The payoff set of any PD variant is a quadrilateral, see those graphs. The algorithm I outlined says the tangent line should be parallel to the slope from (C,D) to (D,C). The point (C,C) is the only Pareto-good point on the first graph where such a line could touch the payoff set, not intersect it. The second graph falls under the NB :-( Since the solution is invariant under scaling in any coordinate, it works for the Eliezer’s True PD just fine—it doesn’t require comparing human lives to paperclips anywhere.
That said, the algorithm has serious flaws. It doesn’t seem to be a likely focal point, and it’s not as continuous as I’d like the solution to be. The part of the post up to “now, there are many ways” is much less controversial.
Can you take us through these, as an example? I’m lost.
The payoff set of any PD variant is a quadrilateral, see those graphs. The algorithm I outlined says the tangent line should be parallel to the slope from (C,D) to (D,C). The point (C,C) is the only Pareto-good point on the first graph where such a line could touch the payoff set, not intersect it. The second graph falls under the NB :-( Since the solution is invariant under scaling in any coordinate, it works for the Eliezer’s True PD just fine—it doesn’t require comparing human lives to paperclips anywhere.
That said, the algorithm has serious flaws. It doesn’t seem to be a likely focal point, and it’s not as continuous as I’d like the solution to be. The part of the post up to “now, there are many ways” is much less controversial.