📚 Research Digest

Finding: Frequent AI chatbot use significantly correlates with undergraduate students' communication, creativity, and critical thinking development, but not collaboration skills — based on a cross-sectional survey of 382 students across four Ghanaian universities.

Why it matters: Directly relevant to teaching about AI's impact on learning, skill acquisition, and human-AI collaboration dynamics, with practical implications for HCI and learning sciences contexts.

Method: Quantitative cross-sectional survey with stratified sampling (n=382); statistical analysis of frequency of chatbot use against 21st-century skill dimensions.

Jargon: *21st Century Skills* = the "4Cs" framework: Communication, Collaboration, Creativity, Critical Thinking — widely used in education research as core competencies for modern work.

Finding: Proposes a "flourishing-centered AI" framework integrating subjective well-being, psychological functioning, character virtues, and psychological richness as design goals for AI systems, explicitly addressing tensions between short-term vs. long-term benefits and individual vs. collective outcomes.

Why it matters: Directly relevant to behavior change design, persuasive technology ethics, and AI's impact on human flourishing — especially the tension between engagement-maximizing design (dark patterns, addiction by design) and genuine user well-being.

Method: Theoretical framework synthesis drawing from philosophy, behavioral science, HCI, and AI design.

Jargon: *Psychological richness* — a dimension of well-being focused on variety, perspective change, and complexity of experience, distinct from hedonic or eudaimonic well-being; *Eudaimonic well-being* — flourishing through meaning, virtue, and self-realization rather than pleasure.

Finding: Stronger AI models selectively involve human partners based on who holds the task advantage (route length), while weaker models show no such selectivity—demonstrating that reduced human involvement can reflect correct division of labor rather than collaboration failure.

Why it matters: Directly relevant to AI-human collaboration design and collective intelligence questions: how should systems decide when to defer to humans vs. act autonomously, and how do we evaluate "good" collaboration when AI capability varies?

Method: Controlled shared-objective maze experiment with variable human/agent route advantages, testing four LLMs (GPT, Command A+, Qwen, Command R) in both solo and human co-play conditions.

Jargon: *Task-conditioned collaboration* = adjusting human involvement based on who has the advantage for a specific subtask; *co-play* = real-time human-AI joint task execution; *route-conditioned human involvement* = deciding to hand off control based on path length comparison.

Finding: AI-driven personalized learning and learning analytics significantly improve student engagement and academic achievement in higher education, with engagement scores and GPAs both higher for students using AI tools versus traditional settings (β = .41, r = .63).

Why it matters: Directly relevant to teaching about AI's impact on learning, skill acquisition, and the ethical dimensions of AI in educational contexts, including data privacy and equity concerns.

Method: Mixed methods with 500 undergraduate students (quantitative regression/correlation) plus 30 semi-structured interviews (qualitative), though labeled "theoretical" which is inconsistent with the described empirical approach.

Jargon: *Learning analytics* = data-driven analysis of learner behavior to improve educational outcomes; *adaptive learning pathways* = AI-adjusted content sequencing based on individual student performance.

Finding: When matchmaking algorithms are cognitively salient during gameplay, players externally attribute losses to the algorithm rather than themselves, reducing enjoyment—but this effect only occurs after losses, not wins.

Why it matters: Directly relevant to HCI (human-algorithm interaction), games/gamification design, and behavioral science (attribution theory, cognitive biases), with practical implications for how algorithmic transparency paradoxically harms user experience.

Method: Exploratory sequential mixed-methods: qualitative interviews generated a process model, tested via quasi-field survey with retrospective recall design.

Jargon: *Algorithm salience* = degree to which an algorithm is cognitively prominent/top-of-mind during use; *matchmaking algorithm* = system that pairs players of similar skill in competitive games; *external attribution* = blaming outcomes on outside forces rather than oneself; *design–experience paradox* = making algorithmic systems more visible/transparent can worsen user experience.

Finding: Otome game engagement (emotional/financial investment, usage frequency) correlates with identity development, social interaction, and romantic expectations among young adult Filipinas, but not with psychological well-being.

Why it matters: Touches on user engagement mechanics, parasocial relationships, and in-game purchase motivation—relevant to game design, gamification, and persuasive technology/engagement design research.

Method: Quantitative correlational design (n=118) using Spearman's rho and multiple linear regression with validated scales.

Jargon: *Otome games*: narrative romance games targeting women with fictional male partners. *Parasocial interaction*: one-sided emotional bond users form with media characters. *Uses and Gratifications Theory*: framework explaining why users choose specific media based on needs.

Finding: Reviews when color in visual design helps vs. hurts usability and for whom, covering color vision deficiency, age-related decline, and evidence quality behind common color-mood-performance claims.

Why it matters: Has marginal relevance to HCI/UX design and interface design principles, but is primarily a human factors/perceptual psychology review with limited connection to the professor's core interests in creativity, teams, AI, behavioral science, or learning.

Finding: Shared autonomy (algorithm-assisted grasp control) increased sense of agency in able-bodied controls but *decreased* it in actual prosthesis users with amputation, suggesting algorithmic assistance can undermine user agency in intended populations.

Why it matters: The finding that algorithmic assistance can reduce rather than enhance user agency is tangentially relevant to HCI debates about human-AI collaboration and the tradeoffs of automation/AI assistance on perceived control and ownership.

Method: Within-subjects VR physics simulation with 58 participants (29 with transradial amputation, 29 matched controls) across four conditions combining control mode × vibrotactile feedback type.

Jargon: *Myoelectric prosthesis* = prosthetic limb controlled by electrical signals from residual muscles; *shared autonomy* = blending user motor commands with algorithmic assistance; *prosthesis embodiment* = psychological integration of the prosthetic into one's body schema; *phantom-limb control* = perceived ability to control the sensation of a missing limb.