explainable ai

Auditable Credit Risk Intelligence for the U.S. Financial System: A Scalable Explainable-AI Framework Reconciling Predictive Performance with ECOA, FCRA, and Model Risk Governance

Machine-learning credit underwriting in the United States operates under three families of obligation that are not reducible to one another: a statutory prohibition on discrimination, a statutory duty to disclose the specific principal reasons for an adverse action, and prudential expectations for model risk management. Prevailing practice reconciles these demands by sacrificing model capacity, deploying low-capacity scorecards whose interpretability is structural rather than earned. This study argues that the United States regulatory realignment of 2025 and 2026 makes that settlement less defensible, not more. The Consumer Financial Protection Bureau withdrew the interpretive circulars governing algorithmic adverse-action practice and amended Regulation B to disclaim disparate-impact liability under the Equal Credit Opportunity Act, while leaving the statutory disclosure duty untouched and preserving liability for the intentional use of facially neutral proxies. The federal banking agencies simultaneously replaced the prescriptive 2011 interagency model risk guidance with a principles-based instrument that expressly disclaims enforceable standards. Fewer obligations are now externally specified, and more must be self-specified, self-justified, and self-evidenced. Because effects-based exposure persists under the Fair Housing Act and state analogues, and private litigation is unaffected by federal guidance withdrawal, the value of verifiable self-generated evidence rises as external specification recedes. The paper develops ACRIS, the Auditable Credit Risk Intelligence Stack, a five-layer architecture in which admissibility, explainability, disparity control, and governance evidence are enforced during training and serving rather than audited afterwards. Its layers comprise a provenance-gated feature-admissibility screen bounding residual proxy information through conditional mutual information; a shape-constrained predictive core; a constrained-optimization layer recording an entire searched alternative-model frontier, including rejected candidates and rejection rationales; an explanation engine deriving principal reasons from a counterfactual approval baseline and gating disclosure on a per-decision stability margin; and an append-only, tamper-evident governance ledger. A finite-sample sufficient condition for top-k reason-set preservation under parameter resampling is proved, with its assumptions and failure modes stated explicitly. The framework is not empirically validated. A pre-registered protocol of falsifiable propositions, corpora, temporal validation design, baselines, metrics, statistical plan, and pre-committed disconfirmation criteria is specified so that every central claim can be tested and, if wrong, refuted.

Advancing Decoding Methods for Enhanced AI Model Performance and Interpretability

The rapid advancement of artificial intelligence, particularly in large language models, has brought significant challenges in decoding methods that balance performance with interpretability. This paper presents a comprehensive analysis of advanced decoding techniques that enhance both model capabilities and explainability. We explore novel approaches including adaptive temperature sampling, nucleus sampling with dynamic thresholds, and hybrid decoding methods that combine multiple strategies. Our research demonstrates that these advanced techniques can significantly improve model performance metrics while maintaining or enhancing interpretability. Through extensive experimentation across diverse datasets, we show that our proposed hybrid decoding method achieves a 12.3% improvement in perplexity scores while maintaining competitive computational efficiency. The findings contribute to the growing body of research on transparent AI systems and provide practical insights for practitioners aiming to deploy more reliable and interpretable AI models in high-stakes applications.

aiDIVA – hybrid AI for rare disease diagnostics using evidence-based, machine learning and language models

aiDIVA is presented, an ensemble-AI combining statistical and machine learning models trained on genomic and phenotypic data to identify causal variants among tens of thousands per patient, and applies a random forest model to classify pathogenicity and generates evidence-based scores for dominant and recessive diseases.

AI as an Aviation Safety Decision Maker

Artificial intelligence is rapidly evolving from an analytical support technology into an active participant in aviation safety decision-making. This paper examines how Safety Management Systems (SMS) must evolve as AI increasingly recommends or executes safety-critical operational decisions within aircraft operations, air traffic management, maintenance, dispatch, and organizational safety processes. Particular attention is given to automation bias, explainability, human override authority, accountability, AI-generated hazards, operational design domains, and continuous safety assurance. The study examines developments involving the FAA, EASA, Airbus, the United States military, and DARPA, while also addressing legal and civil implications arising from shared human–AI decision authority. It argues that future SMS frameworks must treat AI simultaneously as a safety tool, decision-making participant, potential hazard source, and safety control. Successful integration will require explicit authority boundaries, explainable and reconstructable decisions, continuous operational monitoring, enforceable safety constraints, and preservation of human and organizational accountability.

