Learning index2
6.7%
Project #21879 — iteration #20
Minted on November 29, 2022 at 16:56Interactive
Prerendered components
WHISPERS (in code) is part of the fx(hash) interactive minting experience at Performance in Code: Deciphering Value in Generative Art, Art Basel Miami Beach, 2022.
The project is a collaborative performance between the viewer and AI. It reverses the usual approach of using AI as a tool to perform our intents (as in popular prompt-based AI methods) and gives the decision-making power to the AI, where we, humans, receive instructions on how to proceed in the creative process. Interaction happens in a multi-stage manner, in a metaphor for the process of training neural networks. At each step, after the user performs the action, they can send the image for evaluation by the neural network and receive another instruction (similarly to how networks learn by receiving feedback from the loss function). What makes us different from the machines, however, is the presence of free will, and our ability to disobey the machine’s guidance. We are not restricted to its instructions and are free to interact with and break the artwork.
Artwork is built on layers of photography, AI, GAN, diffusion, classification convolutional networks, random pixel manipulations, and user interaction. Like with most artist projects, it starts with a few memories, that are then re-imagined by GAN. The initial building blocks are a combination of photographs and AI outputs. The apparent lack of coherent subject choice in input images signifies how we build the perception of the world around us and apply an individual hierarchy to elements worth remembering. The blocks are then stripped of meaning by reducing them to abstract shapes and applying stochastic image transformations, guided by randomness. The neural network then learns to classify the images and outputs its best guesses on how to improve the pictures.
The goal of the project is to explore the controversial question of authorship through the use of tools incorporating AI. Does it lie with decisive or executive power? Here, the boundaries are blurred and one can no longer distinguish the clear author of a finished artwork - is it the artist, AI, serendipity, or viewer?
- Use your mouse to perform machine instructions.
- Use the "E" key to ask the neural network for instructions.
- Use the "R" key to reset the drawing.
- Use the "S" key to export an image.
The project is a collaborative performance between the viewer and AI. It reverses the usual approach of using AI as a tool to perform our intents (as in popular prompt-based AI methods) and gives the decision-making power to the AI, where we, humans, receive instructions on how to proceed in the creative process. Interaction happens in a multi-stage manner, in a metaphor for the process of training neural networks. At each step, after the user performs the action, they can send the image for evaluation by the neural network and receive another instruction (similarly to how networks learn by receiving feedback from the loss function). What makes us different from the machines, however, is the presence of free will, and our ability to disobey the machine’s guidance. We are not restricted to its instructions and are free to interact with and break the artwork.
Artwork is built on layers of photography, AI, GAN, diffusion, classification convolutional networks, random pixel manipulations, and user interaction. Like with most artist projects, it starts with a few memories, that are then re-imagined by GAN. The initial building blocks are a combination of photographs and AI outputs. The apparent lack of coherent subject choice in input images signifies how we build the perception of the world around us and apply an individual hierarchy to elements worth remembering. The blocks are then stripped of meaning by reducing them to abstract shapes and applying stochastic image transformations, guided by randomness. The neural network then learns to classify the images and outputs its best guesses on how to improve the pictures.
The goal of the project is to explore the controversial question of authorship through the use of tools incorporating AI. Does it lie with decisive or executive power? Here, the boundaries are blurred and one can no longer distinguish the clear author of a finished artwork - is it the artist, AI, serendipity, or viewer?
- Use your mouse to perform machine instructions.
- Use the "E" key to ask the neural network for instructions.
- Use the "R" key to reset the drawing.
- Use the "S" key to export an image.
KT1JcwM...fQ1gSTT
tz2BrRX...tc1Pncf
metadata was signed by fxhash