Generative Adversarial Networks (GANs) are a revolutionary class of Deep\nNeural Networks (DNNs) that have been successfully used to generate realistic\nimages, music, text, and other data. However, GAN training presents many\nchallenges, notably it can be very resource-intensive. A potential weakness in\nGANs is that it requires a lot of data for successful training and data\ncollection can be an expensive process. Typically, the corrective feedback from\ndiscriminator DNNs to generator DNNs (namely, the discriminator's assessment of\nthe generated example) is calculated using only one real-numbered value (loss).\nBy contrast, we propose a new class of GAN we refer to as xAI-GAN that\nleverages recent advances in explainable AI (xAI) systems to provide a "richer"\nform of corrective feedback from discriminators to generators. Specifically, we\nmodify the gradient descent process using xAI systems that specify the reason\nas to why the discriminator made the classification it did, thus providing the\n"richer" corrective feedback that helps the generator to better fool the\ndiscriminator. Using our approach, we observe xAI-GANs provide an improvement\nof up to 23.18% in the quality of generated images on both MNIST and FMNIST\ndatasets over standard GANs as measured by Frechet Inception Distance (FID). We\nfurther compare xAI-GAN trained on 20% of the data with standard GAN trained on\n100% of data on the CIFAR10 dataset and find that xAI-GAN still shows an\nimprovement in FID score. Further, we compare our work with Differentiable\nAugmentation - which has been shown to make GANs data-efficient - and show that\nxAI-GANs outperform GANs trained on Differentiable Augmentation. Moreover, both\ntechniques can be combined to produce even better results. Finally, we argue\nthat xAI-GAN enables users greater control over how models learn than standard\nGANs.\n