Kathrin Paimann, Elizângela Valarini, Sebastian Juhl • Semantic Scholar
Finding: A multi-method study (participatory design, expert review, meta-analysis, interviews) identifies eight core UX principles for human-AI agent interaction in workplace settings, providing actionable design guidelines for practitioners.
Why it matters: Directly relevant to HCI, UX design, and AI's impact on work—offers a structured framework that could anchor teaching on human-centered AI design and enterprise AI adoption.
Method: Notably combines participatory design workshops with meta-analysis and expert review to validate principles, making it both empirically grounded and practically oriented.
Jargon: *AI agents* = AI systems that autonomously perform tasks within workflows (vs. passive tools); *agentic AI* = AI with goal-directed, multi-step autonomous action capabilities in enterprise contexts.
Zeshu Zhu, Natalie Friedman, Kevin Weatherwax et al. (4 authors) • Semantic Scholar
Finding: HARP is a research platform that enables controlled experiments with live LLM-based AI agents, capturing behavioral metrics (keystroke pauses, deletions, prompt revision) alongside self-report measures to study how AI design choices (response length, technical specificity) affect users.
Why it matters: This platform directly enables rigorous HCI research on human-AI interaction, including studying persuasion, engagement, and behavior change in AI systems—relevant to understanding conversational AI design, proactive agents, and how AI output characteristics influence user cognition and retention.
Method: Controlled experiment using configurable live AI agents with behavioral logging (prompt composition time, deletions, keystroke pauses) plus triggered in-situ surveys—a methodological advance over static prototypes or post-hoc transcript analysis.
Jargon: *Response latency*: time between prompt submission and AI response; *model parameters*: settings controlling LLM behavior (temperature, token limits, etc.); *moderated usability sessions*: user testing with a researcher present to observe and probe.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Arnau Marin-Llobet, Steven A. Lehr, M. Banaji • Semantic Scholar
Finding: LLMs don't just reflect human cognitive biases—they amplify them across model generations and can transmit biases back to human users, making AI a net negative for debiasing decisions.
Why it matters: Directly relevant to behavioral science and decision-making courses, especially for teaching how AI tools interact with cognitive biases, persuasive technology ethics, and the dark patterns/ethical persuasion thread.
Method: Theoretical synthesis with proposed countermeasures across diagnostic, regulatory, and operational levels.
Jargon: *Alignment goals* = training objectives designed to make AI behave helpfully and safely, which ironically can embed and mask biases; *sociocognitive biases* = systematic errors in social perception and judgment.
Finding: Simpson's Paradox (where subgroup trends reverse upon aggregation) is reframed as both a statistical and cognitive problem, with behavioral economics concepts (framing effects, representativeness heuristics) explaining human susceptibility to data misinterpretation.
Why it matters: Tangentially relevant to behavioral science and decision-making/cognitive biases coursework, but the paper is a basic literature review with limited empirical or theoretical novelty for HCI or behavior change design contexts.
Jargon: *Simpson's Paradox*: when aggregated data shows the opposite trend from each subgroup individually; *representativeness heuristic*: judging probability by how closely something matches a prototype rather than base rates.
Finding: Adults' gaming addiction patterns vary by demographics (age, gender, marital status, occupation) and gaming habits (platform, genre, screen time), with coping motives and loss of control as central dimensions.
Why it matters: Marginally relevant to games/gamification research and behavior change design, specifically addiction-by-design and ethical persuasion concerns around game industry practices.
Method: Cross-sectional survey (n=391) using the Digital Game Addiction Scale for Adults (DGAS-A) with non-parametric analyses.
Jargon: *Coping* (using gaming to manage negative emotions); *Sad/Angry Withdrawal* (emotional distress when unable to play); *Loss of Control* (inability to regulate gaming behavior).
⚠️ Self-reported data only | No control group mentioned | Single study (not replicated) | Cross-sectional design (no longitudinal data)
Liman Wang, Randy Yang Gao, Minah H. Jung et al. (5 authors) • PsyArXiv
Finding: Chinese online participant pools produce data comparable to American platforms across 21 established behavioral and cultural phenomena, with diverse demographics and high reliability.
Why it matters: Marginally relevant as methodological infrastructure for behavioral science replication and cross-cultural research, but doesn't address any of the core teaching topics directly.
⭐ Pre-registered ⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract