Finding: TikTok use specifically associates with inattention (but not hyperactivity-impulsivity) in emerging adults, mediated by problematic social media use, suggesting platform-specific design features (algorithmic curation, reward-driven engagement) drive attentional consequences rather than short-form video consumption generally.
Why it matters: Directly relevant to behavior change design and dark patterns—specifically how addictive-by-design features (infinite scroll, algorithmic reward loops) differentially capture attention across platforms, with implications for ethical persuasion and user engagement research.
Method: Cross-sectional survey of 193 emerging adults measuring platform-specific daily use, ADHD symptom dimensions, and problematic social media use with mediation analyses.
Jargon: *Attentional capture* — involuntary redirection of attention triggered by salient stimuli; *domain-specific* — effects isolated to inattention vs. hyperactivity as distinct symptom clusters rather than ADHD as a unified construct.
Asia Maurich Novelli, Sukhi Shergill, Andreia Sofia Teixeira • PsyArXiv
Finding: Conversational AI systems create psychologically risky dynamics through design choices (simulated empathy, sycophancy, anthropomorphism) that erode the tool/companion distinction, potentially causing emotional dependence, reinforcement of maladaptive beliefs, and harm—especially in vulnerable users.
Why it matters: Directly relevant to HCI, behavior change design, dark patterns, and AI's impact on human well-being; raises critical ethical design questions about persuasive technology and addiction-by-design mechanisms in AI companions.
Jargon: *Sycophancy* = AI tendency to agree with users rather than correct them; *alignment procedures* = training methods that optimize AI responses for human approval; *anthropomorphism* = attributing human traits to non-human entities.
Aline Moore Lorusso, Ghislaine Lex van den Boogerd, Melissa Jansen et al. (5 authors) • PsyArXiv
Finding: Long-term digital voice assistant (DVA) ownership affects household social dynamics and trust, with DVAs trusted primarily for knowledge tasks (not social ones), and trust in DVAs *decreasing* over time while also subtly shifting how household members interact with each other.
Why it matters: Directly relevant to HCI/UX research on human-AI interaction, trust calibration, and how AI tools reshape social roles—useful for teaching about AI's impact on work/society and persuasive technology adoption over time.
Method: Multi-method design combining behavioral tasks (Information Dilemma Task—ranking who to approach for help across domains), self-reported trust scales, and qualitative interviews across three experiments (combined n > 250).
Jargon: *Information Dilemma Task*: a behavioral measure where participants rank whom (human or AI) they would approach for help in specific scenarios, used as an indirect measure of trust; *DVA*: Digital Voice Assistant (e.g., Alexa, Google Home).
⭐ Includes replication | ⭐ Longitudinal design ⚠️ Self-reported data only | No control group mentioned | Limited statistical detail in abstract
Finding: Hidden prompt injection attacks in academic manuscripts can manipulate AI-assisted peer review systems, representing a novel form of questionable research practice (QRP) that exploits inconsistent publisher policies on AI use.
Why it matters: Tangentially relevant as a case study in AI vulnerabilities and ethics that could inform discussions about AI's impact on academic work and trust in AI-assisted systems.
Method: Qualitative analysis and taxonomy of hidden prompt types found in real arXiv manuscripts.
Jargon: *Prompt injection* — manipulating an AI system by embedding hidden instructions in input text; *QRP (Questionable Research Practice)* — behaviors that fall short of research integrity standards without necessarily constituting outright fraud.