sales and supply chain datasets show significant performance gains over statistical and deep learning baselines.
Empirical evaluation reported on textile sales and supply chain datasets with comparisons to statistical and deep learning baseline models (datasets described broadly; no sample sizes given).
The framework jointly minimizes forecasting error and operational cost through embedding-based feature representation and recurrent neural architectures.
Paper text describing joint objective (minimize forecasting error and operational cost) and the use of embedding-based features plus recurrent networks to accomplish this.
The optimization layer prescribes cost-efficient replenishment and allocation decisions (MILP).
Method description stating the use of a MILP optimization layer to produce replenishment/allocation decisions aimed at cost efficiency.
The LSTM captures temporal and contextual demand dependencies.
Methodological description asserting LSTM's role in modeling temporal and contextual dependencies within the forecasting module.
The paper proposes a Hybrid AI Framework for Demand-Supply Forecasting and Optimization (HAF-DS), which integrates a Long Short-Term Memory (LSTM)-based demand forecasting module with a mixed integer linear programming (MILP) optimization layer.
Paper description of the proposed framework (design/architecture). Reports integration of LSTM forecasting module and MILP optimization layer as the core contribution.
Dynamic combinations of AI and organizational structure can help managers overcome traditional trade-offs between scale and scope, opening pathways for scalable, cross-market expansion.
Managerial implication drawn from the paper's longitudinal case study of ByteDance; qualitative inference from observed organizational practices and AI deployment patterns.
AI transforms the scale–scope nexus from being a trade-off into a source of strategic advantage.
Synthesis and theoretical claim derived from longitudinal case study of ByteDance showing simultaneous scaling and diversification enabled by AI and organizational design.
AI reverses the conventional logic of the resource-based view: rather than valuable resources enabling diversification, diversification amplifies the value of resources.
Theoretical argument supported by the ByteDance case study; paper presents this as a theorized inversion based on observed patterns in the single-case study.
The value of AI learning transfer across domains is contingent on access to structurally related data that allow learning to transfer across domains.
Claim derived from the ByteDance longitudinal case study showing conditions for successful cross-domain AI transfer (qualitative evidence emphasizing data structure/relatedness).
AI evolves and improves through self-learning and cross-fertilization across domains, becoming increasingly valuable as learning accumulates.
Theoretical claim supported by longitudinal observations from the ByteDance case study (qualitative evidence from repeated AI deployments over time).
ByteDance leveraged AI and adaptive organizational design to scale rapidly and diversify across industries and markets without incurring rising costs or coordination complexity.
Longitudinal single-case (qualitative) study of ByteDance described in the paper; method reported as a longitudinal case study of one firm.
Meaningful human oversight of AI agents in knowledge work requires not improved post-hoc review mechanisms, but active participation in decisions as they are made.
Authors' conclusion drawn from the formative (N=8) and summative (N=16) studies and associated observations.
Users reported a sense of co-ownership over the resulting output.
Participant self-reports from the formative and/or summative studies (authors report users expressed co-ownership of outputs when participating in execution).
Users detected errors that post-hoc review would have failed to surface.
Empirical observation reported from the studies (authors report that active participation allowed users to detect errors that would be missed by post-hoc review).
Users identified their own intent reflected in the agent's actions.
Reported participant observations/self-reports from the formative (N=8) and/or summative (N=16) studies; claim presented as a finding of the evaluations.