The Dark Side of Dataset Scaling: Evaluating Racial Classification in Multimodal Models

TL;DR

This study investigates how scaling large multimodal datasets impacts racial and gender biases in vision-language models, revealing increased bias in larger models trained on bigger datasets and discussing mitigation strategies.

Contribution

It provides a comprehensive evaluation of bias changes in 14 visio-linguistic models as dataset size increases, highlighting the complex effects of data scaling on model fairness.

Findings

Bias towards criminal classification increases with dataset size in larger models.

Smaller models show decreased bias with dataset scaling.

Larger models exhibit more racially dehumanizing misclassifications.

Abstract

Scale the model, scale the data, scale the GPU farms is the reigning sentiment in the world of generative AI today. While model scaling has been extensively studied, data scaling and its downstream impacts on model performance remain under-explored. This is particularly important in the context of multimodal datasets whose main source is the World Wide Web, condensed and packaged as the Common Crawl dump, which is known to exhibit numerous drawbacks. In this paper, we evaluate the downstream impact of dataset scaling on 14 visio-linguistic models (VLMs) trained on the LAION400-M and LAION-2B datasets by measuring racial and gender bias using the Chicago Face Dataset (CFD) as the probe. Our results show that as the training data increased, the probability of a pre-trained CLIP model misclassifying human images as offensive non-human classes such as chimpanzee, gorilla, and orangutan decreased, but misclassifying the same images as human offensive classes such as criminal increased. Furthermore, of the 14 Vision Transformer-based VLMs we evaluated, the probability of predicting an image of a Black man and a Latino man as criminal increases by 65% and 69%, respectively, when the dataset is scaled from 400M to 2B samples for the larger ViT-L models. Conversely, for the smaller base ViT-B models, the probability of predicting an image of a Black man and a Latino man as criminal decreases by 20% and 47%, respectively, when the dataset is scaled from 400M to 2B samples. We ground the model audit results in a qualitative and historical analysis, reflect on our findings and their implications for dataset curation practice, and close with a summary of mitigation mechanisms and ways forward. Content warning: This article contains racially dehumanising and offensive descriptions.