Prejudice remains a persistent social issue, yet limited evidence exists on how its predictors differ across countries. Data from Wave 7 of the World Values Survey were used to construct a generalized prejudice index based on self-reported xenophobia, homophobia, and sexism in a sample of 97,220 adults from 66 societies (Mage = 43.18, SDage = 16.58, 52.6% female). Age, gender, education, income, subjective social class, and religiosity were examined as predictors of prejudice, and it was assessed whether these associations varied according to national human development. Bayesian multilevel analyses indicated that higher national human development was associated with lower average levels of prejudice. In the pooled global sample, older age and greater religiosity were positively associated with prejudice, whereas being female and having higher education, income, and subjective social class were negatively associated with prejudice. These relationships were, however, contingent on societal development. Cross-level interactions showed that demographic and religious predictors were weak or negligible in low-development countries but became stronger and statistically significant in high-development countries. In less developed societies, these predictors were largely unrelated to prejudice, suggesting that individuals’ demographic characteristics and religiosity say little about their level of prejudice in such societies. These results suggest that many findings often assumed to be general are in fact more characteristic of high‑development than low‑development contexts.
Brain tumor classification from magnetic resonance imaging (MRI) is an important component of computer-aided diagnosis, yet automated models must remain accurate, efficient, and robust under heterogeneous imaging conditions. Although conventional CNN-based methods have shown promising results, many existing architectures rely on computationally expensive backbones, contain a large number of parameters, or do not explicitly combine lightweight feature extraction with spatial and channel-wise attention. These limitations may reduce their applicability in resource-constrained medical image analysis, particularly when subtle tumor-related patterns must be distinguished from surrounding brain tissues.To address this gap, this study proposes ShuffleUNet-Lite, a lightweight U-Net-based classification architecture enhanced with spatial-channel dual attention for brain tumor MRI analysis. The proposed model integrates ShuffleNet-style feature mixing, pixel shuffle/unshuffle operations, multi-scale feature fusion, and spatial-channel attention to capture both fine-grained local features and broader contextual representations. The spatial attention module emphasizes tumor-relevant regions, while the channel attention module highlights discriminative feature channels, improving the model’s ability to learn informative MRI patterns.Experiments were conducted on a publicly available Kaggle brain tumor MRI dataset containing 7023 two-dimensional MRI images from four classes: glioma, meningioma, pituitary tumor, and no tumor. All images were resized, normalized, and enhanced using Contrast Limited Adaptive Histogram Equalization (CLAHE) before training and evaluation. ShuffleUNet-Lite achieved an accuracy of 98.97% with only 0.33 million trainable parameters, demonstrating a favorable balance between classification performance and computational efficiency.
Conventional gravity compensators capable of handling large loads are often bulky and require hardware modifications for different payload ranges. This study presents a multi-winding gas-spring gravity compensation mechanism (MW-GCM) for high-torque sit-to-stand assistance and evaluates its mechanical feasibility using a sensorized one-axis testbed. The MW-GCM uses four parallel cable paths designed to reduce the nominal load carried by each path while transmitting the gas-spring force and allowing the energy-storage element to be located away from the joint axis. The sit-to-stand assistance target was obtained from a quasi-static two-knee anthropometric model. By adjusting the gas-spring pressure and initial stroke, the mechanism can be retuned for different load conditions without component replacement. Experiments under variable load and speed conditions demonstrated high-torque compensation and showed direction-dependent torque error consistent with losses in the cable-pulley transmission. These results demonstrate mechanism-level feasibility under controlled testbed conditions.
Drawing on the Challenge-Hindrance Stressor Framework (CHSF), we examined how AI-supported work functions as a daily stressor that can elicit both engagement and disengagement responses depending on individual challenge and hindrance appraisals. Using a 10-day diary study design conducted in the US (N = 173), we tested dual pathways through which daily AI-supported work influenced employee behavioral and attitudinal responses. On days when employees perceived having more AI-supported work, they were more likely to appraise it as a challenge, which in turn was associated with higher levels of task crafting and innovative work behavior. At the same time, AI-supported work was also appraised as a hindrance, triggering job replacement anxiety and resulting in higher levels of organizational cynicism. Furthermore, AI system evaluation, a general attitude toward AI systems, moderated the hindrance pathway such that individuals with more favorable AI evaluations were less likely to perceive AI-supported work as a hindrance. Our findings extend the CHSF to the context of AI, highlighting that the same AI stimulus can evoke distinct psychological and behavioral responses depending on appraisal and attitude. We discuss implications for theory and for organizations seeking to implement AI in ways that enhance positive employee outcomes while mitigating adverse effects, thereby contributing to a more human-centric approach to AI integration at work.
