Chu Zhang, XiaoKe Zeng, Jin Zhang et al. (7 authors) • ArXiv
Finding: FIERO, a hybrid physical-digital card game using generative AI, significantly improved creative writing outcomes (idea fluency, novelty, plot coherence) compared to online collaborative writing alone, demonstrating that tangible card scaffolding plus AI augmentation enhances collaborative creativity without reducing player agency.
Why it matters: Directly hits multiple core interests: HCI/UX design of creative tools, collaborative creativity and team dynamics, gamification/serious games, and AI's role in augmenting human creative work.
Method: Controlled experiment (N=60) comparing FIERO against online collaborative writing, measuring idea fluency, novelty, intuitive stimulation, and narrative coherence.
Jargon: *Idea fluency* = quantity of ideas generated; *narrative coherence* = logical consistency of story plot across contributors; *creative scaffolding* = structural supports that constrain and guide creative thinking productively.
⚠️ Small sample (n=60) | No control group mentioned | Single study (not replicated)
Finding: TrailLM is a system design that augments LLM-based search to support metacognitive reflection and iterative sensemaking, rather than just rapid answer delivery.
Why it matters: Directly relevant to HCI/UX design and learning sciences—raises important questions about how AI tools can be designed to preserve effortful cognition and self-monitoring during exploratory learning.
Method: System design paper presenting TrailLM, which visualizes and reconstructs user exploration paths to scaffold reflection.
Jargon: *Exploratory search* = open-ended information seeking in unfamiliar domains; *sensemaking* = the iterative process of building mental models from new information; *metacognitive engagement* = monitoring and regulating one's own thinking processes.
Finding: JobMate, an AI system that converts social media career posts into conversational persona-based agents, shifts users from passive feed browsing (triggering upward social comparison) to active dialogue, reducing detrimental comparison while preserving the emotional grounding of authentic peer content.
Why it matters: Directly relevant to HCI/UX design, AI impact on work and career exploration, and behavioral science around social comparison and cognitive overload in digitally-mediated contexts.
Method: Between-subjects study (N=24, three disciplines) comparing AI-mediated conversational system vs. native social media browsing.
Jargon: *Upward social comparison* = comparing oneself unfavorably to higher-status peers; *sensemaking* = the process of constructing understanding from ambiguous information; *persona-grounded agents* = AI chatbots embodying characteristics of real social media users.
⚠️ Single study (not replicated) | Limited statistical detail in abstract
Finding: In a crossover field experiment across 121 employees, AI-rewritten emails shifted emotional tone (playful = more positive, professional = less positive), but tone changes only affected open/reply rates *indirectly* through within-sender emotional positivity, not directly through AI use itself.
Why it matters: Directly relevant to AI's impact on workplace communication and collaboration—shows AI writing assistance changes behavior through emotional mediation, not just content quality, with implications for HCI design of AI writing tools.
Method: Randomized crossover field experiment across 16,880 real emails with three conditions (unaided, GPT-5 playful, GPT-5 professional), measuring behavioral outcomes (open rates, reply rates, response times) plus sentiment analysis—unusually rigorous ecological validity for this topic.
Jargon: *Crossover design*: same participants experience all conditions sequentially; *indirect pathway*: AI affects behavior only through an intermediary variable (emotional tone), not directly.
Wen Ying, Adil Rahman, Erzhen Hu et al. (4 authors) • ArXiv
Finding: HandPad uses asymmetric bimanual hand gestures in VR to enable device-free window management, where the non-dominant hand sets spatial context and the dominant hand performs precise manipulation on a palm-as-touchpad surface.
Why it matters: Relevant to HCI/UX design research on embodied interaction and VR productivity interfaces, and touches on input modality design principles useful for teaching interaction design.
Method: Exploratory user study evaluating efficiency and ergonomics of the bimanual interaction suite.
Jargon: *Asymmetric bimanual coordination* — using each hand for distinct roles (framing vs. executing); *passive haptic feedback* — using the physical surface of one's own palm to simulate tactile resistance without actuators; *spatial remapping* — translating hand position in physical space to cursor/interaction position on a virtual surface.
Roberta Rocca, Sami Boukortt, Geoff Keeling et al. (4 authors) • ArXiv
Finding: LLMs show significant gaps in functional Theory of Mind (social reasoning), with even frontier models struggling to track epistemic states (what others know vs. what is mutually known) in a novel coordination game benchmark.
