Search Results (978)

Artificial intelligence (AI) may influence economic performance through productivity, innovation, and the diffusion of advanced technologies, but direct country-level measures of AI adoption remain limited for small economies. This study examines the association between AI-related technological capability and economic growth in Cyprus. The descriptive analysis covers 1993–2024, while the econometric analysis uses the common annual sample for 2008–2024 after introducing one-period lags. High-technology exports as a percentage of manufactured exports are treated as an indirect indicator of technological sophistication and absorptive capacity, rather than as a direct measure of AI adoption. A sequential distributed lag ordinary least squares framework introduces current and lagged high-technology exports, merchandise trade, and inflation, with heteroskedasticity-consistent HC3 standard errors. A COVID-19 indicator for 2020–2021 and an observation-exclusion sensitivity analysis are used to assess the influence of exceptional pandemic-era movements. The preferred specification explains a substantial share of annual GDP growth variation, but the small sample requires cautious interpretation. The estimates indicate a positive contemporaneous and a negative lagged association for high-technology exports, a stronger lagged than contemporaneous trade association, and opposite-signed current and lagged inflation coefficients. The COVID-19 indicator is statistically insignificant and does not materially alter the principal coefficient pattern. Complementary DESI, Eurostat enterprise AI use, and ICT employment indicators show that Cyprus has broadly adequate digital inputs while enterprise AI adoption has not yet reached the European Union average, indicating considerable scope for further diffusion. The findings are exploratory statistical associations and do not identify causal effects of AI. The study contributes a country-specific macroeconomic assessment of technological capability, digital readiness, and growth in a small, highly open economy. Full article

With the proliferation of internet-connected infrastructures and the complexity of cyberattacks, cybersecurity and intelligent intrusion detection systems have become more and more critical. Intrusion detection datasets, however, are now highly imbalanced, and conventional machine learning models have become biased towards the majority of benign traffic, misclassifying minority attack classes. This paper introduces a Cost-Sensitive Forest (CS-Forest) approach to enhance the detection of minority attacks in the CSE-CIC-IDS2018 dataset. The proposed framework combines cost-sensitive learning, ensemble-based Random Forest classification, feature selection, and SHAP explainability analysis to boost the performance of intrusion detection and interpretability. Various machine learning algorithms such as Decision Tree, Random Forest, AdaBoost, and XGBoost were tested and compared based on accuracy, precision, recall, F1-score, ROC-AUC, false positive rate, and false negative rate. Experimental results proved that the proposed CS-Forest has excellent performance, with 99.81% accuracy, 99.55% recall, 99.61% F1-score, and 0.998 ROC-AUC, significantly enhancing the performance of minority attack detection and reduced false negatives. The framework learned meaningful and interpretable network traffic behaviors, which was also confirmed using SHAP analysis. The research suggests that future IDS systems should incorporate cost-sensitive learning and explainable AI techniques to ensure improved reliability, transparency, and deployment in the cybersecurity landscape. Full article

Trajectory prediction is a critical component of autonomous driving systems, enabling vehicles to anticipate future motion behaviors and perform safe decision-making in dynamic traffic environments. While recent trajectory forecasting methods achieve state-of-the-art prediction accuracy, many operate as black-box systems and are evaluated primarily on high-end computing platforms, limiting their interpretability and practical deployment feasibility in resource-constrained autonomous driving systems. To address these limitations, this work proposes an interpretable and edge-deployable spatio-temporal trajectory prediction framework for autonomous driving. The proposed architecture integrates a Temporal Convolutional Network with Multi-Head Self-Attention (TCN–MHSA) in ActorNet for selective temporal modeling, a Lane Graph Attention Network (LaneGAT) for structured spatial reasoning, and a multi-stage FusionNet for actor–lane interaction. To improve model interpretability, a comprehensive Explainable AI (XAI) evaluation framework is introduced, including temporal sensitivity analysis, interaction-aware perturbation studies, spatial influence analysis, and gradient-based feature attribution methods. These analyses provide insights into how the model captures temporal motion dependencies, neighboring vehicle interactions, and environmental context during trajectory prediction. To improve the robustness of the interpretability analysis, temporal sensitivity was additionally evaluated over 100 validation scenes, demonstrating that recent observations consistently exert the greatest influence on trajectory prediction, while neighboring interaction effects gradually diminish with increasing spatial separation. Furthermore, practical real-world deployment feasibility is investigated on the NVIDIA Jetson Xavier NX platform using edge-aware optimization strategies, including mixed-precision inference and graph-complexity reduction techniques for efficient resource-constrained inference, achieving 125.74 ms latency at 12.86 W. Additional edge deployment comparisons with HiVT and SIMPL approaches under identical hardware conditions demonstrate that the proposed framework provides a more favorable balance between computational efficiency and embedded deployment performance. Experimental evaluation on the Argoverse 1 dataset demonstrates a minimum Average Displacement Error (minADE) of 0.90 m, a minimum Final Displacement Error (minFDE) of 1.50 m, a Miss Rate (MR) of 0.19, and DAC = 0.95, while establishing an accuracy–deployability operating point under embedded hardware constraints with low power consumption and practical inference throughput. Full article

