A challenging but central question in population genetics is the detection of genomic regions underpinning recent adaptation. To this end, we recently devised a machine learning method, termed S/HIC, which detects both “hard” selective sweeps on de novo mutations and “soft” sweeps on standing genetic variation with high sensitivity and specificity, while being exceptionally robust to demographic model misspecification. We previously applied S/HIC to human population genomic data and uncovered evidence of a large number of recent selective sweeps across the genome, most of which we classified as soft sweeps. A critique of recent efforts to detect soft sweeps, including our own, has made the argument that S/HIC is in fact so vulnerable to demographic misspecification that our analyses with it should be completely discounted. Through a careful consideration of the claims of this critique, we argue that the impact of such misspecification on our analysis in humans is minimal with respect to our conclusions. The critique in question also argued that our false discovery rate in humans was essentially 100%; however we show that this inaccurate claim is due to a regrettable error on the part of its authors. We argue that our scan for selection has produced several interesting observations on recent adaptation in humans that are highly concordant with independent efforts to detect signatures of more ancient positive selection. We conclude that the evidence for the utility of S/HIC, and the validity of our application of it to human data, is highly compelling, and that strictly demographic explanations for our results are clearly untenable.

Update log: fixed both an omission and a typo in the Acknowledgments (one of the authors--cough--mistakenly acknowledged a professional surfer instead of a colleague who has the same given name).