Warning: Objects in this mirror may appear uglier than you wish.
That algorithms trained on human biases exhibits human biases is no
surprise. There are no neutral images. Unless you point cameras
randomly at any and all objects in the world, every single training
image must pass through the filter of human selection. Unlike our ears
and microphones that are omnidirectionally receptive we point our head
and eyes at specific objects, and the camera is a technological
facsimile of that process. When it comes to Art, to paintings, one
could hardly choose a stronger filter for the inherently human
elements of the visual psyche - even if you limit it to bowls of fruit
and sunsets.
> Imagen can be safely integrated into user-facing applications
I am intrigued by the use of the word safely here. Safe for whom,
from what and to what end? Micrological thought control? To hide us
from the truth about our own psyche, individual or collective?
Perhaps categorisation itself is inherently prejudiced? It's part of how the brain works.
Like how some people will say they are colour blind and others that you must see colour, both are about categorising (and the application of) the world.
How does an AI deal with human-like categorisation?
Perhaps in this story, the technology is too advanced for the culture at the moment.
Or maybe it's political - the company has had issues with dealing with stereotypes before and employees. The issue wasn't whether the stereotype was truthful but whether it was politically accepted or not.
In other words categorising things is at least personal and if today's politics is politicising the personal, then categorisation is political.
It may be actually always impossible to have an impartial AI. Maybe the future will be artificial lawyers of AI presenting their biased side and other biased AI deciding which arguement is best? Maybe the future of AI is a replication of humanity, good and bad. Perhaps AI will believe in God rationally.
2 comments
[ 1.7 ms ] story [ 13.6 ms ] threadThat algorithms trained on human biases exhibits human biases is no surprise. There are no neutral images. Unless you point cameras randomly at any and all objects in the world, every single training image must pass through the filter of human selection. Unlike our ears and microphones that are omnidirectionally receptive we point our head and eyes at specific objects, and the camera is a technological facsimile of that process. When it comes to Art, to paintings, one could hardly choose a stronger filter for the inherently human elements of the visual psyche - even if you limit it to bowls of fruit and sunsets.
> Imagen can be safely integrated into user-facing applications
I am intrigued by the use of the word safely here. Safe for whom, from what and to what end? Micrological thought control? To hide us from the truth about our own psyche, individual or collective?
Like how some people will say they are colour blind and others that you must see colour, both are about categorising (and the application of) the world.
How does an AI deal with human-like categorisation?
Perhaps in this story, the technology is too advanced for the culture at the moment.
Or maybe it's political - the company has had issues with dealing with stereotypes before and employees. The issue wasn't whether the stereotype was truthful but whether it was politically accepted or not.
In other words categorising things is at least personal and if today's politics is politicising the personal, then categorisation is political.
It may be actually always impossible to have an impartial AI. Maybe the future will be artificial lawyers of AI presenting their biased side and other biased AI deciding which arguement is best? Maybe the future of AI is a replication of humanity, good and bad. Perhaps AI will believe in God rationally.