This study explores how university students experience conversational artificial intelligence (AI) in the era of generative AI from the perspectives of algorithmic intimacy, social connection, well-being, and autonomy, and identifies their subjective meaning structures. Specifically, it examines how university students perceive conversational AI in relation to emotional regulation, self-understanding, relational boundaries, and autonomy, and how these perceptions relate to intimacy and well-being. Unlike previous studies focused primarily on AI use and acceptance, this study analyses how students interpret their experiences with conversational AI and negotiate relational boundaries. Using Q methodology, we developed 40 Q statements from literature review and in-depth interviews. Fifty-six university students in South Korea with experience using conversational AI completed the Q-sorting process. A four-factor solution explained 55% of the total variance and yielded four distinct types: (1) Relationship-Negating Instrumentalists, (2) Selective Emotional Utilizers, (3) Reflective Distancers, and (4) Autonomy Defenders. Although none of the groups viewed AI as a complete substitute for human relationships, they differed in their acceptance of algorithmic intimacy, use of AI for emotional support and regulation, concerns about dependence, and efforts to protect autonomy. These findings suggest that conversational AI experiences involve negotiating intimacy and boundaries rather than producing uniformly positive or negative outcomes. This study complements variable-centred approaches by identifying distinct subjective meaning structures in university students’ human–AI relationship experiences. Full article

The rapid proliferation of generative artificial intelligence (GenAI) has reshaped university students’ learning practices, yet the association between GenAI-assisted learning and procrastination remains insufficiently understood. Drawing on the Theory of Planned Behavior as an established framework, this study offered a contextual extension of TPB and procrastination research into the GenAI domain by examining the association between students’ use of GenAI for learning task completion and GenAI-assisted learning procrastination among 1243 Chinese university students. Path analysis showed that students’ use of GenAI was negatively associated with GenAI-assisted learning procrastination (β = −0.195, p < 0.001). Behavioral attitude (β = −0.164, p < 0.001), subjective norm (β = −0.171, p < 0.001), perceived behavioral control (β = −0.331, p < 0.001), and behavioral intention (β = −0.137, p < 0.001) were each negatively associated with GenAI-assisted learning procrastination. The indirect associations linking GenAI use to GenAI-assisted learning procrastination through these Theory of Planned Behavior constructs were statistically significant, with the total indirect association accounting for approximately 30.9% of the total association (standardized indirect association = −0.086, p < 0.01). Learning GenAI anxiety moderated the association between behavioral intention and GenAI-assisted learning procrastination, with the negative association being stronger among students with higher anxiety (b = −0.268, SE = 0.040, p < 0.001) than among those with lower anxiety (b = 0.006, SE = 0.052, p = 0.915). The model explained 71.6% of the variance in procrastination in GenAI-assisted learning. Given the cross-sectional design and the sample’s specific characteristics, these findings should be interpreted as conditional associations rather than causal evidence, and their generalizability requires further investigation. Nevertheless, the results highlighted the roles of beliefs, intentions, and emotional experiences in understanding procrastination in GenAI-assisted learning. Full article

