Will the Inclusion of Generated Data Amplify Bias Across Generations in Future Image Classification Models?

TL;DR

This paper investigates whether using synthetic data generated by models for training image classifiers amplifies bias over successive generations, highlighting potential fairness concerns in AI development.

Contribution

The study introduces a simulation environment to analyze bias dynamics across generations when training with generated data, providing empirical insights into fairness impacts.

Findings

Bias can increase over generations with synthetic data

Generative models may reinforce subgroup biases

Fairness metrics vary across datasets and generations

Abstract

As the demand for high-quality training data escalates, researchers have increasingly turned to generative models to create synthetic data, addressing data scarcity and enabling continuous model improvement. However, reliance on self-generated data introduces a critical question: Will this practice amplify bias in future models? While most research has focused on overall performance, the impact on model bias, particularly subgroup bias, remains underexplored. In this work, we investigate the effects of the generated data on image classification tasks, with a specific focus on bias. We develop a practical simulation environment that integrates a self-consuming loop, where the generative model and classification model are trained synergistically. Hundreds of experiments are conducted on Colorized MNIST, CIFAR-20/100, and Hard ImageNet datasets to reveal changes in fairness metrics across generations. In addition, we provide a conjecture to explain the bias dynamics when training models on continuously augmented datasets across generations. Our findings contribute to the ongoing debate on the implications of synthetic data for fairness in real-world applications.