Understanding Generative Adversarial Networks

Generative adversarial networks (GANs) are a new technology impacting social, policy, and security discourses. The rise of GANs enables the creation of artificially generated hyper-realistic human faces. GANs have been around since 2014,1 yet only recently has their quality risen to a level capable of fooling the average person. StyleGANs are GANs trained to manipulate or generate high-quality images.2 StyleGANs are becoming increasingly publicly accessible and enable users to generate human faces with ease. The styleGAN program is open access;2 while this allows for usage in positive ways, it also leads to easy accessibility for those that want to use this technology for malicious purposes. Fake accounts withAI-generated faces as their profile pictures plague social media sites and are often used as tools for misinformation. As styleGANs improve, it becomes more difficult to spot these fake faces. It is important to understand how these styleGANs work in order to best combat these disinformation attempts and understand what to do moving forward.