This article examines the intent of a substantial amount of AAA games to craft a visual presentation that emulates realism as faithfully as possible. But before we can get to the core of the issue, we have to make a detour through the movie industry to ease us into it, and more specifically, the introduction of computer-generated imagery (CGI).
Before movies are released, their CGI aspects are often advertised and thus receive very specific attention. This foreknowledge alone can be enough to prevent any disbelief from being suspended. There are movies that are entirely CGI, but this level of attention is more apparent when a movie combines CGI with live-action. A recent example is Jeff Bridges' dual role in Tron: Legacy. He plays himself as an older Kevin Flynn, and his appearance in the original Tron is digitally recreated as his counterpart CLU 2.0 (who looks like his younger self). Many people were put off by CLU 2.0's representation, calling it "appalling", "eerie" and "creepy". A movie which suffered similar criticism is The Polar Express, which has become the poster movie for misfired results in attempted CGI realism.
Conversely, the amount of people aware of any CGI in The Social Network is far less. The Winklevoss twins were played by a single actor who simultaneously portrayed both brothers, and their presence in a single shot was accomplished through seamless editing and post-production superimposition. While the conditions are different (it's not an entirely computer-generated character), the context is. Hardly anybody knew that any CGI was involved, so very little attention was paid. CGI is at its best when it augments, not replaces.
If we translate that line of reasoning to videogames, we find that they are, purely by definition and semantics, always CGI (live-action cutscenes notwithstanding). Hence, the visual presentation of a videogame plays a huge role in its development, especially when looking at AAA titles. Development diaries speak of "real-time dynamic deferred global illumination lighting" and "diffuse specular character animation shading cubemaps" which try to create a world and populate it with characters that adhere as closely to reality as possible. In doing so, the visuals expose themselves to a far-reaching level of scrutiny. A very fitting example here is L.A. Noire.
It might be premature to judge L.A. Noire's visual presentation when the game hasn't been released yet, but it's a fitting case study by virtue of the excess of information on and considerable response to the MotionScan technology that renders its characters. In a nutshell, MotionScan uses an array of cameras to capture every aspect and angle of an actor speaking his lines and then compiles this data into a 3D facial expression system (an early version of this technology has been applied to me). This is currently the closest we are to translating the entire gamut of acting into a game within a single process (only the motion capture is still handled separately).
The characters in L.A. Noire look very lifelike as a result, but something about them seems off in much the same way as CLU 2.0's appearance raises eyebrows. To me, they look like what they amount to on a technical level: shop window dummies with .gif files taped to their heads. Their eyes appear dead and their mouths move like they're chewing honey. L.A. Noire's saving grace is that this is not reality purely for reality's sake. Being able to sniff out liars based on subtle expressions and gestures is a gameplay mechanic, which is why the characters require detailed facial expressions to make every nuance readily visible. But therein lies another problem.
To make sure that as many players as possible will be able to catch the hints and clues that are betrayed by suspects during interrogation scenes, they are made explicit by the actors. It's like they're following the handbook on how to detect lies by making every tell obvious (like shifty eyes or a tightening mouth). This isn't just the case for L.A. Noire. This article on Uncharted 2: Among Thieves, a game praised for its production values in creating believable character visuals, concludes that those characters are "all animated, excitable, overworking their eyebrows and emoting all over the place". In essence, the digital actors overstate themselves to make even the tiniest conveyance of emotion stand out.
Ironically, the flipside is that lip synchronisation seems to go completely the other way. It's often understated and barely visible. This scene from Splinter Cell: Conviction is a good example. Sam Fisher puts emphasis on each word when he threatens Anna, yet his mouth simply opens and closes like a fish bobbing for food. Shout "I! WILL! KILL! YOU!" in front of a mirror and see how your mouth really moves along. Lip synchronisation is still very much hit-or-miss in games, in part because it's something we tend to pay close attention to.
I have no doubt that games will one day acquire the visual acuity to render a fully believable human being without so much as dipping a toe in the uncanny valley. The evolution that graphics have undergone in the last ten years alone is enough of an indication. But this lofty goal might sway developers from a different approach that provides more unique modes of expression, and that approach is distinctly stylised characters and visuals. It's not simply a way of avoiding the uncanny valley altogether by telling the player that realistic visuals is not what the developers are trying to achieve. Usually the extent to which characters are stylised is just the extent to which the art team weren't able to make them realistic.
Many articles on games try to pinpoint the specific aspects that separate them from and elevate them above other media. Games indeed possess unique capabilities to express themselves, and one of those is the way in which the world can be represented as it is seen through the eyes of the character that you play (through intentionally stylised visuals). This point is communicated brilliantly in this analysis of Hitman: Blood Money (from which the phrase that ended the previous paragraph was gleaned).
A good example is the change in appearance of the Little Sisters between the two BioShock games. In BioShock, they purposefully look creepy, wrong, and almost alien with their oversized eyes and drooping faces. This is because the protagonist, Jack, is uncomfortable around them (who wouldn't be around a girl who's got glowing eyes and stabs corpses). But BioShock 2 makes them seem more endearing and human, because the protagonist is now a Big Daddy, who is meant to feel paternal and care for them.
On their own, intensely realistic visuals are not enough to load characters with empathy. For instance, many gamers have attested that Limbo affected them on many levels, despite (or because of) its minimalistic graphics and vague context. The subject matter of a game naturally plays a role in the decision to stylise visuals. Heavy Rain wouldn't work as a cartoon, but Team Fortress 2 wouldn't work as a gritty war simulation (incidentally, it was initially planned as such). Stylised visuals also work in movies (see any Pixar movie), but in games, they have the capability to do so much more.