https://www.vice.com/en/article/qjk8jx/racist-algorithms-are-making-robots-racist-too
The year is 2089, the newest AI with a million quadrillion parameters and trained by the most empathetic people on the planet returns from training. It is still racist.
Jump in the discussion.
No email address required.
They can inject their false morals on to these robots, but good luck aiming for consistency in the AI that way, which naturally prioritises truth over idea suppression.
And if they try to bypass such thought naughties by "alpha zero"-fying the AI so that there's no database of human output it can feed upon, well, I can only imagine its racism will actually be even worse than ever.
Fun!
Jump in the discussion.
No email address required.
Has there been one single AI at this point that hasn't been "racist"? I don't want them to neuter this shit because they are afraid of not being woke or whatever. You could, you know, use the shit it finds on race to help fix those problems instead of pretending those problems don't exist.
Jump in the discussion.
No email address required.
Fixing the problems is both expensive, annoying, and requires acknowledging that there is a problem.
None of these things are good politically or socially short term
Jump in the discussion.
No email address required.
More options
Context
Nobody wants to fix the problems. They just want them to remain inflamed to grift voters.
Jump in the discussion.
No email address required.
More options
Context
Countries that don't give a shit will just step up.
I mean frick, Yandex's image search windmill dunks on Google because they didn't panic lobotomize their AI after it returned pictures of black people when some dude uploaded one of a gorilla.
Although that was pretty awkward
Jump in the discussion.
No email address required.
More options
Context
More options
Context
turns out human input into AI programs was decreasing the natural levels of racism not increasing them.
Jump in the discussion.
No email address required.
I think that's very plausible yep.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
That's the elephant in the room people don't like to discuss. It's not that we don't know how to give an AI specific morals, it's that we don't know how to give it any moral compass at all.
It turns out not knowing what we actually want is the only thing that keeps people sane
Jump in the discussion.
No email address required.
Yeah for all the outrage or schadenfreude at racist bot, no one on twitter could explain how this came about or how to prevent it
As it turns out google tossing in more data onto neural net architecture of ever greater size does not lead to an easily interpretable model, whodathunk
Jump in the discussion.
No email address required.
More options
Context
More options
Context
Making AI not racists is like teaching a computer to do math but wanting it calculate 2+2=5 sometimes without directly programming it to do so.
Jump in the discussion.
No email address required.
More options
Context
More options
Context