Mass Driver’s recent comment about developing the US Constitution being like the invention of a Friendly AI opens up the possibility of a mostly Friendly AI—an AI which isn’t perfectly Friendly, but which has the ability to self-correct.
Is it more possible to have an AI which never smiley-faces or paperclips or falls into errors we can’t think of than to have an AI which starts to screw up, but can realize it and stops?
Mass Driver’s recent comment about developing the US Constitution being like the invention of a Friendly AI opens up the possibility of a mostly Friendly AI—an AI which isn’t perfectly Friendly, but which has the ability to self-correct.
Is it more possible to have an AI which never smiley-faces or paperclips or falls into errors we can’t think of than to have an AI which starts to screw up, but can realize it and stops?
It’s not feasible to attempt to create a government which both perfect and self-correcting. I’m not sure if the same is true of FAI.