Race, Face, and the Uncanny Valley

Dr. J has been riffing on the phenomenon of the ‘uncanny valley’ lately. ‘ For those who don’t know, the ‘uncanny valley‘  is a theory for why we find some human-like artificial things so weird. My favorite example is dolls, as this image suggests, but masks are also good examples and ultimately Masahiro Mori was concerned with robots and androids for reasons that are closely connected to the Japanese fascination with robotics. The question is: what is it about near-human faces and movements that feels so wrong and abject to us? Why isn’t our comfort level linear?

Dr. J writes that:

The non-human/human divide must intersect at the point of the human corpse, I think, which is the point at which whatever revulsion we feel at the “simulation” of the human– my hypothesis is that this revulsion is motivated by our aversion to “deception”– begins to be replaced by an affectively similar, but categorically different, kind of revulsion, i.e., the kind of revulsion that is motivated by our aversion to morbidity and mortality.

So if I’m reading her correctly, just as we begin to worry that we are being deceived, we transition to a set of concerns around our own mortality that are inspired by viewing a human corpse. But that doesn’t seem born out by the principle distinction between animated and unmoving uncanniness, since there’s much more revulsion at seeing an animated near-human robot than in seeing a still one. (Unless we truly do have an abiding fear of zombies….) That said, Dr. J is much more interested in the way that the ‘uncanny valley’ can help us to think about race and racism, so I think there’s room to tinker with the explanation for why near-human simulations so trouble us.

Here’s my take on the ‘uncanny valley:’ human brains are extremely good at faces. We’re so good at faces, we even see them when they’re not really there. Take the ‘hollow mask’ illusion:

This isn’t a trick video, there’s no special effects here. That’s your brain’s highly tuned facial recognition doing its work. Unless you’re drunk, high, or schizophrenic, you’ll be unable to persuade your eyes that they are seeing a concave face moving from right to left… as the exterior of the mask leaves your view, it will look as if you are seeing a concave face is moving from left to right, instead of the convex side of the mask continuing from right to left. Apparently, the reason schizophrenics and some drunks don’t see the illusion is that the part of their brain that specializes in faces is either permanently damaged or temporarily impaired.

A piece of cheese on toast purportedly showing the Virgin MaryThe way I interpret the ‘uncanny valley’ is as a challenge to our facial recognition faculties. Your brain is working overtime to resolve the image as a human face, but in so doing it’s coming up against an obstacle that it cannot surmount. The brain is so highly attuned to human faces that it will sometimes create false positives, such as the Man in the Moon or the the Virgin Mary in a grilled cheese! In the process of cognitive dissonance that emerges when we can’t easily distinguish a real face from an image, we experience a sense of vertigo.You can provoke a very similar feeling by looking at a loved-one’s face upside-down: at first, their hair line resolves as a smile, but then you start to look at this familiar face in terms of its parts and, well, it’s weird and uncomfortable. The materiality of the face confronts you with your own materiality, and with the artifice that animates even our quotidian interactions. I don’t think this is directly related to the fear of death, but rather to the nausea of ambiguity, in which death is less threatening than meaninglessness.

Fanon described the Syisyphan existence of non-whites in a white-supremacist society as a project of “black skins donning white masks.” Similarly, Bhabha explains the strategy of racial (and racist) discourse as one of ironic farce– in which whites demand of non-whites the impossible task: “Be like me, but don’t be just like me”– such that the life of non-whites becomes defined by and obsessed with an always-imperfect mimicry. The point here is that “whiteness” is set as a universal norm to which everyone is accountable and expected to aspire, only the very idea of “whiteness” requires that not everyone be able to achieve it. Like the uncanny valley, whites accept non-whites the closer and closer they approximate whiteness– up to a point— but when non-whites get close enough in their mimicry to “pass” as “really” white, the integrity and truth of “whiteness” becomes threatened and the mimicry is violently rejected.

As a metaphor, I think this is a spot-on use of the ‘uncanny valley.’ The uncanniness of ambiguous race (or gender or sexuality, for that matter) strikes me to be very similar to the uncanny failure to fully visualize a face when confronted with a robot or mask. But it’s not the same, for the obvious reason that those of ambiguous race or gender are just as human as anyone else. But if it doesn’t feel that way to the majority of viewers, then it will be tempting to ignore the fact of humanity in favor of the feeling of alienation. So much of our discourse on human rights is rooted in a kind of pre-rational indignation at the mistreatment of fellow human beings: when we fail to see the sufferer as a fellow human, the indignity is lost. We might still recognize that the sufferer has rights, but only in a reflective or calculative way. The affective dimension of the rights are lost. That’s why the hooding of the Guantanamo prisoners was so troubling: they weren’t just abused, but every available image of them was designed to foreclose empathy. This might also help to explain why a drunk or high person, having impaired the part of their brain that engages in facial recognition, might also lose track of the empathy for a fellow human’s suffering and so act cruelly or violently.

One other implication of this might be tied to the various attempts to make sense of the “Cross-Race Effect” or “Other-Race Effect.” People of all races tend to be better at recognizing individual faces from members of their own race, while they are better at picking out members of others races when asked to seek the other-race face among a group of own-race photographs. This explains the somewhat prevalent attitude that members of other races ‘all look alike’: we lack practice in easily distinguishing faces outside of our own ethnic phenotype, and when exposed to other-race faces we concentrate on broad ethnic differences rather than the minor variations that allow us to identify individuals.

We know this is a learned behavior because the ‘own-race bias’ has been shown to transfer to the race of adopted parents in interracial adoptions: African-American children adopted by Caucasians share their parents ‘own-race bias’ and more easily identify individuals whites. So if facial recognition is the key to empathy, and most people are better at identifying members of their own race, we have strong evidence of a kind of subconscious racism that doesn’t have an ‘intentional’ angle. We now know that certain kinds of perceptual training can be used to eliminate the facial recognition bias, which is especially important for police since without such training there’s increased risk in using other-race eye-witness testimony in criminal courts. Interestingly, this training in other-race individuation cues has the side-effect of reducing implicit racial bias, specifically the time it takes for whites to associate positive words with black faces. It’s easier to think of Martin Luther King, Jr. or Nelson Mandela as a “leader” after training to recognize individual faces among African-Americans. Implicit Association Tests show that a white person would still come to those conclusions, but it would take longer. Something about learning the specific cues needed to tell members of another race apart also enhances our capacity to think highly of them!

This takes us back to the ‘uncanny valley.’ It seems to me that the biggest problem with masks and dolls is that they are mass-produced. They all look the same, and individuality doesn’t matter much. As they grow closer to human-simulations, however, they transfer that mass-produced anonymity into a space where identity and individuality are of paramount importance to the facial recognition part of our brains. The uncanniness of this human-like simulacrum is its resistance to individuation, even as it presents itself to me as a potentially individual being who will deserve my sympathy and concern.

Is there an analogy in race, here? Dr. J proposes Michael Jackson as an example, but though the rhetoric around his skin color and nose job was racialized, I think he is actually an example of our general distaste for bad plastic surgery. That said, within the framework of own-race individuation bias, one reason that bad plastic surgery might be difficult for us to stomach is the way that it mixes our facial (and racial) taxonomies. If I can’t find the individual markers in a face because I’m forced to bridge my taxonomy for dealing with black and white faces, for instance, I may experience that uncanny ambiguity in the face of the struggle. This interpretation is either challenged or bolstered by several studies that show that mixed race faces tend to be more attractive to members of either race, even when they tend to rate members of their own race more attractive beforehand. (This finding suffers a little bit from a false assumption, as it pretends that races can be unadulterated or pure before they are ‘mixed.’ Still, insofar as it works with viewer-generated racial attributions, it may be salvageable.)

This also raises questions about mimicry and symmetry, since one of the major components of facial beauty is apparently left-right symmetry which is also alleged to identify genetic ‘health,’ and seems to be one of the major components of mixed race beauty. The very mirroring or twinning that troubles us in masks and robots is the wellspring of facial beauty, as well! If you’re thinking of the original uncanny story of the doppelgänger that Freud interpreted to generate his own theories of unheimlichkeit (not-at-home-ness) then I’m right there with you.

In general, I want to maintain that when we think about faces we’re thinking about individuation, and that this individuation is more important that species membership (humanness) or race/gender/sexuality membership, but that those groupings form phenomenological filters that mediate our attempts to individuate others and become individuated ourselves. Ambiguities, whether they be in sexuality, gender, race, or species, threaten us with the artificial and ephemeral grounds for individuation. Each of us can only become an ‘I’ in contrast with some ‘We,’ which is the group from which I prise my individuality, and in whose idiom I express it. For evolutionary biological reasons it makes sense to assume that the human brain is built so as to work best among a small troupe of individuals, between 100-200 people, with whom we share many facial traits. But these groups in which we claim membership are all of them artificial, and the most uncanny valley is the one that separates each ‘We’ from some ‘Them.’ If so, then Dr. J is right: anti-racist causes can be assisted by bridging that valley, but we can’t simply declare ourselves race-blind in order to do it. To make progress in this arena, we’ll first have to work quite hard at seeing and understanding race, first in faces, but eventually in discourses, institutions, and experiences, too.

I leave you with the first in a series of short lectures on the ‘uncanny valley’ by Karl MacDorman:





7 responses to “Race, Face, and the Uncanny Valley”

  1. Dr. J Avatar

    AnPan, thanks for picking up this theme… which I am absolutely obsessed with right now and which has caused me to exhaust the patience of all of my proximate interlocutors! You've got a lot here.

    First, let me remark on this (from you):

    So if I’m reading her correctly, just as we begin to worry that we are being deceived, we transition to a set of concerns around our own mortality that are inspired by viewing a human corpse. But that doesn’t seem born out by the principle distinction between animated and unmoving uncanniness, since there’s much more revulsion at seeing an animated near-human robot than in seeing a still one. (Unless we truly do have an abiding fear of zombies….)

