Can you make a deep CA such that layer 1 evolves the rules itself that apply to layer 2? Visualize the outputs of layer 1 and 2 in tandem then. Following this, apply this update logic to n layers. There are some papers on this that exist that can LLM can help find.