(Created page with "Average Decision Quality is related to the notions of power-law world, model uncertainty, and the prediction horizon. In a world in which the outcomes of actions * Vary widely...") Tags: Visual edit apiedit |
No edit summary Tags: Visual edit apiedit |
||
| Line 5: | Line 5: | ||
You don't know in advance which of your decisions will turn out to be important. You therefore want your average decision to be high quality. Decision quality judgment criteria are domain dependent but might generally include things like |
You don't know in advance which of your decisions will turn out to be important. You therefore want your average decision to be high quality. Decision quality judgment criteria are domain dependent but might generally include things like |
||
* Being well calibrated with regard to evidence |
* Being well calibrated with regard to evidence |
||
| − | * Having good problem- |
+ | * Having good problem-heuristic fit |
* Minimizes loss of option value |
* Minimizes loss of option value |
||
* Is robust with regard to reference class uncertainty (i.e. is foxy) |
* Is robust with regard to reference class uncertainty (i.e. is foxy) |
||
Revision as of 08:20, 6 July 2016
Average Decision Quality is related to the notions of power-law world, model uncertainty, and the prediction horizon. In a world in which the outcomes of actions
- Vary widely in utility
- Have unknown causal structure and/or side-effects
- Are hard to predict
You don't know in advance which of your decisions will turn out to be important. You therefore want your average decision to be high quality. Decision quality judgment criteria are domain dependent but might generally include things like
- Being well calibrated with regard to evidence
- Having good problem-heuristic fit
- Minimizes loss of option value
- Is robust with regard to reference class uncertainty (i.e. is foxy)
- Takes all stakeholder incentives into account
- Cashes out in object level next actions
- Is consistent with terminal goals