My assertion in more detail. First, there aren't big population-level differences amongst random partitions of the single human race. There are population-level differences amongst
geographic partitions, and humans are known to not travel especially far from where they are born, so that we might suspect genetic drift is happening; humans tend to not spread genes very far geographically. Indeed, there are countless examples of genetic drift in humans.
Second, those drift differences are largely cosmetic, and cosmetic on average. Both properties follow from differences being Gaussian distributed with respect to fitness; the typical drifted gene is neither especially fit nor especially unfit.
Third, as gene pools grow, the ability to control for fitness on a per-gene basis shrinks exponentially, so that the behavior of the pool is like a random permutation. To see this, imagine a picky procreator choosing from a list of suitors by genetic partition; at each stage, they will reject some suitors who have the wrong allele. The pool randomly combines each pair of alleles (because it is deep enough to hold every viable combination), so each rejection removes roughly the same proportion of suitors every time. Thus, a linear list of picked alleles requires an exponentially-large list of suitors to choose from!
Aside: This third fact is why breeding dogs gives us pugs with their poor breathing and also why breeding tends to give us lots of hip dysplasia across different breeds. We can only positively select for a relatively small handful of traits and alleles, and each selection allows thousands of selfish genes to join the entourage, giving us patterns of unwanted gene expression.
None of these facts preclude services like 23andMe; we can certainly write down lists of clustered alleles which correspond to drifted genes, and use that information to correlate people with the geographic history of their ancestors, on a probabilistic basis. (Remember, large gene pools are randomly permuted, so these services must be probabilistic.)
None of these facts preclude some genes from being selected for fitness, either. Let's examine sickle-cell anemia more closely. Sometimes a mutation can improve an allele's fitness when one copy of the gene is present, but two copies results in a poor allele. Sickle cell is a very good example of this. Having one copy gives the "sickle cell trait", which resists malaria. It has a bunch of other minor health effects, positive and negative, but resisting malaria is sufficient to allow the gene variant to stay in the gene pool.
What I'm saying is that, just because some people have sickle-cell anemia, that doesn't therefore change the fact that people are all of the same species. If folks want to talk about which individual genes they have, then they should do that, and not talk about race, because genes only stick together in geographic clusters probabilistically, due to drift.