Ping?
Pong?
You jumped online briefly this afternoon, but my ping came too late. I can [~only/[~primarily/jokingly-mostly]] assume that you’re immediate offlineness was due to the fact that you saw that I was online and that you were not at all interested in having another time sink of a Jaap monologue.
(…with [~a/b] being the syntax for: “I initially typed a, then realized that it was too strong of a statement, and therefore replaced it with b - yet leaving the a in (through said syntax) to indicate that I actually think about what I type before I submit it.”)
Man, you know me well… :) :)
I’m fighting a cold. Back in Cranbrook again.
Ah, right on. Once again disconnected from the reality we call the intertubes, and thus forced to exercise your own brain due to lack of blog stimuli. Damn, that must suck. Sorry to hear about your cold.
Your sympathy is not misplaced. I’m kicking its ass… :-)
Hahaha, nice. Hey, are you familiar with Dempster-Schafer Theory?
Not off-hand. Wikipediaing. Interesting. Oddly, reminds me of Kalman filters.
It appears that Bayes Theorem (which I’m mostly motivated to read up on from OB/LW reading) and the Kalman filter (which I’m motivated to read up on from Wiimote tracking) both come up alongside Dempster-Schafer.
Of course, given how Bayes lies much closer to the root of the scientific method than almost any other theorem, one could argue that Bayes is likely to show up in a lot of places.
Oh, reading further.
I just thought it was kinda neat how two fairly distinct interests of mine overlapped today.
Looks kind of like meta-probability…
Hard to formalize meta-notions (call them computational sentiments) that have formed in my brain after “grokking” the Kalman filter (in particular, the two step process of calibration and reinforcement) appear to linger subconsciously in my readings on rationality and AI. Maybe it’s a deceptive bias from my surfing patterns, but there appears to be an resurgence in converging of philosophy and math. Both in philosophy of thought, and philosophy of the universe. I’m sort of glad to see that, worried for a bit that a crystal-hippie interpretation of Godel would lead philosophers to drift away from math.
Did you catch the thing on Reddit the other day about “Why your friends have more friends than you do…”?
No, searching… Got it.
Jason and I were chatting about that article last weekend. Lots of things fall into the same category. For example, the proposition that the best thing you can do in traffic is to change lanes. Sample bias is very cool.
Hey, that’s a great article.
:-)
Thanks for sharing.
No problemo. Was thinking of you when we were discussing it this weekend.
Are you saying I’m biased? You’d better be… ;-)
LOL. Was just about to say the same regarding my statement!
Sample bias is one of these things that occupies a substantial part of my cognitive load, I’d say. Not like I’m always thinking about it, but it really seems to come up quite often.
When you say it’s part of your cognitive load, you mean mostly regarding to identifying it? Or to overcome it? And in other people, or in yourself, or in abstract?
At the moment, mainly identifying it. Not sure it’s possible, strictly, to overcome it, although one certainly can learn to recognize sample bias. For now, perhaps the best description of my involvement is that it makes me giggle every time I see it.
A particular interesting form of sampling bias comes up a lot in the Doomsday argument.
?
You are not familiar with the Doomsday argument? Oh boy, are you in for a treat. Let me find a good reference… http://www.anthropic-principle.com/primer1.html And the particular paper that explicitly talks about sampling bias (in this case, “the self-indication-assumption”): http://www.anthropic-principle.com/preprints/olum/sia.pdf
Reading…
Out of curiousity, is it sampling bias in particular that you are intruiged by, or other forms of bias as well?
Well, sampling bias seems to be at the bottom of many bias issues. What else do you have in mind?
This paper lists some well known ones that I found fun to learn about: http://www.singinst.org/upload/cognitive-biases.pdf Looking at http://en.wikipedia.org/wiki/List_of_cognitive_biases, I’m not sure if it’s obvious to say that sampling bias is at the bottom of many other bias issues (though I could probably be convinced that for most biases, one could define “sampling” such that it could be seen as a sampling bias.) That could be a fun exercise. …. Actually, I take my comment about the obviousness back.
That’s quite a list! I’d certainly be careful about saying that sampling bias was at the bottom of all of those.
