In this context, a random walk happens on a 2D coordinate plane. Your drunk person starts at the origin, (0, 0), and for a “random walk” they move either left, right, up, or down by exactly 1 unit each step. It’s a mathematical fact that this process, taken to its limit where infinitely many random steps are taken, will always have the drunk return to the origin - in fact, for any given integer coordinate on the plane there’s a 100% chance the drunk will eventually visit that coordinate following a random walk.
This doesn’t work in 3D though, where there’s an x, y, and z axis. A random walk there won’t always return to the origin - it only will about 34% of the time. If the drunk gets too far away the probability of ever finding their way back at random quickly drops to 0.