zlacker

> No one has ever been able to demonstrate an "unsafe" AI of any kind

"A man has been crushed to death by a robot in South Korea after it failed to differentiate him from the boxes of food it was handling, reports say."

https://www.bbc.com/news/world-asia-67354709

>>threes+(OP)
This is an "AI is too dumb" danger, whereas the AI prophets of doom want us to focus on "AI is too smart" dangers.

>>threes+(OP)
Oh no do not use that. That was servo based, AI drones, which I think is the real "safety issue"

>>38199233

>>sensei+R3
All robots are servo based.

And there is every reason to believe this is an ML classification issue since similar robots are in widespread use.

>>threes+(OP)
And someone lost their fingers in the garbage disposal. A robot packer is not AI any more than my toilet or a landslide.

>>kspace+K1
This sort of prediction is by its nature speculative. The argument is not--or should not--be certain doom. But rather that the uncertainty on outcomes is so large that even extreme tails has nontrivial weight