I dedicated my first website to Cardcaptor Sakura and decorated it pink and green. Sitting on a computer in my middle school’s library, I spent hours teaching myself HTML, taking furtive looks over my shoulders in case a classmate came in. The code remains, half-working, on a server somewhere very far away—on a blinking machine in a dark, refrigerated warehouse full of other, identical blinking machines, which themselves hold blueprints for webpages about other, better television series. Rooms like these incarnate the internet, but in its physical form, the web is illegible: little circuits configured just so to inscribe the meaning humans make. Rooms like these hold much of my digital life, which is increasingly just my life. And, so, my life increasingly wants for a body.
“The Internet is paradigmatic magic,” Virginia Heffernan wrote in her book Magic and Loss. “It turns experiences from the material world that used to be densely physical–involving licking stamps, say, or winding clocks or driving in cars to shopping centers–into frictionless, weightless, and fantastic abstractions.” Heffernan is right. But, also, the computers that hold my first ever website would certainly crush me if they fell. When I make meaning online, I make a paradox: it’s weightless and forever; it’s on a server and will rot away.
The Irish poet Seamus Heaney, who is dead, wrote a poem called “Blackberry-Picking” about how blackberries rot. In the poem, Heaney’s speaker hoards berries in a byre, but “once off the bush the fruit fermented, the sweet flesh would turn sour.” The poem ends: “Each year I hoped they'd keep, knew they would not.” The American poet Robert Hass wrote a poem called “Meditation at Lagunitas” about how meaning rots. “[B]ecause there is in this world no one thing to which the bramble of blackberry corresponds,” he writes, “a word is elegy to what it signifies.” The poem ends: “There are moments when the body is as numinous as words ... such tenderness those afternoons and evenings, saying blackberry, blackberry, blackberry.”
Life online happens in the gaze of algorithms. They note what I buy, what I listen to, what I write to my friends, what I say to my family. They predict what I’d like to listen to, who I’ll probably talk to, what I’ll say. The trouble is, in order to witness our human world, otherwise incomprehensible to them, algorithms must rot the flesh away first, leaving behind only a numeric husk.
One particularly virulent form of algorithm “embeds” words as numbers. These algorithms are trained on vast storehouses of text. A computer will split, say, all of Wikipedia up into short sequences of words. Then it will, based on a sequence of, say, five words, try to predict the sixth. It does this over and over again, until it knows that the word June is used in similar ways to the word July but differently from the word blackberry. It takes in a word, sucks it dry, and spits out a vector—a long list of numbers.
Think of a vector as something like the x, y coordinates found on a scatter plot. Each number in the vector gives you a little bit of information about the whole. In that two-dimensional vector, one number means something about the horizontal position, and one number means something about the vertical position. The vectors that embed text are generally much longer; they might have hundreds of dimensions. But think about it like this: we often take a three-dimensional object, like a cube, say, and visualize it in two dimensions, like in a drawing of a cube. Obviously, something is lost—namely all that information captured by the third dimension—but much is gained, too: you can frame that cube and hang it on a wall. In the case of word vectors, with their many hundreds of dimensions, we do something similar—reduce the vector down to two or three dimensions, to something visualizable. This sloughs off some kinds of meaning and adds others. Something is always lost and gained in translation. C’est la vie.
Computer languages often define colors using three numbers: one for red, one for green, and one for blue. Each number in this color vector is somewhere between 0 and 255, which makes 16,777,216 possible colors. So, you could take the three numbers that represent a word and make them three numbers that represent a color, in effect translating a machine’s understanding of a word into a machine’s understanding of a color. With this, words are rehydrated into a new kind of meaning.
One computer language, CSS, is used to tell your computer how a webpage should look. It tells your browser what color text should be, or how much space to put between images. CSS has a list of built-in colors with names and corresponding red, green, blue vectors. You could find the closest CSS color vector to your word vector—translating one kind of meaning (all the possible 16,777,216 colors) into another (only the 150-odd CSS colors).
Some of those CSS color names are evocative. (I like cornflower.)
The internet, with its permanent weightlessness, gives us the illusion of a bodiless existence. In reality, it has a body: we just can’t make sense of it—all silicon and steel. It can’t make sense of us, either. Not really. To understand who we are and what we make, it translates us into vectors, into shadows of three-dimensional objects projected on a wall.
Every year, Heaney’s speaker feels like crying after his blackberries rot. But every year, he picks them anyway. The magic of the internet is that, despite its incomprehensibility, we make meaning from the vectors it makes of us. We make meaning from Heaney poems translated into color, and we make meaning from a Facebook algorithm that sows division. In this sense, the internet is nothing new: just a problem of lossy meaning and blackberry decay. But it seems to me that, online, the rot takes hold faster than ever. The solution to this problem might be the same old solution: more and better ways to make meaning. More art. More poetry. More afternoons and evenings, saying blackberry, blackberry, blackberry.