Google’s DeepMind AI just taught itself to walk

I admire how I can watch the model slowly evolve into fitness. The project was created by Google’s DeepMind AI, and it involves teaching various models how to traverse across an obstacle. The model learned how to walk, run, balance, and climb. I am not actually sure if this is a machine learning AI, it could most likely be a genetic algorithm that is making this work. Regardless of the underlying algorithm, it is still an impressive feat of compressing the natural process of evolution.

Ant Behavior

I admire how a swarm behavior is represented here. The nodes simply randomly search the space, and one it found food it starts moving back to its hive leaving behind a trail of chemicals. If other nodes found the trail of chemical they simply followed it. It’s amazing how such a simple rule can produce such complex behavior. One downfall for this piece of “art” is that it is very unaesthetic. The system is implemented via a grid system, which is fine, but it is represented using simply rectangles as if this was the early 90s. Really the aesthetic of just sucks.

http://www.natureincode.com/code/various/ants.html