BLUE
P
palizcat
@palizcat.bsky.social
meow 🍂
40 followers212 following53 posts
Ppalizcat.bsky.social

(Of course there are plenty of exceptions. E.g. Seth Dickinson's description of ML in his new novel Exordia is precise and incisive, but this type of critique seems all too rare.) 4/

And the universal theme of machine learning is thought without understanding. A properly trained machine learning system can build frighteningly accurate models of economic interdependencies in Southeast Asia, or deduce the spread of an epidemic in Africa from changes in satellite photos of crop fields, or predict political unrest in ex-Soviet client states from seemingly unrelated Google searches in Moscow. But it does not know what Moscow is, or what a human being might be, or even that it exists. It is simply building connections between points of data. The logic of those connections is left to the machine itself to devise. It has no prior knowledge of the universe. (Except anything introduced in the network’s initial values, or in data augmentation, but—those are added by human operators.)

The machine might be absolutely and pathologically insane, with a model of the world straight out of Ligotti. But if that insanity produced useful responses, no one could know.
2

Ppalizcat.bsky.social

Instead of old paradigms of totalizing AI, we can take up materially grounded depictions that do not obscure technical limitations & social harms. A neural network is not, after all, a sentient entity but a function with an ungodly number of parameters trained on vast amounts of energy & data. 5/

1
JLjolwalton.bsky.social

Ligotti is a great call

0
P
palizcat
@palizcat.bsky.social
meow 🍂
40 followers212 following53 posts