If we can’t demand perfect metrics then surely we should at least demand metrics that aren’t easily gamed. If people with the quality named “rationality” don’t on average win more often on life-problems like those named, what quality do they even have, and why is it worthwhile?
I understand “rational” people “win” at the goal of believing the truth, but that goal may be in conflict with more familiar “success” goals. So the people around us we see as succeeding may not have paid the costs required to believe the truth.
If rationality is defined as making the decisions that maximise expected utility in a given situation then it is by definition more winninng. The question would be nonsensical.
If another definition of rationality is implied then I don’t think Eleizer demanded that it win.
I do have components of my utility function for certain rituals of cognition (as described in the segment on Fun Theory) but net wins beyond that point would compel me.
If we can’t demand perfect metrics then surely we should at least demand metrics that aren’t easily gamed. If people with the quality named “rationality” don’t on average win more often on life-problems like those named, what quality do they even have, and why is it worthwhile?
I understand “rational” people “win” at the goal of believing the truth, but that goal may be in conflict with more familiar “success” goals. So the people around us we see as succeeding may not have paid the costs required to believe the truth.
Suppose we did the experiments and found other policies more winning than rationality. Would you adopt the most winning policy?
If not, then admit that you value rationality, and stop demanding that it win.
If rationality is defined as making the decisions that maximise expected utility in a given situation then it is by definition more winninng. The question would be nonsensical.
If another definition of rationality is implied then I don’t think Eleizer demanded that it win.
That would be a rational thing to do!
I do have components of my utility function for certain rituals of cognition (as described in the segment on Fun Theory) but net wins beyond that point would compel me.