Inevitably, when we find ourselves swimming in a sea of complex data, one begins to worry about fundamental limits of knowledge. But in many applied fields, important clues to fundamental physics often emerge from complex data, particularly when complex physical processes are at work (cf. beam-line experiments in particle physics). In fields where we are far from understanding key physical processes, as in the study of galaxy formation and evolution, this effort must be worth it.
Astronomers now recognize this: every large survey has unveiled new lines of enquiry or revealed something important about our environment. We talk in terms of data mining, virtual observatories, and so forth. Furthermore, the numerical simulators push down to ever decreasing scales, and work to include new algorithms that capture an important process.
At what stage do we declare that galaxy formation is basically understood? Such a declaration becomes possible when one is able to reproduce the salient features of galaxies today, in a host of different environments, with a theory that is firmly rooted (presumably) in ΛCDM. This same axiomatic theory should be able to reproduce observations of galaxies at different epochs out to high redshift, until we reach an epoch where objects no longer look like modern day galaxies. The HUDF indicates that this appears to happen at about z ≈ 2.
A moderately complete theory of galaxy formation must also tell us whether the Galaxy is typical or unusual in any way. It is often stated that if the Galaxy is pathological, it is hardly worthy of the attention we give it. But in fact, recent observations by R.B. Tully and collaborators suggest that "Local Group" collections of galaxies are common throughout the local universe (see Fig. 1). Therefore, it would be a surprise to discover decades from now that our large-scale surveys were seriously misleading us in our quest to understand galaxy formation and evolution.