But the key fact to notice is that the second law is derivative: it is merely a consequence of probabilistic reasoning applied to Newton's laws of motion.
This leads us to a simple but astounding point: Since Newton's laws of physics have no built-in temporal orentation, all of the reasoning we have used to argue that systems will evolve from lower to higer entropy toward the future works equally well when applied toward the past. Again, since the underlying laws of physics are time-reversal symmetric, there is no way for them even to distinguish between what we call the past and what we call the future. Just as there are no signposts in the deep darkness of empty space that declare this direction up and that direction down, there is nothing in the laws of physics that says this direction is time future and that direction is time past. The laws offer no temporal orientation; it's a distinction to which they are completely insensitive. And since the laws of motion are responsible for how things change--both toward what we call the future and toward what we call the past--the statistical/probabilistic reasoning behind the second law of thermodynamics applies equally well in both temporal directions. Thus, not only is there an overwhelming probability that the entropy of a physical system will be higher in what we call the future, but there is the same overwhileming probability that it was higher in what we call the past.