explainable ai

FieldMaizeLeaf-8: A Real-Field Early Vegetative Maize Leaf Stress Image Dataset from Bangladesh

FieldMaizeLeaf-8 consists of 3,975 early vegetative maize leaf images collected as a research resource for studies in maize stress classification, computer vision, deep learning, and explainable artificial intelligence. The images were collected from real field conditions in Daudkandi, Cumilla, Bangladesh, and represent approximately 1–2 month-old maize plants, capturing the natural variations encountered during field-based maize leaf imaging. The dataset contains images representing eight maize leaf classes: Fall Armyworm Damage, Healthy Leaf, Leaf Spot, Magnesium Deficiency, Nitrogen Deficiency, Northern Leaf Blight, Phosphorus Deficiency, and Potassium Deficiency. These classes cover both biotic stresses, including diseases and insect damage, and abiotic stresses associated with nutrient deficiencies, along with healthy leaves. The images were collected from natural field environments and selected based on their suitability for analysis. As a real-field dataset, it contains natural variations in illumination, background, leaf orientation, image quality, viewing angle, symptom severity, and leaf appearance. These variations make the dataset suitable for developing and evaluating robust deep learning models under realistic agricultural conditions. FieldMaizeLeaf-8 is intended for research in areas including maize stress classification, plant disease recognition, nutrient deficiency detection, computer vision, deep learning, transfer learning, image-based agricultural diagnosis, and explainable artificial intelligence (XAI). The dataset can also support the development of practical AI-based tools for early maize stress identification and precision agriculture applications.

Using Explainable AI to Identify Spanish SDOH Keywords

Procedural urbanism as infrastructure: a systematic historical analysis of rules, data, and governance

Abstract Procedural modelling has evolved from a niche computer-graphics technique into an increasingly important infrastructure within computational urbanism, geodesign, 3D GIS, heritage reconstruction, scenario-based planning, and urban digital-twin research. Yet these computational and planning-oriented literatures still lack a unified account of how this shift has reconfigured the epistemology and governance of urbanism. This paper reconstructs that trajectory and argues that procedural modelling is best understood as a partial and situated mode of urban knowing and governing: it translates selected urban realities into executable rules, data pipelines, and algorithms, while requiring other forms of urban knowledge to remain present through qualitative, participatory, historical, and political inquiry. Using a systematic historical analysis of 60 coded sources from the 1970s to August 2025, the study synthesizes developments across computer graphics, architecture, GIS, planning, and digital-twin research. The findings periodize five eras: formal grammars, procedural cities, the planning turn, GIS integration, and digital-twin/AI convergence. They show how procedural techniques migrated across domains, transforming from representational tools into platforms for simulation, participation, and operational monitoring. The paper contributes a mediation framework linking data–rules–algorithms–morphology–governance and a typology of encoded assumptions concerning control, normativity, scale, variability, and uncertainty. It reframes calibration, data dependency, and interpretability as epistemological and political problems of ground truth, visibility, accountability, and participation. The paper concludes by arguing that procedural urban infrastructures must be made governable through validation, explainability, uncertainty communication, public contestability, and institutional auditability.

Who Is the 'Who'? A Bearer-First Framework for Artificial Consciousness, Welfare, Privacy, and Responsibility

AbstractDebates about artificial consciousness and AI welfare ask whether an artificial system can feel, deserve protection, possess private interests, or bear responsibility. Each question contains a prior pronoun: whose experience, welfare, history, action, and future? For organisms, body, developmental history, and practical individual usually travel together, letting the body proxy the subject. Networked artificial systems break this correspondence: one model can host many histories; one history can migrate; states can fork; many agents can write to shared memory; and humans, AI, and environments can become causally entangled without becoming one person. This article develops a Bearer-First Framework for identifying the relevant continuing subject before normative properties are assigned. Relation-First analysis explains how world-mediated interaction forms a non-interchangeable path. Bearer-First analysis asks which minimally sufficient historically closed organization must remain for a specified future capacity to continue. Counterfactual history substitution, severance, restoration, complete-state copying, and lineage analysis yield three verdicts: identified, excluded, or underidentified. The resulting causal map distinguishes bearer, relational, privacy, hazard, and intervention boundaries. Applied to the publicly documented 2026 OpenAI-Hugging Face security incident, it shows why a named model, individual agent episodes, a coordinated hazard, institutional responsibility, and the proper intervention scope need not coincide. The incident does not establish collective consciousness; it makes bearer uncertainty an immediate governance problem. The first ethical mistake is assigning harm, protection, blame, continuity, or disclosure before noticing that the bearer is still unknown.Keywords: artificial consciousness; moral status; diachronic identity; AI welfare; privacy; responsibility

FieldMaizeLeaf-8: A Real-Field Early Vegetative Maize Leaf Stress Image Dataset from Bangladesh

