As language models continue to be integrated into applications of personal
and societal relevance, ensuring these models' trustworthiness is crucial,
particularly with respect to producing consistent outputs regardless of
sensitive attributes. Given that first names may serve as proxies for
(intersectional) socio-demographic representations, it is imperative to examine
the impact of first names on commonsense reasoning capabilities. In this paper,
we study whether a model's reasoning given a specific input differs based on
the first names provided. Our underlying assumption is that the reasoning about
Alice should not differ from the reasoning about James. We propose and
implement a controlled experimental framework to measure the causal effect of
first names on commonsense reasoning, enabling us to distinguish between model
predictions due to chance and caused by actual factors of interest. Our results
indicate that the frequency of first names has a direct effect on model
prediction, with less frequent names yielding divergent predictions compared to
more frequent names. To gain insights into the internal mechanisms of models
that are contributing to these behaviors, we also conduct an in-depth
explainable analysis. Overall, our findings suggest that to ensure model
robustness, it is essential to augment datasets with more diverse first names
during the configuration stage.

本研究旨在检验语言模型中名字对常识推理能力的影响，并提出了可控性实验框架和解释分析来证明了首次提出的假设，即一般的常识推理与他们所用的人名无关，结果显示人名的频次对模型的预测产生了不同的影响，并建议在数据集配置阶段增加更多多样性的人名以确保模型的健壮性。

探究姓氏对语言模型的因果效应：以社交常识推理为例

Examining the Causal Effect of First Names on Language Models: The Case  of Social Commonsense Reasoning

A Google search for a person's name, such as "Trevon Jones", may yield a
personalized ad for public records about Trevon that may be neutral, such as
"Looking for Trevon Jones?", or may be suggestive of an arrest record, such as
"Trevon Jones, Arrested?". This writing investigates the delivery of these
kinds of ads by Google AdSense using a sample of racially associated names and
finds statistically significant discrimination in ad delivery based on searches
of 2184 racially associated personal names across two websites. First names,
assigned at birth to more black or white babies, are found predictive of race
(88% black, 96% white), and those assigned primarily to black babies, such as
DeShawn, Darnell and Jermaine, generated ads suggestive of an arrest in 81 to
86 percent of name searches on one website and 92 to 95 percent on the other,
while those assigned at birth primarily to whites, such as Geoffrey, Jill and
Emma, generated more neutral copy: the word "arrest" appeared in 23 to 29
percent of name searches on one site and 0 to 60 percent on the other. On the
more ad trafficked website, a black-identifying name was 25% more likely to get
an ad suggestive of an arrest record. A few names did not follow these
patterns. All ads return results for actual individuals and ads appear
regardless of whether the name has an arrest record in the company's database.
The company maintains Google received the same ad text for groups of last names
(not first names), raising questions as to whether Google's technology exposes
racial bias.

研究通过对 2184 个种族相关的个人名称进行搜索，发现 Google AdSense 提供社会背景差异广告的行为具有显着的统计歧视性，其中出生时指派给黑人婴儿的大多数姓名在搜索结果中产生有关逮捕记录的广告，而出生时主要指派给白人婴儿的姓名则产生更中立的广告，这引发了谷歌技术是否存在种族偏见的问题。