Latent Sheep dream

Machine for extrapolated visual inception

2019 - 2020

Keywords
Speculative Design
Human-Machine Expression
Deep Learning
Variational Autoencocoder(VAE)
Publication
DIS'20, Accepted
Teammates
Jiewen Wang
Ollie Hsieh
Jack Gale
Advisor
Prof. Sang-won Leigh
My Role
Early Concept Development
Innovations
Interactive Human-Machine Collaboration:Introduced a stylus-based interface that facilitates real-time interaction between human sketches and machine-generated reinterpretations, investigating the unexplored gap between modes of creation where human intent, agency, or control can be either lost or mitigated.

Generative Creativity Framework:Developed a hybrid system combining a sequence-to-sequence variational autoencoder (VAE) with rule-based generative models, enabling nuanced transformations of human input across both local and global axes to explore diverse creative outputs.

Speculative Exploration of Creative Agency:Visualized the spectrum of human-machine collaboration, from harmonious co-creation to complete divergence, provoking critical speculation on the boundaries of creative ownership and intent.
Overview

Latent Sheep Dreaming is a physical-digital installation exploring cohesion and conflict between human and machine agents in generative processes. 

Taking place in an empty room, a participant walks up to a stylus console and is instructed to draw sheep. Stylus input is transformed into a matrix of locally and globally affected outputs, increasingly unrecognizable with system influence and connoting of how an artist relinquishes creative ownership to machines. The machine's imagination process is projected on a display wall and physicalized via a pen plotter, documenting imagined shapes that emerge from user and machine collaboration.

Procedural Design Model

One of two components of the digital generative system, the rule-based system allows for the definition of a transformation rule to be applied globally or a group of points, with rule types ranging from insertion, translation, segmentation, and interpolation. Shown below is the application of a stacked set of transformations.