From an intuitive and non rigorous way, you could think in this fashion:
Get the evolution of stock price as standard normal variable (N), ie. ln(St) is normally distributed, (St is a geometric brownian motion), reduce it to standard normal form: z= (ln(St)-E[ln(t)])/StdDev[St]. Now say for a digital call option, it would be exercised only if the stock price St >K (strike) or ln(St)>ln(K). Hence the probability of its exercise is the probability that z corresponding to a price St is > z corresponding to K ie. (say: x=lnK-E[ln(t)])/StdDev[St]. ) Since z is standard normally distributed, hence the probability is 1-N(x). Similary for digital put, the probability is z corresponding to St<x, hence N(x).
It is bit roughly presented but to have a better understanding, you might want to have a look at the derivation of black-scholes from the first principle.