Advanced multi-modal Threat Detection: Explainable AI for Early Warning from Human Behavioral Cues Advanced multi-modal Threat Detection Preprint

Autonomous Enterprise Platforms: A Framework for AI-Guided Decision Loops, Predictive Intelligence, and Continuous Organizational Adaptation

Enterprise platforms are evolving from systems that primarily record and analyze business operations into intelligent environments capable of predicting outcomes, recommending interventions, executing decisions, and learning from their consequences. This article proposes a conceptual Autonomous Enterprise Platform (AEP) based on continuous AI-guided decision loops integrating enterprise sensing, contextual intelligence, predictive analytics, decision intelligence, prescriptive policies, autonomous execution, learning, and governance. The proposed framework extends the classical Monitor, Analyze, Plan, and Execute model of autonomic computing by incorporating continuous prediction, intervention, evaluation, and adaptation. The study synthesizes research published between 2000 and 2022 on autonomous agents, autonomic computing, self-adaptive systems, predictive process monitoring, reinforcement learning, and prescriptive analytics. Three key studies provide the conceptual foundation: Kephart and Chess on autonomic computing, Metzger et al. on proactive process adaptation using deep learning, and Kubrak et al. on prescriptive process monitoring. The framework distinguishes operational, learning, and governance loops to support continuous enterprise adaptation. It emphasizes the transition from predicting business outcomes to selecting and executing appropriate interventions. The study also examines challenges involving causal reasoning, intervention timing, resource constraints, model drift, explainability, and human oversight. Overall, AI-guided decision loops provide a foundation for adaptive, intelligent, and governed enterprise platforms capable of continuous decision making, organizational learning, and operational optimization.

Exploring the Relationship Between Visit Frequency and Customer Retention in Study Cafes Using AI-Based Predictive Modeling

This study empirically investigates actual customer repurchase behavior by analyzing behavioral log and payment data collected from study cafe users in an attendance-based learning service environment. Unlike prior studies that primarily relied on survey-based measures, this study integrates attendance records, payment data, and explainable artificial intelligence (XAI) techniques to provide a data-driven understanding of customer retention behavior. It compares the predictive performance of traditional regression models with machine learning approaches. Specifically, logistic regression, random forest, XGBoost, and neural networks were employed, with SHAP analysis applied as an XAI technique. The results indicate that visit frequency has a significant positive effect, supporting H1, while stay-duration variables show only limited and inconsistent effects, providing partial support for H2 and H3. Total payment in May negatively affects subsequent repurchase, suggesting possible saturation or substitution effects, thereby supporting H4. Age demonstrates a negative effect (supporting H5), and regional differences are captured by the XGBoost model (supporting H6). In terms of predictive performance, machine learning models outperformed logistic regression, with XGBoost achieving the strongest overall results among the evaluated models (ROC-AUC = 0.676; PR-AUC = 0.642). Overall, this study contributes to the literature by presenting empirical evidence based on behavioral data and by highlighting the practical interpretability of integrating XAI techniques. From a managerial perspective, the findings provide actionable insights for designing customer retention strategies based on visit frequency, spending behavior, and regional characteristics, thereby supporting AI-driven decision-making in attendance-based service environments.

White-Box Completeness for Artificial Intelligence

Abstract Current research on explainable and white-box artificial intelligence faces prominent issues: conceptual disarray, divergent perspectives, and a disconnect between theory and practice. The foremost priority is therefore to return to the field’s purest objectives and explore its most fundamental questions. To this end, this paper introduces the axiomatic criterion of white-box completeness (WBC). Specifically, an agent is white-box complete if and only if its behavior can be bidirectionally approximated by human interpretable and manipulable mathematical forms with bounded error. It is a formalization for realizing three ultimate objectives for trustworthy AI: discernible learned knowledge and behaviors, knowledge extraction from AI, and human knowledge injection. Therefore, the ultimate pursuit of explainable AI has been transformed from the vague goal of “making AI interpretable” into a precise mathematical problem. Keywords: White-box completeness; Interpretability; Explainable artificial intelligence; Trustworthy artificial intelligence; Knowledge extraction and injection.

Understanding Human Acceptance of Explainable AI in Automated Aesthetic Evaluation: A Multi-Study Investigation of Advice Taking and Explanation Requirements

Explainable Artificial Intelligence (XAI) has been increasingly applied in image aesthetic quality evaluation. However, it remains unclear how perceived explainability influences users’ decision adoption and what specific explanation requirements shape this perception. This study develops and tests a multi-stage model of XAI-supported human–AI decision-making in aesthetic evaluation based on Task–Technology Fit theory with a self-developed AI aesthetic evaluation system and PLS-SEM analysis (N = 494). Our findings reveal that perceived explainability does not exert a direct effect on perceived fairness, but influences it indirectly through two underlying psychological mechanisms: functional understanding and affective satisfaction. Furthermore, perceived fairness serves as a critical mediating mechanism linking explainability perceptions to decision outcomes. Extending this framework, a second online questionnaire study(N = 313) investigates the specific explanation requirements of the users in AI-assisted aesthetic evaluation. The results reveal seven categories of explanation needs and develop a corresponding question bank to guide the design and evaluation of XAI systems. This study contributes to a unified theoretical understanding of explainable human–AI decision-making in aesthetic quality evaluation and provides practical guidance for researchers, developers, and organizations designing XAI-assisted aesthetic quality evaluation systems.

Understanding Human Acceptance of Explainable AI in Automated Aesthetic Evaluation: A Multi-Study Investigation of Advice Taking and Explanation Requirements

Explainable Artificial Intelligence (XAI) has been increasingly applied in image aesthetic quality evaluation. However, it remains unclear how perceived explainability influences users’ decision adoption and what specific explanation requirements shape this perception. This study develops and tests a multi-stage model of XAI-supported human–AI decision-making in aesthetic evaluation based on Task–Technology Fit theory with a self-developed AI aesthetic evaluation system and PLS-SEM analysis (N = 494). Our findings reveal that perceived explainability does not exert a direct effect on perceived fairness, but influences it indirectly through two underlying psychological mechanisms: functional understanding and affective satisfaction. Furthermore, perceived fairness serves as a critical mediating mechanism linking explainability perceptions to decision outcomes. Extending this framework, a second online questionnaire study(N = 313) investigates the specific explanation requirements of the users in AI-assisted aesthetic evaluation. The results reveal seven categories of explanation needs and develop a corresponding question bank to guide the design and evaluation of XAI systems. This study contributes to a unified theoretical understanding of explainable human–AI decision-making in aesthetic quality evaluation and provides practical guidance for researchers, developers, and organizations designing XAI-assisted aesthetic quality evaluation systems.

Can Artificial Intelligence Improve Work-Life Balance? Examining the Relationship between AI Adoption, Workload, Stress and Employee Well-Being

The growing use of Artificial Intelligence (AI) in organisations is changing how employees perform, organise, and experience their work. AI-supported systems can automate routine activities, assist with information processing and improve the speed of task completion. These advantages have created expectations that AI may help employees manage their professional responsibilities more efficiently and achieve a healthier work-life balance. At the same time, increased dependence on AI can introduce new pressures, including higher performance expectations, continuous technological adaptation, work intensification and technology-related stress. The present paper examines the relationship between AI adoption, workload, stress, work-life balance and employee well-being. Drawing on the Job Demands-Resources (JD-R) perspective, the paper conceptualises AI as a workplace resource when it reduces unnecessary effort and supports employees, while recognising that it may become a job demand when it increases pressure and complexity. The paper proposes an empirical framework in which workload and stress explain how AI adoption may influence work-life balance and employee well-being. Recent research indicates that AI can improve task optimisation and work-life outcomes under supportive conditions, while AI-related technostress may contribute to exhaustion and work-family conflict. (DOI) The study provides a foundation for future empirical investigation and highlights the need for organisations to evaluate AI not only through productivity indicators but also through employee health, balance and sustainable work practices.

Explainable Artificial Intelligence for Enhancing Intrusion Detection Systems: A Comprehensive Framework for Transparent Network Security

This research presents a comprehensive framework for integrating Explainable Artificial Intelligence (XAI) into Intrusion Detection Systems (IDS) to address the critical challenge of AI model opacity in cybersecurity. Traditional AI-based IDS models function as "black boxes," limiting trust, accountability, and practical deployment. The proposed XAI-IDS framework combines machine learning-based intrusion detection (Random Forest, XGBoost, Neural Networks) with explainability mechanisms including SHAP, LIME, and feature importance analysis. Experimental evaluation using benchmark datasets (KDD Cup 99, UNSW-NB15, CIC-IDS-2017) demonstrates that the XAI-enabled system achieves high detection accuracy (96-98%) while providing transparent, human-interpretable explanations for each security decision. The framework significantly reduces false positives, enhances auditability, and improves security analyst trust and response effectiveness without substantial performance degradation.