News for hackers. Stuff that matters.
Re:The greatest risk of AI is from the people who control it... by nradov on Wednesday May 31, 2023 @03:34PM (#36143044)
I appreciate your concern, but the same issue would apply with existing expert systems or linear regression models that wouldn't typically be classified as "AI" today. Most power companies are subject to fairly strict regulations so they generally can't cut off customers unless bills are badly overdue, or the customer tampers with utility equipment. In some jurisdictions the power companies are required to report suspicious usage patterns to law enforcement; that may be objectionable on privacy grounds but that's a political issue, not really an AI issue. Where there's a large power imbalance between private citizens and large institutions we should address that with laws and regulations based on impact, fairness, and accountability rather than trying to specify how those institutions can use particular algorithms.
Re:The greatest risk of AI is from the people who control it... by majormajor on Wednesday May 31, 2023 @03:38PM (#36143090)
Unfortunately in the US if your law isn't that specific - and almost any law about "impact, fairness, and accountability" is going to be much harder to tighten up, interpretation-wise, than one about "don't fucking do these things" - then it's a big target for judges who don't like your definition of impact/fairness/accountability, or the definition of the agency you create to oversee it.

That's why the Republican party has been gunning so hard in the past couple decades for the Chevron deference precedent: they want it harder to have legally-enforceable regulations that don't require lawmakers to get into the weeds (and risk being narrow and outdated soon).

Re:The greatest risk of AI is from the people who control it... by ethanbond on Wednesday May 31, 2023 @03:41PM (#36143129)
I don’t think it’s honest to suggest linear regression models fail in the same way current AI models do. Linear regression “fails” when you are an outlier case, whereas AI systems just occasionally truly fuck up in unpredictable ways on standard inputs.
Re:The greatest risk of AI is from the people who control it... by jstarfish on Wednesday May 31, 2023 @04:00PM (#36143429)
> In some jurisdictions the power companies are required to report suspicious usage patterns to law enforcement

It's going to be a dark day when some idiot passes a law forcing AIs to serve as mandatory reporters.