A Method to the Duality:
While it might seem to make more sense to go either one way or the other in a game--either wholly full-detail, or wholly low-resoltion--in truth, both styles of animation serve their purpose, and work well to contrast each other to provide a good balance of impressive visuals and interactivity. The main thing that divides them is that FMVs are made to be watched, to tell a story, while playable models are meant to be interacted with; this requires different kinds of modeling techniques and different levels of detail to achieve each purpose.
Playable Models: Consistency in Real-Time Rendering
The reason that the low-poly models are used in the first place is to preserve smooth frame-by-frame display during gameplay; many gamers don't realize that their gaming console actually has a rendering engine that is constantly creating visual output on a frame-by-frame basis to create the illusion of motion and an expansive, continuous world map. In essence, a gaming console is a mini-computer, but with only one purpose and function; if you think of it that way, and then think of any PC games that you've played where the effort of constantly following your motions and adjusting rendered output in real-time makes the game (or your computer) stall, the need for low-polygon, low-detail models might make more sense. While your gaming consoles will probably stall less than your PC because there are no other processes running and all of their processing power is devoted solely to rendering output, they still wouldn't be able to deliver smooth, real-time gameplay using high-resolution models.
Of course, with the constant improvement of next-gen consoles and development of more advanced modeling and texturing techniques, playable low-poly models grow more and more detailed, but FMVs have also improved at the same rate. In some cases, the playable models of next-gen consoles even surpass the FMV models of the older consoles; part of this is because next-gen consoles pack more power and more efficient rendering engines, but the playable models themselves are also created using more efficient techniques that simulate fully-detailed modeling more seamlessly.
FMVs: Non-Interactive Storytelling Through Video Playback
The reason that FMVs can be so detailed, with better audio and special effects, is because they aren't actually being rendered. Rendering is only necessary for interactive parts of games; for non-interactive areas meant to tell a story or highlight dramatic scenes, the action is pre-rendered and recorded as a movie file that the gaming console merely plays back at the appropriate time. It's like popping a DVD in; you have no control over the action of the FMV.
Not all games employ FMVs for storytelling or other dramatic purposes, but many do. The FMVs give players a chance to see the characters in full detail, and thus get a better picture of the virtual "people" that the playable models represent; also, for many, it's the stunning visual composition of the FMV scenes that lure them to buy a game, and keep them playing in anticipation of each emotionally-charged, breathtakingly beautiful scene. In addition, FMV sequences can often give animators an outlet to display their talents more fully, and perhaps even show a measure of the emotional involvement that goes into creating such detailed, intricate games with plot-rich storylines.
It's said that there's a time and place for everything, and that applies even to types of animation and modeling used in gaming. It's possible that one day a gaming console will be created that can render playable models so quickly and accurately that they'll be indistinguishable from FMV models, but until then both will always have their place in the world of video games.
