A Study on the Types and Ambivalence of Deepfake Content

This study aims to examine deepfake content, which has sparked heated social controversy. Deepfake content utilizes generative adversarial networks to create videos, images, and audio that appear more real than reality itself. As a result, deepfake content is enhancing creative efficiency in the cultural content domain. Conversely, deepfake content generates videos, images, and audio at a level where it is difficult to distinguish between real and fake, leading to criminal exploitation. This ambivalence is related to the fact that deepfake content primarily references real people and deceased individuals as its main subjects, while also being connected to Baudrillard's concept of hyperreality. Based on this premise, this paper attempts to re-categorize deepfake content centered on reference subjects, moving beyond existing discussions that have typologized deepfake content from a technical perspective, and examines the resulting ambivalence. The research findings are as follows. Deepfake content was categorized into existing types, absent types, and virtual types, with detailed subtypes including appearance modification type, speech manipulation type, action synthesis type, personal type, and non-personal type. The majority of deepfake content was found to belong to existing types and absent types. Examining the ambivalence by type, while existing types and absent types brought production efficiency in dramas, films, and advertisements, ethical issues were found to be mainly concentrated in existing types.