Think about how a game treats a block of stone right now. It's a hollow cube with textures pasted on it. It's bounding box is pretty easy to calculate, but a complex object that doesn't match it's own bounding box closely will still clip right through it.
Now, imagine if we replaced that single block of stone with a matrix of tiny blocks. Also, imagine that we've made the bounding box of the complex object fit its geometry much more closely. If the same motion that caused clipping before is carried out, we'll detect it. It's going to take a lot more computation, but we will detect it.
Not only will we detect collisions that we wouldn't have before, we'll be able to handle them in a variety of gameplay enhancing ways. If the stone is meant to be strong stuff, we make it's mass high and the bonds between sub-blocks strong so that the complex object will come to a jarring halt when it makes contact instead of passing into the stone. If the stone is meant to be softer, we dial down the bond strength between the sub-blocks a bit so that they can be broken. The object might dig in a bit and leave a noticeable gouge as some of the sub-blocks are separated from the main block. If the stone is meant to be more like dirt, we make the bond strength very low so that a blow will cause a chain-reaction that will cause the whole block to tumble apart. We can change the probability of blocks sticking together to simulate either wet, clumpy earth or dust that's ready to fly off in all directions when it's touched. Deforming the complex object is also an option. It could be similarly be composed of sub-blocks, but with tweakable levels of stickiness/stretchiness holding them together that would allow us to capture things like flexing or bending.
In short, we really oversimplify the physical nature of objects in video games at present. This greatly reduces computation, but it makes the world seem unrealistic in how objects interact. By splitting things into smaller chunks that more closely resemble real matter we can start to simulate real materials, but at the cost of greater computation that isn't devoted directly towards making things pretty. Prying resources away from the graphics engine has long been something developers just don't want to do.
Voxels, pretty much? Like minecraft, but 10 or 1,000 or 1,000,000 times smaller than the objects/characters in the world.
It might look pretty blocky at lower resolutions, but considering the enduring appeal of "pixel art" (and minecraft's success too), it's kinda surprising it hasn't been done yet. i.e. a voxel physics world at a similar resolution to old 8-bit games (for example, 8x8x8 voxel characters).
If voxel physics can be done largely locally (with long-range effects being a slow ripple), it could be a killer app for massively multicore devices.
So, that's cool and all, but we aren't making simulations--we're making games.
Nobody gives a hoot in hell how realistic the drapes on a window are if the game isn't fun.
Many games have gone for very good realism, and it hasn't really worked out so well.
Doom 3 had per-poly hit detection, and all it did was make many weapons feel less accurate and multiplayer more annoying.
ArmA2, one of the most detailed military sims out there, did very accurate character-world collision detection and as a side-effect actually moving around inside buildings became quite hard. A community mod was created to address that and restore the funner, less realistic behavior.
Red Faction Guerilla tried to make more realistic destructible physics, but it mostly just made more trouble for level designers, because things they built tended to fall apart.
Jarring details do spoil the fun a little. I'm fine with lack of realism. I am not fine with noticeable differences in realism.
My last example comes from Rembember Me, which I have played on the PS3. An excellent game, with very good graphics… except for self shadows: they emphasise the underlying vertices. I'd rather have no shadow and a smooth face, than a "realistic" shadow and a blocky face. Other titles who feature self shadow display a horrible dithering (I have seen this in Mass Effect).
To date, the best consistency I have ever seen comes from Dragon Quest on the PS2. It's not realist at all (its style is Japanese motion picture), but there's next to no mistakes such as clipping. Even more recent titles such as Ni no Kuni aren't as consistent.
I would guess that it would imply that meshes are no longer static, and that they could bend, stretch, and fold based on their properties. In some ways, we already get this in modern games that try their best to simulate cloth in flags and such, but it's mostly still smoke and mirrors.
What does this mean? Like, what do you envision this changing?