Do AI Models “Like" Black Dogs? Towards Exploring Perceptions of Dogs with Vision-Language Models

Abstract

Large-scale, pretrained vision-language models such as OpenAI’s CLIP are a game changer in Computer Vision due to their unprecedented ‘zero-shot’ image classification capabilities. As they are pretrained on huge amounts of unsupervised web-scraped data, they suffer from inherent biases reflecting human perceptions, norms and beliefs. This position paper aims to highlight the potential of studying models such as CLIP in the context of human-animal relationships, in particular for understanding human perceptions and preferences with respect to physical attributes of pets and their adoptability.