As with randomly generated CA, I think the most interesting are the ones that exhibit some large scale characteristic which persists while micro states continue to evolve unpredictably, as for example:
https://maximecb.github.io/Turing-Drawings/#4,4,1,3,2,3,2,0,...
I wonder if Stephen Wolfram has opined on the applicability of his 4 classes of CA behaviour (fixed, periodic, random, complexly structured) to this type of automaton?