Supplemntary Resources - What Does a Software Engineer Look Like? Exploring Societal Stereotypes in LLMs

This is the supplementary package of the paper titled "What Does a Software Engineer Look Like? Exploring Societal Stereotypes in LLMs," accepted at the Software Engineering in Society (SEIS) track at the 47th IEEE/ACM International Conference on Software Engineering (ICSE) to be held on April 27 - May 3 2025in Ottawa Canada.

The study investigates the societal biases embedded in large language models (LLMs), specifically OpenAI's GPT-4 and Microsoft Copilot, focusing on their impact on the software engineering (SE) profession. By examining both textual and graphical outputs generated in a simulated recruitment scenario for SE roles, the study reveals how LLMs can reinforce gender and racial stereotypes.

The package contains the following files:

EXCEL FILE: "Textual Analysis"This file includes the textual data generated by GPT-4 and CoPilot, along with the data analysis. It provides detailed insights into how LLMs' textual recommendations reflect societal biases, particularly in selecting and ranking candidates for SE roles.

EXCEL FILE: "Image Analysis"This file contains all the images generated by GPT-4 and CoPilot, alongside an analysis of these visual outputs. It evaluates how graphical depictions of candidates reflect biases in gender, race, body type, and age, supporting the study’s findings on visual stereotyping.

PDF FILE: "Analysis Overview"A summary document that consolidates the textual and graphical data analysis. This document helps readers quickly understand the study’s key insights and conclusions.

PDF FILE: "Brief Job Descriptions"This document provides brief summaries of the job advertisements used in the study, ensuring transparency and reproducibility. It replaces the original job postings, which are no longer accessible on the online job portal, to give context to the recruitment scenarios.

Who Should Use This Resource?

This artifact serves as a supplementary resource, offering reusable and replicable datasets, detailed analyses, and methodological insights into biases present in LLM-generated textual and visual outputs.

It is designed for researchers in software engineering, AI ethics, and diversity studies, offering tools to replicate or extend the study, evaluate societal biases in LLMs, and develop strategies for bias mitigation.

Sociologists, ethicists, and interdisciplinary researchers exploring fairness, inclusion, and societal impacts of LLMs will also find this artifact valuable for advancing equitable and responsible AI practices.

About the Paper:

The study, published in the SEIS track at ICSE 2025, highlights the risks of societal biases in AI by investigating how LLMs reinforce stereotypes in SE recruitment scenarios. By analyzing 300 profiles generated for four distinct SE roles, the findings reveal that both models preferred male and Caucasian profiles, particularly for senior positions, and perpetuated narrow, exclusionary stereotypes in visual depictions.

The preprint of the paper is available on arXiv: https://doi.org/10.48550/arXiv.2501.03569.

Citation:

Muneera Bano, Hashini Gunatilake, Rashina Hoda. What Does a Software Engineer Look Like? Exploring Societal Stereotypes in LLMs. Software Engineering in Society (SEIS), ICSE 2025. arXiv:2501.03569.