> most bodybuilders go through a lot of trial and error to find the right diet.Scientists sometimes go through a lot of trial and error and fail, despite of using n>1, despite of collecting a lot of data and processing it with tricky statistical methods.
Just going through motions of trial and error doesn't guarantee a success. One can hope that he will do a gradient descent and will find a local minimum, but even this is not guaranteed, because for gradient descent you need to identify all the variables and measure all partial derivatives. And do this multiple times, for each step of gradient descent.
> they exist as a community of millions of people all trying things are reporting what works
Millions of flies cannot be wrong? I'm sorry, I cannot resist the urge to bring this phrase.
Lets try more rational arguments and start with the question: could these millions do a search for a global maximum?
1. Most of them do not track their progress, do not write things they do, and just relying on the innate abilities of their minds to learn the best way. So they reports of their results are not reliable. Probably they can be used to seek truth but one needs to be very careful and skillful to do this properly.
2. Just a few of them can measure the real result with high enough frequency. One needs an access to medical diagnostics to learn what their bulges consist of. Is this a lot of muscle fibers or not so much fibers but they are swelled? Why they are swelled? So really they can vaguely suspect what results they got after several months of adopting new training strategy.
3. They do not track failures. If someone tried their methods and failed, he would leave the scene quietly. So if these methods works only for a small part of self selected population, bodybuilders would remain in a blissful ignorance of the fact. It is a selection bias that can be substantial in bodybuilding.
So we know about some successes but we cannot attribute them reliably to a training strategy, because it may be the result of genes or medication or something else. And bodybuilders have incentives to lie: it is more honorable to get big muscles through pain and hard work than through a lot of medication or due to favorable gene combination. Most of them probably even do not lie consciously, they fool themselves while doing their "research" what works and what doesn't.
So I believe that all these millions of bodybuilders are not a reliable global maximizer. Just looking at data generated by them we cannot judge if they had found a global maximum or something else. Though they probably have found some local maximum, but we do not how good it is for a general population.