I’m kind of thinking of doing a series of posts gently spelling out step by step the arguments for Bayesian decision theory. Part of this is for myself: I’ve read a while back Omohundro’s vulnerability argument, but felt there were missing bits that I had to personally fill in, assumptions I had to sit and think on before I could really say “yes, obviously that has to be true”. Some things that I think I can generalize a bit or restate a bit, etc.
So as much as for myself, to organize and clear that up, as for others, I want to do a short series of “How not to be stupid (given unbounded computational power)” In which in each each post I focus on one or a small number of related rules/principles of Bayesian Decision theory and epistemic probabilities, and gently derive those from the “don’t be stupid” principle. (Again, based on Omohundro’s vulnerability arguments and the usual dutch book arguments for Bayesian stuff, but stretched out and filled in with the details that I personally felt the need to work out, that I felt were missing.)
And I want to do it as a series, rather than a single blob post so I can step by step focus on a small chunk of the problem and make it easier to reference related rules and so on.
Would this be of any use to anyone here though? (maybe a good sequence for beginners, to show one reason why Bayes and Decision Theory is the Right Way?) Or would it be more clutter than anything else?
I have a similar plan—however, I don’t know when I’ll get to my post and I don’t think the material I wanted to discuss would overlap greatly with yours.
Can you characterize a bit more concretely what you mean, by zooming in on a tiny part of this planned work? It’s no easy task to go from common sense to math, and not shoot your both feet off in the process.
Basically, I want to reconstruct, slowly, the dutch book and vulnerability arguments, but step by step, with all the bits that confused me filled in.
The basic common sense rule that these are built on is “don’t accept a situation in which you know you automatically lose” (where “lose” is used to the same level of generality that “win” is in “rationalists win.”)
One of the reasons I like dutch book/vulnerability arguments is that each step ends up being relatively straightforward as to getting from that principle to the math. (Sometimes an additional concept needs to be introduced, not so much proven as much as defined and made explicit.)
I’m kind of thinking of doing a series of posts gently spelling out step by step the arguments for Bayesian decision theory. Part of this is for myself: I’ve read a while back Omohundro’s vulnerability argument, but felt there were missing bits that I had to personally fill in, assumptions I had to sit and think on before I could really say “yes, obviously that has to be true”. Some things that I think I can generalize a bit or restate a bit, etc.
So as much as for myself, to organize and clear that up, as for others, I want to do a short series of “How not to be stupid (given unbounded computational power)” In which in each each post I focus on one or a small number of related rules/principles of Bayesian Decision theory and epistemic probabilities, and gently derive those from the “don’t be stupid” principle. (Again, based on Omohundro’s vulnerability arguments and the usual dutch book arguments for Bayesian stuff, but stretched out and filled in with the details that I personally felt the need to work out, that I felt were missing.)
And I want to do it as a series, rather than a single blob post so I can step by step focus on a small chunk of the problem and make it easier to reference related rules and so on.
Would this be of any use to anyone here though? (maybe a good sequence for beginners, to show one reason why Bayes and Decision Theory is the Right Way?) Or would it be more clutter than anything else?
It’s got my upvote.
I have a similar plan—however, I don’t know when I’ll get to my post and I don’t think the material I wanted to discuss would overlap greatly with yours.
Can you characterize a bit more concretely what you mean, by zooming in on a tiny part of this planned work? It’s no easy task to go from common sense to math, and not shoot your both feet off in the process.
Basically, I want to reconstruct, slowly, the dutch book and vulnerability arguments, but step by step, with all the bits that confused me filled in.
The basic common sense rule that these are built on is “don’t accept a situation in which you know you automatically lose” (where “lose” is used to the same level of generality that “win” is in “rationalists win.”)
One of the reasons I like dutch book/vulnerability arguments is that each step ends up being relatively straightforward as to getting from that principle to the math. (Sometimes an additional concept needs to be introduced, not so much proven as much as defined and made explicit.)
Sounds interesting.