📚 Research Digest

Finding: A longitudinal study with design students found that generative AI use alone doesn't improve creative self-beliefs, but specific AI usage styles can increase creative self-efficacy, though students view AI primarily as an efficiency tool rather than a creativity enhancer.

Why it matters: This directly informs how to teach AI integration in creative disciplines and reveals the gap between AI's potential creative benefits and students' perceptions of their own creativity.

Method: Longitudinal experimental design with three conditions (no AI, mandatory AI, optional AI) measuring Creative Self-Efficacy and Creative Personal Identity across four time points while tracking usage patterns and attitudes.

Jargon: Creative Self-Beliefs (CSB) - one's beliefs about their creative abilities; Creative Self-Efficacy (CSE) - confidence in creative abilities; Creative Personal Identity (CPI) - seeing creativity as part of one's identity.

Finding: Introduces the Translational Literacy Framework (TLF), a pedagogical architecture that teaches students to evaluate and communicate scientific evidence across disciplines and to public audiences, using AI as a critical tool rather than passive resource.

Why it matters: Directly addresses how to design educational experiences that develop cross-disciplinary collaboration skills and critical AI literacy - essential competencies for future team-based work and decision-making.

Method: Proposes a three-component pedagogical framework combining convergent triadic instruction (interdisciplinary co-teaching), modular disciplinary architecture, and AI interrogation methods.

Jargon: Meta-competency = higher-order skill integrating multiple lower-order abilities; Convergent Triadic Instruction = coordinated teaching across natural science, social science, and information science faculty.

Finding: Math games based on cognitive science principles about children's intuitive number and geometry concepts significantly improved learning outcomes in two large randomized trials in Delhi, both immediately and one year later, even when scaled to larger classroom groups led by regular teachers.

Why it matters: This demonstrates how cognitive science insights can be translated into scalable educational games that enhance learning, providing a model for evidence-based game design in educational contexts.

Method: Two randomized controlled trials with nearly 5,000 children across preschools and government schools, comparing game-based interventions to standard instruction with longitudinal follow-up.

Jargon: Ready-to-scale interventions - educational programs designed to be easily implemented widely without specialized resources or personnel.

Finding: Using GenAI for emotional support correlates with higher depression and anxiety symptoms, while using it for learning/exploration associates with better mental health outcomes, with gender-specific patterns in these relationships.

Why it matters: This directly informs how AI tools should be designed for human wellbeing and reveals critical considerations for AI's impact on human collaboration and social connection.

Method: Cross-sectional survey of 1,003 U.S. young adults using path modeling to examine associations between GenAI use motivations and mental health symptoms across demographic groups.

Jargon: Internalizing symptoms - mental health issues like depression and anxiety that are directed inward rather than expressed through external behaviors.

Finding: Generative AI use during adolescence presents a critical developmental tension—while it may offer safe spaces for social practice and identity exploration, it risks displacing essential real-world experiences needed for healthy psychological development including social belonging, emotion regulation, and autonomy.

Why it matters: This directly informs how we design AI systems and educational interventions to support rather than undermine crucial developmental processes, with implications for long-term human-AI collaboration patterns.

Method: Theoretical synthesis of interdisciplinary research on adolescent development and genAI impacts, proposing a developmental science framework for future research.

Jargon: GenAI = generative artificial intelligence tools like ChatGPT; distress tolerance = ability to withstand negative emotions without immediate relief-seeking.

Finding: Higher loneliness among Polish school teachers correlates with stronger anthropomorphization of AI chatbots and more problematic AI use patterns, but anthropomorphization doesn't mediate this relationship, and overall social motivation for AI use was low in this work-focused sample.

Why it matters: This reveals important psychological factors that influence how educators interact with AI tools, informing design of educational AI systems and understanding potential risks of AI dependency in professional contexts.

Method: Survey of 102 Polish teachers examining relationships between loneliness measures, anthropomorphization scales, and problematic AI use indicators.

Jargon: Anthropomorphization = attributing human-like qualities, emotions, or consciousness to AI systems; Problematic AI use = patterns of AI interaction resembling behavioral addiction symptoms.