Deep visual recognition has achieved remarkable success across various domains, including medical imaging, autonomous vehicles, and security systems. However, the black-box nature of deep learning models poses challenges in terms of transparency and trust, especially in critical applications where human understanding is essential. Explainable AI (XAI) seeks to address these concerns by providing human-interpretable explanations for model predictions. This review explores the key techniques for explainability in deep visual recognition, including model-agnostic methods such as LIME and SHAP, model-specific approaches like saliency maps and feature visualization, and intrinsically interpretable models like decision trees and rule-based systems. We also discuss the evaluation of explainability through metrics like fidelity, consistency, and stability, and explore the challenges of balancing model performance with interpretability. Furthermore, we examine applications of XAI in medical imaging, autonomous driving, security and surveillance, agriculture, satellite imagery and remote sensing, industrial inspection, and visual forensics, highlighting how domain-specific data and operational constraints affect the required form and validation of explanations. Finally, we address current research gaps and propose future directions for enhancing the robustness and human–AI interaction in explainable visual recognition systems. As AI continues to be integrated into safety-critical domains, the development of explainable, transparent, and trustworthy AI systems will be crucial for their widespread adoption and ethical use. Full article

Meaning no longer resides where we assumed it did. At the scale of hundreds of millions of daily human–AI encounters, intelligent systems have ceased to function as external tools and have become the constitutive environment through which thought, interpretation, and knowledge formation occur. Yet dominant frameworks remain anchored to a unit of analysis that does not capture what is actually happening: they measure outputs, govern behaviours, and optimise performance, while the encounter itself remains theoretically unnamed. This paper proposes encounter ontology as the conceptual framework that names it, arguing that the encounter constitutes a distinct ontological site irreducible to either human or machine ontology alone. Human–AI interaction is not only transmissive but generative: the encounter produces interpretations, conceptual formations, and altered cognition irreducible to either the user’s original intent or the model’s specified behaviour. Building on new materialist philosophy, assemblage theory, agential realism, and the philosophy of technics, the paper introduces relational intelligence as the emergent co-production of meaning across situated human–AI interactions sustained over time, situated within Floridi’s concept of hyperhistory. The encounter generates epistemic trajectories that are not merely unknown prior to the interaction but genuinely undetermined, brought into form through the relational event itself rather than retrieved from either participant. The paper examines implications for alignment, explainability, and governance, introducing the specification–emergence gap as a constitutive rather than contingent feature of intelligent interaction. Three design orientations are developed toward operationalisation: design for temporal continuity, design for encounter quality, and design for symbolic legibility. The paper contributes a theoretically grounded vocabulary for analysing human–AI co-emergence and extending intelligent system design toward more relationally aware and responsibly governed AI. Full article

This paper investigates adversarial robustness and explainability in AI-generated face detection through the Robust and Explainable Detection (RED) framework. RED unifies CNN and Vision Transformer (ViT) backbones with FGSM/PGD adversarial training and Grad-CAM-based interpretability in one reproducible pipeline. Experiments on the Kaggle real/fake face dataset with stratified 70/15/15 splits report accuracy, F1, AUC-ROC, Adversarial Robustness Index (ARI), and Explainability Fidelity (EF). Xception and ResNet-50 reach 97.53% and 96.64% validation accuracy, respectively, while ViT-B/16 attains 69.08%. RED supports forensic and legal applications requiring both detection accuracy and transparent model behavior. Full article

Generative artificial intelligence (GenAI) is increasingly used in entrepreneurial practice, yet limited empirical evidence explains how sustainability-oriented entrepreneurs appropriate it in relation to Triple Bottom Line (TBL) performance. This study examines the strategic alignment paradox faced by entrepreneurs who pursue accountable sustainability outcomes while using fast-moving, probabilistic, and partly opaque technologies. It investigates whether Sustainability Orientation (SO) is associated with Sustainable Entrepreneurial Performance (SEP) through five theoretically distinct GenAI appropriation dimensions: Adoptive, Customized, Ethical, Integrative, and Interface Appropriation. In this study, Ethical Appropriation is interpreted as ethical-boundary permissiveness because the retained GAIA items capture acceptance of self-serving, improper, or beyond-intended GenAI use rather than broad responsible-AI governance. A cross-sectional survey of 360 GenAI-using entrepreneurs and SME owner-managers in Saudi Arabia was analyzed using PLS-SEM in SmartPLS 4, with a 10,000-resample bootstrapping procedure, MICOM, permutation-based multi-group analysis, and higher-order robustness tests. The empirical results show that SO is positively and significantly associated with Customized, Ethical, and Integrative Appropriation, but has no significant relationship with Adoptive or Interface Adoptive Appropriation or Interface Appropriation. The aggregate model explains 60.8% of the variance in SEP. Mediation analysis indicates that the observed indirect association between SO and aggregate SEP is statistically significant only through the Customized, Ethical, and Integrative pathways. The pillar-specific analysis further shows that the aggregate SEP construct masks important differences across TBL outcomes: Integrative Appropriation is consistently associated with all three pillars; Ethical Appropriation is selectively associated with economic sustainability and should be interpreted as a governance warning rather than as evidence of responsible-AI governance; and Customized Appropriation is associated only with environmental and social sustainability. Finally, multi-group analysis provides no statistically robust evidence of gender-based heterogeneity after Bonferroni correction, suggesting that the SO–GenAI appropriation–SEP relationships are broadly comparable across male and female respondents in this sample. Overall, the findings suggest that GenAI should not be treated as a monolithic adoption construct; its sustainability implications depend on how generative tools are customized for specific tasks, interpreted with attention to ethical-boundary permissiveness, and structurally integrated into business routines. Full article

