Some mathematical models of "group selection" are really just individual selection in the context of groups.[2] The modeler arbitrarily stipulates that the dividend in fitness that accrues to the individual from the fate of the group does not count as "individual fitness." But the tradeoff between "benefiting the self thanks to benefiting the group" and "benefiting the self at the expense of the rest of the group" is just one of many tradeoffs that go into gene-level selection. Others include reproductive versus somatic effort, mating versus parenting, and present versus future offspring. There's no need to complicate the theory of natural selection with a new "level of selection" in every case.
If an additional parameter in the model helps the model to make better predictions then it's good to use it. The results of the predictions justify the parameter.
Maybe Pinker makes this error because he make a wrong assumption about evolution being deterimistic. In real life there a lot of randomness in evolution. Some gene's who are advantageous simply disappear in genetic drift. Some "bad" genes win just because they are for a bunch of time located next to a really "good" genes.
There are simply a lot of mutations. Life isn't fair. The good mutations only win on average. Adding parameters that track individuals or groups help to produce a more accurate model.
Jerry Coyne says that the vague idea of group selection at best gives the same result as the selfish-gene model, with more complicated mathematics, not less complicated - it fails Occam.
This essay at Edge touches on a few possible meanings for the term "group selection." Pinker argues that as a form of memetic theory it has no explanatory power, and that group selection for genes does not fit the evidence. He focuses on humans with some mention of insects that live in hives. So the essay doesn't seem surprising, but it does seem rather Hansonian.