Popular press versions of biological research are often ripe with anthropocentrism, and this is no exception. Evolution by natural selection acts on 'copying mistakes' all the time, whether adding, deleting, or mistaking a single letter, word, sentence, or paragraph (to extend the crappy metaphor). The underlying research reports that a gene duplication event, the sort of thing that has been well characterized for many years, has occurred in a gene that modifies the number of projections that a neuron has. The amazing thing is the connection between the gene and the trait, not the mutation arose by a copying mistake. One could argue that all mutations are copying mistakes.

Appeal to authority, whether the authority is a crowd or a bearded man in a white robe, is a logical fallacy.

The reason that people are paying attention to hight is, in part, that it's a simply measured complex trait. Every study of human genetics under the sun collects basic anthropometrics, and so it's relatively easy to lump everyone together in an effort to increase the power to detect genetic variation that influences height. I think the real interesting part here is that even after collecting a hundred thousand data points, the obvious data analysis methods can account for a relatively low proportion of the total variance in height. That has consequences for studies of other disease traits with complex genetic architecture like diabetes or schizophrenia, which have often have study sizes one or two orders of magnitude lower than this one. In the not so recent past, influential members of the scientific community have suggested that big studies of complex traits in humans might have a profound impact on bedside medical decisions. It's going to take a bit longer than they anticipated. To our collective dismay, biology is still complicated.

There is almost certainly an epigenetic effect. The latest buzz is that there is no genetics without epigenetics. However, the size of the effect must be less than the 80% that is attributable to genes (and yes, that figure is pretty robust). The real problem is having the data to measure the epigenetic effects from studies that have already been conducted. If the study of epigenetic factors wasn't part of the original study design, it's awfully hard to model these effects as an afterthought in a meta analysis.

Services like those mentioned in TFA may be able to provide information on which genetic variants a person carries, but will not interpret those results. Non-scientists, and even scientists seem to over estimate the ability of modern genetics to assign meaning to common genetic variation. Your average M.D. when confronted with a print out of a patients 'mutations' would be completely unable to make heads or tails of them. There are few instances such as cystic fibrosis, where the etiology is well known, and known mutations WILL cause disease. In other cases such as BRCA in breast cancer, 'mutations' are risk factors for disease. In the vast majority of cases, modern genetics has no idea what a 'mutation' at rs39842093 might actually do. These services are expensive, ambiguous, and require a certain measure of vanity on the part of the consumer. If you have a family history of disease X, there may be a small number of 'mutations' for which you might be tested that could actually impact your future health, and those services are provided by someone other than 23andMe. Biology is a bit different than technology in that observing that biology works does not imply that someone knows how it works. (Creationists can bite me.)