White Stapler

These Ethereal Feelings

[I am … inhuman, almost formed]

Real-time generated 3D AI brain models using a custom interactive application. Created from images generated from a machine learning model trained on coronal MRI scans, these 3D models describe the movement of a form across a neural network space. The resulting otherworldly forms are completely machine generated, but still feel organic enough to be accepted as “beings” or part of something previously alive. These “brains” bring human physicality to the idea of machine intelligence. If they could, how would these hypothetical brains think and feel? Would AI ever be able to fabricate their own version of a natural being? How much control would we as humans have over that process? This is an exploration of machine unpredictability and the human collaboration/oversight of their progress.

Two artificially generated “brains” next to each other textured with faded colours

Raw model with inside voxel geometry

Process of generating raw model

This custom application invites users to take part in generating 3D brains from AI images by allowing them some control over speed and seed position within latent space using their mouse.

Demonstration of how the position of the user’s mouse (represented by the intersection of two lines and a dot) changes the AI brain image as it moves through latent space.

These AI generated brain scans are voxelized in real-time to create a 3D model. The distance between larger/smaller brain slices in the latent space is non-linear unlike the source MRI scans so it takes a bit of practice and luck to create “brains” that conform to the completed forms we are used to.

[Process]

[5,000~ Coronal brain scan MRI images] —-> [Pytorch StyleGAN3 trained AI image model] —-> [Custom TouchDesigner real-time interactive latent-walking voxelization application] —-> [3D models of reconstructed brains from AI generated images, PLY/GLTF export]

[List of sources:]

Edlow, Brian L. et al. (2019), 7 Tesla MRI of the ex vivo human brain at 100 micron resolution, Scientific Data, Article-journal, https://doi.org/10.1038/s41597-019-0254-8

Joosten, Robert, img2ply, robertjoosten/img2ply: Convert an image sequence to a PLY point cloud. (github.com)

White Stapler

Experimental digital artist, programmer, and designer working with new technologies. Dangerously obsessed with VR spaces and testing the fallout of context collapse with more than a hundred online identities

Leave a note!!!