As best as I can tell this story merely reprises a blog post about a conference presentation. I don’t rant to minimize AI risks in any form but this one feels off. I’ll go out on a limb and predict that most likely the story will turn out to be highly inaccurate.
@ct_bergstrom I guess "AI-Controlled Drone Goes Rogue, Kills Human Operator in USAF video game" didn't get enough clicks in the A/B test.
@ct_bergstrom How is this different than *any* kind of programming error that might kill people like, you know, software on a 737 Max?
This looks like a badly programmed RL reward function. WTF?
If it had "purpose" that _would_ be diffierent?
@ct_bergstrom @twitskeptic
Oh, that really interesting question is:
For any such construct: How would we or could we know?
@twitskeptic My guess is AI-controlled drone fucks up, shoots human operator in video game, and afterward over beers someone says holy shit what if it did that ON PURPOSE?!!?