Sacred in Flux

Man vs. Machine

Defining what can be understood as "sacred", in a universal form, is a complex task. Sacredness is a concept that involves the realm of spirituality, and demands a deep reflection on the origins and mysteries of the universe. Closely related to religious beliefs, and often involving a connection to something greater than oneself, what each of us defines as holy is determined by variables such as cultural backgrounds, place of birth, and individual experiences.

For Artificial Intelligence machines, which are usually trained to analyze and process large sets of data, grasping the meaning behind terms such as "God", "holy", "spiritual" and even "mysticism", can be extremely challenging. Some experts claim that AI could potentially understand these ideas if it was programmed to learn about it. However, others suggest that these notions may be too abstract and subjective, and that a real understanding about them, by AI systems, is virtually impossible.

In 2016, artist Juliana Freire and Edson Pavoni were caught up in this debate. They started to wonder how AI generated images could understand, within their own limitations, religious concepts dear (and maybe even exclusive) to humanity.

Sacred in Flux flourished out of a desire to co-create with such technologies, which the artists have come to consider "interdimensional beings". This exercise then became an attempt to amalgamate a cosmic dimension to the man vs. machine paradox.

Creative Process

Sacred in Flux used Generative Adversarial Networks (GANs), a recent innovation in machine learning which is capable of creating new data instances that visually resemble the training information which has been fed into its system. In this case, the artists fed into the deep neural networks over 14 millions images, which included mystical symbols, religious icons, representations of saints and deities, temples, mythological figures, and more. From different time periods and locations, these images aimed to give a substantial amount of material for the software to analyze and create patterns of recognition.

After that first phase, the artists went on to develop "seed images": paintings and collages which would be the basis for the works. These images were added to the software, and combined with the large set of images that it had already been familiar with. The seed images were then modified, over and over again, through the artist's commands. By combining terms such as "God", "woman" and "sacred", for example, they were able to create a dialogue with the machine using the vocabulary it already knew.

Even though these processes are recent, and can be difficult to understand, one could compare the use of AI generated images to the instrument of musicians. Just like a piano player needs to select the notes necessary to sound a chord, so do the artists that need to ask the software for the elements required to create the desired images.

The exhibition

First showcased in São Paulo, at the Pãa Gallery, the exhibition consisted of large scale reproductions of the images, set up in metallic structures that resembled construction's scaffoldings. Smaller objects made of wood held three consecutive glasses, each representing a step of the image's transformation. The show used the existing furniture of the gallery as exhibition devices, creating sharp contrasts between their antique, artisanal styles and the futuristic characteristic of the artworks in display.