Main Page Contact Register Log In


You seem to either be arguing the algorithm might be found to be poorly calibrated, in which case I would respond: then change the algorithm, or you're arguing we shouldn't use algorithms to which I give you






To change the algorythm we would need someone whos job is to manage them and then we need some algorithm to decide whether another algorithm is well-calibrated or not.

There is a hybrid system if in a democracy voters are principled voters in that they value that deciders publicly and explictly outline their decision procedures. This doesn't need a change in the state mechanics. If you argue that forcing such a mechanic will bring about such a focus on the process over single outcomes, I would say it is not clear that processes are easier to pick than outcomes. Once you get a good process you benefit more but it is harder to get there. Or there is an equivalent problem that once a metric has been fixed challenging it is hard if there appears a need to change it.

You are not even arguing that that some particular method of picking algorithms would consistently pick good outcomes. A attitude of "math was involved, it must be good" can easily lead to be systematically bad. The content of the math should matter.
55%
Fwiffo
stars0
Replies (1)


This link: http://lesswrong.com/lw/3gv/statistical_prediction_rules_outperform_expert/
65%
FrameBenignly
stars0
Replies (0)