AI IN GENDERED SOCIAL MEDIA MODERATION: ADDRESSING BIAS AND ENHANCING INCLUSIVITY

Abstract

In this work, we explore the intersection of AI, gender and social media moderation coinciding with how (which) automation deals with gendered expressions, bias and inclusivity. With the rise of AI in social media moderation, waves are being made in how we monitor and control content on our platforms. Although AI has been successful in tending to big platforms, the same cannot be said of its use for gender-sensitive content moderation. Related work has also problematized AI bias – specifically in regard to gender –, with algorithms that over-flag or under-correct gender-based harassment (Noble, 2018; Eubanks, 2018). The contribution of this work is to consider AI's role in mediating gendered expression on social media, considering technical and sociotechnical aspects. We take a mixed-methods approach in which we quantitatively analyze gendered language data and investigate anthropomorphizing of machine learning models used for content moderation, its effect on such systems, if any. Results show that many current AI systems embody biases and imbalances around gender, resulting in both underreporting of harms against marginalized genders and overchilling of some forms of language. Such research highlights the importance of broad training sets, transparency in algorithmic decision-making and human oversight to promote fair practices that are inclusive.

Keywords: Artificial Intelligence, Gendered Content, Social Media Moderation, Algorithmic Bias, Inclusivity, Machine Learning, Gender-Based Harassment.