Coding Gender: Exploring the Presence of Gender Stereotypes within ChatGPT

Abstract

Recently released in November 2022, OpenAI’s latest tech innovation, Chat Generative Pre-Trained Transformer (ChatGPT), has taken the world by storm, using machine learning algorithms to create human-like responses to any given input. With the rapid integration of ChatGPT into numerous social domains and day-to-day tasks, it is imperative to understand this program’s limitations and predispositions. Therefore, this paper examines whether ChatGPT and ChatGPT+ demonstrate a reliance on traditional gender stereotypes in their responses and, if so, how this level of gender bias relates to comparable Artificial Intelligence (AI) programs. This author centers their testing (performed in the Summer of 2023) and conclusions on ChatGPT’s language translation service, asking it to translate gender-ambiguous English sentences into five gendered languages (French, Spanish, Ukrainian, Russian, and Arabic). Additionally, this study examines ChatGPT’s ability to answer open-ended questions and tell stories. Findings reveal a notable correlation between traditional gender stereotypes in both ChatGPT’s translations and open-ended responses, as well as identify minimal differences in this level of reliance between ChatGPT and ChatGPT+. Moreover, this research emphasizes the importance of making continual efforts to mitigate biases in the proof of concept and development stage of Language Learning Models.

Ähnliche Arbeiten