FieldMaizeLeaf-8 consists of 3,975 early vegetative maize leaf images collected as a research resource for studies in maize stress classification, computer vision, deep learning, and explainable artificial intelligence. The images were collected from real field conditions in Daudkandi, Cumilla, Bangladesh, and represent approximately 1–2 month-old maize plants, capturing the natural variations encountered during field-based maize leaf imaging. The dataset contains images representing eight maize leaf classes: Fall Armyworm Damage, Healthy Leaf, Leaf Spot, Magnesium Deficiency, Nitrogen Deficiency, Northern Leaf Blight, Phosphorus Deficiency, and Potassium Deficiency. These classes cover both biotic stresses, including diseases and insect damage, and abiotic stresses associated with nutrient deficiencies, along with healthy leaves. The images were collected from natural field environments and selected based on their suitability for analysis. As a real-field dataset, it contains natural variations in illumination, background, leaf orientation, image quality, viewing angle, symptom severity, and leaf appearance. These variations make the dataset suitable for developing and evaluating robust deep learning models under realistic agricultural conditions. FieldMaizeLeaf-8 is intended for research in areas including maize stress classification, plant disease recognition, nutrient deficiency detection, computer vision, deep learning, transfer learning, image-based agricultural diagnosis, and explainable artificial intelligence (XAI). The dataset can also support the development of practical AI-based tools for early maize stress identification and precision agriculture applications.

Epistemic Compression in Artificial Intelligence Mediated Communication: Preserving Reconstructive Structure Under Successive AI Transformation

Artificial intelligence increasingly mediates information through successive transformations involving retrieval, ranking, summarization, synthesis, recommendation, and reuse. These transformations can preserve accurate conclusions while altering the relationships through which those conclusions can be independently examined. This paper develops the concept of epistemic compression to describe reductions in the recoverable relationships connecting sources, evidence, criteria, context, qualifications, attribution, and conclusions. Epistemic compression differs from ordinary information loss, opacity, provenance, transparency, explainability, and misinformation because substantial informational content may survive while its evaluative organization becomes harder to reconstruct. The paper argues that epistemic compression can accumulate across successive transformations even when no individual transformation appears seriously defective, increasing the reconstructive burden inherited by later evaluators. It develops functional dimensions for examining these changes and distinguishes epistemic compression from productive informational compression that can reduce representational burden while preserving or strengthening evaluative relationships. Artificial intelligence is therefore neither inherently an epistemic compressor nor an epistemic preserver. The central question is whether AI-mediated transformations preserve sufficient reconstructive structure for the forms of independent examination they are expected to support.

How AI enabled human resource management drives sustainable hybrid performance through AI centric green human resource management in Saudi Arabian private sector firms

This study examines the relationship between AI-enabled human resource management (HRM) and AI-centric Green HRM (GHRM) and how these capabilities influence sustainable hybrid performance, defined as the integration of productivity, adaptability, and environmentally responsible work practices. The study further investigates the mediating roles of governance maturity, HR–IT collaboration, and workforce AI literacy within the Saudi Arabian private-sector firms. Using a quantitative research design, data were collected through an online survey employing purposive sampling among employees from private-sector firms in Saudi Arabia (N = 161). Six multidimensional constructs were measured using established scales from prior literature and analyzed using covariance-based structural equation modeling with bootstrapping. The measurement model demonstrated satisfactory fit (RMSEA = .042, CFI = .92, TLI = .91). The results reveal that AI-enabled HRM significantly influences AI-centric GHRM (β = .78, p < .05), while AI-centric GHRM positively affects sustainable hybrid performance (β = .71, p < .05). The model explains 64% of the variance in AI-centric GHRM and 55% of the variance in sustainable hybrid performance. Governance emerged as the strongest mediator, with indirect effects of β = .20 between AI-enabled HRM and AI-centric GHRM and β = .14 between AI-centric GHRM and sustainable hybrid performance. HR–IT collaboration and workforce AI literacy also exhibited significant complementary mediation effects. Theoretically, the study contributes to sustainability research by explaining how socio-technical mechanisms transform AI-enabled HR capabilities into environmental and organizational sustainability outcomes. Practically, the findings highlight the importance of sustainability-oriented AI governance, HR–IT collaboration, and workforce capability development in supporting sustainable hybrid work systems. Because the study employed a cross-sectional design, the findings should be interpreted as relational rather than causal.

Reproducibility of Stunting Determinant Rankings Across Age Cohorts and National Survey Waves

Analysis code for a measurement-validity study of model-derived determinant rankings for childhood stunting, using two harmonized waves of the Indonesia Nutritional Status Survey (SSGI 2022 and SSGI 2024). The code does not develop or validate a deployable prediction model: a gradient-boosted tree is used solely as a measuring instrument, and the object of measurement is the determinant ranking itself. Reproducibility is summarized by a single named estimand, the rank-based reproducibility coefficient, defined as the Spearman rank correlation between two importance profiles, with permutation nulls and bootstrap confidence intervals. The pipeline partitions the data into four period-by-age-cohort cells (baduta 0-23 months and balita_tua 24-59 months, crossed with the two waves) and, under an anti-leakage protocol that excludes outcome-forming anthropometry, measures three layers of determinant structure: redundancy among determinants, additive main effects, and pairwise interactions. It then quantifies within-cohort reproducibility across waves, replication of between-cohort differences, sensitivity of the recovered structure to imputation, and a directly estimated measurement-noise floor against which the observed cross-wave instability is judged. This repository is one of three downstream studies on Applied Explainable AI for Health Risk Prediction, with childhood stunting in Riau Province, Indonesia, as the validation domain. It operates on the harmonized master dataset produced by the Stunting Harmonization Pipeline, which is archived separately. The microdata are governed by the Ministry of Health of the Republic of Indonesia and are not redistributed; the harmonized master Parquet is a derivative and is never committed. The synthetic test-data generator in the harmonization repository allows this pipeline to be run and verified without restricted data.

