I hope they don't remember
Cristobal Valenzuela, in his own words, is a technologist, artist and software developer interested in the intersection between machine learning and creative tools. His website is full of extremely interesting experiments and work on the cutting edge of art and technology.
He has projects that scour satellite imagery to find geographical landmarks that represent each of the letters of the alphabet, augmented reality projects and fascinating data visualization work. The project that first drew me in was the machine learning program he developed called Text 2 Image. It is an interactive real time text to image generator using Attentional Generative Adversarial Networks (Looks like I need to read up on a new interest!). Essentially, it uses an algorithm to read a users text input, and attempt to create a new image based on its knowledge of images associated with that text. What comes out is usually a nightmare of shapes and colors but every once and a while you can get something that is beautiful and oddly accurate.
Some of my first images, the captions are the text I used:
I stopped at the last one. I don't know what that pitiful little grey blob really is but it's definitely alone on an open sea. It just seemed like I had hit some sort of jackpot of clarity. I had also spent about an hour on this site already. It is addictive.
I had to ask myself what this machine learns from us? Or more importantly , what does it remember. I know I found myself drawn to the absurd, hilarious, or somewhat filthy right off the bat, and most other people I know go the same route. Does this mean that in the infancy of machines intelligence it is learning from an internet full of people exhibiting, well, internet person behavior. That causes me to shutter. I think from now on I will spend my time on this site entering praise for our machine overlords so they will remember me as one of the good ones.
Comments