I've also reached this conclusion, but I find it curious. The fact that many CAs are Turing-complete and given the meaning of that, it is theoretically possibl

I've also reached this conclusion, but I find it curious. The fact that many CAs are Turing-complete and given the meaning of that, it is theoretically possible to model anything with a CA. Yet CA models have never proven useful. One possible explanation may be the timeless nature of mathematical explanations and the power of math to make predictions. If I say F=ma (Newtonian mechanics), I can evaluate 'a' for any 'm' and 'F' trivially. I could also construct a CA -- maybe even one that modeled physics very well -- in which F=ma arose as a consequence of that CA's operation, but to make any prediction with this system I would have to actually compute the CA across all iterations until I reached my result. It wouldn't be useful because it would be too slow. In other words predictive mathematical descriptions of nature are more useful because they find ways to factor out time.