📚 Research Digest

Finding: In a longitudinal field experiment, visible AI identity labels reduced engagement with AI posts by ~3.5 percentage points and increased criticism of AI contributions by ~13.3 percentage points compared to unlabeled AI accounts in a Reddit-like social platform.

Why it matters: Directly relevant to AI impact on human collaboration and HCI—reveals how disclosure design choices (labeling vs. hiding AI identity) shape social dynamics in mixed human–AI communities, with implications for transparency policy and persuasive technology ethics.

Method: Longitudinal field experiment (21 days, 36 humans + 72 LLM-powered AI accounts) using Bayesian multilevel models on 2,143 human replies across labeled vs. unlabeled communities.

Jargon: *Credible interval* = Bayesian equivalent of confidence interval; *YSocial* = custom Reddit-like simulation platform for human–AI interaction research; *social uptake* = whether a post receives replies/engagement from others.

Finding: A prediction-based intervention (strengthening people's expectations that they will act goal-consistently in tempting situations) reduced self-reported impulsive behaviors more than go/no-go training over a 3-week smartphone-based study.

Why it matters: Directly relevant to behavior change design—offers a novel mechanism (predictive self-modeling) for habit formation and impulse control interventions, with implications for persuasive technology and app-based behavior change tools.

Method: Two preregistered studies; Study 2 used ecological momentary assessment (EMA) via daily smartphone surveys over 21 days with a 3-condition design (prediction intervention vs. go/no-go training vs. waitlist).

Jargon: *Go/no-go training* — a cognitive inhibition task where users practice stopping responses to certain stimuli; *inferential theory* — the view that behavior is guided by inferences/predictions about what one will do, not just impulses; *EMA (ecological momentary assessment)* — repeated real-time self-reports in daily life contexts.

Finding: Proposes a multiplayer game prototype ("Impression, Sunrise") that forces cross-role negotiation by distributing control of perceptual parameters across three distinct production roles, using Impressionist painting principles as a design framework for collaborative sensemaking.

Why it matters: Directly relevant to games/gamification, team collaboration, and HCI/UX design — the paper addresses how serious game mechanics can scaffold cross-disciplinary vocabulary and shared mental models in design teams.

Method: Unity 2D networked prototype used as a "boundary negotiating artifact" — a design object that mediates communication between people with different expertise.

Jargon: *Boundary negotiating artifact* — an object that helps people from different disciplines coordinate and translate across their differing knowledge frameworks; *graded legibility* — the degree to which visual information is rendered recognizable vs. atmospheric/ambiguous; *phenomenological modes* — ways of experiencing or perceiving that precede conscious conceptualization.

Finding: This panel paper examines how generative AI is reshaping the collaboration dynamics between academic CSCW researchers and tech industry practitioners, exploring opportunities, challenges, and strategies for this relationship going forward.

Why it matters: Directly relevant to AI's impact on work and human collaboration, and touches on HCI research and education — core teaching interests.

Method: Panel/position paper format (CHI 2026 + CSCW 2025 panels); not empirical.

Jargon: *CSCW* = Computer-Supported Cooperative Work, a research community focused on how technology supports group work and collaboration; *HCI* = Human-Computer Interaction.

Finding: CollabPEAS is a collaborative VR system that teaches AI literacy by having learner pairs jointly specify AI agent behavior using the PEAS framework (Performance measure, Environment, Actuators, Sensors) in shared virtual reality with block-based coding.

Why it matters: Directly relevant to HCI/UX design, collaborative learning, and the learning sciences—specifically how VR and collaborative interfaces can scaffold conceptual understanding of AI systems, which connects to AI literacy as a growing educational concern.

Method: Design paper presenting a system with three integrated components (PEAS-guided design flow in VR, block-based coding, collaboration supports); no empirical evaluation yet conducted.

Jargon: *PEAS framework* — a structured decomposition of AI agents into Performance measure, Environment, Actuators, and Sensors; *AI literacy* — the competency to understand, evaluate, and work with AI systems.

