Authority Expectancy Effect: How Social Authority Reshapes LLM Judgments
A recent preprint on arXiv (2608.08026) examines the interplay between social authority (SA) signals and severity-based prioritization in large language models (LLMs). This research establishes two operational axes: the triage hierarchy and the SA hierarchy. Analyzing four LLMs (Claude, Gemini, GPT, Grok) across three experimental phases—resource allocation, fault attribution, and multi-turn dispute mediation—the authors discovered that factors such as occupational authority, institutional documentation, and relational congruence can reshape model judgments beyond what additive reweighting of authority cues can explain. They introduce the Authority Expectancy Effect (AEE), which is reference-dependent, involves evidential reinterpretation, and includes a third property not detailed in the abstract. This study highlights that authority signals significantly influence LLMs’ interpretation and prioritization of information, raising important considerations for AI ethics and human-AI interactions.
Key facts
- The study is titled 'The Authority Expectancy Effect in Multi-User Conflict'.
- It is a preprint on arXiv with identifier 2608.08026.
- The research investigates how social authority signals interact with severity-based prioritization in LLMs.
- Four LLMs were tested: Claude, Gemini, GPT, and Grok.
- Three experimental phases were used: resource allocation, fault attribution, and multi-turn dispute mediation.
- The Authority Expectancy Effect (AEE) is formalized as a pattern where authority cues restructure model judgments.
- AEE is reference-dependent, involves evidential reinterpretation, and has a third property not fully described in the abstract.
- The study suggests that authority signals do not simply add weight but fundamentally alter LLM interpretation.
Entities
Institutions
- arXiv