One of the interesting things about complexity is that while the future evolution of the system may be essentially unpredictable, generally the statistics of it aren’t. Thus we still have system characteristics. Another interesting thing is the concept of universality classes.
The idea here is that the typical behaviour characterized by the behavioural statistics, depends on a relatively small number of features of the system, such as its symmetry or dimensions. The importance being that the global behaviour may be quite insensitive to a large number of variations in the rules governing the micro-behaviour of individual cells. This is what gives some hope for representing vastly more complex systems in terms of relatively simple automata.