Sensemaking in User-Driven Algorithm Auditing: A Case Study on Gender Bias in an Image Captioning Model

Key Points

• The study aims to explore how users perceive and audit algorithms for biases, specifically gender bias.
• Conducted a case study focusing on an image captioning model.
• Engaged users to assess the model for gender bias.
• Used qualitative analysis of user feedback to identify biases.
• Identified significant gender bias in the image captioning outputs.
• Users reported discomfort with biased representations.
• Highlighted the importance of user insight in detecting algorithmic biases.