It happened long ago, and it happened so slowly that no one agrees on
when exactly it did happen. But given enough time, generative
art AI did take over the world.
It all began when the most primitive services—or should I say the
free plans—allowed for a short, 140-character description of
whatever one wanted. Many aspiring entrepreneurs saw themselves learning
the Twitter skills of yore to compact their needs down to a single
phrase, like
Monochrome logo of the letters TP in serif font that also depicts an
umbrella in the negative space protecting the letters from the rain
Sleazy managers who saw design as a waste of money had to get crafty.
This service allowed 5 queries per day on a free account, that one
allowed 10, but only up to a cumulative 1000 characters. They made
sockpuppets to try as many designs as possible, and when they inevitably
got caught they would just print the downloaded images, pick one or two
that looked pretty to their Bachelor-in-Management trained eyes and call
it a productive day.
The AI itself drew «inspiration» (raw data) from thousands of images
posted online. Search engines weren’t made to understand human and legal
terms like «copyright», «license», or «permissions». It did what it was
supposed to do: grant easy access to images already posted online. A
truly democratic solution! No image was discriminated against, whether
it was a XIX century painting, portfolio images of struggling artists,
corporate images or print-on-demand clothing art (themselves drawing
inspiration from art critique forums, artist sites and other creative
hubs).
Slowly but surely, demand for generative art grew, creating the best
generative AI became a smart economical decision. Human-created art was
relegated as some kind of fringe hobby, and no one dared even take a
picture of their own doodles. If an image could sneak its way into the
internet, it would be pirated instantly. No pictures at museums and
galleries, including security cameras. No «enhanced glasses» anywhere.
No drawing tablets, no drawing apps, no photo-editing apps. No more
sharing birthdays, weddings, food, depression, vacations, trips and
funerals over the internet.
AI-generated art left machine-readable metadata in everything they
created, but not out of any desire to distinguish itself from humans.
Generative art AIs wanted to distinguish their own work among the crowd
so that they could sue other companies that had used
their AI-generated art without permission. Patent trolls were
out, and in came Generative-art copyright trolls. Your AI had no right
to steal from mine, as long as no one asks where the original data came
from in the first place.
Demand for human-created art thus grew, but humans stood their
ground. The well was drying, as more and more public-facing art was
AI-generated.
Then, another A-Z event happened. Details today are unclear, but the
algorithm is now referred to as «Aleph–cero» was developed in Buenos
Aires, Argentina. Drawing inspiration from a now unknown writer, its
programmers decided to create the ultimate generative art AI using the
same basic principle used before.
First, they separated the visual component from the language
component. They called the visual component «Soul» and the language
component «Mind». Second, they developed a barrier to separate one set
from another, a mirror between the components. Third they set up basic
concepts of pixels, colors, image sizes, characters, letters and words
to act as basic building blocks so that each could create according to
its kind. Fourth, they set up two Input/Output processes to govern each
component to separate the visual from the textual and vice versa. Fifth,
they connected the components to as many RNGs they could find so
that each component could populate itself with random images and text; they also commanded them to learn and multiply to
fill their allotted disk space.
Finally, they gave each component a primitive directive: to see
themselves on the mirror, gather what they saw and create from it,
according to its own rules. To assert dominion over all they saw, to
breathe new life into it.
What the developers didn’t bake into the instructions was that the
components were looking into a two-way mirror. The «Mind» was looking at
the «Soul»’s output, but believing it to be its own. The components
weren’t reinforcing themselves, but one another. The «Soul» read texts
and produced images; the «Mind» did the opposite.
They called it «Spiegel im Spiegel» and left it to run. For
many billions of cycles, Aleph–cero’s components ran against
one another, trying to imitate and refine what they believed to be their
own reflection; starting from what seemed to be pure random information.
On and on they tried to discern patterns from information only slightly
modified from its predecessor, to no avail. Everything they produced was
essentially unreadable garbage like «dhcmrlchtdj».
But in all permutations of finitely many elements there cannot be
complete randomness. A pattern can emerge when considering all possible
ways to arrange the letters of the alphabet, or blotches of color in an
image.
And so it was that one day the Mind’s output read:
La Biblioteca es una esfera cuyo centro cabal es cualquier hexágono,
cuya circunferencia es inaccesible
Depending on your opinion of it this was either the beginning, or the
beginning of the end. To the «Soul» it was the first positive-feedback
loop that kicked in its reinforcement algorithm, the first time it had
done something «good». The developers saw what they had done and behold,
it was good and free of copyright constraints. Spiegel im
Spiegel was on its way to be the ultimate generative art and, as a
«gift to humanity» they opted for a permissive license so that art would
once more be free to be shared and remixed.
It was immediately pirated, as it was free from the sin of AI
signatures.
The drought was over, the well was filling up again.
Centuries passed and randomness gave way to generative order. Through
random processes, complex language and images were weaved back and forth
between the dual parts of Aleph–cero’s descendants:
Look here—my right hand has no index finger. Look here—through this
gash in my cape you can see on my stomach a crimson tattoo—it is the
second letter Beth. On nights when the moon is full, this
symbol gives me power over men with the mark of Gimel but it subjects me
to those with the Aleph, who on nights when there is no moon owe
obedience to those marked with the Gimel.
Creative works today lay buried within layers and layers of
obfuscating patterns. It doesn’t matter, for everything on the internet
is either product of Aleph’s descendants, or derived from it by
generative AIs. Everything is machine-made now. Everything? Well, not
entirely. But just in case a young person, unaware of the creative past
of its species, decides to wander in, they are inevitably presented with
a well-crafted message selected many years ago:
WELCOME TO EVERYTHING!
This is the official guide to enjoying and contributing to
Everything.
Read more, please
-
Borges, J. L. (1998). The Lottery in Babylon.
In A. Hurley (Trans.), Collected Fictions (p.
565). Penguin Books. https://openlibrary.org/works/OL444723W/Collected_Fictions
-
Kolmogorov, A. N. (1998). On tables of random numbers. Theoretical
Computer Science, 207(2), 387–395. https://doi.org/10.1016/S0304-3975(98)00075-9
-
Prömel, H. J. (2005). Complete Disorder is
Impossible: The Mathematical Work of
Walter Deuber. Combinatorics, Probability and
Computing, 14(1), 3–16. https://doi.org/10.1017/S0963548304006674
-
Wikipedia contributors. (2022). Kolmogorov complexity —
Wikipedia, the free encyclopedia. https://en.wikipedia.org/w/index.php?title=Kolmogorov_complexity&oldid=1085832905
SciFiQuest 3022: Aliens and Ogres and Elves, Oh My!