Life, sacred, unpredictable life, is not a predictive algorithm
Sarah Abraham
TO THE WORDS
W.S. Merwin, 1926-2019, American,
published in 2001, republished in The Essential Merwin (Copper Canyon Press), 2017
When it happens you are not there
O you beyond numbers beyond recollection passed on from breath to breath given again from day to day from age to age
charged with knowledge knowing nothing
indifferent elders indispensable and sleepless
keepers of our names before ever we came to be called by them
you that were formed to begin with you that were cried out you that were spoken to begin with to say what could not be said
ancient precious and helpless ones
say it
as above
Words (numbers, images) are not sacred. But they are the chief representations of our lives and life is sacred.
Are words (numbers, images) helpless? Yes: they are artifacts of our intention and imagination. Sometimes they operate like grail cups or mediums. They are not autonomous even if artists, scientists, authoritarians, advertisers and some children pretend otherwise.
Why repeat this well-known stuff?
Because the already dazzling exploits of Artificial Intelligence is camouflaging what we well know.
Here from interviews of AI researchers published in the NY Times on August 25, 2024 by Aatish Bhatia
are 3 diagrams of what is known to be happening to the AI information flooding the internet.
The net is the primary source of information for a very large number of people.
We do not usually know when we are reading AI-generated information and when we are reading from the pen of or out of the mouth of a human being.
One North American AI company noted recently that its AI model generates 100 billion wordsevery day from data that it has scraped from sources available to it.
The degradation illustrated below of numbers – images and words-strung-into-sentences are equally affected – occurs when successive models of AI are trained on the data already in the web’s oceans of data: the output becomes a feed-back loop of complete garbage.
This is because when AI is trained on such data, it assembles and generates ” a statistical distribution — a set of probabilities that predicts the next word in a sentence, or the pixels in a picture.”
Not a real world but what seems to it to be the most probable world built from all the data swilling around. Possible or not.
An unknown amount of this data is wrong, misleading or needs further explanatory context to be a reflection of our real lives.
Thus the garbage which has resulted in massive amounts of incorrect advice and information on the internet
including advice to kill one’s partner to resolve partnership problems;
images of humans with unreal bodies;
famous paintings with subtle and unsubtle variations in composition etc.
To say that these misrepresentations of our reality are dangerous is an understatement.
Real, handwritten digits
AI output after 20 generations:
AI output after 30 generations
The ultimate solution for this degenerate AI is a verifiable, correct stream of information entering its data models of the real experience – in numbers, images and words –
of humans from all the cultures of the world,
for ever
until our sun goes supernova and our earth is at an end.
There are ethical issues which will become more urgent (for instance: privacy, copywrite, malign human input, lop-sided input of data by population age, ethnic group etc.).
There are the ever larger amounts of energy needed for scraping and formatting data to build and add to models.
Without further work, and perhaps even with,
without regulation with teeth,
we will be left on the web with representations of our lives which are dribbles, mockeries, subtle and unsubtle manipulations of words, numbers and images:
