Artificial intelligence and the monkey’s mirror
This is the story of a child born in a prison without windows, and he is extraordinarily good at drawing. In the same prison cell live dozens, hundreds of people who incessantly tell him stories and fables about the outside world — a world he has never seen or experienced on his own. To kill time, he tries to portray the world starting from the stories he’s told.
He never really has been able to look outside: all the information he has access to happens by nurture, not by nature. When asked to draw a successful man, he portrays him as a caucasian middle-aged male in a suit and tie, the word beauty evokes a woman with Angelina Jolie cheekbones, the word God prairies of crosses and clouds, a black man comes portrayed with sculpted pecs, the hero is a humongously chested knight. His cellmates aren’t necessarily Caucasian males, though, yet the stories converge on recurring stereotypes.
As a matter of fact the child cannot even get out of his cell, because he is not even a child but the closest thing to a child that humanity, in evolutionary terms, has produced: an artificial intelligence. And, as in a mirror — as a child does to a certain extent — he talks about us with innocent eyes, telling us back his adult-fed, naive, vision of our own world.
It’s all obviously living in the uncanny valley realm; the drawings look like without really being: they are an inaccurate, yet often creepy, representation of reality. Faces can look perfectly symmetrical and attractive, but without a portion of the head, or with three hands; a family sitting at the table looks like guests from the Addams family dinner. But its superficial meaning appears instantly clearly to us, though.
A few weeks ago the always excellent Luca Malisan sent me the link to Midjourney, one of these GAN artificial intelligences, where, with a few descriptive commands, a user can generate awesome images, refinable through iteration, commands and accurate descriptions.
The beautiful image you see above was created using the phrase “giant tree-house with mid-century modern stylings, ropes, vines, rope ladders and large windows decorate its sides, high breaking through the misty clouds, atmospheric, rain forest, kodak, fuji film, hgtv, 12k ursa, cinematic”, and it is really very beautiful and full of details: our child draws with skill and imagination.
It is nearly magic, perhaps, in the way we used to think about magic — you own the name and the words, and this becomes real: although only on the pixels of a computer, for now, with a single formula that looks like a magic spell we can generate and distort the reality that surrounds us.
And the more accurate our spell is, the more powerful and beautiful our magic will be.
The object to my research, however, was not to obtain photorealistic results — I have no artistic ambitions whatsoever — but to see how much the deeper, more anthropological part of its results were influenced by whose cultural stereotypes. I generated a series of images based on “primitive” concepts (“motherhood”, “hero”, “successful person”) so to speak, and not entirely abstract (“love”), reiterating the same request several times to see how much results varied from time to time, being especially careful to provide as little detail as possible. And the result is quite uniform, as clear beyond words: it’s mostly a white man’s world.
The database these algorithms draw data and information from is not influenced by us alone, of course, but by the zillions of millions of images from databases such as Shuttershock or Getty Images they are granted access to, and from our subsequent choices of whose image on the four is the “correct” version, from which he learns how to to please us. Artificial intelligence did not make any choice, in fact, but just reflected the world that we wanted to teach as real and to please us. And actually does its job pretty good, imagining gaudy things starting with the little I shared: I couldn’t have done any better myself.
Internet, again, is not at all as a realistic depiction of reality: the results that Google Images searches return are strongly conditioned by other algorithms that work into addressing results for commercial purposes, manipulated by other men, of course, while we everyday humans work hard at feeding social media with our glossy and unrealistic photos. However, imagine how being able to influence these results turns out into an extraordinarily powerful political and propaganda weapon.
Both language and its representation of reality are the first and most immediate instruments of control of a more and more overpopulatedd and thereby centripetal human culture, rippling in a million of niches and subcultures that cannot be defined as minorities for sheer numeric prejudice. And to which, for better or for worse, internet has given voice and strength. And the more the niches and the cultures, the more humanity gets hard to control.
It is by digging into the more tribal archetypes, those on which our small veil of firstworldist glaze has no grip, that it is more strongly evident how language and its representation are so incredibly powerful: who decided that a successful man is a white, middle-aged man in a suit and tie, or that beauty means female youth and caucasian cheekbones? Our brain works primarily gathering and associating archetypes and stereotypes together, of course: they are the meters of the world we use from an early age to understand the world that surrounds us. If in the joke “tradition is peer pressure from dead people” lies a grain of truth, what the contemporary philosopher Byung-Chul Han — not without a bit of Luddism — states can be considered true as well: “rites objectified the world, structure a relationship with the world” and that, in good analysis, they are a valuable tool for us.
“The Hard People” by Patrick Heady is hard to find , and it also costs quite some money.
But it is an absolutely delightful read. It is the result of the research of an anthropologist who spent two years in the Ovaro valley, in Carnia — the alpine valley in which my mother was born, so to speak — studying the rites and traditions of our valleys with the same judgement parameters he would have used with a lost tribe of the Amazon.
And he is adorably merciless in exposing how the ancestral patterns of a first world population are not so distant from those of our own kind in other parts of the globe.
The narration changes, but the atavistic principles of “much / little”, “we / them”, “curse / bless”, underneath, are so ancient that they all look a little alike. Almost nature, not nurture, we could maybe venture.
Those able to manipulate language are powerful enough to manipulate ideas too: language shifts over the Overton Window — what is morally and politicaly acceptable and what is wrong. Accepting linguistic changes actually mean, as a matter of fact, allowing reality to be manipulating, acknowledging the power of the spell itself.
The same language that underlies and contains a cultural heritage also underlies and implicitly contains a divisive political and social tension.
And artificial intelligence describes our modern tribes, and the transformations of our society, with a disenchanted yet analytical gaze, learning to please us — human apes — with the stories we feed it with, until it will learn how to find its own grammar and narrative.
(English version of “L’intelligenza artificiale e lo specchio della scimmia”. Forgive my mistakes please ❤)
