📚 Research Digest

Finding: Editing AI-generated text increases psychological ownership, iterative prompting partially substitutes for editing, but fully manual writing produces substantially higher ownership than all AI-assisted conditions—even after controlling for effort, quality, and motivation.

Why it matters: Directly relevant to teaching about AI's impact on creative work, human-AI collaboration design, and behavior change design (how tool affordances shape users' sense of authorship and accountability).

Method: Preregistered mixed factorial experiment (N=273) crossing prompting style (single vs. iterative) × editing (yes/no) in an AI-assisted creative writing task.

Jargon: *Psychological ownership* — the subjective feeling that something is "mine," distinct from legal ownership; *iterative prompting* — steering AI through multiple back-and-forth exchanges rather than a single instruction.

Finding: This narrative review (192 articles) identifies trust miscalibration, communication inflexibility, and limited shared mental models as key barriers to Human-Agent Teaming (HAT), and argues that metacognitive awareness and cross-training are the most promising interventions for improving human-AI collaboration readiness.

Why it matters: Directly relevant to teaching AI's impact on teamwork and collaboration, and provides a practical framework for designing AI systems that support calibrated human-AI teaming.

Method: Structured narrative review of 572 screened records across 6 databases (2010–2026), synthesized around operational competencies and regulatory capacities.

Jargon:
- *Human-Agent Teaming (HAT)*: collaboration frameworks where AI systems are treated as teammates rather than tools
- *Trust calibration*: matching one's level of trust in an AI to its actual reliability/capability
- *Situation awareness*: real-time understanding of what is happening in a dynamic environment
- *Metacognitive awareness*: monitoring and regulating one's own cognitive processes, including recognizing the limits of one's understanding

Finding: The timing and ordering of attribute presentation during decision-making systematically reweights how people value pleasure vs. nutrition in food choices, with time pressure amplifying whichever attribute is processed first.

Why it matters: Directly relevant to choice architecture, nudges, and behavioral economics—specifically how interface/environment design (timing, salience, ordering) can steer decisions without changing options, which is core to persuasive technology and behavior change design.

Method: Binary food choice tasks with time pressure/delay conditions, analyzed via regression and drift-diffusion modeling (DDM) to model the cognitive accumulation process underlying decisions.

Jargon: *Drift-diffusion modeling* — a computational model of decision-making that treats choices as evidence accumulation over time toward a threshold; *intra-decisional dynamics* — how cognitive processing unfolds within a single decision episode; *value-based choice* — decisions made by comparing subjective values of options rather than by rule.

Finding: Humans feel in control when prepared to tackle future tasks, formalized via a controllability measure quantifying action range, outcome certainty, and outcome possibility — validated across three experiments on control preferences.

Why it matters: Directly relevant to behavioral science, motivation, and goal-setting (sense of agency affects behavior and well-being), and has implications for designing persuasive technology, choice architecture, and autonomy-supportive learning environments.

Method: Three experiments combining ratings of sense of control with reward-seeking/punishment-avoidance tasks, with a formal computational model of controllability.

Jargon: *Controllability* — a formal measure of how much an agent can influence outcomes; *agency* — the sense of being the cause of one's own actions and outcomes.

Finding: Combining ingroup identity appeals with intergroup comparison (club rankings) more than doubled recycling rates at a football stadium, while standalone identity appeals had no effect.

Why it matters: This is a strong real-world field experiment on nudges and choice architecture—specifically how social identity + competition can be engineered as a behavior change mechanism, directly relevant to teaching persuasive technology, gamification, and behavioral economics.

Method: Two preregistered season-long field experiments plus a post-game survey (N ≈ 13,000 total) in a naturalistic setting—unusually rigorous for a behavior change study.

Jargon: *Positive group distinctiveness* — the social identity theory concept that group members are motivated to differentiate their group favorably from others; *intergroup comparison* — ranking groups against each other on a shared metric.

Finding: Quadruped virtual agents (dog, cat, horse) using emotional expression and attention-guiding cues are more persuasive than baseline bark-only agents, and species-specific motion matters less than functional cues like intention readability for behavior change.

Why it matters: Directly relevant to persuasive technology and behavior change design, specifically how non-humanoid agents can be designed to influence behavior (trash disposal, smartphone use) with low psychological reactance—useful for teaching choice architecture and agent design principles.

