AI's perception and response to gender discrimination language

Jeongbok Lee

Daegu University

Abstract

This study evaluates how three AI models—ChatGPT, DeepSeek, and Clova X—detect and respond to gender-biased expressions in Korean. Clova X exhibited the highest accuracy in identifying discriminatory language, followed by ChatGPT, while DeepSeek performed the poorest. While terms like “kimchi girl” and “doenjang girl” were correctly recognized, phrases such as “female doctor” and “maiden work” were often misinterpreted. ChatGPT and DeepSeek occasionally provided inaccurate definitions, raising concerns about misinformation. Interestingly, DeepSeek performed best when interpreting sexist proverbs, although the overall differences across models were minor. All three models generally succeeded in recognizing biased expressions in conversational contexts, but DeepSeek struggled with non-standard sentence formats, leading to delays or missing responses. These results reveal current limitations in generative AI’s ability to process culturally specific and nuanced language. This study emphasized the need to incorporate more diverse Korean language data and up-to-date linguistic research in AI training. As generative AI tools become more integrated into everyday communication, improving their ability to detect and respond to gender biases is crucial for fostering fair and responsible language technologies.

성차별 언어에 대한 인공지능의 인식과 대응

초록

Vol. 40 No. 4 (2022.12)

Journal Abbreviation j. Korea Saf. Manag. Sci.
Frequency Published four times annually in March, June, September, and December
Doi Prefix 10.14353/sjk.
Year of Launching 1993
Publisher The Sociolinguistic Society of Korea

Indexed/Tracked/Covered By