Essay talk:LessWrong Bayes and Rationality

You forgot the part where that net negative effect on the world goes on to be a point of pride and a way to feel better about oneself. Don't get me wrong, there are not-goofy people on LW, and I lurk more than occasionally... but there comes a point when I stare into it and think, 'wow, this all exists because some people want to feel smarter than some other people.' ±KnightOfTL;DR^{longissimus non legeri} 23:56, 7 June 2012 (UTC)

If that impresses you, just wait 'til you see Mensa! Nebuchadnezzar (talk) 00:01, 8 June 2012 (UTC)

And they all have brains for their logos, as if that's a universal sign that it makes them super-clever. What's with that? ^postate 01:09, 8 June 2012 (UTC)

Everyone loves braaaaains. Dmytry (talk) 07:47, 8 June 2012 (UTC)

I've did some editing to make it sound less rant-y and more technical, as in, geared towards potential LW participants, and improved the brevity of how deficiencies are being exploited. On the AI issue, my overall impression is that they (SI) don't even care where the truth is, they just work to generate arguments that results in them getting the money, and none of them are curious enough to do anything beyond that. Nowhere on their proposed agenda did I see 'make sure the risk exists' or 'make sure we are not increasing the risk'. Beyond lying, there is this utter and complete disregard for truth. Dmytry (talk) 07:47, 8 June 2012 (UTC)

my thoughts

Dmytry I like this your essay. I think one place where the LW crowd go wrong, is in seeing Bayesianism as rationality itself, as opposed to a useful model, a useful theoretical tool, in our thinking about our own rationality (our metarationality), but not a model which is guaranteed to be perfectly correct. If you look in the literature, there is a fair amount of argument about whether and how Bayes' rule can actually be justified, and about potential alternative (albeit mostly related) formulations. So simplistically identifying rationality with Bayesianism is jumping ahead a fair bit, and in itself less than completely rational. (((Zack Martin)))^™ 08:41, 8 June 2012 (UTC)

Math like Bayes rule is not the problem. It is perfectly good when you apply it within the correct context. It is however not the rule for making partial inference graph full of cycles and inaccuracies perform the best. Dmytry (talk) 15:19, 8 June 2012 (UTC)

Bayes rule is perfectly valid as a piece of mathematics; the question is, how to justify that we should use Bayes rule instead of some alternate possible rule? It is difficult to prove Bayes rule - there are several different arguments for it, some of which are clearly flawed; many arguments put forward to support Bayes rule also support other rules, and the set of premises needed to produce Bayes rule uniquely are questionable. [1] (((Zack Martin)))^™ 23:40, 8 June 2012 (UTC)

Well it's not a question about such philosophical / foundation of mathematics point. It's a question of those guys not having a clue how to use Bayes rule in the first place. LW isn't some mathematical community where people would post questions and get answers (such as mathoverflow), so you can't really see direct evidence of them not knowing the mathematics, but I have very strong suspicion that pretty much anyone from there (EY specifically), given concrete problems, would simply fuck up math. If you wish, it's my prior, if I wish, it's result of years of observation that meta-addicts whom talk about math or physics can't actually solve jack shit and will fuck up anything but most trivial problems. Dmytry (talk) 10:02, 10 June 2012 (UTC)

As to Yudkowsky's arguments based on Kolmogorov complexity for MWI vs. Copenhagen — given that Kolmogorov complexity is uncomputable, we can never know with any degree of certainty what the Kolmogorov complexity of any non-trivial theory is, all we can produce is imprecise (and very often far from rigorous) approximations to it. Given, how can an argument based on estimated Kolmogorov complexity give us confidence in choosing MWI over Copenhagen? (((Zack Martin)))^™ 08:41, 8 June 2012 (UTC)

Well, the whole argument is pretty damn vague, I don't think they understand clearly what K-complexity is. But it is the case that you can expect a theory which is neatly a subset of another theory to have lower K-complexity. The issue is that K-complexity of a theory is not the point. He thinks that scientific process should work like K-complexity, by finding the simplest explanation for the world. If you seek simplest 'explanation', e.g. seek the simplest theory that contains yourself and your experience, that'll reasonably be something like an iterator over all possible theories, and you're back to square one, you just put complexity in your complexity so you can philosophize while you philosophize. If you find simplest predictor for your personal input, shut up and calculate style, that is guaranteed to be relevant. 'Explanations' in the philosophical sense are not what science seeks. edit: and then if you are seeking prediction of input, there isn't even much distinction between MWI and CI, that's why they're called interpretations. I guess one distinction would be that in MWI ideally the Born probabilities are a derived law, ala thermodynamics. That's it. This whole stuff about the worlds 'being out there' isn't interesting unless the laws would permit talking to those worlds. Dmytry (talk) 15:11, 8 June 2012 (UTC)