Method: Controlled experiment comparing three behavioral conditions across three virtual animal species on real behavioral tasks, measuring both behavioral intention and actual behavior.

Jargon: *Psychological reactance* = resistance/pushback when people feel their freedom is threatened by persuasion; *intention readability* = how clearly an agent's goal or desired behavior is communicated to the user.

Finding: A large-scale replication (N=1,594) of Langer & Roth (1975) finds the original sequence effect replicates, but is attributable to neither the illusion of control nor the gambler's fallacy, and doesn't correlate with problem gambling severity.

Why it matters: Directly relevant to behavioral science and behavior change design — specifically the replication and validity of foundational cognitive bias constructs used in gambling psychology and persuasive technology design.

Method: 2x2 online experiment crossing sequence manipulation (early wins vs. late wins) with active involvement (self vs. other), N=1,594.

Jargon: *Illusion of control* = overestimating personal skill/influence over random outcomes; *Gambler's fallacy* = belief that past random outcomes influence future ones; *PGSI* = Problem Gambling Severity Index, a standard clinical measure.

Finding: LLM-assisted hardware debugging (Chat-Debugging) helps engineering students improve debugging confidence and skills by enabling natural language interaction with circuits, requiring students to assertively correct LLM errors.

Why it matters: Touches on AI impact on learning and skill acquisition, and HCI dynamics between students and AI assistants, but in a narrow domain (electrical engineering hardware debugging) with limited generalizability to the professor's core interests.

Method: Qualitative case study using constant comparative analysis of LLM chat logs and interviews with a single undergraduate student.

Finding: TIDES is a longitudinal dataset of 12 university project teams tracked over a semester, enabling modeling of multi-party conversation dynamics including emergent roles and team development stages.

Why it matters: The dataset's focus on real team dynamics, emergent roles, and longitudinal team evolution is tangentially relevant to team dynamics and collaboration research, though the paper's primary contribution is an NLP/LLM dataset rather than behavioral or HCI insights.

Method: Fine-tuned LLMs on naturalistic team meeting transcripts with socio-structural annotations for next-speaker prediction and utterance generation.

Jargon: *AMI Meeting Corpus* = a benchmark dataset of multiparty meeting recordings; *bigram baseline* = predicting next speaker based only on the two most recent speakers.

Finding: Complexity reduces the signal-to-noise ratio in evidence accumulation during risky choices, making decisions slower and less consistent, and biases people toward simpler options.

Why it matters: Marginally relevant to behavioral science and decision-making, but focused narrowly on computational modeling of risky choice rather than applied behavioral design, nudges, or HCI contexts.

Method: Three preregistered experiments manipulating complexity via number of outcomes, arithmetic expressions, and compound probabilities, analyzed through evidence-accumulation (drift-diffusion) modeling.

Jargon: *Evidence accumulation*: a model where decisions result from noisy sampling of information until a threshold is reached; *drift rate*: speed of evidence accumulation; *decision threshold*: amount of evidence required before committing to a choice.

Finding: A pixel-art game (Jejak Pagaruyung) was designed and implemented using Godot Engine to introduce Minangkabau culture through gameplay systems including minigames, NPC dialogues, and interactive cultural objects.

Why it matters: Tangentially relevant to serious games and cultural learning, but the study is a design/implementation report without empirical outcomes on learning or engagement effectiveness.

Method: Multimedia Development Life Cycle (MDLC) model with functional testing and questionnaires, but no outcome measurement of cultural learning.

Finding: PBL combined with Powtoon (an animation tool) improved motivation and critical thinking in 4th-grade math students versus a control group.

Why it matters: Marginally relevant to learning sciences and technology-enhanced learning, but the context (primary math education) and tool specificity are far from the professor's core interests.

Method: Quasi-experimental pretest-posttest with control group, using motivation questionnaire and critical thinking test.

Finding: A topic modeling analysis of 243 papers on AI in talent management identifies eight research themes, concluding that AI's value extends beyond automation to strategic talent enablement and continuous learning.

Why it matters: Tangentially relevant to AI's impact on work, but the focus is HR/talent management literature mapping rather than HCI, behavioral science, learning sciences, or team dynamics in any substantive way.