Here's an fun anecdote from a biography of Fermi:

“When Dyson met with him in 1953, Fermi welcomed him politely, but he quickly put aside the graphs he was being shown indicating agreement between theory and experiment. His verdict, as Dyson remembered, was 'There are two ways of doing calculations in theoretical physics. One way, and this is the way I prefer, is to have a clear physical picture of the process you are calculating. The other way is to have a precise and self-consistent mathematical formalism. You have neither.' When a stunned Dyson tried to counter by emphasizing the agreement between experiment and the calculations, Fermi asked him how many free parameters he had used to obtain the fit. Smiling after being told 'Four,' Fermi remarked, 'I remember my old friend Johnny von Neumann used to say, with four parameters I can fit an elephant, and with five I can make him wiggle his trunk.' There was little to add.”

This implies the observations should be accounted for parsimoniously by the least possible number of variables. But there is a limit to this, and we find this repeatedly expressed across our use of science vis-a-vis the decisions of our ordinary life: what if we have a parsimonious theory that accounts for the better part of experience, but leaves out things that the individual regards as crucial to their life? And any theory that accounts for the nuance would be so complex that there is no reduction of manifold forms of evidence into a few general principles.

In this latter scenario, which I think almost everyone has experienced, suggests that the "best" science is often inadequate to human purposes. This may be the "best" science in terms of predicting the greatest number of outcomes with the smallest number of variables or principles, but they are not "best" at everything. There are a great many ways that this could be framed, but you probably get what I'm driving at.