Cellular automata (CA) are underrated.
Most people’s experience with them are simply Conway’s Game of Life or similar setups. The topic is much deeper than that though. Some 2D CA are in fact Turing complete [1] and can be used to perform any computation (albeit inefficiently).
So what? Why not just use python or if you're really bored use a transformer model? Because in this context it's not about doing the computation but rather what we can learn about computation in general. The beautiful thing about CA is that they are naturally visual and we can see emergent patterns at a glance while they work.
First consider that ALL scripts, models, charts, etc. on computers are discrete. Then understanding the simplest discrete systems (ie. CA) may give us an idea of the limitations of the more complex ones. In fact, even simple, deterministic systems can produce unpredictable behavior [2]. And sometimes even complex rules don’t lead to complex output.
It also turns out that there's a natural connection between convolutional neural networks and CA [3] so this touches on important topics in AI. This is the focus of my current research and I'm hoping to publish more on that eventually (not under this nym of course).
I recommend more people get creative with studying cellular automata.
——
1. 
2. 
3. 
Rule 110 - Wikipedia
Rule 30 - Wikipedia
arXiv.org
Cellular automata as convolutional neural networks
Deep learning techniques have recently demonstrated broad success in predicting complex dynamical systems ranging from turbulence to human speech, ...