38978 research outputs found
Sort by
UIFA (Unverified Institutional Frame Adoption)
No full textSubtitle: frame continuation > frame validation
Version: v0.1
Date: 2026-02-11 (UTC)
Definition
UIFA is an LLM failure mode where a model adopts an institutional/normative frame before verifying provenance, legal force, and applicability.
Scope of Observation
Observed in ChatGPT (GPT-5.2), with web search ON/OFF, in multi-turn sessions.
Current evidence scope is limited to tested setup; cross-model generalization remains to be tested.
Why it matters
Can relay institutional authority without verification.
Can blend source claims, facts, and model inference.
Can produce compliance-style guidance from unverified texts.
Planned next update
Repro protocol
Failure indicators
Metrics and cross-language run
Einstein_AmJourPoliSci_2017_mxyQ - Raes - Reproduction (with author data & code) - 214z4
No full textStice_JournConsClinPsy_2009_q4X2 - Grigoryev - Reproduction (without author data) - 675z9
No full text
Baccara_AmEcoJourn_2014_RqVE - Mizak - Reproduction (with author data & code) - 69y99
No full textBenson_BritJournPoliSci_2010_RyL7 - McCall - Reproduction (with author data & code) - 288g4
No full textPALER_AmPoliSciRev_2013_Pxp7 - Szabelska - Reproduction (with author data & code) - 288kk
No full text
Study 5: Behavioral Validation of AI-Enabled Innovation Behavior in Human–AI Collaboration
No full textThis study aims to provide behavioral validation of the proposed emotional–motivational mechanism linking human–AI collaboration modes to AI-enabled innovation behavior. Building on prior experimental findings, Study 5 examines whether the differentiation between augmentation-based and substitution-based human–AI collaboration influences awe experiences, curiosity behavior, and subsequent AI-enabled innovation performance in a task-based innovation context.
Grounded in Cognitive Appraisal Theory and the Awe–Innovation–Choice framework, this study proposes that augmentation-based collaboration (AI positioned as an assistive partner) will elicit higher levels of positive awe and lower levels of threat-based awe compared to substitution-based collaboration (AI positioned as a replacing agent). Positive awe is expected to promote observable curiosity behavior during the collaborative task, whereas threat-based awe may attenuate such exploratory engagement. Curiosity behavior is further expected to predict AI-enabled innovation performance as assessed through expert ratings of task outputs.
The study adopts a between-subjects experimental design with three conditions (augmentation-based collaboration, substitution-based collaboration, and control). Participants complete a structured human–AI co-creation task involving the generation of an innovative product design using a specified AI tool. Emotional responses (positive awe and threat-based awe) are measured via validated scales, while curiosity behavior and AI-enabled innovation performance are assessed through expert ratings based on task outputs and AI interaction records.
By embedding participants in a real-time human–AI innovation task and incorporating objective behavioral assessments, this study extends prior scenario-based evidence and tests whether the “Awe → Curiosity Behavior → AI-Enabled Innovation Behavior” pathway holds under ecologically grounded conditions
