"By the end of the nineteenth century, on the shoulders of Maxwell, Boltzmann and many lesser giants, physics had realised that there is a deep puzzle behind the familiar phenomena described by the new science of thermodynamics. On the one hand, many such phenomena show a striking temporal bias. They are common in one temporal orientation, but rare or non-existent in reverse. On the other hand, the underlying laws of mechanics show no such temporal preference. If they allow a process in one direction, they also allow its temporal mirror image. Hence the puzzle : if the laws are so even-handed, why are the phenomema themselves so one-sided ?"

"If it is conventional which direction counts as positive time, then it is also conventional whether entropy increases or decreases. It increases by the lights of the usual convention, but decreases if we reverse the labelling. But this may seem ridiculous. Doesn't it imply, absurdly, that the thermodynamic asymmetry is merely conventional ?

No. The crucial point is that while it's a conventional matter whether the entropy gradient slopes up or down, the gradient itself is objective. The puzzling asymmetry is that the gradient is monotonic it slopes in the same direction everywhere (so far as we know)."
p.121