Comments by "Jim Luebke" (@jimluebke3869) on "Elites FEARMONGER About AI So They Can Censor Online Speech MORE: Michael Shellenberger" video.

"Sometimes even Google Engineers don't know what the AI is doing" That's because of how the math works -- many AI approaches are sort of the ultimate "by committee" decision-making. Responsibility for a given decision is so fractionated among the various "neurons" that you can't really tell why it makes the decision it makes.
7
@fmayer1507 This is not to say that when every neuron-equivalent makes its contribution, the whole lot doesn't come up with the right answer. It just means, that when it's wrong, you have no idea why it's wrong. It's one big reason we haven't seen completely self-driving cars yet, and may never.
1
@__Rob To be fair, they now have an approach called "attention" that involves metrics called "weighting", which allows you to see what the AI thinks is important. Still, when you tell an AI paralegal of this kind to write you a brief on a particular subject, the AI will do that to the best of its abilities -- even if it means fabricating relevant case law. This obviously has drawbacks. While an AI could fairly easily "drive the pace of operations faster than a human could hope to match", there's still a fairly high probability that it will make decisions that are not based on material reality.
1
@__Rob I have to differ with you on "equality" -- the only way to get quality of life to increase across the board is productivity, not "equality", which can only be truly achieved by pounding people down. That said, I agree with you that disembodied computers trained just to look convincing, will only be convincing, not truthful. It's a real problem.
1