the paper mentions the similarity of the managing the news/outcome pumps problem with edt vs cdt issues, and this old comment by nate is the do-ops version of the attempt in their paper, namely
U(a1, o, a2) := { UN(a1, o, a2) + E[US|do(O in Press)] if o not in Press ; US(a1, o, a2) + E[UN|do(O not in Press)] else }
likewise Stuart Armstrong’s old paper has an indifferent agent paper that uses dynamically adjusting utilities conditioning on do ops.
It looks like MIRI also considered a couple cases with do()-ops[1], however the ones I’ve seen don’t have subagents.
the paper mentions the similarity of the managing the news/outcome pumps problem with edt vs cdt issues, and this old comment by nate is the do-ops version of the attempt in their paper, namely
U(a1, o, a2) := { UN(a1, o, a2) + E[US|do(O in Press)] if o not in Press ; US(a1, o, a2) + E[UN|do(O not in Press)] else }likewise Stuart Armstrong’s old paper has an indifferent agent paper that uses dynamically adjusting utilities conditioning on do ops.