Finding: This structured integrative review argues that GenAI in software engineering education risks creating "comprehension debt"—where AI-assisted production outpaces learners' ability to understand, test, and justify the code—and proposes a framework linking AI delegation to evidence and accountability obligations.
Why it matters: The comprehension debt concept and human-AI collaboration framing have moderate relevance to learning sciences, skill acquisition, and AI's impact on work, though the domain focus (software engineering education specifically) is narrow for this professor's interests.
Method: Structured integrative review of research and practice from 2023–2026, supplemented by established literature on AI literacy, technical debt, and human-AI collaboration.
Jargon: *Comprehension debt* = deferred learning/maintenance cost when AI-generated artifacts accumulate faster than users can understand them; *AASEE* = AI-Augmented Software Engineering Education framework.
Alberto Monge Roffarello, Luigi De Russis, Massimiliano Pellegrino • Semantic Scholar
Finding: A multimodal LLM prompting technique embedded in a Figma plugin can detect attention-capturing dark patterns (infinite scroll, autoplay) in UI prototypes and suggest wellbeing-oriented alternatives, with designers finding it useful for early-stage reflection.
Why it matters: Directly addresses dark patterns, ethical persuasion, behavior change design, and HCI/UX practice—all core to teaching about persuasive technology, user engagement mechanisms, and the ethics of attention engineering.
Method: Multimodal prompting technique evaluated by 4 deceptive design experts, then integrated into a Figma plugin and tested in a formative lab study with 15 UI/UX designers.
Jargon: *ACDPs (Attention-Capturing Damaging Patterns)* — UI features like infinite scroll and autoplay designed to maximize engagement at the cost of user wellbeing; *Digital wellbeing by design* — proactively embedding wellbeing considerations into the design process rather than treating them as afterthoughts.
Caterina Moruzzi, Charlotte Bird, L. Herman • Semantic Scholar
Finding: AI-powered Creativity Support Tools (CSTs) are being appropriated by creative professionals in complex ways across workflow stages, with mixed impacts on accessibility, expertise, and creative agency in collaborative contexts.
Why it matters: Directly addresses AI's impact on creative work, HCI/UX design, and human-AI collaboration—core teaching areas for this professor, and provides practitioner-grounded insights on how generative AI reshapes creative workflows.
Method: Multi-session exploratory study with creative professionals, allowing longitudinal observation of tool use across workflow stages rather than single-session snapshots.
Jargon: *CSTs (Creativity Support Tools)*: software designed to augment human creative processes; *appropriation*: how users adapt tools in ways beyond their intended design.
Finding: Students using constructive cognitive strategies (conceptual combination, semantic) with GenAI produce more original ideas, while AI-assisted profiles (even AI-Reliant Reproducers) produce higher quality ideas than largely independent thinkers.
Why it matters: Directly informs how to design AI-collaborative learning environments that support rather than undermine creativity, and provides a typology of human-AI collaboration patterns relevant to teaching creativity and understanding AI's impact on cognitive work.
Method: Lab-based observational study (n=146 undergrads) combining deductive/inductive coding of cognitive strategies with latent profile analysis to identify collaboration archetypes.
Jargon: *Latent profile analysis* — statistical method that identifies subgroups of individuals based on patterns across multiple variables; *conceptual combination* — merging distinct concepts to generate novel ideas; *assumption reversal* — challenging default assumptions to spark creative alternatives.
Finding: College students in a participatory design study rejected guilt-based, streak/leaderboard gamification for wellness apps, preferring flexible, low-pressure, peer-anchored mechanics that support meaning-making without penalizing inaction.
Why it matters: Directly addresses gamification design, dark patterns in persuasive technology (guilt-based feedback, streaks), behavior change ethics, and participatory design methods—all core to HCI, behavioral science, and games teaching.
Method: Used narrative tabletop RPG-adapted focus groups (N=30) as "playful research probes" to elicit design preferences; analyzed transcripts with reflexive thematic analysis interpreted through the MDA (Mechanics–Dynamics–Aesthetics) framework.
Jargon: *MDA framework*: a game design lens connecting rules (Mechanics) → emergent patterns (Dynamics) → emotional experience (Aesthetics). *Reflexive thematic analysis*: qualitative method where researcher positionality is treated as part of the analytical process, not a bias to eliminate.
