인공지능 기반 이미지 생성 알고리즘과 사진 = AI-based Image Generation Algorithm and Photography

다국어 초록 (Multilingual Abstract)

This article deals with how images produced using artificial intelligence-based algorithms differ from existing photography. A generative adversarial network (GAN) is an algorithm that calculates fake data through balanced learning of generators and discriminators. When using original photos as learning data, it creates images that are visually indistinguishable from photos. There are two issues raised by this. First, the algorithmic image cannot be called a photograph from the point of view of index theory, but it is no different from the existing digital photograph in that it creates a new image by changing the pixel value of the original photograph. Second, the algorithmic image is an advanced form of program automatism and human exclusion, which Vilem Flusser defined as the core of the technical image. Humans are excluded from the production process of the GAN algorithm image. In that sense, the generator of the GAN is a black box. As the automaticity of the program increases, humans do not control image production and become simple consumers. Therefore, it is time for humans to think about how to turn a black box called a program into a “transparent box.”

참고문헌 (Reference)

1 박상우, "박상우의 포톨로지: 베르티옹에서 마레까지 19세기과학사진사" 문학동네 2019

2 "https://generated.photos/"

3 Radford, Alec, "Unsupervised Representation Learning with Deep Convolutional Gererative Adversarial Networks"

4 Zhu, Jun-Yan, "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks"

5 Flusser, Vilém, "Towards a philosophy of photography" Reaktion Books Ltd 1983

6 Choi, Yunjey, "StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation"

7 Bazin, André, "Qu'est-ce que le cinéma?" Cerf 1945

8 Karras, Tero, "Progressive Growing of GANs for improved quality, stability, and variation"

9 Ledig, Christian, "Photo-Ralistic single image Super-Resolution using a Generative Adversarial Network"

10 Gunning, Tom, "La retouche numérique à l'index" 19 : 96-119, 2006

11 Isola, Phillip, "Image-to-Image Translation with Conditional Adversarial Networks"

12 Parker, James R., "Generative art: algorithms as artistic tool" Alberta 2020

13 Goodfellow, Ian J., "Generative Adversarial Nets"

14 Berthelot, David, "BEGAN: Boundary Equilibrium Generative Adversarial Networks"

15 Karras, Tero, "A Style-Based Generator Architecture for Generative Adversarial Networks"

16 McCarthy, John, "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence" 27 (27): 12-14, 1955