Deepfake Sockpuppets: The Toxic “Realities” of a Weaponised Internet

Deepfakes combine the latest in machine learning and Artificial Intelligence (AI) to create hyper-realistic audio-visual forgeries. Deepfakes materialise from a specific type of deep learning—hence the name—in which sets of algorithms compete in a generative adversarial network or GAN (Goodfellow et al. Advances in Neural Information Processing Systems. Cambridge, MA: MIT Press, 2014). Deep learning is a form of AI in which sets of algorithms (or neural networks) learn to simulate rules and replicate patterns by analysing large data sets. The generator algorithm creates content modelled on source material, which is already readily available (social media content, e.g.), whilst the discriminator algorithm seeks to detect flaws on the fake. This is an iterative process allowing for rapid improvements which address the flaws, meaning the GAN is able to produce highly realistic fake video content. This field of AI continues to advance at a rapid rate thus it is ever easier to create sophisticated and compelling videos. This has sparked concerns from politicians and international organisations about how these types of videos may be weaponised for malicious ends. These fears have fed into the post-truth era in the second decade of the early twenty-first century, where social media and information platforms have become increasingly uncanny spaces creating “abtruth” environments that, often toxify individual and collective memory to, destabilise the idea of provable facts and the nature of the “real” world.