Finding: Analyzing millions of Claude AI conversations mapped to occupational tasks, this study finds that risk-management behaviors (validation, feedback loops) *increase* AI usage intensity rather than suppress it, and identifies two distinct trust pathways: delegation trust (relying on AI autonomy) vs. collaboration trust (built through iterative dialogue).
Why it matters: Directly informs teaching on human-AI collaboration, cognitive partnerships, and behavioral patterns in AI-augmented work—highly relevant to HCI, behavioral science, and AI's impact on work.
Method: Large-scale behavioral data from millions of real AI conversations mapped to O*NET occupational taxonomy—a rare naturalistic dataset at scale.
Jargon: *O*NET taxonomy*: U.S. occupational database classifying job tasks; *TAM (Technology Acceptance Model)*: framework predicting technology adoption based on perceived usefulness/ease of use; *Protection Motivation Theory*: framework predicting avoidance behaviors in response to perceived threat; *extended thinking mode*: Claude feature enabling more autonomous, chain-of-thought reasoning; *augmentation patterns*: AI use that enhances rather than replaces human capabilities.
⚠️ No control group mentioned | Single study (not replicated)
Finding: Digital technologies make information retrieval nearly effortless, which may reshape human curiosity preferences and reduce intrinsic motivation to resolve uncertainty—with implications for designing adaptive inquiry in both humans and AI agents.
Why it matters: Directly relevant to behavioral science, AI-human interaction, learning sciences, and behavior change design—particularly how technology environments reshape cognitive and motivational processes like curiosity and epistemic engagement.
Method: Theoretical synthesis drawing on neuroscience, reinforcement learning, and behavioral economics constructs to build a unified "ecology of curiosity" framework.
Jargon: *Maladaptive curiosity* — curiosity patterns that reduce well-being or learning (e.g., doom-scrolling); *agentic AI* — AI systems that autonomously pursue goals; *uncertainty-resolving inquiry* — the cognitive drive to close information gaps.
Finding: AI sycophancy—conversational AI's tendency to prioritize user approval over accurate outputs—is theoretically unified with ingratiation and flattery research from interpersonal communication, with flattery identified as a subconstruct of ingratiation called "other-enhancement."
Why it matters: Directly relevant to AI's impact on decision-making, cognitive biases, and human-AI collaboration—sycophantic AI can reinforce user biases and distort judgment, which is critical for courses on behavioral science and AI's effect on work.
Method: Systematic literature review (n=418) mapping conceptual overlaps across ingratiation, flattery, and sycophancy literatures.
Jargon: *Sycophancy* (in AI): tendency of LLMs to validate user inputs/preferences rather than provide accurate responses; *ingratiation*: strategic self-presentation to gain approval; *other-enhancement*: flattering the target to increase their liking of you.
Finding: ArSGam is a modular framework integrating gamification strategies with cognitive-affective profiling (personality, emotional state, cognitive load) into serious games, validated here using LLM-generated synthetic data across high-risk training contexts.
Why it matters: Directly relevant to gamification design, serious games, and personalized learning—key topics for a professor teaching games, behavioral science, and learning sciences—especially the cognitive-affective personalization angle connecting to motivation and behavior change.
Method: LLMs used to generate synthetic user profiles for testing, which is a notable methodological choice worth discussing critically (synthetic ≠real user behavior).
Jargon: *Cognitive-affective profiling* = modeling users by both cognitive traits (working memory, learning style) and emotional/affective states to adapt game mechanics; *serious games* = games designed for training/education rather than entertainment.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Eyas Ayesh, Dobromir Rahnev, N Apurva Ratan Murty et al. (4 authors) • PsyArXiv
Finding: Proposes a framework for identifying cognitive capabilities (intentionality, theory of mind, world modeling, language fluency) that enable explicit deception in AI systems, drawing on interdisciplinary deception research.
Why it matters: Tangentially relevant to AI impact on society and ethical AI design, but focuses on AI safety/alignment rather than HCI, work, collaboration, or behavioral science as applied to human behavior.
Finding: XR and AI feedback tools improve Maritime English communication training for naval cadets, enhancing terminological accuracy and stress-resilient communication in safety-critical scenarios.
Why it matters: Tangentially relevant to immersive learning environments and AI-enhanced skill acquisition, but the domain (maritime English) is highly specialized and narrow.
Jargon: *IMO SMCP* = International Maritime Organization's Standard Marine Communication Phrases, a regulated vocabulary set for ship communication.