On a paradox of probability theory