From a purely functional, non-ethical standpoint, the problem here is that you can't select the smartest embryo from a pile. You can only select the embryo with the most genes identified as correlating with intelligence.

Even assuming that this works, when you do that, you immediately run into the same problem as every other eugenics program: you are reducing the diversity of the population to seek a local maximum. In the analogy for optimization problems of the cows in the field standing on hills and in valleys, you're having all the cows stand on a single hill. You improve the current fitness of the population, according to your metrics, at the cost of the future fitness - you'll never find higher hills to stand upon, and if the fitness function changes, your population will be less equipped to handle it.