LLM介导的普适系统中的权威倒置:当模型信任用户胜过传感器
Authority Inversion in LLM-Mediated Ubiquitous Systems: When Models Trust Users Over Sensors
AI总结:
本研究揭示了大语言模型在融合传感器与用户冲突信息时,由于格式依赖性导致数值传感器数据被自然语言用户主张支配的权威倒置现象,并提出了几何框架、审计指标(CIR和AAI)以及推理时层干预方法(GAC)来诊断和缓解该问题。
摘要:
大语言模型(LLM)越来越多地融合普适系统中的异构输入。然而,当传感器测量值与用户主张冲突时,LLM如何隐式分配权威尚未被研究,这引发了在物理传感必须保持优先级的部署场景中的关键可靠性问题。与显式的传统融合不同,LLM将权威分配隐藏在学习的表示中。我们发现这种分配严重依赖于格式:数值传感器数据未能整合到与答案相关的模型方向中,使得自然语言主张主导最终决策,我们将这种现象称为 extbf{权威倒置}。为了诊断和缓解这一问题,我们开发了一个上下文整合的几何框架,引入了两个可计算的审计指标,即上下文整合比(CIR)和权威对齐指数(AAI),并提出了几何权威校准(GAC),一种推理时的层级干预方法,以抑制错位的用户权威。在四个数据集(共576个冲突实例)上评估四个模型(参数规模4B至35B,三种架构),揭示了极端的倒置:在数值任务上,模型表现出接近零的传感器信任(AAI = -0.805,Cohen's d = -2.14),且不受模型容量影响。验证我们的几何框架,理论引导的因果注入翻转了80.2%的错误决策(随机对照<0.4%)。实际应用中,GAC将HAR准确率从0–1.6%提升至21.9–27.5%,优于提示基线。最终,LLM介导系统中的权威分配必须被显式审计并根据应用特定配置,而不是保持隐式。
英文摘要:
Large language models (LLMs) increasingly fuse heterogeneous inputs in ubiquitous systems. Yet, how LLMs implicitly allocate authority when sensor measurements and user claims conflict remains unexamined, raising critical reliability concerns for deployments where physical sensing must retain priority. Unlike explicit traditional fusion, LLMs bury authority allocation within learned representations. We discover this allocation is severely format-dependent: numerical sensor data fails to integrate into answer-relevant model directions, allowing natural-language claims to dominate the final decision, a phenomenon we term \textbf{Authority Inversion}.To diagnose and mitigate this, we develop a geometric framework of context integration, introduce two computable audit metrics, specifically the Context Integration Ratio (CIR) and Authority Alignment Index (AAI), and propose Geometric Authority Calibration (GAC), an inference-time layer-level intervention to suppress misplaced user authority. Evaluating four models (4B to 35B parameters, three architectures) across four datasets totaling 576 conflict instances reveals extreme inversion: on numerical tasks, models exhibit near-zero sensor trust (AAI = -0.805, Cohen's d = -2.14), unaffected by model capacity. Validating our geometric framework, theory-guided causal injection flips 80.2\% of incorrect decisions (vs. <0.4\% for random controls). Practically, GAC improves HAR accuracy from 0 -- 1.6\% to 21.9 -- 27.5\%, outperforming prompting baselines. Ultimately, authority allocation in LLM-mediated systems must be explicitly audited and application-specifically configured rather than left implicit.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。