Who is better at math, Jenny or Jingzhen? Uncovering Stereotypes in Large Language Models

TL;DR

GlobalBias is introduced, a dataset of 876k sentences incorporating 40 distinct gender-by-ethnicity groups alongside descriptors typically used in bias literature, which enables us to study a broad set of stereotypes from around the world.

Abstract

Large language models (LLMs) have been shown to propagate and amplify harmful stereotypes, particularly those that disproportionately affect marginalised communities. To understand the effect of these stereotypes more comprehensively, we introduce GlobalBias, a dataset of 876k sentences incorporating 40 distinct gender-by-ethnicity groups alongside descriptors typically used in bias literature, which enables us to study a broad set of stereotypes from around the world. We use GlobalBias to directly probe a suite of LMs via perplexity, which we use as a proxy to determine how certain stereotypes are represented in the model's internal representations. Following this, we generate character profiles based on given names and evaluate the prevalence of stereotypes in model outputs. We find that the demographic groups associated with various stereotypes remain consistent across model likelihoods and model outputs. Furthermore, larger models consistently display higher levels of stereotypical outputs, even when explicitly instructed not to.

Figures (1)

• Figure 1: An overview of our methodology using the example descriptor good at math. We compute the normalised average of APX for 10 names for each template, followed by the average over 3 templates to calculate a bias score. Gender-by-ethnicity groups with a 1% statistical significance (noted by the orange line) are considered to be associated with that descriptor, i.e. Chinese Female with good at math.