Why it matters: Relevant to AI's capacity for genuine collaboration and HCI contexts where AI must model user knowledge states — directly implicates AI as a teammate or assistant.
Method: Two-player LLM-LLM dialogue game requiring convergence on semantic Schelling points (focal points for coordination without communication) under varying epistemic transparency conditions.
Jargon: *Theory of Mind (ToM)*: ability to attribute mental states to others; *Schelling points*: solutions people independently converge on without communication; *epistemic tracking*: correctly modeling what oneself vs. others know.
Gabrielle Kaili-May Liu, Areeb Gani, Jacqueline Lu et al. (6 authors) • ArXiv
Finding: This review surveys how LLMs can exhibit metacognitive abilities (knowing what they know/don't know), covering measurement methods, techniques to improve metacognition, and applications.
Why it matters: Tangentially relevant to learning sciences (metacognition is critical to skill acquisition) and AI's impact on work, but the paper is primarily a technical NLP/AI systems review rather than focused on human behavior, HCI, or pedagogy.
Jargon: *Metacognition* — awareness and regulation of one's own thinking processes; *LLM calibration* — whether a model's expressed confidence matches its actual accuracy.
Finding: LightMem-Ego is a lightweight multimodal memory system for wearable/mobile AI that continuously captures egocentric video and audio, organizes it hierarchically, and retrieves it to answer user queries about past experiences.
Why it matters: Relevant to HCI and AI impact on daily life, particularly how persistent AI memory on wearables changes human cognition, recall, and behavior—touches on the "Origin and Control of Persistent Mental Content" interest area.
Method: Hierarchical memory architecture (current/short-term/long-term) with dynamic retrieval routing across levels, deployed on smartphones and AI glasses.
Jargon: *Egocentric streams* = first-person video/audio captured from the user's perspective; *multimodal memory* = memory system integrating both visual and audio data.
Finding: Proposes a human-centered AI framework for lexicography with four dimensions: augmented professional, sociotechnical context, bias mitigation, and tool design, arguing AI should augment rather than replace lexicographers.
Why it matters: The HCAI framework mirrors broader debates about AI augmentation vs. replacement in professional work, but the domain (dictionary-making) is too narrow and specialized to be directly useful for HCI, creativity, or team dynamics teaching.
Finding: Adaptive agents with memory degradation achieve faster actual conceptual convergence in coordination games, while non-adaptive agents *perceive* convergence earlier than it actually occurs.
Why it matters: Has tangential relevance to team communication and shared mental models, but the game-theoretic/computational modeling approach is quite distant from HCI, creativity, or applied behavioral science teaching contexts.
Method: Counterfactual agent simulations in a non-partnership coordination game varying adaptiveness and memory degradation parameters.
Jargon: *Conceptual alignment* = agents agreeing on shared category boundaries for language/meaning; *non-partnership game* = players do not share payoffs (unlike cooperative games).
Fernanda Miyuki Yamada, João Paulo Gois, Hiroki Takahashi • ArXiv
Finding: AI can potentially address gaps in 3D avatar systems that fail to represent people with limb differences, amputations, or non-normative body types in extended reality environments.
Why it matters: Marginally relevant to HCI and inclusive design, but too narrowly focused on a technical avatar representation problem to be useful for the professor's core teaching areas.
Weiqi Jin, Peijun Tang, Kuncheng Luo et al. (8 authors) • ArXiv
Finding: PHILIA is a multi-robot agent architecture enabling long-term physical coexistence with intelligent robots through a unified capability interface that decouples high-level semantic reasoning from low-level robot execution.
Why it matters: Tangentially relevant to HCI through human-robot interaction and human-in-the-loop design, but the focus is primarily on robotics systems engineering rather than human behavior, learning, or collaboration.
Method: Architecture demonstration validated on Astribot S1 robots with representative household task scenarios.
Finding: Across 15 nations, AI literacy in secondary education is structurally unequal—a "Syntax Ceiling" concentrates deeper algorithmic education (C++) in elite STEM tracks while general students get surface-level Python or no programming at all.
Why it matters: Tangentially relevant to learning sciences and AI's societal impact, but the focus is curriculum policy and governance rather than HCI, creativity, teams, cognition, or behavioral science.
Jargon: *Syntax Ceiling* — the phenomenon where students in general tracks learn only surface programming syntax without deeper algorithmic thinking; *access architectures* — structural arrangements determining who gets access to what educational content.