[stracer]

Ok, but notice this argument is based on purely probabilistic assumptions and that all micro-states are equally probable to occur and do occur in time. There is still the question whether these assumptions are valid in a given physics system.

In physics we know much more about evolution of systems, for example in classical mechanics model of gas, there are deterministic equations of motion which also conserve some quantities like energy. So not all thinkable chaotic states are accessible, and the evolution of state is not a repeated random choice out of the whole set of states.