= 27d567bc2d48d9f40e9ccc20ea370b15)
Decius comments on How To Have Things Correctly - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (218)
Consider it as a basic optimisation algorithm. You evaluate the expected utility of your life if you buy the cloak and compare it to the expected utility when you do not buy the cloak. Answering the first question requires recursing to the optimisation of how you will live your life assuming you have the cloak. Because naturally the utility you wish to estimate is the utility of having the cloak assuming you use it well, not the utility of the cloak assuming you use the cloak to suffocate babies.
Here some of the principles of "Min-Max" come into play. That means not assuming that we have perfect control of all future actions but instead making a realistic assessment of how we will behave, with other parts being partially modeled as agents with different incentives in a game.
For instance if I happened to assign ten quintillion utility to me making muffins every day for the next sixty years and yet know that even if I buy a muffin tin I'm still not going to make muffins I will choose to not buy a muffin tin. This is the same reasoning that applies when I am playing chess and white opens with F3---I don't instantly move E5 under the expectation that white will move to G4, because that isn't likely to happen.
But if I know that if I buy a kettlebell I will do kettlebell swings regularly as well as show it off to my sports-nerd friends and family then I will choose to buy kettlebell.
Overall this means "Make purchases where you predict that you will change into (or already be) a person who gets a net benefit from having made that purchase".
If you assign some amount of utility to making muffins, and then choose not to make muffins, then you are either failing to optimize for utility, or assign some larger amount of utility to something which is mutually exclusive with making muffins.
What you have just described is correctly predicting your future actions, not deciding what your future actions will be for the purpose of reducing the negative effects of an action which would otherwise be a short-term benefit followed by a long-term harm. I predict that I would benefit greatly from buying a cloak, but I would be harmed more in the long term from having to keep and maintain a cloak that I rarely used. Without hacking myself, I would rarely use a cloak, and that is why I haven't purchased one.
I thought I saw someone who had managed to to hack their own future preferences for a purpose similar to my own, and was trying to confirm if that was the case and if so gather a data point from which to evaluate the expected results of imitation. It appears as though I just found someone who made the same decision based on simple evaluation of the expected results, without a prior attempt to alter the expected results.
And while my opinions regarding cloaks are literally true, my final goal isn't to decide whether to buy a cloak, but to practice in a low-risk environment like garb the skills required for high-risk behavior with multiple complicating factors.