Quantifying Societal Bias Amplification in Image Captioning
This addresses the issue of standardized bias measurement for image captioning models, which is crucial for developers and researchers aiming to reduce societal biases in AI, though it is incremental in proposing a specific metric.
The paper tackles the problem of societal bias amplification in image captioning by proposing LIC, a new metric that considers the whole context beyond protected attribute prediction, and finds that existing bias mitigation models unexpectedly amplify bias.
We study societal bias amplification in image captioning. Image captioning models have been shown to perpetuate gender and racial biases, however, metrics to measure, quantify, and evaluate the societal bias in captions are not yet standardized. We provide a comprehensive study on the strengths and limitations of each metric, and propose LIC, a metric to study captioning bias amplification. We argue that, for image captioning, it is not enough to focus on the correct prediction of the protected attribute, and the whole context should be taken into account. We conduct extensive evaluation on traditional and state-of-the-art image captioning models, and surprisingly find that, by only focusing on the protected attribute prediction, bias mitigation models are unexpectedly amplifying bias.