Three years ago I was studying for a masters in Evolutionary and Adaptive Systems at the University of Sussex. That’s just UK academic speak for ALife. I was building evolutionary simulations of complexity increase in ecosystems .. basically a bunch of numbers doing interesting things if you thought about it or visualized it in the right way. Since graduating, I’ve been obviously focused on other things .. very very interesting and exciting things very much of the times. But I do miss the simulations and the ALife, and I have an ever growing list of things I’d like to hack in my tiddlywiki tech tickler file. Anyway, I had a couple ideas recently which I’d like to do more with than just stuff away for another day .. so I’ll take “action” and blog them!
Urban/Countryside Auto-classification of Geograph images
Geograph contributors are photographing every grid square in the UK. With human look at these photos, it’s simple to classify that area as urban or countryside, or any number of finer grain categories, and that seems like useful thing and some nice maps could made of the results. Automating that doesn’t seem that tricky .. take the color palette of a group of training images and run those through a machine learning algorithm. Bet it’s very learnable and a fun little hack.
School that Can See Itself
At Lift, Ben Cerveny blew everyone’s mind with his talk The Luminous Bath, and we’re all waiting for the video to be posted since no one really quite understood it the first time. There were bits and pieces, phrases, which stuck in my head, illuminating all sorts of things .. one of which was “individuals in a school of fish can’t see themselves” .. alluding to how our own social tools reveal our own higher order behaviors, and suggesting just how different a dynamic that will create.
Flocking behaviors are one of the classic algorithms of ALife. So easily, each individual could get additional input, some information about the state of the flock as a whole. Say the group was trying to elude some predators .. what happens if every member of the group has non-local knowledge of predator proximity to any other member? What happens to the flocking dynamics .. I can only guess there will be new stabilities and instabilities, periodic behaviors, maybe chaos.