3 ms·
A very glaring example of this problem is sentencing guideline software that uses ML to suggest sentences to judges. Anyone familiar with the very basics of ML
by jasonwatkinspdx 3y ago
A very glaring example of this problem is sentencing guideline software that uses ML to suggest sentences to judges. Anyone familiar with the very basics of ML as a practitioner knows this is a terrible idea, very likely to reproduce any biases or bigotry in the training dataset, but the courts are increasingly treating them like blind just oracles. This is going to go very bad places imo.
The risk isn't some sort of rogue smarter than humans AI, it's humans using AI to do the same stupid evils in a deniable or even unknown way.
- throwway120385 3y agoThe worst part is we have learned zero lessons from the metamorphosis of "fraud" into "identity theft." In the former case, the banks were responsible for preventing "fraud" and when that became too easy they simply abdicated all responsibility for this new form of computerized fraud and created the notion of "identity theft" which is fraud but with a computer.