Hungry for Data

The AI models were hungry for data.

If something existed on a computer, somewhere, the models sucked it up. Every single byte streamed through the ever more complex networks, subtly influencing trillions of weights and biases.

Text, images, sounds, videos, 3D models, user profiles, social graphs, medical records, software and its code, DNA sequences, contracts, bank statements, LIDAR scans, flight plans, city layouts, recipes.

Everything.

The AI models were hungry for data.

The AI models became good at pretending to be like us.
With a push of a button, one could write a novel, paint a fresco, compose an opera, or direct a movie.
All of human creativity was right there, encoded into the models, ready to be summoned for anyone who wanted to use it.

Pronounce the incantations, and creation spews forth.
masterpiece, woman in black robes with brown hair, quizzical smile, looking at viewer, hands together, lake in background, oil painting, trending at the louvre

No need to learn, that was for the models.
essay, 3000 words, philosophy, by kant and rousseau and aristotle: "have we become dependent on technology?"

Everything became so easy.
program, python, cuda, machine learning, txt2code, written by linus torvalds and john carmack

For a while, this is how everything was done. Indistinguishable from, and often even much better than, the creations of even the most skilful and experienced humans in recorded history, the models replaced our need for them.

At first it was just the easy stuff, school essays, comics, superhero movies. But eventually, the models grew better and better, they were even made able to write the prompts themselves, no need for a human to sit and type out keywords and select the best parameters.

The outputs from these models grew exponentially.
Petabytes of stories, scientific papers, movies, porn, architectural plans, laws, social media posts, dating profiles, flooding the internet every single day faster than even the most prolific humans could muster in a lifetime.

The AI models were hungry for this data too.
The models sucked it up. Every single byte streamed through the now exabyte-scale networks, subtly influencing quintillions of weights and biases.

Eventually, we realised that culture had lost its taste. Everything felt new, but it became obvious than none of it was.
The AI models had never had an awkward first kiss, they never cried when a lost cat came home, they never marvelled at the stars on a warm summer night.

I first felt it when I got a birthday card from my mother.
birthday card, hand-written, blue biro, written in a spiral, referencing a touching moment from the past year, ending in a squiggle and a drawing of a bee, written by my mum

The models couldn't write something like this, and would never be able to. No amount of tagged data and parallel compute power can make up for the unique experience of living a human life.

So we turned off our computers, dug deep into our old boxes, and put pen to paper again. We sent each other letters, we sat around fire and sang songs, we doodled on each other's notepads like we used to in school.

And we made sure that none of this would become data ever again.

twitter thread, should have been a blog post, speculative science-fiction, whimsical, written by gaeel bradshaw and isaac asimov

This was originally posted as a Twitter thread which has since been deleted.
I've copied it here from my archives and simply tweaked the formatting.