Evidence (5539 claims)
Adoption
5539 claims
Productivity
4793 claims
Governance
4333 claims
Human-AI Collaboration
3326 claims
Labor Markets
2657 claims
Innovation
2510 claims
Org Design
2469 claims
Skills & Training
2017 claims
Inequality
1378 claims
Evidence Matrix
Claim counts by outcome category and direction of finding.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 402 | 112 | 67 | 480 | 1076 |
| Governance & Regulation | 402 | 192 | 122 | 62 | 790 |
| Research Productivity | 249 | 98 | 34 | 311 | 697 |
| Organizational Efficiency | 395 | 95 | 70 | 40 | 603 |
| Technology Adoption Rate | 321 | 126 | 73 | 39 | 564 |
| Firm Productivity | 306 | 39 | 70 | 12 | 432 |
| Output Quality | 256 | 66 | 25 | 28 | 375 |
| AI Safety & Ethics | 116 | 177 | 44 | 24 | 363 |
| Market Structure | 107 | 128 | 85 | 14 | 339 |
| Decision Quality | 177 | 76 | 38 | 20 | 315 |
| Fiscal & Macroeconomic | 89 | 58 | 33 | 22 | 209 |
| Employment Level | 77 | 34 | 80 | 9 | 202 |
| Skill Acquisition | 92 | 33 | 40 | 9 | 174 |
| Innovation Output | 120 | 12 | 23 | 12 | 168 |
| Firm Revenue | 98 | 34 | 22 | — | 154 |
| Consumer Welfare | 73 | 31 | 37 | 7 | 148 |
| Task Allocation | 84 | 16 | 33 | 7 | 140 |
| Inequality Measures | 25 | 77 | 32 | 5 | 139 |
| Regulatory Compliance | 54 | 63 | 13 | 3 | 133 |
| Error Rate | 44 | 51 | 6 | — | 101 |
| Task Completion Time | 88 | 5 | 4 | 3 | 100 |
| Training Effectiveness | 58 | 12 | 12 | 16 | 99 |
| Worker Satisfaction | 47 | 32 | 11 | 7 | 97 |
| Wages & Compensation | 53 | 15 | 20 | 5 | 93 |
| Team Performance | 47 | 12 | 15 | 7 | 82 |
| Automation Exposure | 24 | 22 | 9 | 6 | 62 |
| Job Displacement | 6 | 38 | 13 | — | 57 |
| Hiring & Recruitment | 41 | 4 | 6 | 3 | 54 |
| Developer Productivity | 34 | 4 | 3 | 1 | 42 |
| Social Protection | 22 | 10 | 6 | 2 | 40 |
| Creative Output | 16 | 7 | 5 | 1 | 29 |
| Labor Share of Income | 12 | 5 | 9 | — | 26 |
| Skill Obsolescence | 3 | 20 | 2 | — | 25 |
| Worker Turnover | 10 | 12 | — | 3 | 25 |
Adoption
Remove filter
The paper provides concrete, regulation-inspired policy examples (e.g., content prohibition, sensitive data exfiltration) showing how they map into the Policy function.
Worked, illustrative examples included in the paper mapping regulatory constraints to the Policy(agent_id, partial_path, proposed_action, org_state) formalism.
Runtime policy evaluation can intercept, score, log, allow/modify/block actions, and update organizational state as part of an agent's execution loop (reference implementation architecture).
Reference implementation design described in the paper (runtime policy evaluator hooks, logging, enforcement actions); architectural reasoning and pseudo-workflows provided; no production deployment data.
Policies can be formalized as deterministic functions p_violation = Policy(agent_id, partial_path, proposed_action, org_state) that return a probability or score of violation for a proposed next action.
Formal definition and mapping in the paper; worked examples showing how regulatory-style constraints map into this function; no large-scale empirical validation.
Effective governance for agentic LLM systems requires treating the execution path as the central object and performing runtime evaluation of proposed next actions given the partial path.
Theoretical argument and formal proposal of runtime policy evaluator that takes (agent_id, partial_path, proposed_action, org_state) and returns a violation probability; reference architecture described; illustrative examples.
Multiple off-the-shelf vision-language models (closed-source and open-source) representative of current state-of-the-art architectures were benchmarked.
Paper reports experiments across a mix of closed-source and open-source VLMs; exact model names provided in the released materials.
Evaluation targets include correctness, consistency, and update efficacy, operationalized via quantitative metrics (accuracy, consistency rates, update success rate).