Finding: Haptic feedback significantly enhances people's perception of causality between events, with the largest benefit coming from adding a second sensory modality, and both the number and physical realism of multisensory cues boost causal judgments.

Why it matters: This directly informs HCI design by showing how to make interactive systems feel more natural and causally coherent through strategic use of haptic feedback alongside visual and auditory cues.

Method: Three psychophysical experiments using visual launching paradigms combined with force-based haptics, vibrotactile feedback, and audio, modeled with Bayesian multimodal inference.

Jargon: Haptic feedback = touch-based sensory information; launching paradigm = experimental setup where one object appears to cause another to move; psychophysical = measuring perception of physical stimuli.

Finding: Personality traits predict employee turnover risk as effectively as employment history (AUC = 0.896 vs 0.894), with explainable AI revealing complex non-linear patterns like V-shaped effects for agreeableness and threshold effects for other traits.

Why it matters: This demonstrates how AI can uncover complex behavioral patterns in decision-making contexts and highlights the predictive power of personality over traditional resume metrics, relevant for understanding human behavior in organizational settings.

Method: Used XGBoost classifiers with nested cross-validation on 2,126 job seekers and SHAP (SHapley Additive exPlanations) framework to interpret the AI model's decision logic.

Jargon: SHAP - a method for explaining individual predictions by computing the contribution of each feature; XGBoost - an ensemble machine learning algorithm; AUC - Area Under the Curve, a performance metric for binary classification.

Finding: State rumination causally disrupts working memory performance and increases perceived cognitive effort through cognitive interference, with specific impairments in inhibitory control during non-match trials.

Why it matters: This provides crucial evidence for how mental states affect both objective performance and subjective effort in cognitive tasks, informing design of interfaces and learning environments that account for users' cognitive load and mental state.

Method: Preregistered online experiment with random assignment to rumination vs. distraction conditions, using n-back working memory task and drift-diffusion modeling to analyze underlying decision processes.

Jargon: Rumination - repetitive negative thinking patterns; Drift-diffusion modeling - computational method to separate decision accuracy from response caution in cognitive tasks; n-back task - working memory test requiring identification of stimuli from n trials back.

Finding: Researchers used a Delphi method with clinical and scientific experts to create a "Matching Matrix" that links specific cognitive behavioral therapy modules to treatment targets for chronic pain, aiming to personalize treatments while avoiding intuitive biases.

Why it matters: This demonstrates a structured approach to expert consensus that could inform personalized intervention design in HCI systems and addresses cognitive biases in clinical decision-making.

Method: Used an adapted group Delphi technique with 22 clinicians across multiple rounds plus expert validation to systematically match interventions to targets.

Jargon: Delphi method - structured communication technique using expert panels to reach consensus on complex issues.

Finding: This paper introduces a computational model (Multinomial Processing Tree) that formalizes the two-stage theory of sentence production - hierarchical structure building followed by linearization - and validates it using brain imaging data from individuals with aphasia.

Why it matters: Provides a formal computational framework for understanding how humans encode language structure, which could inform HCI design for language interfaces and contribute to behavioral science understanding of cognitive processing stages.

Method: Uses Multinomial Processing Tree modeling to analyze syntactic error patterns in aphasia patients, correlating computational estimates with neural substrates through brain imaging.

Jargon: MPT (Multinomial Processing Tree) = statistical model for decomposing cognitive processes into component stages; agrammatism/paragrammatism = specific types of language production deficits in aphasia.

Finding: Students with Special Educational Needs and Disabilities (SEND) in England have significantly higher risks of becoming NEET (Not in Education, Employment, or Training), with odds ratios ranging from 1.34 to 2.54 compared to students without SEND, even after controlling for sociodemographic factors.

Why it matters: This identifies vulnerable populations in educational systems and could inform inclusive design principles for educational technologies and interventions targeting at-risk students.

Method: Large-scale longitudinal analysis of 3.9 million students using England's National Pupil Database with fixed and mixed-effects models.

Jargon: NEET = Not in Education, Employment, or Training; SEND = Special Educational Needs and Disabilities; Statement/EHCP = formal educational support plans in the UK system.