Finding: This meta-analysis of 42 studies shows AI systems achieve 87.5% sensitivity and 84.2% specificity for diabetic retinopathy screening, with performance dropping when externally validated compared to internal validation (AUROC 0.90 vs 0.94).
Why it matters: Demonstrates how AI diagnostic tools face the challenge of generalizability across different populations and settings, illustrating key principles of AI system deployment and validation in real-world human-computer interaction contexts.
Method: Meta-analysis using bivariate random-effects models across 521,568 retinal images from multiple databases.
Jargon: AUROC = Area Under Receiver Operating Characteristic curve (measure of diagnostic accuracy); QUADAS-2 = Quality Assessment tool for Diagnostic Accuracy Studies.
Rui Liu, Jan R. Wiersema , Lara Bardi β’ PsyArXiv
Finding: Mixed autistic-trait dyads (one high, one low autism traits) showed reduced interpersonal adjustment and communication coordination compared to matched-trait dyads, but communication improved when partners accommodated each other's mental states during real-time coordination tasks.
Why it matters: This provides empirical evidence for designing inclusive team collaboration tools and practices that account for neurodivergent communication differences, directly informing HCI design for diverse teams and collaborative interfaces.
Method: 152 participants formed different trait-matched dyads and completed real-time conceptual coordination tasks where they developed novel signaling systems to coordinate choices, measuring interpersonal adjustment dynamics.
Jargon: Double Empathy Problem (DEP) - theory that communication breakdowns between neurotypes result from mutual mismatches rather than deficits in one group; interpersonal adjustment - dynamic process of accommodating partner's communication style.
β οΈ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Mikko Sams, Einat Amir, Enrico Glerean et al. (8 authors) β’ PsyArXiv
Finding: Customer feedback kiosks provided equally valid audience experience data as detailed questionnaires but with much higher participation rates (67% vs 24%), and revealed significant variation in collective audience response across identical performances, identifying three distinct audience engagement profiles.
Why it matters: This demonstrates effective methods for gathering user experience feedback in live collaborative settings and shows how brief, immediate feedback mechanisms can capture collective experience variations that inform both design and team dynamics.
Method: Combined immediate kiosk ratings with post-performance questionnaires across 20 identical theatre performances, using cluster analysis to identify audience profiles and qualitative interviews with creative teams.
Jargon: Customer feedback kiosks - immediate digital rating systems for collecting audience responses.
β οΈ Self-reported data only | No control group mentioned | Single study (not replicated)
Finding: Students view AI as an optimization tool for educational tasks while academics emphasize cognitive risks and threats to academic integrity, but humans maintain the leading role as research subjects despite AI integration into scientific inquiry.
Why it matters: This directly addresses how AI impacts human work in academic settings and highlights important cognitive biases and decision-making differences between student and faculty populations regarding AI adoption.
Method: Mixed-methods study combining theoretical analysis with survey data from 90 students and faculty, analyzed using correlation analysis, factor analysis, t-tests, and reliability testing.
Jargon: Digital research culture - norms and practices for responsible AI use in academic research contexts.
β οΈ Small sample (n=90) | Self-reported data only
Finding: This paper proposes an explainable AI system for anomaly detection in industrial settings, combining feature selection with variational autoencoders to improve trustworthiness and interpretability in human-machine collaboration for Industry 5.0.
Why it matters: The focus on explainable AI for human-machine collaboration directly relates to HCI principles and the broader challenge of building trust in AI systems that humans must work with.
Method: Combines robust scalar data preprocessing, recursive feature elimination for feature selection, and variational autoencoders with attention-based bidirectional LSTM for anomaly detection.
Jargon: XAI (explainable AI) - AI systems designed to provide human-understandable explanations for their decisions; VAE (variational autoencoder) - a type of neural network that learns compressed data representations; RFE (recursive feature elimination) - iteratively removes features to identify the most important ones.
Finding: This multi-rater study of over 22,000 participants found that personality traits predict pro-environmental attitudes much more strongly than previous single-method studies suggested (correlations up to .60 vs typical .20-.30), with narrow personality "nuances" being better predictors than broad Big Five domains.
Why it matters: Demonstrates how methodological choices can significantly underestimate behavioral predictors, which is crucial for understanding decision-making biases and designing effective behavior change interventions.
Method: Used multi-rater, multi-trait, multi-sample design with both self-reports and informant reports to control for measurement error and estimate "true" correlations across three cultural samples.
Jargon: Nuances = narrow personality traits that are more specific than broad domains like the Big Five; rtrue = correlation corrected for measurement error.
β Includes replication | β Longitudinal design β οΈ Very small sample (n=20) | Self-reported data only | No control group mentioned
Njabulo Ndlovu, P. Sifolo, N. Tshipala β’ Semantic Scholar
Finding: This bibliometric review of 373 articles reveals that while AI research in tourism is growing rapidly (especially post-2019), very few studies specifically address how small and medium tourism enterprises can effectively adopt AI tools within their resource constraints.
Why it matters: It highlights a research gap in designing accessible AI tools for smaller organizations, which is relevant to understanding AI's broader impact on work and collaboration across different organizational contexts.
Method: Bibliometric analysis using Scopus database with VOS Viewer software for network mapping and thematic clustering of publication trends from 2014-2025.
Jargon: SMTEs = Small and Medium Tourism Enterprises; bibliometric analysis = statistical analysis of publications to identify research trends and patterns.
Hyoju Kim, Jamie Klein-Packard, Keith Baxelbaum et al. (5 authors) β’ PsyArXiv
Finding: A 3-year longitudinal study of 263 children found that spoken word recognition involves five distinct dimensions, with activation rate showing the strongest developmental changes and links to language ability across school-age years.
Why it matters: This provides insights into how cognitive processing systems develop over time, which could inform educational technology design and learning sciences approaches to language instruction.
Method: Used Visual World Paradigm with eye-tracking and principal component analysis to identify orthogonal dimensions of lexical competition dynamics.
Jargon: Visual World Paradigm - eye-tracking method where participants look at pictures while hearing words to study real-time language processing; lexical competition - how the brain processes multiple possible word matches when hearing speech.