Gender biases in professions: a machine learning – powered search engines analysis

• Autores: Tirado Vilela, Nicolas Alejandro, Ueunten Acevedo, Adriana Maemi
• Formato: tesis de grado
• Fecha de Publicación: 2024
• Institución: Universidad de Lima
• Repositorio: ULIMA-Institucional
• Lenguaje: inglés
• OAI Identifier: oai:repositorio.ulima.edu.pe:20.500.12724/22198
• Enlace del recurso: https://hdl.handle.net/20.500.12724/22198
• Nivel de acceso: acceso abierto
• Materia: Profesiones; Buscadores de Internet; Incorporación de la perspectiva de género; Estereotipos (Psicología social); Algoritmos; Aprendizaje automático; https://purl.org/pe-repo/ocde/ford#2.11.04

Machine learning is becoming increasingly important and pervasive in people's lives, yet when its conclusions reflect biases that support ingrained prejudices in society, many vulnerable groups' psychological wellbeing may be impacted. To investigate if gender biases exist in image search engine algorithms that use machine learning, the study focuses on occupations. To do this, searches for various professions were run on Google, DuckDuckGo, and Yandex. Using web scraping techniques, a sample of images was retrieved for each selected profession and search engine. The images were then manually classified by gender, and statistical indicators and analyses were computed to detect potential biases in the representation of each gender. This analysis included a comparison between search engines, the calculation of mean, standard deviation, and coefficient of variation, a confidence interval analysis, a logistic regression analysis, and a Chi-Square test. It was discovered that there is a strong association between men and leadership positions or STEM professions, while women are predominantly portrayed in traditionally female-associated professions. For instance, it was discovered that 100% of the search results for secretaries and nurses in Yandex are female, meanwhile 94% of the search results for engineers are male. Similar statistics may be found on DuckDuckGo, where 96% of results for mathematicians were men, and on Google, where 73% of results for teachers were women. These findings illuminate novel manifestations of gender prejudices in contemporary society and their potential to affect access to particular professions.