I think American techbros, who can't answer if it's okay for people to be racist, probably shouldn't be delivering AI platforms that speak for people without supervision. Just a thought.
@GossiTheDog They would claim that since the choice is done by an algo, it is math and therefore not racist/sexist.
@rrb
They would be fucking idiots who don't understand how their own tools work at even the most basic level if they said that.
@GossiTheDog
I wish @rrb was joking, but that's word for word the justification used by LeCun as to why algorithms can't be racist circa 2018-2019, on Twitter.
Ok, this exchange actually put me over a line, and I finally finished my essay about the superstar scientists, their ego, eugenics, and current trends in ML and AI research:
https://mastodon.social/@andrei_chiffa/110215451244246319
(or a direct link: https://andreikucharavy.com/L3Cache/eugenics-and-longtermist-effective-altruism-share-the-same-original-sin/)
@andrei_chiffa @dymaxion @GossiTheDog I like anything that discusses Goedel. Do you know how he died?
Tbh going full Beautiful Mind is the way out for a lot of brilliant mathematicians, but having lived through the rise of Nazism in the 1930s and getting a close friend assassinated by an ex-student in the process, despite not being Jewish, is a valid trigger for phobia. And in the absence of modern psychiatric medicine, it is surprising he made it to 71.
@andrei_chiffa @dymaxion @GossiTheDog Einstein said the only thing he liked about Princeton was walking home from work with Goedel.
There is a good BBC documentary on Cantor, Goedel, and Turing. They mentored each other and all committed suicide.
Tbh, understand the fundamental limitations of what is knowable to humans with the rolling drums of celebrating human progress and ideological supremacy of science anticipating the future (and the ramifications of it - namely eugenicism) was certainly a rather depressing experience. Rays Solomoff seems to have been a rather happy hippy in the 50s-90s.
For Turing it was really not being let live with probably the most important aspect of his personality.
That last part still makes me utterly mad every single time I have to touch statistics and probability.
We are using an utterly suited formal tool to do them - algebraic mathematics, in which we artificially massage confidence intervals, significance tests, and co so that they give us a sliver of capabilities in working with non-continuous, discrete, and finite reality.
And later works of Turing contain indices on how to get around it.
Except some high-ranking prude smart-ass decided that his gayness was more important than him being probably the only person in that century that would be able to formalize those ideas properly.
While Shannon's theory of information is good and Solomonoff's algorithmic probability are good tools, there is currently a gaping hole where an appropriate formal system to reason about generalizations from observations should be.
So instead, we have to invent and teach magic recipes for doing the same thing by hand - not unlike Romans for division.
Don't take me wrong - I am happy that statistical tests, differential privacy, and ML generalization tricks exist, but it really feels like we are using a sledgehammer to do jeweler's work.
@andrei_chiffa @dymaxion @GossiTheDog I think a lot of it is laziness. People take short-cuts because that is what they learned in school and they do not want to think. There is a lot you can do, once you get the right mind-set.
The issue is mast people want to be spoon-fed rather than think for themselves.
I have some assignments where I point out a security issue, give them code with a vulnerability, and the assignment is "surprise me." It is really depressing to hear them respond with: "What do I have to do to get a B"
