Babak Hemmatian, Anita Keshmirian, Yijun Lin et al. (12 authors) • ArXiv
Finding: A controlled two-player Alternate Uses Test platform shows that originality with a GPT-4 partner is statistically equivalent to a human partner, with approach motivation (BAS Drive) moderating whether partnership benefits creativity and cognitive outsourcing predicting lower originality in human-human dyads.
Why it matters: Directly tests human-AI vs. human-human co-creation on a classic creativity task with experimental rigor, offering actionable insights for teaching creativity, designing AI collaboration tools, and understanding how individual differences shape creative output.
Method: Controlled two-player interactive platform that decomposes performance into participant traits, partner perceptions, and content dynamics — a methodological advance over typical independent-agent comparisons or uncontrolled field studies.
Jargon: *Alternate Uses Test (AUT)* — standard divergent thinking task where participants generate novel uses for everyday objects; *BAS Drive* — Behavioral Activation System Drive, a trait measuring reward-seeking motivation; *cognitive outsourcing* — offloading thinking to a partner rather than generating ideas independently; *seeding* — prior exposure to highly creative ideas that improves subsequent creative performance.
⚠️ Small sample (n=62) | Self-reported data only | No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Marcos Economides, Paul M. Sacher, Samuel Salzer et al. (6 authors) • ArXiv
Finding: Current AI evaluation frameworks miss "psychological competence" — how AI systems shape user cognition, emotional interpretation, trust calibration, and decision-making — and this paper proposes a conceptual framework for assessing these effects.
Why it matters: Directly relevant to HCI, behavioral science, and AI's impact on human cognition and collaboration, offering a framework professors can use to discuss responsible AI design and the psychological dimensions of human-AI interaction.
Method: Conceptual framework developed from behavioral science and HCI literature, with proposed assessment via scenario-based probes, structured human evaluation, and model-assisted evaluation.
Jargon: *Psychological competence* = an AI system's capacity to appropriately support user cognition, emotional interpretation, and behavioral decision-making given context and purpose; *framing effects* = how presentation of information shapes perception and choice.
Jessica Woodhams, Amy Burrell, Wanyin Li et al. (12 authors) • ArXiv
Finding: Analysts using an AI crime-linkage tool selectively trusted AI predictions and frequently cross-validated them against behavioral evidence, revealing partial trust and reliance on established practices even when AI explanations were available.
Why it matters: Directly addresses how humans calibrate trust in AI decision-support systems in high-stakes contexts—central to HCI, explainable AI, and the broader question of how AI reshapes human judgment and collaboration.
Method: Mixed-methods in-situ usability study combining direct observation, eye-tracking, mouse-tracking, and surveys—an unusually rich multimodal evaluation approach worth highlighting in HCI teaching.
Jargon: *Crime linkage analysis* = identifying crime series likely committed by the same offender based on behavioral/situational patterns; *model features as explanations* = XAI (explainable AI) outputs showing which input variables drove the AI's prediction.
Finding: ArtMine is a framework that uses abductive AI reasoning to reconstruct and formalize artistic production processes from fragmented historical evidence, modeling the *how* of creation rather than just the finished artifact.
Why it matters: Directly relevant to creativity and design process research—formalizing iterative artistic decision-making has implications for creative education, human-AI co-creativity, and understanding how visual thinking unfolds over time.
Method: A Peircean abductive agent synthesizes heterogeneous evidence (archival records, preparatory studies, correspondence) into a compositional graph and rendering prompt, then uses self-reflection to optimize against reference artworks.
Jargon: *Peircean abduction*: inference to the best explanation (from philosopher C.S. Peirce), here used to reason backward from evidence to plausible process steps. *Compositional graph*: a structured representation of sequential artistic decisions and operations.
Finding: HeadRoom is a lightweight pipeline for smart glasses that routes notifications to whichever sensory channel (visual or auditory) is less occupied, reducing response time under high perceptual load.
Why it matters: Directly relevant to HCI and UX design for wearable/immersive systems, and touches on attention management and cognitive load—concepts central to behavioral science and learning sciences.
Method: Controlled user study (N=25) manipulating perceptual load and notification channel routing using egocentric video/audio processing on-device.
Jargon: *Edge-deployable* = runs locally on the device rather than in the cloud; *perceptual load* = how much cognitive/sensory capacity a task demands; *adaptive routing* = dynamically selecting the delivery channel based on context.
Finding: fog is a function-composition framework that uses AI-generated code to create expressive motion animations (verbs, adverbs, emotions), evaluated with 452 animations at 68% recognition accuracy—2.68x above chance—and shown to support rapid iteration in a user study.