Finding: Online training with expert video feedback significantly improves novice social workers' decision quality (especially risk assessment and information evaluation) compared to training alone, suggesting expert feedback accelerates expertise development in high-stakes, high-turnover fields.

Why it matters: Directly relevant to learning sciences and skill acquisition, with implications for how deliberate practice and expert modeling can be designed into training systems—connects to behavioral science (decision-making under uncertainty) and HCI (video-based feedback tools).

Method: Experimental design (n=83) with feedback vs. no-feedback conditions using realistic child protection scenarios and video feedback from experienced practitioners.

Finding: A replication and qualitative extension testing a Goal-Directed Predictive Processing (GDPP) intervention for impulse control showed heterogeneous results—only 2 of the GDPP participants showed reliable behavior reduction—with qualitative themes revealing barriers like motivational issues, craving, and poor fit between intervention rationale and target behavior.

Method: Single-case experimental design with daily behavioral tracking across baseline/training/post-training phases, plus semi-structured interviews analyzed thematically.

Why it matters: Directly relevant to behavior change design (habit formation, impulse control interventions, personalization of nudges), and the replication framing with mixed results offers honest teaching material about intervention efficacy and the limits of uniform behavior change approaches.

Jargon: *Goal-Directed Predictive Processing (GDPP)*: a framework positing that behavior change occurs by building predictive inference chains linking risk situations to goal-consistent responses, rather than relying on willpower alone. *Single-case analysis*: statistical method evaluating treatment effects for individual participants rather than group averages.

Finding: VR-based guided-discovery instruction outperforms lecture for science learning outcomes, with engagement, motivation, and collaboration in VR showing small positive effects on science process skill development.

Why it matters: Directly relevant to learning sciences, educational technology, and the pedagogical value of immersive simulations—useful for teaching about HCI, serious games, and technology-enhanced learning environments.

Method: Pretest-posttest nonequivalent control group design (n=342) with PLS-SEM and exploratory factor analysis to model predictors of skill development.

Jargon: *SDG4* = UN Sustainable Development Goal 4 (quality education); *PLS-SEM* = Partial Least Squares Structural Equation Modeling, a statistical method for testing complex causal models; *guided-discovery instruction* = learning approach where students explore content with structured guidance rather than direct instruction.

Finding: A structured math teaching intervention integrating cognitive and metacognitive strategies (based on Zimmerman's self-regulation model) significantly improved high school students' self-regulation, attitudes toward math, and motivation over 8 weeks.

Why it matters: Directly relevant to learning sciences, skill acquisition, and goal-setting — particularly the role of metacognitive strategies, self-monitoring, and goal-setting in academic motivation and behavior change.

Method: Mixed-methods quasi-experimental design (n=11th graders, 8 weeks) using Self-Regulation, Attitude, and Motivation scales plus semi-structured interviews analyzed via content analysis.

Jargon: *Zimmerman's cyclical model* — a self-regulated learning framework with forethought, performance, and self-reflection phases; *metacognitive monitoring* — awareness and tracking of one's own comprehension and strategy effectiveness.

Finding: Positive emotions drive diverse, exploratory creative behaviors while negative emotions promote refinement and structural control in clay art creation; action diversity partially mediates the effect of emotion states on originality.

Why it matters: Directly informs understanding of how emotional states shape creative cognition and behavior strategies, relevant to creativity research, design processes, and behavioral science of creative performance.

Method: Within-subject experiment with physiological measures, computer vision extraction of hand movement features, linear mixed-effects modeling, and mediation analysis — a notably multimodal measurement approach to studying creative behavior.

Jargon: *Linear mixed-effects modeling* — statistical method accounting for repeated measures within the same participants; *mediation analysis* — tests whether a variable (action diversity) explains the pathway between cause (emotion) and outcome (originality).

