I have to agree with Dennis to a large extent here. But I think we're arguing TOOLS rather than ART, and that's the wrong argument to be having.
CGI is one "brush" in the artist's pallet, while models are another. Hell, cell animation is another (aka "rotoscoping").
Basically, all of these are there for use where and when they do the job best. Each has advantages, and yes, each has disadvantages. It's up to the artist to determine which tool is right for which job.
There are times that CGI gets used where it's not the best tool for the job (but it's the CHEAPEST or EASIEST tool) and in those cases, we often find ourselves noticing "hokey CGI." As I said earlier, this is mainly an issue with chaotic phenomena... smoke, flame, water, fine particles... and current-day work with those usually involves "fudging" the features by the use of filmed live-action elements. If you try to do those with pure CGI, it's almost always "fake" to the naked eye. Which is NOT to say that this will always be the case... the tools are always improving. But so far, computers just don't do chaos very well.
On the other hand, well-ordered systems (like, say, spacecraft flying in space) are VERY well-suited towards CGI. You can model things that would be difficult if not impossible to do in a real model (such as having a full interior, properly lit, with real little people inside!). You can move the camera without having to worry about avoiding "mounting stand points" on the physical model or lighting cables or whatever. You can adjust the camera to be, effectively, in the same scale as the model (meaning that the model will "photograph" as though it were full-sized). You can totally eliminate the issue of atmospheric scattering that is UNAVOIDABLE with real models (unless you shoot your models in a vacuum chamber!). You can make your lighting be like the real lighting conditions in space would be (even if that would make audiences say "that looks fake").
Basically, with the exception of certain damage scenarios involving chaotic phenomena (the original Death Star explosion was far better than the later CGI-ed one, for example), there's not much justification for spacecraft modeling to be done any other way but CGI these days.
The trick is to realize that the CGI is just the brush. The artist should still be trying to accomplish the exact same thing either way. If someone makes something dramatically different in CGI than they would have done with a model, "just because I can"... that's an ideal example of "bad art," I think.
The idea shouldn't be altered just because the tool lets you... the tool should be used to best approximate the original idea.
Otherwise, you get annoying "kewl CGI things" like funky organic shapes that are totally impractical (but quite easy to model in CGI), abusive levels of use of things like lens-flares or "shakeycam" stuff... things that don't add to the presentation, but are just there because the artist said "I CAN, SO I WILL."
Make sense?