A lot of systems we have already made are super fucked up. this is true. a lot of them were designed to fuck shit up, and be generally evil. we do that sometimes.
these systems only serve to magnify them. see, there’s been a massive marketing push is to call these things “artificial intelligence”. they’re not. they tell you it’s all to complex to explain, but type something on your phone. no, really, do it. like a sentence or two. anything.
you just used the small easily comprehensible version of a large (thing) model. the problem is, as you try to scale complexity on these, both accuracy and compute resources grow exponentially, because it’s literally the same kind of algorithm as your software keyboard uses to autocorrect, but with a bunch of recursion in it and much larger samples to reference every time someone hits a key.
there are some philosophical implications to this!
see, there is no neutral. there is no such thing as a view from nowhere. which means these systems are not. they need to be trained on something. you don’t just enter axioms. that would be actual AI. this, again, isn’t that. these are tools for making statistical correlations.
there’s no way to do this that is ‘neutral’ or ‘objective’. so what data do you think these tools get fed? Lets say you’re a bank, let’s say you’re wells fargo, and you want to make a large home-loan-assessment model. so you feed it all the data from your institution going back to the day your company was founded. back in stagecoach and horse times.
so you have names of applicants, and house statistics, and geographic location, and all sorts of variables to correlate and weigh in deciding who gets a home loan.
which is great if your last name is, for example: hapsburg. less good if your last name is, for example: freeman. and you can try to find ways to compensate, if you want to. keeping in mind that the people who made this system may actively want to stop you. but it’s possible. but these systems are very very good at finding secret little correlations. they’re fucking amazing at it. it’s kind of their shit. this is the thing they’re actually good at. so you’ll find weird new incomprehensibly cryptic markers for how to be a racist piece of shit, all of which will stay within the black box and be used to entrench historical financial bigotry.
trougnouf@lemmy.world 1 day ago
What do you think the AI would be trained on?