autocorrelation definition

Question

There is something that is not clear for me in the definition of an autocorrelation and how to calculate it physically. Namely, following a definition for a random process with real values $x(t)$: $$ R(t_1,t_2) = E [x(t_1)\cdot x(t_2) ]$$ So we calculate the expected value of the product of the values at time $t_1$ and $t_2$. The question is however how to calculate this value? I mean that from a process we have only two values $x(t_1)$ and $x(t_2)$ and there is no statistics in order to average this values. Or does this mean that we perform thought experiment and restart a process from the beginning and in this case we will get another values $x(t_1)$ and $x(t_2)$ in accordance to the distribution of the random process x?