This comment is I think an essential couterbalance to the post's valid points. To expand a little, the book Good Calories, Bad Calories by Gary Taubes argues that bad nutritional recommendations were adopted by leading medical and then governmental associations, partly justified by the above advice (we need recommendations to help people now, can't wait for full testing). So someone could refer to this as an example of why the comment above is dangerous in areas that are harder to test than the efficacy of steel production (which I presume they knew worked better than other procedures, whereas some nutritional effects have long term consequences that aren't clear or it's not clear which component of the recommendation is affecting what). However, Taubes also shows that this was also used to justify overlooking flaws in the evidence, and he points to a group heuristic bias (if that's the right term) of information cascades. There are other biases and failures of rationality (how certain statistical evidence was interpreted) in the story as well. So all this to say, while trial and error give give faster and as effective results, the less clear the measurement of the results are, the more care required interpreting them. When stated, it sounds obvious and I almost feel dumb for saying it, yet it's one of those rules honored more in the breach as they say. In the field of nutrition, you'll have headlines that say "Meat causes cancer" based on a study that points to a small statistical correlation between two diets which have very many differences other than type and amount of meat and itself concludes that more studies are called for to examine possible links between meat and cancer but not other possible causes that are just as much pointed to by the study.