Finding: TiltTogether, a novel cross-device tilt-based interaction for co-located multiplayer mobile games, promotes collaborative problem-solving and enhances social gameplay experience, as evaluated through a playtest workshop with 16 participants.

Why it matters: Directly relevant to HCI/UX design, collaborative technology, and games research—particularly around designing interaction mechanics that support teamwork and shared goals in co-located settings.

Method: Playtest workshop (n=16) gathering qualitative insights on usability, collaboration, and social experience.

Jargon: *Proxemics* = study of how physical space between people affects interaction; *cross-device interaction* = coordination across multiple separate physical devices; *affordances* = perceived action possibilities of an interface element.

Finding: Integrating a generative AI learning agent into cognitive apprenticeship-based problem-based learning (AI-CAS-PBL) improved medical students' knowledge outcomes, reduced cognitive load, and increased self-efficacy compared to conventional PBL and lecture-based learning.

Why it matters: Directly relevant to AI's impact on learning, skill acquisition, and instructional design—particularly how AI tools can scaffold novice learners in complex problem-solving contexts.

Method: Quasi-experimental design comparing three instructional conditions (AI-CAS-PBL, conventional PBL, lecture-based) using NASA-TLX for cognitive load and technology acceptance measures.

Jargon: *Cognitive apprenticeship* — a framework where expert thinking is made visible to learners through modeling, coaching, and scaffolding; *NASA-TLX* — NASA Task Load Index, a multi-dimensional scale measuring perceived workload; *Coze platform/Doubao 1.5 Pro* — a Chinese AI development platform and large language model used to build the learning agent.

Finding: AI-enabled fraudulent participation (bots, AI-generated responses) is undermining human-subjects research integrity, and existing bot-detection methods are increasingly ineffective against these new tools.

Why it matters: Directly relevant to HCI researchers and those running behavioral studies or online experiments—any empirical work in these areas is now vulnerable to this problem, affecting research validity in behavior change, gamification, and learning sciences studies.

Method: Workshop format with breakout discussions by research method type, producing community guidelines for the CSCW/SIGCHI community.

Finding: Black online communities experience disproportionate content suppression and demonetization by personalization algorithms, and this paper proposes studying these experiences through Afropessimism, Afrofuturism, and Digital Black Feminism frameworks within HCI/CSCW.

Why it matters: Directly relevant to HCI and the ethics of algorithmic systems, touching on dark patterns, algorithmic bias, and the tension between engagement-optimizing platforms and equitable user treatment.

Jargon: *Afropessimism* — a theoretical framework examining anti-Blackness as a structural condition; *CSCW* — Computer-Supported Cooperative Work, a subfield of HCI focused on how technology supports groups; *Critical algorithmic studies* — interdisciplinary examination of how algorithms embed social power and bias.

Finding: AI acceptance (via performance and effort expectancy) predicts career goal-setting through serial mediation of AI trust and career confidence, with intrinsic motivation amplifying the trust-to-confidence link.

Why it matters: Touches on AI's impact on work/career preparation, goal-setting, and motivation — but at a surface level that doesn't deeply advance any of these areas for teaching purposes.

Method: Survey of 539 Chinese undergraduates using serial mediation + moderation analysis (UTAUT + SCCT framework).

Jargon: *UTAUT* (Unified Theory of Acceptance and Use of Technology) — framework predicting tech adoption via performance expectancy, effort expectancy, etc.; *SCCT* (Social Cognitive Career Theory) — career development model emphasizing self-efficacy and outcome expectations; *serial mediation* — two mediators in sequence linking predictor to outcome.

Finding: AI-enabled smart wearable learning systems improved academic performance (62.4 to 78.6) and engagement (87.3%), with student-centered design and learning analytics effectiveness being the strongest predictors of achievement.

Why it matters: Touches on AI impact on learning, personalized education, and learning analytics—relevant to learning sciences and AI's role in education, though the wearable/sensor angle is peripheral to the professor's core interests.

