📚 Research Digest

Finding: Decision complexity causes people to rely on heuristics, leading to systematic biases and errors across domains like financial products and tax reform; the dissertation examines this in three empirical chapters.

Why it matters: Directly relevant to behavioral science, cognitive biases, decision-making under uncertainty, and choice architecture/nudges—core topics for teaching behavioral economics and persuasive technology design.

Jargon: *Decision heuristics* = mental shortcuts that simplify complex choices; *choice architecture* = how the structure of options influences decisions (e.g., defaults).

Finding: The SCAFFOLD framework steers LLM behavior toward developmentally appropriate learning interactions for adolescents, combining technical reliability layers with principles from developmental psychology, learning science, and HCI; a pilot with 12–16-year-olds using an LLM-powered social robot for collaborative mnemonic design showed more engagement and on-topic participation compared to prompt-only LLMs.

Why it matters: Directly relevant to AI's impact on learning environments, collaborative learning design, and HCI for educational technology, with a co-creation/team task element.

Method: Pilot classroom study (n = adolescents, ages 12–16) using a social robot running an LLM, comparing SCAFFOLD prototype vs. prompt-only baseline on engagement and post-test knowledge scores.

Jargon: *SCAFFOLD* = Steered Contextual AI Framework for Orchestrating Learning Dialogue, a modular system wrapping LLM output with verification, repair, and fallback layers; *model-agnostic* = works with any underlying LLM; *sensitive periods* = developmental windows where experiences have outsized impact on brain/skill formation.

Finding: Social influence shapes when people stop sampling information and what they choose, but not how they sample; people copy others only when personal information is ambiguous, leading to mild information cascades where group decisions become more homogeneous than solo decisions.

Why it matters: Directly relevant to behavioral science, collective intelligence, and decision-making—explains how social information propagates through groups and distorts individual judgment, with implications for team decisions, nudge design, and choice architecture.

Method: Preregistered experiment with free sampling paradigm (participants draw samples before deciding), complemented by computational modeling and agent-based simulations to link individual rules to collective dynamics.

Jargon: *Information cascade* — when individuals abandon private information to follow others' choices, causing a chain of imitation. *Experience-based decisions* — choices where probabilities must be inferred from sampling outcomes rather than given explicitly. *Stopping rule* — the decision about when to stop gathering information and commit to a choice.

Finding: Describes the development process of a mobile smoking cessation app for youth using an interdisciplinary integration of behavioral science (e.g., delay discounting, cue devaluation) and UX design methods, including iterative playtesting with 380+ users and a youth advisory board.

Why it matters: Directly relevant to behavior change design, persuasive technology, and the practical challenges of translating behavioral science into engaging digital interventions—useful for teaching habit formation, nudges, and ethical persuasion design.

Method: Iterative, interdisciplinary co-design process using focus groups, playtests, and qualitative questionnaires with adolescent and young adult users across development cycles.

Jargon: *Delay discounting* = tendency to prefer smaller immediate rewards over larger future ones; *smoking cue devaluation* = reducing the appeal of stimuli associated with smoking; *Bloombox Toolkit* = a framework for integrating science and design in intervention development.

Finding: AI can facilitate opinion change, but only for people with higher intellectual humility and greater trust in AI — both independently and jointly predict acceptance of counter-attitudinal information.

Why it matters: Directly relevant to behavioral science, persuasive technology, and AI's societal impact — particularly for understanding when and for whom AI-driven persuasion works, with implications for behavior change design and ethical AI deployment.

Method: Two-study online experiment (N=294, British sample) measuring attitude change on immigration after brief GPT interactions, with dispositional measures of intellectual humility and AI trust as moderators.

Jargon: *Intellectual humility (IH)* — recognition of the limits of one's own knowledge and openness to other viewpoints; *epistemic vigilance* — the tendency to critically evaluate information sources before accepting claims; *counter-attitudinal information* — content that challenges one's existing beliefs.

Finding: An integrated pedagogical framework combining flipped learning, exemplar-based reference, and GenAI-assisted visual prototyping in a knit yarn design course showed exploratory improvements in creativity, design skills, and problem-solving compared to a historical control cohort.

Why it matters: Directly addresses AI's impact on creative/design education and skill acquisition, with practical implications for how GenAI tools fit into studio-based learning workflows.

Method: Quasi-experimental comparison between historical control (N=12) and intervention cohort (N=16), supplemented by questionnaires and interviews — small sample, no random assignment, explicitly framed as exploratory rather than confirmatory.

Jargon: *Flipped learning* = students consume instructional content (videos) before class, freeing class time for hands-on practice; *GenAI-assisted visual prototyping* = using generative AI image tools to explore design ideas before physical sampling.

