This project was developed to crowdsourced reinforcement learning to identify cellular automata rulesets that appear to mimic organic growth. We currently have three families of cellular automaton games - Conway, RedVsBlue, and Rhomdos - each of which has a corresponding space of rulesets that determines how the states of cells interact. For each game, volunteers can rate generated rulesets (www.biotaornada.com) to help inform our machine learning model about what types of rulesets look interesting or organic.

Our machine learning models (found in the ruleset_learning.py python file) are continuously trained by human feedback on our website (linked above). Eventually, once we have aggregated enough data to be effective, we will be adding a convolutional neural network which we hope to use for the purpose of automatically scoring rulsets as "interesting" or not. This step of automation will allow for drastic acceleration in our ability to develop, test, and search for optimal rulesets in a variety of cellular automata games.

We don't currently have a simple, well documented way to run our code locally on your own machine. We're working on tidying up some of our core classes and wrapping them into simple examples that can be continued to be developed by others. In the meantime, we recommend interacting with the experiment on our website.