This is interesting, but I don't understand one thing: how can we use a Rule 30 or Rule 110 Cellular Automaton if it's computationally irreducible? How can we understand and program it to do whatever we need it to do if the only way to see what it does is to run every step of its computation? I see only computationally reducible automatons as useful tools to perform predictable computations, the kind of computations that we understand and can use. Where am I mistaking?

The experiment has to be run to know the outcome.