Ian David Moss has a post on this forum arguing for things along the lines of ‘EA for the rich country fine arts’ and other such restricted scope versions of EA.
My biggest objection to this is that to stay in line with people’s habitual activities the rationales for the restricted scope have to be very gerrymandered (perhaps too much to be credible if stated explicitly), and optimizing within that restricted objective function may be pick out things that are overall bad, e.g. the recent media discussion comparing interventions purely in terms of their carbon emissions without taking anything else into account, suggesting that the existence of a member of a society with GDP per capita of $56,000 is bad if it includes carbon emissions with a social cost of $2,000 per person.
Ian David Moss has a post on this forum arguing for things along the lines of ‘EA for the rich country fine arts’ and other such restricted scope versions of EA.
Thanks for the link! I did a quick search to find if someone had already said something similar, but missed that.
My biggest objection to this is that to stay in line with people’s habitual activities the rationales for the restricted scope have to be very gerrymandered (perhaps too much to be credible if stated explicitly), and optimizing within that restricted objective function may be pick out things that are overall bad,
I’m not sure whether the first one is really an issue—just saying that “these are general tools that you can use to improve whatever it is that you care about, and if you’re not sure what you care about, you can also apply the same concepts to find that” seems reasonable enough to me, and not particularly gerrymandering.
I do agree that optimizing too specifically within some narrow domain can be a problem that produces results that are globally undesirable, though.
Ian David Moss has a post on this forum arguing for things along the lines of ‘EA for the rich country fine arts’ and other such restricted scope versions of EA.
My biggest objection to this is that to stay in line with people’s habitual activities the rationales for the restricted scope have to be very gerrymandered (perhaps too much to be credible if stated explicitly), and optimizing within that restricted objective function may be pick out things that are overall bad, e.g. the recent media discussion comparing interventions purely in terms of their carbon emissions without taking anything else into account, suggesting that the existence of a member of a society with GDP per capita of $56,000 is bad if it includes carbon emissions with a social cost of $2,000 per person.
Thanks for the link! I did a quick search to find if someone had already said something similar, but missed that.
I’m not sure whether the first one is really an issue—just saying that “these are general tools that you can use to improve whatever it is that you care about, and if you’re not sure what you care about, you can also apply the same concepts to find that” seems reasonable enough to me, and not particularly gerrymandering.
I do agree that optimizing too specifically within some narrow domain can be a problem that produces results that are globally undesirable, though.