For a serious take on their point: With parachutes, we have detailed physical models of how they work.[0] With anecdotal medicinal practices, however, AFAIK the

For a serious take on their point: With parachutes, we have detailed physical models of how they work.[0] With anecdotal medicinal practices, however, AFAIK there's little other than "this works sometimes."[1] We don't understand the body half as well as we understand physics, so I really don't think it's fair to compare the two. If we had no observational evidence for something, but we instead had an underlying model that made many correct predictions, then it would be fair to draw their conclusion that clinical trials aren't the end-all-be-all. But when you're operating solely on raw data with little to no understanding, you need data.[2] Of course, I am most decidedly an amateur in all the fields I've mentioned, so take all this with a grain of salt :-) [0] I don't actually know how parachutes were discovered, so maybe they were discovered purely experimentally. I'm inclined to doubt it, given a presumed lack of willing test subjects—but if I'm wrong, hopefully my comment will have value nonetheless. [1] To be fair, my understanding is that this applies to mainstream medicine too (isn't "how do general anesthetics work" still an open question?). [2] I feel this is a massive problem with current AI/ML trends as well. My undergrad "intro to ML" class briefly covered the statistics of it (VC dimensions and all that). The prof. basically said, "here's the amount of data you need to get statistical guarantees of correctness... and in practice, you'll have orders of magnitude less data than this." And yet we put this half-baked tech into cars, because it works on the happy cases, and totally ignore the lack of statistical confidence regarding the bad cases.