📚 Research Digest

Finding: Sidekick, a multimodal feedback system for Computer Use Agents (CUAs), significantly improves multitasking performance, progress awareness, and error traceability compared to text-only feedback by using ambient cues, visual summaries, and verbalized reasoning at different interaction stages.

Why it matters: Directly relevant to HCI/UX design and AI-human collaboration research, with implications for how interface design shapes human oversight, trust, and cognitive load when working alongside autonomous AI agents.

Method: Formative studies with CUA experts + GenAI users, followed by a 30-participant controlled study comparing Sidekick against two baseline conditions (chat-based and ambient text display).

Jargon: *Computer Use Agents (CUAs)*: AI agents that autonomously operate GUIs to complete multi-step tasks; *Ambient cues*: low-attention background signals indicating system state without requiring direct focus; *Long-horizon collaboration*: human-agent tasks that unfold over extended time with many sequential steps.

Finding: Psychology students and staff in UK universities express concerns about GenAI in assessment centered on three themes: human-machine ambiguity, cognitive overreliance/weakening, and erosion of the social trust contract underlying higher education.

Why it matters: Directly relevant to AI's impact on learning, skill acquisition, and human collaboration in educational contexts—particularly the "augmenting vs. eroding human capital" tension seen in papers the user flagged as relevant.

Method: Qualitative focus groups (N=25 across students and staff from 3 UK universities) with thematic analysis, including reactions to comparative human vs. AI-generated feedback examples.

Jargon: *Social contract* = the implicit mutual obligations between students and educators (effort, authentic feedback, trust); *GenAI* = Generative AI tools like ChatGPT used to generate or grade written work.

Finding: A reanalysis of Lin et al. (2026) argues that the reported negative emotion-to-performance link in self-regulated learning is a temporal confound from practice effects, not a real causal relationship; after modeling learning curves, the effect disappears.

Why it matters: Directly relevant to goal-setting and self-regulated learning research, and exemplifies the kind of methodological replication/reanalysis work the professor values (similar to the Gino et al. reanalysis in their relevant papers list).

Method: Reanalysis introducing a log(trial) learning curve into Bayesian autoregressive panel models, replacing WAIC model selection with exact leave-future-out cross-validation (LFO-CV).

Jargon: *Coasting hypothesis* — the idea that positive emotions cause people to reduce effort toward goals; *WAIC* — Widely Applicable Information Criterion, a Bayesian model comparison metric; *LFO-CV* — leave-future-out cross-validation, a method that respects temporal ordering when assessing model fit in time-series data.

Finding: Brief interaction with an autonomous drone improves mental model accuracy, which is positively associated with generalized trust at follow-up (2-3 weeks later), suggesting trust formation is mediated by accurate mental representations rather than anthropomorphism.

Why it matters: The mental model → trust → behavioral intention pathway is directly relevant to HCI, AI adoption research, and designing human-AI collaboration systems.

Method: Small longitudinal study (N=19) with pre/during/post/follow-up measures in a warehouse mock-up, including semantic representation analysis.

Jargon: *Mental model*: user's internal representation of how a system works; *Generalized trust*: dispositional trust in the system across contexts (vs. situational trust in specific moments); *Intralogistics*: internal warehouse/supply chain operations.

Finding: Individual differences in achievement goal orientation (approach vs. avoidance goals) predict neural responses to negative feedback, with high approach-goal individuals showing less negative ventral striatum activation to negative feedback.

Why it matters: Directly relevant to goal-setting and motivation research—specifically how achievement goal orientation moderates feedback processing, which has implications for learning design, feedback systems, and motivational interventions in educational and HCI contexts.

Method: fMRI scanning during a word association learning task following a competence threat, correlating BOLD signals in the ventral striatum with individual goal orientation scores.

Jargon: *BOLD response* = Blood-oxygen-level-dependent signal, the neural activity measure used in fMRI; *ventral striatum (VS)* = brain region central to reward processing; *approach goals* = motivation oriented toward achieving positive outcomes (vs. avoidance goals oriented toward avoiding negative outcomes).

Finding: The quirky questions task is a new validated measure for humor production that correlates with openness to experience, response length, and time spent—replicating findings from other humor production tasks.

Why it matters: Marginally relevant to creativity research methodology, but the focus is narrow psychometric validation of a humor task rather than HCI, teams, behavior change, or AI topics central to this professor's interests.

Method: Latent variable modeling of human-rated funniness scores across 320 adults completing six humor prompts with individual difference measures.

Finding: Adding more sensory-motor modalities (beyond two) during letter-sound learning *decreases* performance, challenging multimodal enrichment theories and supporting cognitive load explanations.

Why it matters: Has some relevance to learning sciences and skill acquisition, but is narrowly focused on literacy/phonics acquisition in adults rather than HCI, creativity, teams, or behavioral science topics the professor covers.

Method: Two experiments comparing bimodal, trimodal, and quadrimodal learning conditions using novel pseudoletter-syllable associations, with sequential vs. simultaneous modality engagement.

Jargon: *Multimodal enrichment theory* — the idea that engaging more sensory systems improves memory encoding; *cognitive load* — the mental effort required by working memory during learning.

Finding: Boredom unfolds heterogeneously across individuals during sustained tasks, and the *trajectory* of boredom (increasing vs. decreasing trend) predicts task persistence better than average boredom level.

Why it matters: Has tangential relevance to engagement, motivation, and task persistence in learning/work contexts, but is focused on physical exercise rather than HCI, learning, or behavioral design.

Method: Functional principal component analysis (fPCA) to decompose individual boredom trajectories over time — a methodologically interesting approach for studying dynamic states.

Jargon: fPCA (functional principal component analysis): a technique for analyzing curves/trajectories as data, decomposing variability in how a variable changes over time across individuals.

Finding: Successful unassisted quitters internalize external motivations (health scares, social roles, finances) into personally meaningful, identity-aligned reasons for quitting, with intrinsic enjoyment largely absent but autonomous motivation still achievable.

Why it matters: The finding that behavior change is identity-based and relational, not purely intrinsic/extrinsic, has some relevance to behavioral science and persuasive technology design, but the smoking cessation context is narrow and peripheral to the professor's core interests.

Method: Reflexive thematic analysis of 32 semi-structured interviews, theoretically informed by Self-Determination Theory and the Health Belief Model.

Jargon: *Internalisation* (SDT term): process by which external motivations become personally endorsed and integrated with one's values and identity; *Autonomous motivation*: acting from self-endorsed rather than pressured reasons, even if originally externally prompted.

Finding: Teacher authority is not disappearing with AI integration but transforming from knowledge-giver to moderator/mentor, reshaping pedagogical relationships toward more collaborative interaction.

Why it matters: Tangentially relevant to AI's impact on human roles and collaboration, but the focus is narrow (teacher authority in K-12/traditional education) and the theoretical contribution is thin.

Method: Purely theoretical; no empirical data or novel framework.