Finding: A conceptual model explaining how AI-assisted decision-making can narrow (or widen) the intention-behavior gap in sustainable consumption, identifying three enabling mechanisms (diagnostic simplification, calibrated confidence, self-regulatory reinforcement) and three countervailing ones (algorithmic bias, miscalibrated reliance, dependency/moral licensing).

Why it matters: Directly relevant to behavior change design, choice architecture, nudges, and behavioral economics—especially how AI-mediated persuasion/nudging can succeed or backfire, and the ethics of automation in shaping human decisions.

Method: Conceptual/theoretical paper (mislabeled as empirical) proposing testable propositions and a three-stage empirical programme (experiments, longitudinal field study, algorithmic audits).

Jargon:
- *Intention-behavior gap*: the well-documented disconnect between intending to act sustainably and actually doing so
- *Calibrated reliance*: appropriately trusting AI neither too much nor too little
- *Automation bias*: over-trusting automated recommendations at the expense of independent judgment
- *Algorithm aversion*: distrust/underuse of algorithmic recommendations
- *Moral licensing*: using one virtuous act (e.g., getting AI help to recycle) as psychological permission to behave less virtuously elsewhere
- *Bias migration*: individual cognitive errors shift to systemic errors embedded in AI data and filtering

Finding: Visual attention during decision-making activates attributes associated with the attended option, and the valence of those activated attributes (positive vs. negative) depends on contextual factors like stimulus desirability and behavioral goals, explaining when gaze-choice links are stronger or weaker.

Why it matters: Directly relevant to behavioral science and decision-making — this mechanistic account of how attention shapes choice has implications for choice architecture, nudges, and persuasive design (e.g., how product placement or UI layout directs gaze to influence decisions).

Method: Eye-tracking combined with think-aloud protocol plus a new computational model of gaze-driven attribute activation.

Jargon: *Attribute activation* — the process by which looking at an option brings specific features/attributes of that option to mind; *gaze-choice link* — the empirical regularity that people tend to choose options they look at longer.

Finding: Otome game engagement (frequency, emotional investment, financial investment) is positively associated with identity development, real-life social interaction, and romantic expectations in young adult Filipinas, but does not predict psychological well-being.

Why it matters: Relevant to games/gamification research and behavior change design—specifically how parasocial relationships and in-game purchase mechanics drive emotional investment and identity formation, touching on engagement design and potentially dark patterns (monetization hooks).

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

Jargon: *Otome games*: narrative romance games targeting women featuring relationship-building with male characters. *Parasocial interaction*: one-sided emotional relationships users form with media characters. *Uses and Gratifications Theory*: framework explaining why people actively choose specific media to satisfy psychological needs.

Finding: A systematic mapping of 84 chess research families spanning human players, classical engines, neural/RL systems, LLMs, and hybrid approaches reveals that planning, metacognition, and human–AI collaboration are understudied relative to move evaluation and action selection.

Why it matters: The chess-as-model-domain framing connects directly to AI impact on human collaboration and decision-making, and the identified gaps in human–AI complementarity and metacognitive calibration are relevant to teaching about collective intelligence and skill acquisition.

Method: Systematic mapping study (not meta-analysis) classifying 84 study families by agent type, strategic-reasoning stage, and evaluation dimension.

Jargon: *Metacognitive calibration* = accuracy of one's self-assessment of one's own reasoning quality; *grounded explanation* = AI explanations tied to verifiable, meaningful domain features rather than post-hoc rationalizations; *human–AI synergy* = the specific claim that human+AI together outperforms either alone (distinct from merely showing improved human performance with AI assistance).

Finding: Proposes the BRIDGE model for human-centered digital transformation in higher education, identifying three tensions (upskilling vs. deskilling, autonomy vs. dependence, individualization vs. collaboration) as central challenges for Generation Alpha entering universities.

Why it matters: Touches on AI's impact on learning, skill acquisition, and collaboration in educational contexts, but remains at a high-level agenda-setting level without empirical findings or actionable design insights.

Jargon: *Generation Alpha*: cohort born ~2010–2025, growing up with smartphones and generative AI as defaults; *Community of Practice*: group of practitioners sharing expertise around a domain.

Finding: Chess players allocate thinking time resource-rationally — spending more time when computation is more valuable and when they have more time available, resulting in better moves.

Why it matters: Provides empirical evidence about human planning efficiency under constraints, relevant to decision-making, cognitive resource allocation, and expertise research.

Method: Large-scale naturalistic data analysis of online chess games examining move time, move quality, and move entropy across time control conditions.

Jargon: *Resource-rational analysis* — framework evaluating whether cognitive effort allocation matches the value of additional computation given constraints; *move entropy* — measure of variability/uncertainty in move selection.

Finding: Proposes that employees develop "conditioned appraisal meanings" — learned associations between recurring workplace cues (meetings, supervisor phrases, performance reviews) and specific emotional/cognitive appraisals that shape workplace emotion and behavior.

