5 ms·
Most questions of this sort remind me of "Can submarines swim?" If we build an AI that can interact via natural human language, and ask it about its internal s
by Razengan 8y ago
Most questions of this sort remind me of "Can submarines swim?"
If we build an AI that can interact via natural human language, and ask it about its internal state, and it replies "I am happy.", is it happy?
And whatever criteria we use to decide that machines cannot really be happy the way we can be, could a hypothetical species that thinks at a "higher level" than us, use similar criteria to determine that humans can't really think?
- visarga 8y agoAn AI would have a goal, even if it is simple curiosity, and its happiness is related to the distance from its goal. It would evaluate each situation and action with regard to the goal and decide on actions that lead to it, learning from past mistakes. 'I am happy' is possible for us because we have an innate goal - to live, to reproduce, to maximise our rewards, rewards which have been programmed into us by evolution. AIs can have rewards and can learn to act towards them, thus they can have emotion which is just a map from the present situation to the goal.
- natosaichek 8y agoI disagree with basically everything you just said. An AI might not have a goal. If it did, its happiness may or may not be related to the distance from the goal. It might not evaluate every situation and action with regard to it's (presumed) goal. I don't think we do have a (single) innate goal. Everyone has different goals, and they are not innate. Emotions are not a map from a present situation to a goal. AI might not have emotions.
- visarga 8y ago> I don't think we do have a (single) innate goal. Survival. And that includes many subgoals such as feeding, safety, autonomy, being part of a group, curiosity, and making offspring, all preprogrammed into the brain by evolution. Artificial agents would have goals preprogrammed by humans instead, such as 'winning at Go'.
- natosaichek 8y agowhat about suicidal people? Are you saying they are not consious? Artifical agents _might_ have goals preprogrammed by humans, but I don't see that as a prerequesite for intelligence or self awareness.