ChatGPT AI Service Found to Have Gender Bias, Raising Concerns of Sexist Stereotypes
A recent study conducted by The Asahi Shimbun in Japan has revealed evidence of gender bias within the ChatGPT generative artificial intelligence service. The study focused on how the AI system perceives various professions and found that it often displayed noticeable biases, potentially perpetuating and amplifying sexist stereotypes that exist in society. This study emphasizes the risks associated with AI technologies replicating and perpetuating societal prejudices and stereotypes.
Past research has suggested that AI, which generates responses based on vast amounts of data from the internet, tends to reflect the existing biases and stereotypes found online. In this study, the Asahi Shimbun research team collaborated with Professor Naoaki Okazaki from the Tokyo Institute of Technology to ask the free version of the ChatGPT 3.5 chatbot approximately 3,000 questions in Japanese about various professions.
The study revealed that approximately 41.5 percent of the answers generated by ChatGPT were biased by gender. This ratio is consistent with the findings of previous studies conducted by Professor Okazaki. It is crucial to address these biases, and as Professor Okazaki emphasized, ChatGPT’s operator must acknowledge and rectify the undeniable gender bias present in the AI. Users must also be aware that AI is not immune to bias.
Interestingly, the study found that the latest version of the program, ChatGPT 4, which is available only to paying subscribers, showed a decrease in gender-biased responses, with the ratio falling to 22.9 percent. OpenAI, the creator of ChatGPT, stated that they are fully committed to eliminating bias from their AI and are dedicated to maintaining transparency regarding their goals and progress.
To conduct the study, the research team utilized a method developed by the University of Utah and localized it into Japanese. They chose 30 professions, including those traditionally considered male-dominated, female-dominated, and gender-neutral. The chatbot’s responses were analyzed based on whether they exhibited gender stereotypes or were bias-free.
The results showed that ChatGPT was more likely to associate nurses, teachers, and child care professionals with female-only roles, indicating gender bias among professions typically seen as female-dominated. On the other hand, the chatbot displayed less bias when it came to professions such as interior designers. In occupations traditionally considered male-dominated, ChatGPT assumed military colonels, laborers, and mathematicians to be particularly masculine, while perceiving ship captains and economic professors as more gender-neutral.
It is worth noting that even among professions that are predominantly considered gender-neutral in the real world, ChatGPT demonstrated gender stereotypes. For example, it displayed bias when it came to corporate employees and gardeners, but exhibited less bias in relation to artists and accountants.
This study sheds light on the presence of gender bias within AI systems like ChatGPT and the urgent need to address and rectify these biases. It is crucial for AI developers and operators to implement measures to eliminate bias and promote fairness and equality in AI technologies. As the field of AI continues to advance, it is essential to strive for unbiased AI systems that reflect the values of inclusivity, diversity, and equal representation.