Method: Sequential explanatory mixed-methods design (N=320 students, 24 teachers) using PLS-SEM, Pearson correlation, and multiple regression alongside thematic analysis of interviews and focus groups.

Jargon: *PLS-SEM (Partial Least Squares Structural Equation Modeling):* a statistical technique for modeling relationships between latent variables; *Learning analytics:* data-driven measurement and reporting of learner behavior to optimize learning.

Finding: Math instruction designed around Bloom's Digital Taxonomy significantly improved 8th-grade students' academic achievement, with students reporting mostly positive experiences in technology-integrated learning environments.

Why it matters: Relevant to learning sciences and educational technology design, particularly how taxonomic frameworks can scaffold technology-integrated instruction.

Method: Explanatory sequential mixed-methods design with n=20, combining a researcher-made achievement test with qualitative open-ended questionnaires — small sample limits generalizability.

Jargon: *Bloom's Digital Taxonomy* — an adaptation of Bloom's classic cognitive hierarchy updated to include digital tools and activities at each level (e.g., remembering → googling; creating → programming).

Finding: People's intuitions about counterfactual reasoning are systematically biased and fall far below normative standards, even with feedback; decisions follow coherently from flawed estimates rather than from flawed decision logic itself.

Why it matters: Relevant to behavioral science and cognitive biases coursework, particularly around decision-making under uncertainty and limits of learning from experience.

Method: Three experiments (N=354) with trial-by-trial feedback conditions, comparing novices vs. domain experts (statisticians, economists, data scientists).

Jargon: *Counterfactual reasoning* = inferring what would have happened under a different action; *observational vs. experimental data* = naturally occurring data vs. randomized controlled data.

Finding: Implementing the STAD (Student Teams Achievement Division) cooperative learning model in 5th-grade Indonesian language classes improved mean scores from 64.31 to 87.59 and mastery rates from 44.83% to 93.10% across two action research cycles.

Why it matters: Provides evidence for structured cooperative learning's effectiveness on academic outcomes, tangentially relevant to team dynamics and learning sciences topics.

Method: Classroom Action Research (CAR) with two cycles (plan-act-observe-reflect), n=29 students, using observation checklists and cognitive achievement tests.

Jargon: *STAD*: a cooperative learning model where heterogeneous teams study together but are individually assessed, with team scores based on individual improvement; *Classical learning mastery*: percentage of students meeting a minimum score threshold.

Finding: Individual differences (fatigue, pain sensitivity, sex, agreeableness, motion sickness susceptibility) predict cybersickness severity and tolerance in VR, explaining ~25-29% of variance in outcomes.

Why it matters: Marginally relevant to VR/HCI design—cybersickness is a real barrier to VR adoption, and these findings suggest designers should account for user state and individual differences when deploying immersive systems.

Method: Hierarchical regression across N=252 participants exposed to one of five VR games, measuring both self-reported symptoms (SSQ) and behavioral tolerance (time in VR).

Finding: This theoretical paper redefines motivation as socially situated and interactional, integrating productive disciplinary engagement and expansive framing frameworks with mainstream motivation theories (self-determination, expectancy-value, achievement goal theories).

Why it matters: Touches on motivation and learning sciences, but the focus is narrowly on equity pedagogy and educational psychology rather than goal-setting, behavior change, or HCI-relevant motivation mechanisms.

Finding: This conceptual paper proposes a modular active learning framework (MALI) for vocational chemistry education, integrating cognitive load theory, constructivism, and self-determination theory into ten structured 90-minute modules.

Why it matters: The theoretical synthesis touches on learning sciences and instructional design, but the narrow focus on vocational chemistry in Chinese higher education limits transferability to the professor's core teaching areas.

Jargon: *Implementation fidelity* = degree to which an intervention is delivered as intended; *cognitive-affective perspective* = framework treating cognition and emotion as jointly influencing learning outcomes.