Breaking into Binaries: Exploring AI’s Constructions of Women and Feminist Perspectives on AI Interaction

Gender biases entrenched within artificial intelligence (AI) models have the potential to build upon and worsen the conceptualizations of women and reinforce misogynistic power structures. This happens through the biased datasets that are used to train AI, the feminization and sexualization of AI, and the interactions users can have that build upon harmful conceptualizations of gender. By building chatbot models and training them on data collected from Reddit, I examined how differently biased datasets and model configurations could produce various constructions of gender by analyzing the model’s outputs. Connecting this with the development of AI girlfriends like Replika, this thesis underscores the potential for users to reinforce, create, and/ or internalize various constructions of women. Considering the current trajectory of AI, I put forth that applying feminist theories of gender performativity to users’ interactions with AI not only provides a feminist approach to AI but also works to disrupt the construction and performance of gender as a whole.