I model this as a case where the Matt Simpson system has a network of preference-systems (where a “preference system” = propositions + algorithms + environment), and some of those preference systems are usually in the minority in the action recommendations they give. Matt Simpson the reflective agent would have less stress if he could eliminate those preference systems in the “loyal opposition”.
Then, replace each preference system with a system that puts as much of its content as possible into propositions, so you can optimize algorithms.
You might find, after doing this, that those preferences in the “loyal opposition” were of great value in a small number of situations. The values we’re ashamed of might be the “special teams” (American football term) of values, that are needed only rarely (yet vote all the time). I’m just speculating.
If that’s still what you want to do, it’s not honest to call the values left over “human values”. Human values are the values humans have, in all their meaty messiness.
And you’re still faced with the problem, for the preference systems that remain, of deciding what results they should give in new environments.
If that’s still what you want to do, it’s not honest to call the values left over “human values”. Human values are the values humans have, in all their meaty messiness.
Fair enough
And you’re still faced with the problem, for the preference systems that remain, of deciding what results they should give in new environments.
To put it simply (and more generally) I would just say that I still don’t fully know my preferences. In particular, even after figuring out which parts of the Matt Simpson system are me, my values are still underdetermined (this is the point of my “partially?” in the grandparent).
I’m not disagreeing with you at all, btw, just clarifying some terms.
With that out of the way, are you suggesting that there is no correct/best way to go from the underdetermined preferences to a consistent set of preferences? Or in your terms, to decide what results the remaining preference systems should give in new environments? (I think I know your answer to this—its just an understanding check).
I model this as a case where the Matt Simpson system has a network of preference-systems (where a “preference system” = propositions + algorithms + environment), and some of those preference systems are usually in the minority in the action recommendations they give. Matt Simpson the reflective agent would have less stress if he could eliminate those preference systems in the “loyal opposition”.
Then, replace each preference system with a system that puts as much of its content as possible into propositions, so you can optimize algorithms.
You might find, after doing this, that those preferences in the “loyal opposition” were of great value in a small number of situations. The values we’re ashamed of might be the “special teams” (American football term) of values, that are needed only rarely (yet vote all the time). I’m just speculating.
If that’s still what you want to do, it’s not honest to call the values left over “human values”. Human values are the values humans have, in all their meaty messiness.
And you’re still faced with the problem, for the preference systems that remain, of deciding what results they should give in new environments.
Fair enough
To put it simply (and more generally) I would just say that I still don’t fully know my preferences. In particular, even after figuring out which parts of the Matt Simpson system are me, my values are still underdetermined (this is the point of my “partially?” in the grandparent).
I’m not disagreeing with you at all, btw, just clarifying some terms.
With that out of the way, are you suggesting that there is no correct/best way to go from the underdetermined preferences to a consistent set of preferences? Or in your terms, to decide what results the remaining preference systems should give in new environments? (I think I know your answer to this—its just an understanding check).