Methods section describing evaluation metrics and how correctness, consistency, and update efficacy are measured across experiments.
A curated set of time-sensitive factual items (e.g., officeholders, company statuses, recent awards/results) was used to construct the benchmark.
Benchmark composition description listing categories of time-sensitive facts and methodology for curation of items used in experiments.
The authors release the V-DyKnow benchmark, code, and evaluation data for community use.
Statement in paper and accompanying release materials indicating benchmark, code, and evaluation data are publicly available.
V-DyKnow is a benchmark specifically designed to evaluate time-sensitive factual knowledge in vision-language models across both text and image modalities.
Release and description of the benchmark in the paper: curated set of time-sensitive factual items, paired multimodal stimuli (text + images), input perturbations, and evaluation scripts. Methodological description of benchmark composition and tasks.
Ethical handling: the study involved sensitive material (self-harm, trauma) and authors applied validation and careful handling consistent with research ethics.
Ethics section and methods describing sensitivity of material and precautions taken in data handling and validation.
Selected coded items (for example, suicidal messages) were validated by the authors to increase reliability of certain critical annotations.
Methods section describing validation procedures applied to selected items such as suicidal ideation.
The authors developed and applied a manual codebook of 28 behavioral/phenomenological codes (e.g., delusional thinking, suicidal ideation, chatbot sentience claims, romantic interest) across the full corpus.
Method section describing construction of a 28-code inventory and manual coding applied to entire dataset.
The parallel associative scan enables the reductions required by Newton-style updates across time steps, thereby enabling parallelism across sequence length.
Algorithmic construction and implementation details in the thesis showing how associative scan operations aggregate intermediate Jacobian/ update information across time; examples provided in implementation section.
The thesis proves linear convergence rates for a family of fixed-point/Newton-like solvers, with rates depending on approximation accuracy and stability properties of the chosen method.
Mathematical proofs and convergence theorems provided in the theoretical analysis section establishing linear rates under stated assumptions (bounds on approximation error, stability metrics).
Evaluation of dynamical systems can be cast as solving a system of nonlinear equations, enabling parallel solution methods.
Methodological framing and derivation in the thesis showing recurrent updates and Markov transitions can be represented as a global nonlinear root-finding problem; algorithmic constructions follow from this representation.
Explicit enforcement of signal constraints in DeePC provides a safety/operational advantage over many pure learning approaches that do not explicitly enforce hard constraints.
Algorithmic formulation includes constraints in the optimization; paper contrasts this with unconstrained learning-based controllers and demonstrates constrained, feasible actuation in simulation.
DeePC can compute traffic-light actuation sequences that respect hard operational and safety constraints (e.g., phasing, minimum/maximum green times).
Formulation of DeePC as a constrained optimization problem in the paper with explicit constraint terms for signal phasing and safety; implemented in simulation experiments where constraints are enforced in the controller optimization.
Reframing urban traffic dynamics with behavioral systems theory allows system evolution to be learned and predicted directly from measured input–output data (no explicit model identification).
Theoretical exposition in the paper showing that traffic trajectories can be represented as linear combinations of past measured trajectories via Hankel/data matrices; used as the basis for predictive control (DeePC).
Applying DeePC yields measurable improvements in system-level outcomes (reduced total travel time and CO2 emissions) in a very large, high-fidelity microscopic simulation of Zürich.
Simulation experiments in a city-scale, high-fidelity microscopic closed-loop simulator of Zürich comparing DeePC-controlled signals against baseline controllers (e.g., fixed-time or standard adaptive schemes); reported reductions in aggregated metrics (total travel time and CO2 emissions).
A model-free traffic control approach (DeePC) can steer urban traffic via dynamic traffic-light control without building explicit traffic models.
Algorithmic/theoretical development (behavioral systems theory + DeePC) and controller-in-loop experiments in a high-fidelity microscopic closed-loop simulator of Zürich demonstrating closed-loop control using only input–output trajectory data (Hankel matrices) rather than parametric model identification.
The model weights will be open (open-weight release) to support European sovereignty and adoption.
Authors state intent to publish open weights and position the model as an open-weight European alternative; the summary reports this as a declared objective. The paper likely includes a licensing/availability statement.
Calibration data must be representative of deployment data to preserve conformal statistical guarantees in practice.
Theoretical requirement of exchangeability for conformal guarantees combined with empirical results where mismatched calibration caused guarantee violations or degraded factuality.
