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 |
SECaaS gives firms access to specialized expertise and up-to-date threat feeds they might not maintain internally.
Vendor offerings and industry analyses; surveys reporting reliance on external expertise and threat intelligence services.
SECaaS provides scalability and rapid deployment of new defenses compared with building equivalent in‑house capabilities.
Industry reports and vendor benchmarks on deployment times and scalability; case studies and surveys of firm experiences (no single pooled sample size reported).
Processing and using 3D volumetric data requires substantial storage and GPU/TPU compute, creating demand for cloud compute services and managed ML platforms.
Authors note the resource requirements of 3D volumetric data processing as a practical consideration; general technical knowledge supports this claim though no resource-consumption measurements are provided in the paper.
The dataset and its standardization are intended to support automated segmentation, landmarking, feature extraction, and benchmarking for computer-vision and ML methods on biological 3D data.
Authors describe the acquisition and metadata design as 'automation-ready' and suitable for downstream automated/ML workflows.
Phenomic (3D scans) data are linked/paired to ongoing genome sequencing projects to create multimodal phenome–genome resources.
Paper reports links to genome projects where available and describes pairing of phenomic data with genome sequencing efforts.
Sampling is global and broadly covers ant phylogeny.
Authors state global sampling and intended phylogenetic breadth; taxonomic counts across genera/species presented to support breadth.
The field needs standard evaluation metrics and benchmarks for XAI in EEG; such standards will reduce information asymmetry, lower transaction costs, and facilitate market growth.
Recommendation motivated by recurring heterogeneity in evaluation practices and lack of reproducible metrics across reviewed studies.
Developing robust, clinically validated XAI increases upfront R&D costs but can accelerate adoption, reduce downstream monitoring costs, and enable higher reimbursement.
Economic reasoning and cost–benefit projection offered in the review; not backed by quantified cost or reimbursement data in the paper.
Funding and commercial interest should prioritize robustness, clinical validation, and domain-aligned XAI development rather than focusing solely on accuracy benchmarks.
Policy/recommendation arising from identified evaluation and validation gaps in the literature.
Explainability materially affects the economic value and adoption of EEG AI tools: transparent and clinically credible models are more likely to be adopted, reimbursed, and integrated into care pathways, increasing market size.
Economic argument and synthesis presented in the paper; reasoning links explainability to clinician/regulatory trust and reimbursement potential (no direct market-data empirical test provided).
Clinical and research EEG applications require explanations as much as raw predictive performance to enable clinician trust, regulatory acceptance, and safe deployment.
Argument and rationale presented in the paper drawing on regulatory and clinical adoption considerations discussed in the literature (no single quantified empirical test provided).
XAI techniques have become central to EEG analysis because interpretability is necessary for clinical adoption.
Synthesis/argument in the review based on surveying contemporary EEG-AI literature and the stated motivation that clinicians and regulators require explanations alongside performance; no single empirical study cited for centrality.
Legitimacy economies matter: public trust and stakeholder legitimacy influence willingness to share data and participate in collaborative research, with direct economic consequences for data‑intensive innovation.
Argument grounded in coded references to stakeholder legitimacy in the documents and theoretical literature linking legitimacy/trust to participation; the paper does not present empirical measures of trust or sharing behavior.
Extending civil‑rights liability to vendors provides a clear regulatory signal that discrimination risks in algorithmic systems are materially consequential, which could spur broader governance practices across AI product markets.
Policy argument about regulatory signaling effects; theoretical, not empirically tested in the Article.
Treating vendors as recipients would internalize externalities by shifting responsibility for discriminatory harms from schools onto EdTech firms, aligning private incentives with nondiscriminatory product design.
Policy and economic reasoning (theoretical argumentation about incentives), not empirical measurement.
Most EdTech vendors can be brought within the scope of federal financial assistance rules under three theories: (1) direct recipients (federal contracts/grants), (2) intended indirect recipients (intended beneficiaries of pass‑through federal funds), and (3) controllers of a federally funded program (firms exercising controlling authority).
Close reading of statutory language and administrative/judicial precedent applied to procurement and control relationships; doctrinal reasoning and illustrative examples (no empirical sampling).
Treating EdTech vendors as recipients would make the companies themselves directly liable for discrimination harms in schools.
Statutory interpretation of nondiscrimination obligations (Title VI/Title IX/Section 504) and precedent about recipient obligations; doctrinal reasoning and illustrative case law.
EdTech companies that provide tools like automated grading or plagiarism detection can — and should — be treated as “recipients” of federal financial assistance under existing federal education civil‑rights statutes.
Doctrinal legal analysis and policy argumentation drawing on statutory text, administrative guidance, and illustrative case law (no empirical dataset or sample size).
Policy interventions (public investment in open models/data, licensing regimes, standards, workforce retraining) can influence equitable diffusion and mitigate concentration risks.
Policy recommendations grounded in economic and governance analysis; not empirically tested within the paper.
Markets may demand certification, auditing services, and standardized benchmarks for AI-driven experimental systems, creating potential third-party validation/compliance markets.
Economic and policy argument about demand for assurance services in response to risk; no market-evidence or adoption rates provided.
Open-source LLMs and community datasets could serve as counterweights to concentration and influence pricing, innovation diffusion, and access.
Observation of open-source effects in the broader AI ecosystem and policy argument; no empirical evidence specific to microscopy domain adoption provided.
Experimental data, protocol metadata, and provenance logs will become critical assets for fine-tuning models and benchmarking, and ownership/sharing arrangements will affect competitive dynamics.
Conceptual argument about the role of data for model training and benchmarking; supported by analogies to other data-driven industries, no direct empirical evidence in microscopy.
Firms that combine instrumentation with proprietary LLM stacks or exclusive datasets could capture larger economic rents, encouraging vertical integration and platformization.
Argument based on network effects and data-as-asset logic; no firm-level empirical evidence in microscopy provided.
Value will shift toward software, data infrastructure, and integration layers relative to hardware; microscopes may become platforms that generate ongoing subscription or model-related revenues.
Market-structure reasoning and analogies to platformization trends in other industries; no market-share or revenue data presented.
LLM-driven orchestration could lower the marginal cost and time per experiment by automating protocol design, instrument tuning, and analysis, thereby raising lab-level productivity.
Theoretical economic reasoning and analogy to automation benefits; no randomized trials or empirical throughput measurements provided.
LLMs can integrate contextual knowledge, experimental intent, and multi-step reasoning to coordinate sensors, actuators, and analysis tools.
Conceptual argument supported by literature on LLM context modeling and tool orchestration; some proof-of-concept integrations mentioned in related work but no systematic evaluation or sample sizes.
Potential applications of LLM orchestration in microscopy include conversational microscope control, adaptive experimental workflows, automated data-processing pipelines, and hypothesis generation/exploratory analysis.
Illustrative use cases and system-architecture proposals synthesized from related work and authors' analysis; these are proposed applications rather than empirically demonstrated at scale.
LLMs offer emergent capabilities in reasoning, abstraction, and tool coordination that make them natural interfaces between users and complex experimental systems.
Review of foundation-model literature demonstrating emergent reasoning and tool-use behaviors and conceptual arguments about fit with instrument orchestration; no experimental validation in microscopy contexts provided.
LLMs enable conversational control and multi-step workflow supervision that go beyond task-specific ML models.
Argument based on documented emergent LLM capabilities (reasoning, tool use) and illustrative prototypes from the literature; no controlled comparisons to task-specific ML models provided.
Large language models (LLMs) can serve as cognitive and orchestration layers for modern optical microscopy, bridging experiment design, instrument control, data analysis, and knowledge integration.
Conceptual synthesis and perspective drawing on recent literature about LLM capabilities, computational imaging, and illustrative proof-of-concept integrations reported in related work; no controlled experimental evaluation or quantitative sample size reported.
Research priorities for economists should include assembling integrated datasets (strain performance, TEA/LCA, patents/funding, compute/data assets) and building scenario TEA/LCA models under varying yield/productivity and regulatory assumptions.
Prescriptive recommendation based on identified gaps in the literature and the heterogeneity of existing case studies; justified by the review’s mapping of missing cross‑disciplinary datasets and methodological heterogeneity.
High‑throughput screening, microfluidics, and automated lab infrastructure materially increase the throughput of DBTL cycles and reduce time per iteration.
Aggregate experimental reports demonstrating use of droplet microfluidics, automated liquid-handling, and high-throughput assays enabling larger combinatorial libraries to be tested more rapidly in several published studies.
Integration of synthetic chemistry with engineered biology enables hybrid chemo‑bio manufacturing routes that can fill gaps where biological access alone is insufficient.
Examples in the review where biological steps produce advanced intermediates that are then completed by chemical steps (or vice versa), improving overall route efficiency or enabling transformations difficult for either domain alone.
Cell‑free synthetic platforms provide rapid prototyping and a decoupled route for bioproduction that can shorten design timelines.
Reports of cell-free pathway prototyping enabling quick testing of enzyme combinations, kinetics, and pathway flux before cellular implementation; experimental demonstrations at bench scale described in reviewed literature.
Machine learning and AI methods (sequence-to-function, phenotype prediction) significantly accelerate DBTL cycles and improve hit rates in strain optimization.
Cited studies using ML models to predict enzyme activity, rank pathway variants, and prioritize constructs for experimental testing; reported reductions in screening burden and improved selection of productive variants across several examples.
