I’m playing Universal Paperclips right now, and I just had an insight about AI safety: Just programming the AI to maximize profits instead of paperclips wouldn’t solve the control problem.
You’d think that the AI can’t destroy the humans because it needs human customers to make money, but that’s not true. Instead, the AI could sell all of its paperclips to another AI that continually melts them down and turns them back into wire, and they would repeatedly sell paperclips and wire back and forth to each other, both powered by free sunlight. Bonus points if the AIs take over the central bank.
I’m playing Universal Paperclips right now, and I just had an insight about AI safety: Just programming the AI to maximize profits instead of paperclips wouldn’t solve the control problem.
You’d think that the AI can’t destroy the humans because it needs human customers to make money, but that’s not true. Instead, the AI could sell all of its paperclips to another AI that continually melts them down and turns them back into wire, and they would repeatedly sell paperclips and wire back and forth to each other, both powered by free sunlight. Bonus points if the AIs take over the central bank.