⚠️ Small sample (n=30) | No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Shi Yin, Ru-Qiao Wu, Xue-Jun Liu et al. (4 authors) • Semantic Scholar
Finding: GenAI dependence (functional vs. existential) positively predicts research creativity in graduate students, mediated serially by critical thinking and human-AI collaboration quality, with functional dependence having a much stronger effect than existential dependence.
Why it matters: Directly relevant to AI's impact on creative work, human-AI collaboration, and learning sciences — especially for teaching about how students engage with AI tools in research contexts.
Method: SEM + PROCESS Model 6 (serial mediation) with 1,157 graduate students and 5,000 bootstrap resamples.
Jargon: *Functional dependence* = task-oriented reliance on GenAI for specific outputs; *Existential dependence* = deeper psychological reliance where one's sense of capability is tied to AI availability; *Serial mediation* = indirect effect where Variable A → B → C → outcome in sequence.
⚠️ Self-reported data only | Cross-sectional design (no longitudinal data)
Laura Allen, Caitlin Mills, Angela Stewart et al. (7 authors) • PsyArXiv
Finding: Proposes a taxonomy of five canonical roles AI can play in educational technology (Cognitive Extender, Assessor, Social Partner, Environment, Orchestrator), arguing that treating "AI" as a monolith has produced contradictory and unusable evidence about its educational effectiveness.
Why it matters: Directly useful for teaching about AI's impact on learning, skill acquisition, and HCI design — especially the "Cognitive Extender" and "Social Partner" roles which map onto creativity support tools and collaboration research.
Method: Theoretical framework development grounded in distinct learning science traditions for each role.
Shuyan Feng, Martin J. Koch, Astrid Carolus • PsyArXiv
Finding: A triadic AI Mindset framework (dispositional factors, contextual characteristics, technology appraisals) significantly predicts teachers' AI competence and AI use, with habit showing the strongest effect and fear showing a paradoxical positive direct but negative indirect effect.
Why it matters: The psychological variables studied (habit, hedonic motivation, fear, effort expectancy) directly map to behavioral science concepts relevant to behavior change design, technology adoption, and AI impact on professional work in educational settings.
Method: Structural equation modeling with 224 teachers testing a theoretically derived framework integrating social-cognitive and psychological theories.
Jargon: *Hedonic motivation* = intrinsic enjoyment-based motivation to use technology; *effort expectancy* = perceived ease of use; *social facilitation* = support from social environment enabling behavior.
María Camila Samper Meza, Alexander Bustamante-Martínez, E. Galvis-Lista et al. (4 authors) • Semantic Scholar
Finding: A scientometric analysis of 9,346 documents (1982–2025) on usability and UX evaluation reveals that SUS and heuristic evaluation remain dominant methods, while empirical and technology-assisted approaches are growing, with gaps identified when applying these frameworks to open data visualization systems requiring visual reasoning and sense-making.
Why it matters: Provides a useful map of the HCI evaluation landscape—relevant for teaching UX methods and for identifying where current evaluation frameworks fall short for complex, visualization-heavy systems.
Method: Bibliometric/scientometric analysis using keyword co-occurrence in VOSviewer on 9,346 Scopus-indexed documents—a reasonable method for tracking field evolution but limited in depth per paper.
Jargon: *Scientometric analysis* = quantitative study of scientific publications to map research trends; *SUS (System Usability Scale)* = standardized 10-item questionnaire measuring perceived usability; *ODVS (Open Data Visualization Systems)* = publicly accessible platforms for exploring and interpreting datasets; *VOSviewer* = software for visualizing bibliographic networks.
David Gonçalves, J. Godinho, Pedro Trindade et al. (6 authors) • Semantic Scholar
Finding: Modular multiplayer game design—where players engage in distinct but interconnected gameplay loops tailored to individual play styles—can accommodate diverse group preferences but risks fragmenting shared experience and social cohesion.
Why it matters: Directly relevant to games/gamification research and raises important questions about how collaborative experiences can be personalized without undermining the social bonding that makes group play valuable—applicable to team dynamics and engagement design.
