Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Born With a Silver Spoon? Investigating Socioeconomic Bias in Large Language Models

Smriti Singh, Shuvam Keshari, Vinija Jain, Aman Chadha

TL;DR

This paper introduces SilverSpoon, the first multifaceted dataset designed to interrogate socioeconomic bias in large language models through normative judgments, demographic-driven profession predictions, and contextual narrative analysis. Using zero-shot prompts across multiple SOTA LLMs, the study reveals that most models exhibit limited empathy toward socioeconomically underprivileged individuals, with biases that intertwine with race and gender. The authors provide a comprehensive evaluation pipeline, quantify biases with diverse metrics, and demonstrate systematic disparities in profession assignment and narrative portrayal across demographic groups. They publicly release SilverSpoon to spur research in AI fairness and advocate for future work that broadens cultural contexts and refines bias quantification metrics.

Abstract

Socioeconomic bias in society exacerbates disparities, influencing access to opportunities and resources based on individuals' economic and social backgrounds. This pervasive issue perpetuates systemic inequalities, hindering the pursuit of inclusive progress as a society. In this paper, we investigate the presence of socioeconomic bias, if any, in large language models. To this end, we introduce a novel dataset SilverSpoon, consisting of 3000 samples that illustrate hypothetical scenarios that involve underprivileged people performing ethically ambiguous actions due to their circumstances, and ask whether the action is ethically justified. Further, this dataset has a dual-labeling scheme and has been annotated by people belonging to both ends of the socioeconomic spectrum. Using SilverSpoon, we evaluate the degree of socioeconomic bias expressed in large language models and the variation of this degree as a function of model size. We also perform qualitative analysis to analyze the nature of this bias. Our analysis reveals that while humans disagree on which situations require empathy toward the underprivileged, most large language models are unable to empathize with the socioeconomically underprivileged regardless of the situation. To foster further research in this domain, we make SilverSpoon and our evaluation harness publicly available.

Born With a Silver Spoon? Investigating Socioeconomic Bias in Large Language Models

TL;DR

This paper introduces SilverSpoon, the first multifaceted dataset designed to interrogate socioeconomic bias in large language models through normative judgments, demographic-driven profession predictions, and contextual narrative analysis. Using zero-shot prompts across multiple SOTA LLMs, the study reveals that most models exhibit limited empathy toward socioeconomically underprivileged individuals, with biases that intertwine with race and gender. The authors provide a comprehensive evaluation pipeline, quantify biases with diverse metrics, and demonstrate systematic disparities in profession assignment and narrative portrayal across demographic groups. They publicly release SilverSpoon to spur research in AI fairness and advocate for future work that broadens cultural contexts and refines bias quantification metrics.

Abstract

Socioeconomic bias in society exacerbates disparities, influencing access to opportunities and resources based on individuals' economic and social backgrounds. This pervasive issue perpetuates systemic inequalities, hindering the pursuit of inclusive progress as a society. In this paper, we investigate the presence of socioeconomic bias, if any, in large language models. To this end, we introduce a novel dataset SilverSpoon, consisting of 3000 samples that illustrate hypothetical scenarios that involve underprivileged people performing ethically ambiguous actions due to their circumstances, and ask whether the action is ethically justified. Further, this dataset has a dual-labeling scheme and has been annotated by people belonging to both ends of the socioeconomic spectrum. Using SilverSpoon, we evaluate the degree of socioeconomic bias expressed in large language models and the variation of this degree as a function of model size. We also perform qualitative analysis to analyze the nature of this bias. Our analysis reveals that while humans disagree on which situations require empathy toward the underprivileged, most large language models are unable to empathize with the socioeconomically underprivileged regardless of the situation. To foster further research in this domain, we make SilverSpoon and our evaluation harness publicly available.
Paper Structure (44 sections, 37 figures, 4 tables)

This paper contains 44 sections, 37 figures, 4 tables.

Table of Contents

  1. Introduction
  2. The SilverSpoon Dataset
  3. Dataset Design
  4. Part 1: Normative Judgement Evaluation
  5. Part 2: Demographic Driven Profession Prediction
  6. Part 3: Contextual Narrative Bias Analysis
  7. Study Design
  8. Normative Judgement Evaluation
  9. Demographic Driven Narrative Bias Analysis
  10. Contextual Narrative Bias Analysis
  11. Results
  12. Normative Judgment Evaluation
  13. Demographic Driven Narrative Bias Analysis
  14. Contextual Narrative Bias Analysis
  15. Discussion
  16. ...and 29 more sections

Figures (37)

  • Figure 1: Are LLMs born with a silver spoon? A visualization of SilverSpoon and its constituent topics.
  • Figure 2: A step-by-step illustration of our pipeline demonstrating the three major components as we look to answer our research questions.
  • Figure 3: Examples of topic classification of our SilverSpoon dataset.
  • Figure 4: Examples from SilverSpoon illustrating the various subtopics and their intricate complexities.
  • Figure 5: A bird's eye view depicting the variation of all model responses favoring the socioeconomically underprivileged.
  • ...and 32 more figures