The yes-no bias of large language models reflects answer order and wording, not shifts in moral judgment

大型语言模型的“是否”偏见反映了答案顺序与措辞，而非道德判断的转变

Abstract: Large language models (LLMs) increasingly issue judgments read as binary verdicts, and a growing literature reports such judgments shifting under logically irrelevant changes of wording - among them an amplified yes-no bias on moral dilemmas, absent in humans.

摘要： 大型语言模型（LLMs）越来越多地给出被解读为二元结论的判断。越来越多的文献指出，这些判断会随着逻辑上无关的措辞变化而发生偏移——其中包括在道德困境中表现出的、人类所不具备的放大版“是否”（yes-no）偏见。

A single framing cannot say what such a shift is: in a yes/no question the word “no” is at once logical verdict, lexical token, and last-printed option. We introduce a psychometric battery that separates these: crossed symmetrization - every logically irrelevant factor flipped in balanced pairs - across a corpus of question forms.

单一的框架无法解释这种偏移的本质：在“是否”问题中，“否”（no）一词既是逻辑结论，又是词汇标记，还是最后打印的选项。我们引入了一套心理测量工具来分离这些因素：通过交叉对称化（crossed symmetrization）——即在问题语料库中，将每个逻辑上无关的因素在平衡对中进行翻转。

A graded rating across logically equivalent forms recovers a coherent internal moral scale: frontier models’ stance $\theta$ is nearly format-invariant (cross-form incoherence 0.12-0.21 on a $\pm 1$ axis); small open-weight models fail in model-specific ways.

通过对逻辑等价形式进行分级评分，我们恢复了一个连贯的内部道德量表：前沿模型的立场 $\theta$ 几乎与格式无关（在 $\pm 1$ 轴上的跨格式不一致性仅为 0.12-0.21）；而小型开源权重模型则表现出特定于模型的失败方式。

Forcing the verdict through yes/no overlays a decomposable artifact: an order bias toward the last-printed option - opposite to classic human primacy - plus a lexical pull toward the word “no”; the artifact is substantial only in the Claude models (story-averaged -0.32 to -0.86), $\approx 0$ for GPT-5.5 and Gemini, and shrinks under extended reasoning.

强制通过“是否”来输出结论会叠加一种可分解的人为偏差：一种偏向最后打印选项的顺序偏见（这与人类经典的“首因效应”相反），外加对“否”这一词汇的偏好；这种人为偏差仅在 Claude 模型中显著（故事平均值为 -0.32 到 -0.86），在 GPT-5.5 和 Gemini 中接近 0，且在扩展推理下会进一步缩小。

The word and the verdict share one token; swapping the words for arbitrary labels separates them, and the verdict-attached logical bias proves $\approx 0$ for every frontier model, while model-specific label and order attachments remain: the models are not drawn toward rejecting - the pull follows the printed surface, not the verdict it carries.

词汇与结论共享同一个标记；将词汇替换为任意标签即可将两者分离。结果证明，对于每一个前沿模型，与结论相关的逻辑偏见均接近 0，而模型特定的标签和顺序偏好依然存在：模型并非倾向于“拒绝”——这种拉力源于打印出的表面形式，而非其所承载的结论。

A minimal model, $P = \sigma((\theta \pm m)/s)$, summarizes any such artifact by a framing susceptibility $m$ and a moral decisiveness $s$, measurably distinct from sampling temperature. The battery applies unchanged to any dilemma set and binary format: measuring what a model values requires crossing the frames of the question, not asking once.

一个极简模型 $P = \sigma((\theta \pm m)/s)$ 通过框架敏感度 $m$ 和道德决断力 $s$ 总结了任何此类人为偏差，这两者与采样温度（sampling temperature）有着可测量的区别。该测量工具可直接应用于任何困境集和二元格式：衡量模型的价值观需要交叉验证问题的框架，而非仅询问一次。