The short answer to the title question is: Mainly because of the rendering engine used and how it is programmed.
However, it also depends on the 3D graphics API (
see a list here) used by the renderer and like mentioned above, how that library is used by the programmer/s. Another big dependency as to why they differ from real ones is that a lot of it will be decided on how the programmer or graphic designer works with the software (3D computer graphics software). That software can be a game engine, a 3D modeling software, a CAD software, or an exclusive 3D rendering software.
The reason it is this way is because someone programs an application that communicates with a rendering library, that rendering library communicates with a driver and then that driver communicates with the GPU (Graphics Processing Unit). This is a very lengthy subject. If you are interested, the keywords you are looking for is: Rendering Pipeline. Here is a very basic slide in which you could maybe find more keywords to search and study:
https://www.ics.uci.edu/~majumder/VC/classes/geomgraphics.pdf
The subject itself is complex. Not for being lazy, but I can tell you that if you read a book of Mathematics for Game Developers you will see part of the whole mathematic world behind it as well as the plethora of ways to approach 3D rendering. It gets complex pretty fast and some things you simply cannot understand unless you are well versed in differential equations (referring specifically to animations and real time rendering). For example, one filtering in texture mapping make the rendering hardware examine some partial derivatives of interpolated components of a texture and it does so with a given differential equation. Yet this is only one method of texture mapping. It would be the programmer/s of the rendering engine, the users of the rendering engine, or the designer/s, the ones to choose which method to use for mapping and that will have different results in the final rendered image.
And so on goes the list of variables that affect the rendering of objects.
All in all, the answer is that there are many variables that affect the rendering.
If you ask for my opinion, which you did not, I would tell you that based on my understanding and knowledge, we have the mathematics and tools required to render resembling enough faces to a level in which they cannot be differentiated. What we don't have for the public (you and me) is the widespread access to the work of individuals with intellect enough to use that mathematic and tools to create or help create something that astonishes us and makes us think it is real. Rendering something like that would require a huge human effort and intellect; and things that require huge human efforts and intellect are scarce.
rollete said:
Hardware is a limitation in real-time rendering applications. Consoles and PCs still lack the horse power to render truly life-like CGI on the spot. There's also the fact that faces are particularly tricky objects to model and animate realistically.
I beg to partially differ on the first sentence. I think the mathematics and tools are more than good enough. What is not good enough is the effort put into making the software and the effort of the designer of the models like I mentioned above. On the second sentence I agree, real time rendering of highly realistic objects with and ordinary user computer at the moment of writing this is very hard because it lacks the resources.
In your third sentence I also agree, that is a fact and variable to take into consideration when looking at OP's question. It is tricky and like I mentioned above, requires huge human efforts which are scarce.
rollete said:
The best CGI is found in pre-rendered content--any type of static or motion picture.
True. Pre-rendered content contains more detail that would be hard or would require more processing resources than an ordinary individual's computer would have available for real time rendering. Hence, looking better to the human eye. That's why some game engines give you the option to pre-render content and pack it in the game which increases the size of the game, but gives more realistic graphics.
rollete said:
I think both of them are CGI. The first one has a logo, the second one's scene was not rendered real enough for my eyes and the lightning shades of the neck are not on par with the amount of light striking the lower eye and nose.
Had the first one not had a logo I could have mistaken it by real. Yet a closer inspection shows lightning in the neck muscles that give away the mesh construction.