Biological production routes can achieve higher product specificity (e.g., for complex stereochemistry) than many traditional chemical syntheses for certain targets.
Case studies and examples where biosynthetic pathways produce stereochemically complex natural products and chiral intermediates that are difficult or multi‑step to access by classical chemistry; comparisons in the review between biosynthetic access and synthetic-chemistry challenges.
Experimental results on ICML and ACL 2025 abstracts produced coherent clusters that map to problem formulations, methodological contributions, and empirical contexts.
Reported experiments on ICML and ACL 2025 abstracts with qualitative analyses and cluster-coherence evaluations showing clusters aligning with problem types, methods, and empirical settings. (Exact counts/metrics not provided in summary.)
The framework treats an LLM as a fixed semantic inference operator guided by structured soft prompts to normalize abstracts into compact semantic representations that reduce stylistic variability while preserving conceptual content.
Described pipeline step: application of an LLM with structured soft prompts to transform raw abstracts into normalized semantic representations; qualitative claims about reduced stylistic noise and preserved core concepts (no quantitative metrics reported in summary).
Prompt-driven semantic normalization using large language models, combined with geometric (embedding + density-based clustering) analysis, provides a scalable, model-agnostic unsupervised framework that discovers coherent, human-interpretable research themes in large scientific corpora.
Method implemented and demonstrated on ICML and ACL 2025 abstracts using: (1) LLM-based semantic normalization with structured soft prompts; (2) embedding of normalized representations; (3) density-based clustering; evaluation via qualitative and cluster-coherence analyses. (Number of abstracts not specified in provided summary.)
Practical outputs include open-source tooling (Neural MRI), standardized reporting formats (M-CARE), and clinical-style indices for behavioral profiling released alongside the paper.
Authors report open-source toolkit and standardized instruments in the paper (implementation and release claimed).
Combined imaging (Neural MRI) and profiling can localize dysfunctions in models and support predictive claims about future model behavior, as shown in the case-based demonstrations.
Four clinical case studies plus analyses within the Agora-12 experimental domain demonstrating localization and predictive uses of imaging + profiling.
A behavioral genetics approach decomposes variance in agent behavior into heritable (Core) versus environmental and Shell-level influences, formalized in the Four Shell Model.
Analytical method described and applied to the Agora-12 dataset (variance-decomposition analyses analogous to behavioral genetics).
Neural MRI was validated on four clinical case studies that showcase imaging, comparison, localization, and prediction capabilities.
Case-based demonstrations reported in the paper (n = 4 clinical cases used to validate the toolkit and diagnostic pipeline).
The Four Shell Model (v3.3) explains model behavior as emergent from interactions between a Core and multiple Shell layers.
Theoretical formalization (behavioral-genetics-style framework) plus empirical grounding using analyses from the Agora-12 program (see supporting experiments).
On the supply side, digital platforms reduced intermediaries and enabled direct, flexible gigs, increasing platform-mediated cultural work.
Evidence from inferred measures of platform-mediated activity and interaction effects between digital infrastructure indicators and treatment status on employment outcomes in the DID models (280 cities, 2008–2021).
On the demand side, combined government funding and digital channels boosted cultural consumption, increasing labor demand.
Analysis of government funding/procurement measures and digital channel proxies interacting with employment outcomes in the city-level panel; DID identification with fixed effects across 280 cities (2008–2021).
Fiscal-Digital Synergy: government funding combined with digital platforms amplified cultural demand and disintermediated supply, driving employment effects.
Mechanism tests linking fiscal transfers/procurement variables and measures of digital infrastructure/usage to employment outcomes within the DID framework; interaction/heterogeneity analyses showing larger effects where digital infrastructure and procurement intensity are higher (280 cities, 2008–2021).
Growth manifested through flexible, platform-enabled labor and government-procured gigs rather than firm-based expansion (termed 'De-organized Growth').
Inferred platform-mediated work activity and analysis of government procurement patterns in the city-panel data; mechanism tests linking increases in government funding/procurement and proxies for platform-mediated activity to cultural employment gains (2008–2021, 280 cities).
Firms, regulators, and asset managers can operationalize complaint-topic and sentiment monitoring for early risk detection, prioritizing investigations, and as complementary features in forecasting or factor models.
Practical takeaway informed by empirical results showing complaint features predict short-term returns and topic-specific signals indicate reputational/operational risk; recommendations provided but no deployed field trial.
Including complaint-derived features in supervised machine-learning models improves out-of-sample prediction of abnormal returns relative to models using standard financial predictors alone.
Supervised learning experiments compare baseline financial-predictor models to augmented models that add complaint volume, topic prevalences (LDA), and aggregated VADER sentiment; augmented models show higher out-of-sample predictive accuracy for abnormal returns.