The Sacramento–San Joaquin Delta (Delta) is a critical hub for California’s water supply and ecological health, necessitating rapid and accurate modeling for decision-support. This paper reviews three decades of machine learning (ML) applications in the Delta, tracing the evolution from early artificial neural networks (ANNs) developed in the 1990s for salinity estimation to modern integration within operational planning tools like CalSim. Recent advancements highlight a shift toward deep learning and hybrid physics-informed models, expanding applications to water quality forecasting, ecosystem monitoring, toxigenic and disaster risk assessment, and infrastructure resilience. Despite these successes, widespread operational deployment remains constrained by critical gaps in model interoperability, generalizability, interpretability, and operationalization. To address these challenges, we propose a strategic roadmap emphasizing standardized ML development protocols, multi-modal data fusion, digital twins, generative artificial intelligence (AI), explainable AI, and AI agents. This roadmap aims to transform isolated research ML models into a cohesive, transparent, and adaptive decision-support framework for sustainable water and environmental management in the Delta. Full article

This study develops and applies a three-factor matrix model for assessing counterparty risk within Anti-Money Laundering (AML) measures. The research problem arises from the need to transform qualitative features related to ownership, effective control, economic logic of operations, geographical links, documents and review behavior into a comparable and documented risk score. The model uses three factors: counterparty exposure, transaction impact and control vulnerability. Each factor is assessed on a five-point ordered scale, and the overall risk score is obtained by multiplying the three values. The methodology includes scoring rules, rating interpretation, an application algorithm and a link between the profile, due diligence, ongoing monitoring and updating. The model was applied in 2026 to 25 anonymized counterparty profiles selected for methodological demonstration across different risk configurations. Identifying data, personal information and confidential business information were removed, while the relevant risk characteristics were preserved. The results demonstrate how the model distinguishes between low, moderate, increased, high and critical risk and how it creates a traceable link between established facts, numerical assessment and follow-up control actions. The study does not claim predictive accuracy, detection of money laundering or empirical validation against external AML outcomes. Its contribution is methodological and applied: it provides a transparent and explainable framework for structured, documented and proportionate management of counterparty AML risk, including potential use in AI-assisted monitoring under human oversight. Full article

Agentic artificial intelligence (AAI) systems—capable of autonomously executing multi-step instructional tasks with minimal human oversight—are reshaping higher education while exposing the limitations of adoption- and continuance-centred models that treat value and risk as independent predictors. In an environment of widespread AI availability, the more meaningful question concerns how deeply educators integrate delegation-capable systems into teaching, assessment, and instructional workflows. Drawing on affordance theory, institutional theory, and automation risk perspectives, this study develops and tests a dual-pathway framework that conceptualises pedagogical integration as a dynamic value–risk trade-off embedded within institutional and contextual conditions. Survey data from 338 higher-education educators across Global North and South contexts, all of whom underwent a structured screening process to verify exposure to agentic rather than conventional generative AI systems, were analysed using PLS-SEM with robustness and endogeneity checks. The model explains substantial variance in pedagogical integration (R² = 0.619), pedagogical value (R² = 0.621), and autonomy risk (R² = 0.615). Institutional AI capability is positively associated with pedagogical value and negatively associated with autonomy risk, whereas perceived AI complexity shows the opposite pattern. Higher pedagogical value is also associated with lower perceived autonomy risk (β = −0.445, p < 0.001), suggesting that educators may interpret system autonomy as a manageable pedagogical feature rather than solely as a source of concern. Regulatory clarity weakens the complexity–risk relationship, while AI legitimacy strengthens the value–integration relationship. Given the cross-sectional design, findings should be interpreted as associative rather than causal. Full article

