Death… It's A Living

At the end of a previous post, I wrote that familiarity is a key component of beauty. This likely makes some sense to even those of us who don’t think a lot about what makes someone or something attractive. Nevertheless, almost all of us have the experience of having something grow on us. I didn’t care for coffee or beer the first time I had them, but they’ve grown on me (perhaps unfortunately). I thought I hated Bowie’s Low the first time I heard it, but now I think it’s insane when I don’t see it listed among the greatest albums of all time. I even confess that I don’t recall having any warm or fuzzy feelings when I first met the girl who would become my wife. Rather, it wasn’t until a few months of living on the same dorm floor, after she told me she thought it was cool I listened to the Who while studying calculus that I thought “Am I nuts, or is the girl really cute?” Fortunately, she grew on me.

However, I also wrote that familiarity matters to beauty even when seeing something the first time. For this statement to make any sense, I need to distinguish between two types of feelings of familiarity: objective and subjective. Something is objectively familiar when one has actual experience with something. When people use the word “familiar”, they are almost always referring to this type of familiarity. However, when people says that something “seems familiar”, they are referring subjective familiarity. In many of these cases, it feels as if we must have seen/heard/felt this thing before, we just cannot recall when or where. However, it is also possible for something to be subjectively familiar without any experience at all. Here, people experience an uncanny sensation likely caused by the memory’s inability to accurately reconstruct reality. This might occur most obviously when a person meets a new person who looks like (or behaves similar to) a person they know. They aren’t literally familiar, but they feel familiar.

On a computer, it is possible to create an image that will be judged as subjectively familiar through a process called “prototype abstraction”. A prototype is defined as the archetypal exemplar of a particular category; in other words, it is the best example of that particular category.

Prototypes exist in our minds without direct experience. The best way I know to demonstrate this fact is to imagine that you are playing a game of Pictionary. In this game, two teams take turns drawing cards with illustration prompts. When it is a particular team’s turn, one of their teammates (let’s imagine it’s you) draws a card. Now, you must create a doodle that resembles your prompt, and your teammates must correctly guess what your doodle is supposed to be before that plastic hourglass (minuteglass?) runs out of white sand.

You’ve picked up a card, and you’ve gotten lucky: It says “dog”. Although you are not much of an artist, you think you can draw a dog that can be guessed. The other team has flipped over the timer. Go!

Now, if you think about it, there are many directions in which you might artistically interpret “dog”. Walking through the park just today with my King Charles Spaniel, I can recall seeing a Great Dane, a beagle mix, a Yorkshire Terrier, a French bulldog, and three manic but happy corgis. All of these animals were dogs, but their range of appearance was wide to say the least.

But you probably won’t think about it, you’ll just draw. And even though I don’t know exactly what your dog doodle will look like, I have a pretty good idea. (I know because I have my students draw dogs in class for 30 seconds, with the only condition being that everyone in the class has to agree the artist’s intent was “dog”.) You will draw the same dog that my students always seem to draw. It will have a long snout. It will have floppy ears. It will have a wagging tail (maybe you even drew wag-motion lines). In other words, you drew something that looks like, not just a dog, but a labrador retriever.

This might seem odd that everyone should almost always draw the same type of dog when they could have drawn any other type. But people draw labs because they are the best examples of dogs. They are the most dog-like dogs. No one ever draws a pug. Even owners of pugs don’t draw pugs. Pugs are adorable in their special puggy way, but they are bad examples of dogs. Dogs are supposed to have long snouts and long tails and long ears. Pugs have none of these things. A pug looks more like the offspring of a pig and a cat. Pugs are not the prototype, whereas labradors are.

Prototypes are Familiar, Liked, and Beautiful

The reason for this likely has to do with frequency of observation. Not all of the dogs I see on my daily trip to the park have long ears, but I see those types of ears more often than pointy corgi ears. Not all of the dogs I see have long snouts, but I see more long snouts than I see the smooshy faces of Frenchies. I’m not even sure if I saw a labrador today, but if it were possible to average all the dogs I did see into one dog, it would probably look like a labrador.

That’s the best way to think about what a prototype is: not just an archetype, but the categorical average. There are a couple of strange things about prototypes. One, they tend to be judged as familiar, even when participants have never seen them. Second, they tend to be liked more. As discussed above, familiarity and liking are highly correlated. But they aren’t the same thing: It is of course possible to be familiar with something and not like it and vice versa. Nevertheless, prototypes tend to be liked. It may be no coincidence that the most popular dog breeds (retrievers, spaniels, mutts) tend to be prototypical. Third, prototypicality is an underlying component of beauty.

That last statement might seem odd because prototypicality is, as described above, a synonym for “averageness”. How could averageness equate to beauty? Isn’t beauty something that is not ordinary but extraordinary? Yes and no. As research has shown, faces that have the quality of averageness are commonly judged to be greater than average in terms of attractiveness. But what exactly does an average face look like? In the case of dogs, it is somewhat easy to picture a labrador as a very average dog, as a dog-like dog. But what is a face-like face? Excluding cases of facial trauma, don’t we all have face-like faces? As research has shown, some of us have more face-like faces than others.