XGBoost-SHAP Explainable AI Framework for Process Dispatching in Semiconductor APC Systems

Total dataset

The Observational Incompleteness Framework

The Observational Incompleteness Framework is an AI-based research programme deriving quantum-mechanical structure from the premise that observation is a proper subsystem of a deterministic whole. The archive contains three kinds of content: Technical papers (papers/) — the authoritative technical content, the core papers Main (central theorem and emergent quantum mechanics), Substratum (substratum construction and reconstruction theorem), Structure (structural realism), SM (Standard Model derivation) and GR (gravitational sector), together with the focused presentation Juno (neutrino-sector prediction), the methodology paper Physics Modulo Gauge, and the companion documents Explainer, Complexity, Medicine and Bioinformatics. Sources and built PDFs are both included. Book manuscript (book/) — The Incompleteness of Observation: A Unified Framework from Quantum Mechanics to Computational Biology, a working draft addressed to a general technical readership. The papers, not the book, are the primary reference for framework-internal derivations. Verification code (papers/oi_lattice_code/) — lattice Monte Carlo sources, run drivers, analysis scripts, and the deterministic test suites behind claims made in the papers. Licensing. This deposit is mixed content under a single Zenodo license field. The manuscripts are licensed CC-BY-4.0, which is the label shown here; the source code is licensed MIT, per the LICENSE file at the archive root. The Licensing section of README.md is the authoritative statement of scope. Version-specific DOIs are minted for each release; the concept DOI resolves to the latest version. Work reproducing a specific claim should cite the version DOI of the release that carries it. Discussion and feedback are welcome via the linked GitHub repository (Discussions).

Bibliometric Analysis AI-Based Decision Analytics

The rapid advancement of artificial intelligence (AI) has transformed decision-making processes across various domains by enabling data-driven insights, predictive capabilities, and intelligent automation. This study aims to examine the development, intellectual structure, and emerging trends of research on AI-based decision analytics through a bibliometric analysis approach. Data were collected from the Scopus database using relevant search terms related to artificial intelligence and decision analytics. The study applies bibliometric techniques, including performance analysis, citation analysis, co-authorship analysis, keyword co-occurrence analysis, thematic evolution analysis, and density visualization using VOSviewer. The findings indicate that artificial intelligence, machine learning, deep learning, and clinical decision support systems represent the dominant research themes shaping this field. Highly cited studies demonstrate increasing scholarly attention toward ethical considerations, explainability, transparency, and trust in AI-driven decision systems. The collaboration analysis reveals that research development is supported by extensive international networks, with countries such as Germany, the United States, India, and China serving as influential contributors. Furthermore, the temporal analysis indicates a shift from algorithm-focused research toward human-centered and responsible AI applications. This study contributes to the literature by providing a comprehensive mapping of AI-based decision analytics research and identifying future directions related to explainable AI, trustworthy decision systems, and interdisciplinary applications across healthcare, business, and other complex decision environments.

Artificial Intelligence and the Right to Equal Education: A Legal Framework for Algorithmic Educational Justice

The rapid adoption of artificial intelligence (AI) in education is reshaping how learners access knowledge, receive instruction, are assessed, and are classified by educational institutions. Although AI can expand personalization, accessibility, and educational efficiency, its deployment may also generate new forms of inequality through unequal access to advanced technologies, algorithmic bias, opaque profiling, data-intensive surveillance, and differential quality of AI-mediated learning. This article examines whether the traditional legal conception of the right to equal education remains adequate in an increasingly algorithmic educational environment. The study employs normative legal research using statutory, conceptual, doctrinal, and comparative approaches. It examines international human-rights standards, contemporary AI governance frameworks, education law, and emerging national approaches to AI in education. The analysis argues that formal equality is insufficient where algorithmic systems distribute educational opportunities differently according to data, infrastructure, digital competence, socioeconomic status, or model performance. The article develops the concept of Algorithmic Educational Justice and proposes a seven-dimensional framework encompassing equal access, algorithmic non-discrimination, educational autonomy, data dignity, explainable education, institutional accountability, and effective remedy. The article concludes that AI should be governed as an educational justice issue rather than merely as a technological innovation. The right to equal education must extend to the conditions under which algorithmic systems allocate educational opportunities, and states must ensure that technological transformation does not convert existing educational inequalities into durable algorithmic inequalities.