Latent Space And Error-prone Machines:Image Narratives In Generative Adversarial Networks

Generative adversarial networks are an emerging approach to image generation,which uses a zero-sum game between two machines to statistically explore the commonality of a class of images and then use that commonality to generate images.From the moment the first AI work was auctioned,the discussion on "AI art" has never stopped.While people marvel at the continuous and surreal imagination of the machines,they also question the spiritual rhythm of their works,copyright and the human right to speak about creativity,and other "art ethics" issues."The technology,due to its own algorithmic design,is not only a good way to make the artwork,but also a good way to make the art.Since its inception,this technology has been widely understood as a mysterious machine that replaces human creativity because of the "inexplicable" nature of its algorithmic design and the usual mentality of rejecting the process and accepting the result.However,without a perspective on the technology itself,the understanding and application of the technology will remain stagnant.In this paper,by deconstructing the technical points of generative adversarial networks,we search for the fundamental reason why AI images make people feel alienated and mysterious: the artist transfers a part of the control of creativity and focuses more on guiding and selecting the algorithm.The author also tries to extract the potential aesthetic meanings of the machine’s operation: A.the wandering and gaming in the continuous computable semantic space;B.the "intelligence" evolved on errors,noise and chance;C.the digital sensibility reflected by the unconsciousness of the algorithm.The author proposes that these aesthetic experiences are not only the method of operation of GAN as a media art creation tool,but also can be used as the content of image narratives to deepen the re-examination and expression of human-machine relationship in media art in the era of human-machine integration.