Method: Mixed-methods study with 32 gamers and 2 game developers evaluating a proof-of-concept prototype.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Finding: Different team climate configurations (not a single best climate) facilitate entrepreneurial opportunity recognition across distinct creative stages (preparation, conception, elaboration), with each stage having different cognitive demands requiring different contextual conditions.
Why it matters: Directly relevant to teaching team dynamics, collective creativity, and collaboration — particularly how to design team environments that support creative work at different stages.
Method: Multi-method design combining PLS-SEM, fuzzy-set qualitative comparative analysis (fsQCA), and necessary condition analysis (NCA) on 88 entrepreneurial teams — notable for capturing non-linear, asymmetrical relationships that traditional linear models miss.
Jargon: *fsQCA* = analyzes combinations of conditions (not single variables) that produce outcomes; *NCA* = identifies conditions that are necessary (not just correlated) for an outcome; *configurational lens* = focuses on how combinations of factors together produce results, rather than independent effects.
Hu Meilan, Gabriel Rongyang Lau, Adalia Yin Hui Goh et al. (6 authors) • PsyArXiv
Finding: A practical tutorial for psychological researchers on building and deploying customized LLM-based chatbots, covering no-code through code-based approaches with attention to ethics, reproducibility, and data governance.
Why it matters: Tangentially useful for HCI/behavioral science researchers wanting to deploy LLM agents as research tools (similar to the relevant PerceptUI paper), but this is a methods tutorial for psychologists rather than a substantive finding about behavior, cognition, or human-AI interaction.
Method: Compares no-code (chatbot builders), low-code (visual platforms), and code-based (APIs, local models) development approaches across dimensions like control, cost, and reproducibility.
Sara Morgado-Nunes, C. Patino-Alonso • Semantic Scholar
Finding: A statistical framework for analyzing repeated multimodal HCI data (e.g., physiological sensors) that accounts for within-person dependence, showing naive OLS produces 30.8% false-positive rates vs. 6.2% for proper mixed models.
Why it matters: Directly relevant to HCI researchers designing and analyzing studies with repeated measures or physiological/behavioral data—the methodological lesson about pseudoreplication is important for rigorous HCI research design.
Method: Multilevel mixed-effects modeling applied to a stress-recovery dataset (35 participants, 175 observations) with simulation under null effects to demonstrate false-positive inflation from naive analysis.
Jargon: *Pseudoreplication*: treating non-independent observations (e.g., multiple readings from same participant) as if independent. *Intraclass correlation (ICC)*: proportion of variance attributable to between-person differences. *False-discovery-rate correction*: statistical adjustment to control spurious findings across multiple comparisons.
⭐ Includes replication | ⭐ Longitudinal design ⚠️ Small sample (n=35) | No control group mentioned
Alemayehu Wubale, Alemayehu B. Mengstie, Dagnachew Kebede • Semantic Scholar
Finding: Game-based training (GBT) produces significantly greater task cohesion (group integration around collective goals) than drill-based training in youth volleyball over 12 weeks, though social cohesion develops similarly under both methods.
Why it matters: Relevant to gamification and serious games research—demonstrates that game-structured practice environments strengthen task-oriented collaboration, which connects to both team dynamics and games/gamification literatures.
Method: Randomized pretest-posttest comparison (n=40) using the Group Environment Questionnaire; effect size is large (Cohen's d = 1.08) but small sample warrants caution.
Jargon: *Group Integration-Task (GI-T)*: a subscale measuring members' perceptions of unity around the team's collective task goals, distinct from social attraction dimensions.
⚠️ Very small sample (n=20) | Self-reported data only | Single study (not replicated)
Ariane S. Willems, Katharina Dreiling • Semantic Scholar
Finding: Students' motivational profiles (interest, self-concept, goal orientations) systematically predict how they perceive adaptive learning support from teachers, with more favorably motivated students reporting higher perceived support across six dimensions.
Why it matters: Relevant to goal-setting (mastery vs. performance orientations), motivation theory, and learning sciences, though focused narrowly on secondary school language instruction rather than HCI or design contexts.
Method: Latent profile analysis on 807 students to identify motivational clusters, with two-wave survey design.
Jargon: *Adaptive learning support (ALS)*: teacher adjustments to instruction based on individual student needs; *Latent profile analysis*: statistical method identifying subgroups with similar response patterns across multiple variables.