In the complex landscape of Industry 5.0, traditional management systems for smart manufacturing struggle to harmonize high-speed production with the rapid integration of AI and digital technologies. Crucially, these legacy frameworks often fail to capture tacit human knowledge or ensure trustworthy AI and trusted sharing of sensitive industrial data. This paper proposes a novel Decision Support Framework (DSF) that addresses these challenges through a multi-layered approach. At its core, the framework utilizes Data Spaces to enable secure, sovereign data sharing, ensuring that organizations maintain control over their assets. To handle the inherent ambiguity of industrial data, the system employs fuzzy logic and DAG-based root-cause-oriented investigation to provide robust recommendations, helping users distinguish descriptive correlations from plausible structural dependencies that require expert validation. Furthermore, the framework integrates eXplainable AI (XAI) services and AI-driven visual analytics, transforming complex algorithmic outputs into transparent, intuitive insights. By synthesizing data sovereignty with interpretable machine intelligence, this framework empowers trusted data sharing and human-centric decision-making, providing an advanced platform for achieving operational excellence within the Industry 5.0 vision. The proposed DSF is validated in three different pilot cases with end-users to be a milk industry, an automotive supplier and a machine manufacturer. Full article

Traditional sequential recommender systems have primarily relied on implicit pattern recognition in user interaction sequences, achieving strong performance but functioning as “black boxes” that lack transparent reasoning. This paper introduces LLM-RS, a novel framework that leverages Large Language Models to enable explicit reasoning chains in sequential recommendation. Our approach transforms the recommendation task from mere pattern matching to interpretable reasoning by developing a multi-stage architecture that: (1) extracts structured preference profiles from user interaction sequences, (2) generates explicit reasoning chains analyzing candidate items against inferred preferences, and (3) produces persuasive explanations alongside recommendations. We propose three model variants—fine-tuned reasoning, retrieval-augmented generation, and hybrid ensemble—that integrate LLM-based reasoning with traditional collaborative filtering. The framework addresses key challenges in modern recommender systems by providing transparent, persuasive rationales while maintaining competitive performance, marking a significant step toward more interpretable and trustworthy recommendation systems. Comprehensive evaluations across the Amazon Reviews, MovieLens, MIND, and KuaiSAR datasets demonstrate that LLM-RS not only matches state-of-the-art methods in accuracy but also significantly enhances explanation quality, user trust, and recommendation diversity. Our findings reveal that reasoning-enabled recommendations increase user adherence in online experiments and improve long-term engagement metrics. Full article

Explainable artificial intelligence (X-AI) techniques aim to make the actions and decisions of autonomous systems understandable to humans interacting with these systems. In human–AI teams, explainability supports individual understanding and coordination, shared mental models, and collective decision-making among humans and AI agents. Research has shown that X-AI enhances trust in autonomous systems, improves human–AI team performance, and supports collaboration across domains including aviation, finance, healthcare, hospitality, and sports. However, X-AI technologies face difficult challenges, including a lack of transparency and interpretability due to complex underlying models, also known as the “black-box” nature of AI systems. These technologies also lack any universally accepted evaluation metrics and have limited generalizability across applications. One deficit in the X-AI literature is that most frameworks focus on individual-level outcomes, with limited attention to team-level processes. The current study conducted a systematic literature review adhering to PRISMA guidelines and the SALSA framework. This study introduces the Implementation-Design (I-D) framework that organizes X-AI approaches along two dimensions: implementation, ranging from visual to interactive approaches, and design, ranging from isolated explanations to workflow-integrated systems. This framework captures lower-level engagement, involving individual users, to higher-level understanding that is necessary for teams and collectives. Findings indicate that visual explanation approaches support user engagement, while interactive workflow approaches promote deeper understanding, appropriate reliance, and distributed cognition within human–AI teams. Implications highlight the need for team-oriented explainability grounded in shared mental models, transactive memory systems, and collaborative X-AI artifacts. Practical guidelines are included to support researchers and practitioners in selecting appropriate X-AI techniques based on their context and level of analysis. The I-D framework is offered as a conceptual organizing model to guide research and practice, and empirical validation is identified as a priority for future work. Full article