I'm not the person to explain the whole theory. You can ask Google or wait for another post here. But...
The answer here is essentially the same as this experiment: Take a box and carefully place many red and blue balls into the box so that all of the red balls are on one side and all of the blue balls are on the other side. Then shake the box. Observe the location of the different colors. If you imagine the red ones as hot molecules and the blue ones as cold molecules, this is a pretty good analogy.
Notice that the red didn't "move to the blue" and vice-versa, they just averaged out, moving at random to a new part of the box. It's not that the blue balls attract the red ones. It's just the nature of the world that when things are mixed up they tend towards uniform distributions.
Edit: Plus, it's the nature of the world that things do tend to get mixed up over time.
Consider a container with a removable wall in the middle. On the right are many gas molecules moving relatively slowly (cold). On the left are many gas molecules moving relatively quickly (hot). If we remove the wall the hot and cold molecules will begin to collide with each other. The hot molecules, upon collision with their cold counterparts, will transfer some of their speed and momentum and thus cool down. This means, of course, that the cold molecules heat up.
In order for heat to flow from the cold to the hot these collisions would require that the slow molecules move even slower after a collision and the fast molecules move even faster. This is an impossibility in isolated collisions, and a rarity in multi-body collisions.