If we’re talking about beliefs, and likelihood or “truthiness” of a belief is determined through something Bayes-like, then by definition the only way to skew the result is to skew the input. And since the input is a data-set, it’s by definition a sampling error. Unfortunately, for most topics of interest, the size of the input data set required to obtain objective degrees of validity is so large, that sampling bias is almost inevitable. The trick is to become efficient at avoiding the major errors. Treating every error as a sampling error would be impractical. It’s like saying that every computer is a one-directional single tape Turing machine, which is true - but my 8-core X86 is hella faster than the Turing machine… ;) You did totally give me a new insight though… thank you for that! I must think about it more, but the notion that bias is always a sampling issue if you dig deep enough really appeals to me. It’s like a Kalman filter - there’s a sampling error, and you try to overcome it.
Yeah. I’m curious about it, actually.
From ”I’m not sure if it’s obvious to say that sampling bias is at the bottom of many other bias issues.” to the complete opposite in less than ten MSN messages. Nice!
:-)
And only just now did I notice your comment: ”I’d certainly be careful about saying that sampling bias was at the bottom of all of those.” LOL, did we just flip stances?
I think I may be back on the sample bandwagon. The first bias I picked - the base rate fallacy - is certainly a sampling issue. Jason and I had actually discussed that one in terms of false-positive AIDS tests. :) :)
The space between “false positive AIDS tests” and “:):)” is too large for me to infer what you were laughing about. The sheer amount of (in-)appropriate comments I could make about AIDS make the whole thing rather intractable. ;)
lol. “The sheer amount of jokes I could make about AIDS make the whole thing rather intractable.” -> There’s sample bias in there somewhere, I’m sure.
“There’s sample bias in there somewhere, I’m sure.” -> Are you saying I have chlamydia? Snap! Padamtam tjing!
LOL
I think theoretical AI that operates on the fringe of both math and philosophy is very much about finding statistical representations for cognitive biases.
You know, though… For example: http://en.wikipedia.org/wiki/Denomination_effect A lot of cognitive biases seem more related to psychology than to statistical anomalies. It [~would/will] be really interesting to see what cognitive biases are exhibited by an AI.
Going down the bias list some more I’m further led to believe that indeed it’s all about sampling in the end. Look at it this way… For any finite IO port (i.e., no omniscient gods), there exists an upper limit on universe size and resolution beyond which the IO can’t avoid sampling bias. … Man, I think I just said something really deep and awesome…. Though it’ll probably be deep and awesome in the way that waking up the next day with a hangover makes deep and meaningful things from last night sound like moronicness. :)… I often feel there’s a correlation between how deep something sounds, and (provided it’s not crystal-worshipping ambigious trite) how utterly obivous it is once you see it.
Hey, interesting point.
At any rate, imagine an AI that plays finite games of Pacman (finite meaning there’s only N levels so you can “beat” the game - and yes, the Monte Carlo AIXI prompted this example…)
Then as long as you can increase the IO bandwidth (from only being able to see one ‘tile’ forward and backward, to seeing the whole level and positions of ghosts, to knowing all possible paths that ghosts could take, and simulating those (sort of like Deep Blue plays chess, once you use “ghosts-don’t-bang-their-heads-into-the-wall” heuristics)), you can reach a point where there’ll be no sampling bias anymore.
Note that not until Pacman can read the proverbial ghost’s mind (e.g., have access to the pseudo-random-number-generator that determines their paths) is all sampling bias removed entirely. There’s a sharp distinction between knowing what chess Grand Master’s are likely to do, and actually reading Kasparov’s mind.
Further note that even in the presence of reading minds, Pacman has not yet become God. Without the “game-is-beatable” guarantee, it seems possible for a particular ghost pattern to inevitably corner Pacman. In the abscence of super-natural powers (i.e., that transcend the original game, like walking through walls) it will be game over. That’s just a long way of saying that AI would still have to obey the laws of physics (the actual ones, not our current descriptions thereof.)
This feels related to my belief that even AI would be fundamentally throttled by its interaction with the real world. For example, it doesn’t have an inside track on physical theory, really, so has to do experimentation just like anyone else (although I think we can expect its theories to be more powerful).
Of course, I don’t have to mention to you that when I say IO port, I’m obviously referring to your eyes and ears and skin and so forth just as much as anything.
:-) Of course.
I point out that I don’t have to point that out, to momentarily relish in the fact that I actually happen to know somebody personally that I have that great privilege with….
Hey, I’m gonna run to the cafetaria and get a snack, you gonna be here for a bit, or leaving soon?
I’ll be here, I think.