There’s something to this school of thought but the logical counter-factual stuff just seems like such a dead end
I don’t understand why the better approach isn’t some kind of type-theoretic style answer - just say that a first-order decision algorithm takes in a problem description and returns a choice, and a second-order one takes in a problem and returns a first-order algorithm
Then say your decision algorithm is a second-order one that argmaxes over which first-order one performs best on the input problem. There’s no logical counter-possibility issues because you’re not having to imagine “what if my algorithm behaved differently”, the fact that it necessarily returns a particular first-order algorithm doesn’t contradict it being able to evaluate them
This is quite fun. From a very different starting point, the idea of Team Reasoning is another non-standard way of thinking about rationality, which also starts from the idea of "shared commitments":
> LDT (logical decision theory) seems to be an umbrella decision theory that only requires the use of logical counterfactuals, leaving the iteration type and updatelessness unspecified. So my understanding is that UDT1, UDT1.1, UDT2, FDT, and TDT are all logical decision theories.
>Because of Löb's theorem which is magic. I mean, I could try to explain it, but it would be a distracting detour and you'd need to be at least a little good at math to understand what the heck I was talking about.
I should have listened. Now after many online explanations and going deeper into the rabbit hole I'm not ever sure what a number is anymore.
The fact that you forgot the link doesn't matter, because as soon as you know that if you had the guide you'd understand Löb's theorem, you understand Löb's theorem.
A solution that predicts one-boxing is good. At least that's what I think. Curiously, everybody but me in my academic community seems to lean towards two-boxing, which is one of the rare cases when I have persistent troubles understanding others' intuitions. I mean, I can understand two-boxing as a recommendation in a one-off case, but for me any general solution to a decision problem needs to be acceptable and successful in arbitrary repeated applications.
I recommend reading the introduction for philosophers, by the way.
If I remember the setup correctly, my reasoning is that two boxing is rational,anti-coercive and implementable, while one boxing requires accepting some magical predictor and also opens up the door for a bunch of "arguments" that boil down to magic (violation of causality, or "trust this black box predictor" bro).
"Fuck you, I'll not play your game and take 1000" seems like a robust heuristic given a finite time on earth and a lot more charlatans trying to fool you than reliable predictors around
24 comments
[ 5.2 ms ] story [ 27.1 ms ] threadhttps://wikipedia.org/wiki/Socratic_method
[0] https://en.wikipedia.org/wiki/Socratic_dialogue
> A: Again, this is a theory of decisions about logic, not a theory about logical decisions. [...]
I don’t understand why the better approach isn’t some kind of type-theoretic style answer - just say that a first-order decision algorithm takes in a problem description and returns a choice, and a second-order one takes in a problem and returns a first-order algorithm
Then say your decision algorithm is a second-order one that argmaxes over which first-order one performs best on the input problem. There’s no logical counter-possibility issues because you’re not having to imagine “what if my algorithm behaved differently”, the fact that it necessarily returns a particular first-order algorithm doesn’t contradict it being able to evaluate them
https://www.tandfonline.com/doi/abs/10.1080/1000200309853874...
The paper underlying this discussion is here, by the way, and Eliezer Yudkowsky is an author:
https://arxiv.org/abs/1401.5577
[https://www.lesswrong.com/posts/QPhY8Nb7gtT5wvoPH/comparison...]
I should have listened. Now after many online explanations and going deeper into the rabbit hole I'm not ever sure what a number is anymore.
I recommend reading the introduction for philosophers, by the way.
"Fuck you, I'll not play your game and take 1000" seems like a robust heuristic given a finite time on earth and a lot more charlatans trying to fool you than reliable predictors around