📚 Research Digest

Finding: People rate less effective pro-environmental behaviors as more moral than highly effective ones (a "morality-effectiveness disconnect"), but accurate labeling of climate impact can reverse this bias and increase policy support for more effective behaviors.

Why it matters: This is a behavioral science finding directly relevant to cognitive biases, decision-making miscalibration, and behavior change design—specifically how framing and information architecture (nudges/choice architecture) can correct moral intuitions that undermine collective welfare.

Method: Five main + six supplemental studies (N=9,247) across five countries, including an experimental condition testing impact-category labeling as an intervention.

Jargon: *Morality-effectiveness disconnect* = the inverse relationship between a behavior's objective climate impact and how morally virtuous people perceive it to be.

Finding: TikTok sessions frequently shift in purpose (42% end with a different function than they began), and well-being effects depend more on what function users end with than what they started with—escape-oriented use shows between-person but not reliable within-person associations with lower well-being.

Why it matters: Directly relevant to behavior change design, dark patterns, engagement mechanisms, and addiction-by-design research—the within-session dynamics reveal how app design (autoplay, recommendation algorithms) may pull users toward entertainment/escape loops regardless of initial intent.

Method: Experience sampling methodology (ESM) aligned with natural app session boundaries over 7 days, with preregistered multilevel modeling separating within- vs. between-person effects across 3,943 sessions.

Jargon: *Experience sampling*: real-time self-report triggered during daily life rather than retrospective surveys. *Multilevel modeling*: statistical approach that separates individual-level patterns from group-level trends. *Behavioural function*: the psychological purpose a behavior serves (here: entertainment, information access, social attention, escape/avoidance).

Finding: Proposes a closed-loop adaptive digital platform (NIPE) that personalizes psychoeducation based on real-time brain-behavior states (arousal, fatigue, attention) to improve learning, metacognition, and self-regulation in mental health contexts.

Why it matters: The adaptive personalization logic, behavior change mechanisms, and learning science principles (consolidation, skill transfer, readiness to learn) overlap with the professor's interests in behavior change design and learning sciences, though the clinical mental health framing makes it a weak fit.

Method: Conceptual framework combining rule-based logic, ML-based state estimation, and constrained generative AI for content delivery.

Jargon: *Closed-loop learning system*: system that continuously adjusts outputs based on real-time feedback from the user; *Neuromodulatory interventions*: treatments that directly alter brain activity (e.g., TMS, neurofeedback); *Psychoeducation*: structured patient education about their condition and treatment.

Finding: LLMs represent social networks using "multistep abstraction" (integrating direct and indirect connections) in ways that closely mirror human social reasoning, even when this strategy is suboptimal for the task.

Why it matters: Tangentially relevant to AI's human-like cognition and team/social dynamics research, but focused narrowly on LLM benchmark behavior rather than HCI, collaboration, or applied behavioral science.

Jargon: *Katz communicability*: a graph-theory model weighting all paths between nodes by length, used here to predict which social connections people perceive as "closer."

Finding: People seek social information more when stakes are higher, and risky others exert stronger influence than safe others on subsequent decisions under uncertainty, with individual differences in risk sensitivity predicting both baseline choices and social information-seeking behavior.

Why it matters: Tangentially relevant to behavioral science and decision-making, particularly around social influence and choice architecture, but focused on cognitive mechanisms rather than design applications.

Method: Computational modeling with egocentric discounting parameters across two large online replication samples (N≈400 each).

Jargon: *Egocentric discounting* = weighting one's own prior preference more heavily than social information; *risk/ambiguity task* = paradigm where risk = known probabilities, ambiguity = unknown probabilities.

Finding: Affirming mental capacities in robots creates more uncertainty than denying them, with facial features affecting both agency and experience perception while body features primarily affect agency; this was captured via mouse-tracking to reveal decision uncertainty beyond explicit ratings.

Why it matters: Tangentially relevant to HCI and human-robot interaction, but the focus is narrow experimental psychology of mind perception rather than interaction design, collaboration, or behavior change.

Method: Mouse-tracking to capture implicit decision uncertainty alongside explicit ratings — a methodologically interesting behavioral measure of cognitive conflict.

Finding: Social media algorithms and AI act as functional equivalents of religious structures by exploiting human vulnerabilities (needs for identity, perfection, salvation) to drive engagement through "rabbit holes" and individualized content delivery.

Why it matters: Tangentially relevant to dark patterns, addictive design, and persuasive technology—specifically how algorithmic architecture exploits psychological vulnerabilities—but the religious framing is not central to the professor's teaching areas.