It would actually be cool to see a program emit random responses/actions and then based on a punishment/reward system learn to emit a pattern.