    If what we are talking about it some kind of implicit/non-reflective measurement of "humanness," which is what I suggested in my original post, then the bottom of the uncanny valley will be occupied by the things that seem the most "uncannily" "human" to us. If there IS a non-human/human divide intersecting the bottom of the valley, then that means that the MOST "lifelike" (in terms of both appearance and animation) simulation will be the creepiest possible thing on the "nonhuman" side of the bottom of the valley. Something like an almost-perfect android. Despite all of its movement and characteristics, we will still want to say something like, "but it's not REALLY human." So what's just on the other side of that divide? That is, what is the thing about which we say "yes, it's REALLY human, but an almost infinitely degraded version of the human, only barely what we mean when we say 'human.'"?

    I think that would be a corpse. Non-animated, non-moving, non-breathing… but also NOT a "simulation."

    So the "prinicple distinction between animated and unmoving uncanniness" that you reference is still in play here. Yes, we're more revulsed by an animated, non-human robot than a still one, because the more "life-like" movement that a robot simulates, the more danger we are in of being deceived by it. But on the other side of the valley (the side leading back "up" to the "real" human), the opposite is true. There, more movement equals more "life" (as opposed to more "life-like") and hence the LESS a "real" human exhibits this kind of movement (or exhibits it artifically, as we might say would be the case with artificial limbs), the less "fully" human it seems. So, THAT'S why I think the still corpse is at the bottom of the valley, of which we will say "yes, that's a human, but without almost all of the things that make a human "human," namely, the movements and expressions of LIFE." But the "uncanniness" of the corpse is qualitatively different that the uncanniness of the very-lifelike robot, because we have changed the standards of judgment when we traversed the bottom of the valley. It's like the difference between how we might judge a really good Beatles cover-band and how we might judge the (real) Beatles on a really bad music day. If the standards of those judgments were shared, then we would have to say that the cover-band would win… but they aren't shared. There's a fundamental difference between how we think about images and how we think about the real.

    More later…

  2. Dr. J Avatar

    Re: individuation

    I think you're right that one of the things we're concerned about in our aversion to things in the uncanny valley is individuation… but not for the same reasons that you cite. That is, I agree with you that one of the things that makes a human being "human" is its singularity (which is why we find twins kind of creepy, like Freud said), but this possiblity is not prohibited to masks or dolls or androids. (You can make a one-of-a-kind doll, too.) So, although I think we are looking for "REAL humanness" and one of our ways of judging that is by individuating (or recognizing individuation), I still think that when we are almost-fooled by a doll/robot/android/whatever, that what inspires our aversion to it is that we were almost fooled. Our aversion is to deception, not because it's not possible that this particular doll is individuated, but because once we know it is a doll (a simulation, a product of techne, a manufactured thing) then we also know (as you noted) that it could be mass produced, that is, it could "fool" us into thinking that its individuality is really singular. Which, of course, it isn't.

    To that end, I think our aversion to (good or bad) plastic surgery is motivated by the same aversion to deception. The "plastic" in "plastic surgery" indicates that we are contaminating the realms of physis and techne, that we are in danger of taking the unnatural for the natural, the fake for the real, the strange for the familiar.

  3. Joshua Avatar

    Just a quick note because I see that we are cross-posting:

    I don't think humanness can be the foundational category of either advantage or responsibility. Species-membership strikes me as clearly a synthetic judgment. Instead, the foundational experience would likely be the face. Not in the transcendental way that Levinas describes it, but in a neurologically pre-recognized way: human experience is structured to seek and imprint faces, which is why that's the first thing infants learn to do.

    So the pre-rational line shouldn't be a human-inhuman differentiation, but a recognizable-unrecognizable differentiation, with additional cognitive dissonance when a face is liminal and almost-but-not-quite recognizable.

    I'd also ask whether it makes sense to talk of deception-aversion in something pre-rational as well. That's why I tilt towards cognitive dissonance as an explanation: I only worry about deception from someone who is already a candidate for in-group membership. Bridging this uncanny valley seems to be a prerequisite for that candidacy.

  4. Joshua Avatar

    One last thing: the 'almost fooled' judgment would be reflective. I have to notice myself 'falling for it' and then pull back, and generate disgust at both myself and deceiver. But if the doll or the android disgusts me it does so pre-reflectively. So I can't be castigating the doll for fooling me because I was never fooled and never in danger of being fooled.

  5. Brunson Avatar

    "Belief and disbelief differ from uncertainty in that both provide information that can subsequently inform behavior and emotion. The mechanism underlying this difference appears to involve the anterior cingulate cortex (ACC) and the caudate. While many areas of higher cognition are likely involved in assessing the truth-value of linguistic propositions, the final acceptance of a statement as “true,” or its rejection as “false,” seems to rely on more primitive, hedonic processing in the medial prefrontal cortex and the anterior insula. Truth may be beauty, and beauty truth, in more than a metaphorical sense, and false propositions might actually disgust us."


  6. […] ought to be rooted out. Genetic identity is a stupid thing to object to, rooted in all our worst doppelganger fears. There’s nothing a priori immoral about cloning, and certainly there’s nothing a priori […]

  7. […] a basic picture of what and where various parts of an empathic response occur. (I’ve tried to work on the problem of cross-race implicit bias before, but I’ve learned a lot more of the science since then.) There is research that suggests […]

Second Opinions