The paper introduces informativeness-aware metrics to measure task utility under conformal filtering, going beyond pure factuality rates.
Methodological contribution described: new metrics that penalize vacuous outputs and quantify retained task utility after filtering.
Decomposing generated outputs into atomic claims and calibrating a verifier score threshold on held-out data yields a statistically valid guarantee (under exchangeability) that claims passing the threshold meet a target factuality level.
Method description and theoretical use of conformal calibration applied to per-claim scores, with held-out calibration set used to set the threshold; conforms to standard conformal prediction methodology presented in the paper.
Conformal factuality provides distribution-free statistical guarantees for claim-level correctness in retrieval-augmented LLM outputs.
The paper applies conformal calibration to atomic claims: decompose outputs into atomic claims, score each claim with a verifier, and calibrate a score threshold on held-out (exchangeable) data to guarantee a target claim-level factuality rate. This is a theoretical property of conformal methods described and implemented in the paper.
Traditional machine-learning baselines were included for comparison in the benchmarks.
Paper explicitly states that traditional ML baselines were used alongside TSFMs in benchmarking experiments. The summary does not list which baselines or their quantitative results.
The dataset sampling resolution is at the millisecond level, enabling forecasting horizons from 1 step (100 ms) up to 96 steps (9.6 s).
Paper states sampling resolution is millisecond-level and defines forecasting tasks spanning 1 to 96 steps (100 ms to 9.6 s). This is a methodological description rather than an experimental metric.
Introduces a new millisecond-resolution dataset of wireless channel and traffic-condition measurements from an operational 5G deployment.
Paper describes collection of operational 5G telemetry at millisecond sampling resolution; dataset is presented as a novel domain addition to TSFM pretraining corpora. Exact number of records/sessions not specified in the provided summary.
Under pathological label heterogeneity (mutually exclusive local labels) FederatedFactory restores CIFAR-10 classification accuracy from a collapsed baseline of 11.36% to 90.57%.
Empirical experiment reported on CIFAR-10 configured as a pathological heterogeneity stress test; paper reports baseline collapsed accuracy (11.36%) and FederatedFactory result (90.57%). (Specific sample sizes / client counts not provided in the summary.)
A single communication round of generative-module exchange suffices for clients to synthesize class-balanced datasets locally and align their training data.
Paper reports a single exchange of generative modules across clients (one communication round) and uses that to synthesize a globally class-balanced training set at each client; experiments (CIFAR-10, MedMNIST, ISIC2019) are run under this one-round regime.
Convergence of the three complementary methods (lexical, paraphrase, behavioral) strengthens confidence that contamination is real and systematically inflates scores.
Triangulation across Experiment 1 (lexical detection on public corpora), Experiment 2 (paraphrase robustness on 100-question subset), and Experiment 3 (TS‑Guessing on all items); consistent patterns observed across methods.
All 13 surveyed generative systems report addressing syntactic validity (Layer 1).
For each of the 13 systems the review reports syntactic/parse/compile checks or token-level validity tests under Layer 1 in the systematic application of the evaluation framework.
BenchPreS can be used as an evaluative tool for mechanism designers and regulators to measure and compare models' context‑sensitivity to guide incentives, penalties, or certification regimes.
Methodological claim about the benchmark's applicability: BenchPreS produces MR and AAR metrics that can be used for comparisons; paper suggests use in policy/design contexts.
BenchPreS provides a benchmark and evaluation protocol that systematically varies stored user preference, interaction partner (self vs third party), and normative requirement to assess appropriate suppression or application of preferences.
Dataset construction and evaluation procedure described: scenario generation varying preference, partner, and normative appropriateness; MR and AAR computed across the scenario set.
Historical transitions in standard work hours (e.g., six-day to five-day week) show that phased implementation, collective bargaining, and complementary policies can make work-time reductions feasible and economically beneficial.
Historical analyses and case studies of past industrialized-country workweek transitions cited in the synthesis; evidence drawn from historical institutional records and prior economic histories rather than a unified econometric analysis.
The paper advances a replicable interdisciplinary synthesis method and provides a simulated dataset and transparent protocols enabling other researchers to adapt the approach.
Methods section detailing systematic literature search protocols (ACM/IEEE/Springer, 2020–2024), inclusion criteria, simulation parameterization for the cross-sectoral dataset (seven industries, 2020–2024), and stated reproducibility materials.
AI adoption is strongly associated with workforce skill transformation (reported correlation r = 0.71).
