Sean Carroll in the New York Times:
Nothing makes scientists happier than an experimental result that completely contradicts a widely accepted theory. The scientists who first invented the theory might not be tickled, but their colleagues will be overjoyed. Science progresses when a good theory is superseded by an even better theory, and the most direct route to building a better theory is to be confronted by data that simply don’t fit the old one.
Nature is not always so kind, however. Fields like particle physics and cosmology sometimes include good theories that fit all the data but nevertheless seem unsatisfying to us. The Hot Big Bang model, for example, which posits that the early universe was hot, dense, and rapidly expanding, is an excellent fit to cosmological data. But it starts by assuming that the distribution of matter began in an incredibly smooth configuration, distributed nearly homogeneously through space. That state of affairs appears to be extremely unnatural. Of all the ways matter could have been distributed, the overwhelming majority are wildly lumpy, with dramatically different densities from place to place. The initial conditions of the universe seem uncanny, or “finely tuned,” not at all as if they were set at random.
Faced with theories that fit all the data but seem unnatural, one can certainly shrug and say, “Maybe that’s just the way it is.” But most physicists take the attitude that almost none of our current models are exactly correct; our best ideas are still approximations to the underlying reality. In that case, apparent fine-tunings can be taken as potential clues that might prod us into building better theories.