Deep learning and synthetic media

Deep learning algorithms are rapidly changing the way in which audiovisual media can be produced. Synthetic generated with deep learning—often subsumed colloquially under label “deepfakes”—have a number of impressive characteristics; they increasingly trivial to produce, and indistinguishable from real sounds images recorded sensor. Much attention has been dedicated ethical concerns raised by this technological development. Here, I focus instead on set issues related notion synthetic media, its place within broader taxonomy how techniques differ more traditional approaches synthesis. After reviewing important etiological features pipelines for manipulation generation, argue that “deepfakes” produced such do not merely offer incremental improvements over previous methods, but challenge taxonomical distinctions, pave genuinely novel kinds media.