What ChatGPT Tells Us about Gender: A Cautionary Tale about Performativity and Gender Biases in AI

Large language models and generative AI, such as ChatGPT, have gained influence over people's personal lives and work since their launch, and are expected to scale even further. While the promises of generative artificial intelligence are compelling, this technology harbors significant biases, including those related to gender. Gender biases create patterns of behavior and stereotypes that put women, men and gender-diverse people at a disadvantage. Gender inequalities and injustices affect society as a whole. As a social practice, gendering is achieved through the repeated citation of rituals, expectations and norms. Shared understandings are often captured in scripts, including those emerging in and from generative AI, which means that gendered views and gender biases get grafted back into social, political and economic life. This paper's central argument is that large language models work performatively, which means that they perpetuate and perhaps even amplify old and non-inclusive understandings of gender. Examples from ChatGPT are used here to illustrate some gender biases in AI. However, this paper also puts forward that AI can work to mitigate biases and act to 'undo gender'.

Automated summary

Large language models and generative AI like ChatGPT have had a significant impact on people's personal lives and work, with the potential for even greater influence in the future. However, these technologies also harbor gender biases that perpetuate stereotypes and disadvantage diverse genders. This paper argues that large language models perform in a way that reinforces old and non-inclusive understandings of gender, but also suggests that AI can be used to mitigate biases and promote gender equality.