Why it matters: Relevant to behavioral science and decision-making, particularly how prior experience shapes automatic cognitive responses — with implications for habit formation, nudge design, and behavior change in organizational contexts.

Jargon: *Conditioned appraisal meaning*: a learned association where a recurring cue reliably triggers a specific appraisal (e.g., threat, challenge) due to repeated past experience; *appraisal*: cognitive evaluation of an event's relevance to one's goals or concerns.

Finding: Indebted women farmers in the Philippines make financial decisions through socio-cognitive frameworks where seemingly "irrational" debt behaviors are actually rational adaptive strategies under structural gender constraints, synthesizing Behavioral Finance Theory with Feminist Economics.

Why it matters: Directly relevant to behavioral science and cognitive biases—specifically challenges deficit-framing of financial decision-making and reframes "bias" as contextually rational behavior, which is useful for teaching behavioral economics and decision-making under constraint.

Method: Interpretative Phenomenological Analysis (IPA) with 12 participants; qualitative/phenomenological approach.

Jargon: *IPA (Interpretative Phenomenological Analysis)*: qualitative method focused on participants' lived experience and meaning-making. *Behavioral Finance Theory*: study of psychological influences on financial decisions. *Feminist Economics*: framework analyzing how gender structures economic outcomes and access.

Finding: Proposes a Bayesian multilevel time-varying autoregressive statistical framework for evaluating intensive longitudinal interventions (ILIs) that goes beyond simple pre-post comparisons to capture when, how, and for whom intervention effects emerge and decay.

Why it matters: The methodological framework is potentially useful for researchers studying behavior change interventions (habit formation, persuasive technology), but the paper is primarily a statistical methods contribution rather than a substantive finding about behavior change, HCI, or any of the core topic areas.

Method: Bayesian multilevel time-varying autoregressive models within a generalized additive mixed model (GAMM) framework, demonstrated on a mindfulness-based ILI targeting negative affect.

Jargon: *Intensive longitudinal interventions (ILIs)*: interventions that deliver repeated touchpoints in daily life while collecting frequent psychological measurements (e.g., via smartphone). *TV-AR models*: autoregressive models where parameters (like the influence of past states on current states) change over time. *GAMM*: generalized additive mixed models, a flexible regression approach for nonlinear and multilevel data.

Finding: Structure learning and metacognition co-evolve in context-dependent ways — how well people monitor their own learning accuracy depends heavily on stimulus type and environmental complexity, not just general metacognitive ability.

Why it matters: Marginally relevant to learning sciences and skill acquisition, particularly around how people calibrate confidence during unsupervised learning, which touches on self-regulated learning and feedback design.

Method: Within-subjects, two-session design (N=63) with exposure-test cycles measuring both accuracy and confidence judgments across two stimulus types (fractals vs. line drawings) in a probabilistic learning environment.

Jargon: *Metacognitive sensitivity* = ability to accurately distinguish correct from incorrect responses using confidence ratings; *structure learning* = inferring statistical regularities (co-occurrences, sequences) from the environment without explicit instruction.

Finding: Employees experience digital (im)balance with work ICTs in 8 distinct ways, and high connectivity doesn't automatically cause imbalance—emotional experience depends on configuration of connectivity processes and organizational context.

Why it matters: Marginally relevant to behavior change design around technology use and digital well-being, but the organizational/HR framing is far from the professor's core interests in HCI, creativity, teams, or behavioral science.

Method: Ideal Type Analysis on 40 qualitative interviews—an abductive typology-building approach.

Jargon: *Digital Connectivity Processes (DCPs)*: mechanisms like interference, boundary blurring, exposure, and time displacement through which W-ICTs affect employees. *Ideal Type Analysis*: a qualitative method that constructs theoretically coherent "pure types" from empirical variation.

Finding: This review distinguishes between "parameter uncertainty" (known unknowns) and "structural uncertainty" (unknown unknowns about which model of the world applies), proposing "hypothesis spaces" as a framework to explain anxious cognition like worry, catastrophizing, and overgeneralization.

Why it matters: The structural uncertainty / hypothesis space framework has peripheral relevance to decision-making under uncertainty and cognitive biases, but the paper is primarily a clinical anxiety theory, not a behavioral science or HCI paper.

Jargon: *Hypothesis space*: the set of candidate world-models a person considers plausible; *structure learning*: inferring which generative model governs an environment; *intolerance of uncertainty*: a trait linked to anxiety disorders.

Finding: Culturally adaptive AI platforms align with language educators' professional capital (human, social, decisional) across under-resourced multilingual contexts, with high engagement rates across feedback, collaboration, and analytics features.

Why it matters: Tangentially relevant to AI impact on work and learning sciences, but focused narrowly on language teacher professional development rather than HCI, creativity, teams, or behavioral science as the professor teaches them.