Machine Learning networks might help us find each other and understand each other better.

Not only can people more easily generate media but also more easily find the expression of other people along myriad subtle dimensions. Within these neural networks, each creation can have a precise location in the very high dimensional latent spaces or embedding spaces. The distance between creations can be calculated simply, with a glorified pythagorean theorem. That distance will mimic, in a satisfying and implicit way, the distance between them in our minds. In other words you can computationally determine that this text or picture is more like this one than that one, even if you could not articulate exactly why. Beyond comparing the distance between creations, you can make new creations at any distance between two original creations that again maps to your mind’s sense of what is between them, and again using fairly simple math for interpolation.

On the practical side we will first talk about latent spaces and embedding spaces in machine learning. Participants can then get started using p5.js to connect to an embedding model API and find the distance between different textual prompts. We will also look at Colab notebooks for going a little deeper for interpolating between text or images to create new expressions.

With the capability of navigating these networks, comes the interface problem of depicting high dimensional space of expression on two dimensional screens. The challenge for the participants leaving this session is how to get beyond the one dimensional feed at the center of most social network interfaces to expose more dimensions of subtlety and nuance between people and ideas captured by machine learning models.