What are the limitations of using probability to predict the actual offspring of a single cross?

Punnet square can be described as a diagram which shows the probability of the genotypes and phenotypes of the offsprings that might result by the mating of two parents.

The limitation of this method is that one cannot particularly tell which offspring might possess the genotype or phenotype which we are looking for. For example, in parents who are a carrier of a disease we can predict that some of the offsprings will actually have that disease. But we cannot predict which one of the pregnancies would produce normal offsprings and which one will produce the diseased offsprings.