Why it matters: Relevant to HCI/UX design and creativity tooling—specifically how AI-assisted interfaces enable expressive, open-ended creative exploration for both novices and professionals.
Method: Mixed-methods study combining perceptual evaluation (n=452 animations) with a user study of professionals and novices using direct manipulation + dynamically generated UI.
Jargon: *Heider-Simmel animations* — simple geometric shape animations that humans interpret as having social/emotional meaning (classic social psychology stimulus); *function composition* — combining simple functions to build complex behaviors.
Finding: Aleena is an AI agent that helps research software engineering teams maintain alignment across distributed collaboration artifacts (Slack, GitHub, meetings) by preserving decision rationale and tracking project state.
Why it matters: Directly relevant to team dynamics, collaboration, and AI's impact on knowledge work—particularly how AI tools can support shared mental models and reduce coordination failures in technical teams.
Method: System design paper grounded in university research software engineering center experiences, with illustrative lifecycle scenarios.
Jargon: *Alignment agent* = AI system that tracks and reconciles divergent understandings across stakeholders; *multi-modal stakeholder interactions* = inputs from different communication channels (chat, code reviews, meetings).
Finding: The Behavioural Reflection Test (BRT) outperforms the classic Cognitive Reflection Test (CRT) in measuring reflective reasoning, predicting better evidence evaluation and source quality judgment in decision-making contexts.
Why it matters: Relevant to behavioral science and decision-making curricula—particularly for teaching cognitive biases, epistemic vigilance, and how individuals differ in overriding intuitions when processing information.
Method: Online study (N=473) using open-ended scenarios plus a bespoke 4-item CRT variant, with linguistic analysis of responses as behavioral markers of reflective reasoning.
Jargon: *Cognitive Reflection Test (CRT)*: A classic measure of tendency to override intuitive but wrong answers with deliberate reasoning. *Epistemically charged*: Situations involving truth/belief evaluation. *Convergent validity*: A measure correlates with other tests measuring the same construct.
Victor Jüttner, Xenia Wagner, Christoph Jahn et al. (4 authors) • ArXiv
Finding: LLMs can simulate smart home residents by generating structured, executable device interaction schedules via persona-based multi-stage pipelines, enabling privacy-conscious HCI research without real-world data collection.
Why it matters: Tangentially relevant to HCI and AI's impact on research methodology, but focused narrowly on smart home security/privacy datasets rather than creativity, teams, cognition, or behavioral science.
Method: Multi-stage LLM pipeline generating socio-technically grounded household personas across five dimensions, executed on physical smart home testbeds.
⚠️ No control group mentioned | Single study (not replicated)
Aklile Seyoum Mamo, Amanuel Kebede, Anny Christelle Irakoze et al. (4 authors) • ArXiv
Finding: Authentication complexity in time-constrained USSD systems creates a "Success Cliff" — a nonlinear threshold beyond which session success rates sharply decline, revealing a fundamental security-usability tradeoff.
Why it matters: The security-usability tradeoff and the concept of a nonlinear failure threshold are relevant to HCI/UX design, particularly for designing authentication workflows and understanding how interaction complexity affects user success.
Method: Simulation-based framework modeling failure dynamics across variables of authentication complexity, user response time, and network round-trip time.
Jargon: *USSD* — a mobile protocol used for real-time session-based interactions (common in mobile banking in developing regions); *Success Cliff* — a nonlinear drop in task completion beyond a critical complexity threshold.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Finding: AutoPersonas introduces a system architecture for long-running AI persona agents that separates environment events, observations, and persona state to prevent behavioral convergence ("self-locking") while maintaining identity continuity.
Why it matters: Marginally relevant to AI's impact on human collaboration or HCI only insofar as persona modeling touches UX research methods, but this is primarily an AI systems engineering paper about agent architecture.
Jargon: *Self-locking*: failure mode where a generative loop collapses to repetitive, high-probability outputs; *OSO loop*: Occurrences → observations → State update cycle; *context gravity*: tendency of accumulated context to pull generation toward familiar patterns.
Shyam Agarwal, Courtney Miller, Christian Kästner et al. (4 authors) • ArXiv
Finding: AI coding agents change code review dynamics (less review, faster merging, less discussion), but team expertise and process structure determine whether this helps or hurts software quality.
Why it matters: Tangentially relevant to AI's impact on collaborative work and team dynamics, but focused narrowly on software engineering practice rather than the professor's core areas.
Method: LLM-assisted coding of 3,100 grey-literature documents (blogs, Reddit) to build a causal model with 26 constructs and 67 relationships.