IO

GameWorld

I had some time to continue my experiments on interactive sound & reinforcement learning. Since pretrained agents are fascinating in terms of behaviour, I am thinking about a strange relationship between these skills and the human player. Competing against these artificial agents directly would be too direct and unfair (they are dumb in the beginning, but they get quicker than humans slowly during the training process), so I decided to build another kind of interaction with them: the user should help one of them achieving its goal, while constantly avoiding the other achieving the same goal. 

It is like an indirect interaction, derived from games like Lemmings, SimCity or other strategic games, where the user sometimes reconfigures the environment, then acts like an observer, leaving space for causalities and possibilities to happen within those directed constrains.

I like the word we use for making music: musicians usually play their instruments, so it is also part of these experiments to make music (or event based sonification if you prefer) not in terms of realizing musical ideas, but more like traces of problem solving steps within an abstract playground. 

This is a very basic, minimal & constrained version that has many possibilities to turn this thing into a cute mere sonic toy, by designing levels, involving components such as time & speed etc. Let's see how this idea unfolds in the future. 

Play with the experiment in your browser here, or get the code from the research repository. Built with p5js (using webaudio) and reinforcejs.