Finding: A physician-scientist's autobiographical account argues that generative AI functions as an "intellectual companion" that addresses structural barriers to scholarly productivity (unavailability of collaboration at the moment of need) rather than cognitive ones.
Why it matters: Touches on AI's impact on knowledge work and collaboration, and raises questions about how AI changes solo vs. collaborative creative/scholarly processes—relevant to AI impact on work and HCI, but only tangentially.
Method: Single-case autobiographical narrative offered explicitly as hypothesis-generating rather than evidence.
Finding: MVCAP is a motion-based CAPTCHA framework that exploits temporal visual processing (motion segregation) to create a human-agent perception gap, achieving 99.6% human accuracy vs. 16.8% for best GUI agents.
Why it matters: Peripherally relevant to HCI (CAPTCHA UX, accessibility) and AI's limitations in human-like perception, but not directly tied to the core teaching/research areas.
Ni Made Citariani, K. Agustini, I. K. Suartama et al. (6 authors) • Semantic Scholar
Finding: An AR framework combining brain-based learning principles with Balinese cultural values (Tri Kaya Parisudha) improved prosocial behavior, critical thinking, and communication in early childhood settings.
Why it matters: Tangentially relevant to learning sciences and educational technology design, but the early childhood character education focus is far from the professor's core interests.
Method: ADDIE instructional design model with quasi-experimental evaluation and expert validity review.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Larry John D. Sumalpong, Erniel Ghrizcer G. Ecle, Wenceslao A. Coronado et al. (5 authors) • Semantic Scholar
Finding: A quasi-experimental study showing that think-pair-share (TPS) instructional materials developed via the ADDIE model improved statistics achievement, formative scores, and problem-solving in grade 11 STEM students.
Why it matters: Marginal relevance to collaborative learning and pedagogy, but too narrowly focused on K-12 statistics instruction to be directly useful for HCI, creativity, teams, behavioral science, or the professor's core teaching areas.
Jargon: *Think-pair-share (TPS)*: a collaborative learning strategy where students think individually, discuss with a partner, then share with the class. *ADDIE model*: instructional design framework (Analyze, Design, Develop, Implement, Evaluate).
⚠️ No control group mentioned | Single study (not replicated)
Waad Gharbi Saleh, Ahmed Najm Abdullah, Basim Majid Mohammed • Semantic Scholar
Finding: VR-assisted training significantly improves cognitive-motor skill acquisition and floor exercise technique in gymnastics students compared to conventional instruction.
Why it matters: Marginally relevant to learning sciences and skill acquisition via immersive technology, but the narrow domain (gymnastics technique in a small school sample) limits applicability to HCI, behavior change, or AI topics the professor focuses on.
Method: Small RCT (n=26, split into experimental/control), with motor performance evaluated via observation, interview, and video recording.
⚠️ Very small sample (n=13) | Single study (not replicated)
P. Ramola, K. Veni, S. J. S. Jeyamalathi et al. (5 authors) • Semantic Scholar
Finding: Adaptive learning tools used as scaffolded instruction aligned with Zone of Proximal Development (ZPD) produced significant pre-to-post improvements in English speaking and writing proficiency among 40 undergraduate students over six weeks.
Why it matters: Tangentially relevant to learning sciences and skill acquisition, but narrowly focused on English language teaching with weak methodology (no control group, n=40, one-group pre-post design).
Method: One-group pre-test/post-test design with paired sample t-tests — notably weak without a control group, making causal claims unreliable.
Jargon: *ZPD (Zone of Proximal Development)*: Vygotsky's concept of the gap between what a learner can do independently vs. with guidance. *Scaffolded instruction*: temporary support structures that help learners reach higher performance levels.
⚠️ Small sample (n=40) | Single study (not replicated) | Convenience sample (e.g., MTurk, students)
Finding: Psychiatry's clinical expertise on unreliable self-report data (cognitive distortions, clinical presentation gaps) is largely absent from AI development, creating a "blind spot" in how AI systems model human behavior.
Why it matters: Tangentially relevant to behavioral science and AI's impact on society, but focused narrowly on psychiatric data reliability rather than HCI, behavior change design, or human-AI collaboration.