It looks like me, but it isn’t me: On the societal implications of deepfakes

Abstract

Deepfakes are images or videos generated using deep learning technology to change the original conditions of a piece of media. Potential uses of this technology range from satirical content, depicting public figures in comical scenarios; to generating audio to mimic a specific voice; to inserting the face of an unknowing individual into potentially embarrassing content, for example, a pornographic scene. As deepfake technologies, such as autoencoders and generative adversarial networks (GANs), become more advanced and accessible, deepfakes become easier to create and more believable. This poses some serious threats to individual and institutional safety, as scenes can be modified to alter public perception. Although technology to identify deepfakes does exist, it is essential that these methods progress as rapidly as deepfake technology so that they remain accurate. Additionally, legal consequences for deepfakes are limited, so continued advocacy for increased protection against false content is crucial.