Artificial intelligence developers increasingly rely on synthetic data to train algorithms when real-world information is scarce or too sensitive to use. While this computer-generated data promises to reduce bias, it ultimately remains limited by the original data used to create it.