I want to be using an algorithm that would’ve gotten the right answer on slavery, even given 19th century inputs.
This dosen’t constrain algorithm space as much as you may think. There are plenty of algorithms that would get the
wrong answer to 19th century inputs, the right answer for 12th century inputs and the right answer for 20th century inputs.
Also lets remember what 12th and 19th century inputs are, these are basically partially reconstructed (by interested parties even!) incomplete input sets. When evaluating 21st century inputs and which algorithm to use, sure I’d probably put some weight in favour of an algorithm that gives me the right result for the input sets mentioned, but I’m not sure this is a necessary precondition.
I mean remember it’s not like we have a time machine. Sure TDT may complicate stuff somewhat, but it isn’t magic.
BTW My post on moral progress seems somewhat relevant.
This dosen’t constrain algorithm space as much as you may think. There are plenty of algorithms that would get the wrong answer to 19th century inputs, the right answer for 12th century inputs and the right answer for 20th century inputs. Also lets remember what 12th and 19th century inputs are, these are basically partially reconstructed (by interested parties even!) incomplete input sets. When evaluating 21st century inputs and which algorithm to use, sure I’d probably put some weight in favour of an algorithm that gives me the right result for the input sets mentioned, but I’m not sure this is a necessary precondition.
I mean remember it’s not like we have a time machine. Sure TDT may complicate stuff somewhat, but it isn’t magic.
BTW My post on moral progress seems somewhat relevant.