Creating a Face-like Face

Now over 30 years ago, Judith Langlois and her research team used (from a the perspective of 2024) very rudimentary image manipulation software to create an averaged face. Conceptually, the process is straightforward. If you have ever played with Photoshop or a similar image creation/editing software, then you might know that when any image is converted to grayscale (i.e., a black-and-white image) each pixel that comprises the image holds a value from 0 (perfect black) to 255 (perfect white), with 254 shades of gray (way more kinky than 50) in between. Because each pixel holds a numeric value, that allowed the research team to literally average two images together (that is, adding the pixel values of the two images and dividing by two). When faces are averaged together, the result appears to be a new person who somewhat resembles its “parent” images. But what is most remarkable about this average face is that it tends to judged by others as more attractive than the two images used to create it. Furthermore, and perhaps even more strangely, the more faces that are added into the average (i.e., 4 faces, 8 faces, 16 faces, etc.), the more attractive the face becomes.

Note that the 64 face morph shown above strongly resembles the 32 face average that came before it. There is an obvious reason for this visual oddity, and its simple arithmetic. You might recall from middle school that an average is better representative of its dataset when the dataset is of a sufficient size. As the sample size approaches infinity, the average of the sample approaches the mean of the population (this is known in statistics as “central limit theorem”). Put in more layperson’s terms, the individual characteristics of one face—those features than make our faces our own—are averaged out of the sample when they are less common. For example, the 64-face morph shown above is a relatively attractive face, but it’s not particularly special. She (it?) has a medium-sized nose, medium-sized mouth, etc… there is nothing unique about her. An averaged face removes all personality; but it retains a strange sort of beauty.

Although Langlois et al.’s discovery was well publicized, it wasn’t immediately well-accepted. Most of the criticisms of this work seemed to be based on a misunderstanding. What the researchers had showed was that a facial average was rated as more attractive than most of the individual exemplars from which it was comprised. As more faces were added, the more attractive it became. Some early critics took that to mean Langlois was arguing that averageness was synonymous with beauty. But this clearly does not make sense: Although the 64-face morph might have been attractive, she was no Helen of Troy. In fact, Langlois never believed (and I should know, as she trained me during graduate school) that she had created the most beautiful face. She had only created the most face-like face. She had created the labrador retriever of faces.

In fact, she argued that averageness was merely one component of beauty. That said, she also argued it was the most important component of beauty: that it was the only component of beauty that was both necessary and sufficient. Here, I steal the analogy that she liked to use: faces are like cakes. Averageness is the sugar, butter, eggs, and flour. Without these ingredients, there is no cake. If we only use these ingredients, our cake will be still be delicious… but, as cakes go, it might it be pretty boring too. We can make this cake more delicious if we add frosting, but the presence of frosting is not necessary for a cake to be a cake. Frosting might be your favorite part of the cake, but if you just eat the frosting, you aren’t eating cake. Likewise, all beautiful faces begin with a baseline of prototypicality. We can enhance a prototypical face with make-up or a better haircut or fuller lips in women or a squarer jaw in men. But we can’t make it more face-like.

Now, as it would turn out, a face-like face is also judged to be very subjectively familiar. In fact, one of my favorite study anecdotes is this: As part of the Langlois research team nearly 2 decades ago, my fellow graduate students and I were in charge of obtaining many sets of facial ratings from undergraduate students in exchange for course credit. This typically involved sitting 20-30 students in a room and showing them hundreds of pictures, having them rate the faces for attractiveness, typicality, masculinity/femininity, likeability, you name it: whatever data the current study called for at the time. Many of the images shown were actually fellow undergrads who had volunteered to use their image for research. Although it was unlikely that a rater would be asked to judge a friend’s face at a school like the University of Texas (with nearly 60,000 undergrads at the time), because the faces were of real people it was possible. Furthermore, because people tend to rate friends differently than they rate strangers, we asked that our raters abstain from rating any face they recognized, leaving the scantron sheet blank and moving to the next row. At the end of the session, we collected the scantrons from everyone and thanked them for their participation. Because many of our research questions at the time concerned how people reacted to averaged faces vs. faces of real people, we would usually have a quick glance at the data for the averaged face. At least once, a friend of mine noticed that a participant had left that row blank. She was quick to ask the participant,

“Oh, I see you skipped face 47… did you just miss that one? Did we go too fast?”

“No, but you said not to rate people you know.” (Keep in mind the face in question is NOT A REAL PERSON.)

“I see. So you know that person? Are they a friend of yours?”

“No, I guess I don’t really know them, but they sit behind me in math.”

Again, NOT A REAL PERSON!

But the quality of prototypicality confuses us. It feels like we could know them. They feel familiar, even when it is not possible. As a result, people with prototypical faces feel safe. Unfortunately, this feeling of safety is absent when viewing faces that are less prototypical, and this is likely a root cause of the discriminatory behavior directed at those of us who are less attractive. But that depressing news will have to wait for another post.

-The Plague Doctor