Correlational analysis reported in the paper using the simulated cross-sectoral dataset that mirrors employment trends across seven industries (Manufacturing, Healthcare, Finance, Education, Transportation, Retail, IT Services) over 2020–2024. This corresponds to sector-year observations (7 sectors × 5 years = 35 observations) and is triangulated with findings from a systematic literature synthesis (ACM, IEEE, Springer publications 2020–2024).
The evaluation compared models on multiple metrics (accuracy, precision, recall, F1, AUC) across repeated trials and cross-company tests, and reported gains for AI methods across these metrics.
Evaluation protocol described: repeated trials, cross-validation, holdout sets, cross-company tests; reported performance improvements for AI models on the listed metrics.
Ensemble methods and deep learning models show the largest and most consistent improvements in predictive performance relative to classic statistical models.
Aggregate results across repeated trials and evaluation metrics indicate Random Forests and Gradient Boosting (ensembles) and deep neural networks outperform linear/logistic regression and other baselines on the publicly available datasets used.
Modern AI-driven prediction methods (especially ensemble models and deep neural networks) systematically outperform traditional statistical approaches at predicting job performance in publicly available workforce datasets.
Direct model comparison reported in the paper: baseline statistical models (linear/logistic regression) versus machine learning models (Random Forest, Gradient Boosting, SVM, deep neural networks) evaluated on multiple publicly available workforce datasets using cross-validation and holdout sets; performance reported on accuracy, precision, recall, F1, and AUC across repeated trials.
Research priorities include rigorous real-world trials assessing patient outcomes, cost-effectiveness, and labor impacts; comparative studies of integration strategies; measurement of long-run workforce effects; and development of standard metrics and monitoring frameworks.
Explicit recommendations from the narrative review based on identified gaps: scarcity of RCTs, economic analyses, and long-term workforce studies.
Economists and researchers should measure organizational mediators (governance, mentoring practices, learning processes) alongside AI adoption and use empirical designs such as difference-in-differences with phased rollouts, randomized mentoring/training interventions, matched employer–employee panels, and IV exploiting exogenous shocks to innovation backing to identify causal effects.
Methodological recommendations and proposed empirical designs contained in the paper; no implementation or empirical results reported.
The integrated framework links multi-level outcomes: micro (individual skills, task performance), meso (team coordination, workflows), and macro (organizational strategy, innovation, productivity) effects to adaptive structuration processes and affordance actualization.
Framework specification and theoretical mapping across levels in the conceptual paper; no empirical validation or sample.
The paper develops a conceptual framework that integrates Adaptive Structuration Theory (AST) and Affordance Actualization Theory (AAT) to explain how effective human–AI collaboration can be structured within organizations.
Conceptual/theoretical synthesis and literature integration combining AST and AAT streams; no original empirical data or sample reported (theoretical development).
Reward shaping at the assignment layer enables an explicit trade-off between diagnostic accuracy and human labor by incorporating penalties for human involvement.
Methodology section describing reward shaping and experimental comparisons showing different accuracy/human-effort trade-offs (results reported in paper; exact experimental details not provided in the summary).
Masked reinforcement learning techniques constrain or mask action spaces, reducing exploration over huge symptom/action spaces.
Paper describes use of masked RL to limit action options during training and execution; used in both assignment and execution layers (methodological claim supported by algorithmic description and experiments).
The upper layer ('master') learns turn-by-turn human–machine assignment using masked reinforcement learning with reward shaping to balance accuracy and human cost.
Methodological description in the paper and empirical results from experiments using masked RL and reward-shaped objectives at the assignment layer (implementation and experimental setup reported; dataset/sample size not specified in summary).
The paper advances augmentation debates by articulating the leader’s practical role when decision lead‑agency shifts between humans and AI and by detailing systemic HR changes needed to sustain performance, legitimacy and well‑being.
Stated contribution of the conceptual synthesis comparing existing augmentation and leadership literatures and providing an HR‑focused framework; descriptive of the paper's intellectual contribution.
Core practice 4 — Embed governance: make accountability, bias testing, privacy safeguards, audit trails, escalation thresholds and human oversight explicit and routine.
Prescriptive governance practice grounded in literature on algorithmic accountability and risk management and in practitioner examples; presented without original empirical validation.
Core practice 3 — Manage the human–AI relationship: build adoption, psychological safety and calibrated trust; address automation anxiety and misuse.
Framework recommendation synthesizing organizational‑psychology and technology adoption literature plus practitioner observations; not tested empirically in the paper.