Obvious comment: Looks like a strong argument against approval directed AI alignment approach.
Maybe less obvious one: Assumption that “people want to be happy” may be wrong. People often have reasonable goals which—and they know it—will make them less happy by all typical measures, but they still want it and approve it. Examples: perusing scientific carrier or having children.
Obvious comment: Looks like a strong argument against approval directed AI alignment approach.
Maybe less obvious one: Assumption that “people want to be happy” may be wrong. People often have reasonable goals which—and they know it—will make them less happy by all typical measures, but they still want it and approve it. Examples: perusing scientific carrier or having children.