Public Pastes
Pastebin PRO Accounts AUTUMN SPECIAL! For a limited time only get 40% discount on a LIFETIME PRO account! Offer Ends Soon!
SHARE
TWEET

Untitled

a guest Jan 1st, 2011 74 Never
  1. [Jan 01 2011, 04:00:21 UTC] <Ralith> moshez: wholesale disregard for copyright is making my kindle great!
  2. [Jan 01 2011, 04:00:22 UTC] <Ralith> :D
  3. [Jan 01 2011, 04:01:10 UTC] <HHELLD> “love for dead trees and cheap costly information” all over in other places.
  4. [Jan 01 2011, 04:01:58 UTC] <moshez> ralith: even disregarding copyright yourself doesn't make your kindle "great" as it could be
  5. [Jan 01 2011, 04:02:14 UTC] <moshez> e.g., it's still more painful than it should be to load one's own text on it
  6. [Jan 01 2011, 04:03:17 UTC] <HHELLD> ereaders… like, a 7-years-old PDA device with linux and fbreader on it is able to perform better than defective-by-design e-readers stuff.
  7. [Jan 01 2011, 04:04:33 UTC] <Ralith> moshez: er, no it's not
  8. [Jan 01 2011, 04:04:43 UTC] <Ralith> moshez: cp ~/mytext /mnt/kindle/
  9. [Jan 01 2011, 04:04:44 UTC] <moshez> hhelld: yep, the defective-by-design is a product of copyright
  10. [Jan 01 2011, 04:04:48 UTC] <Ralith> moshez: how is that painful?
  11. [Jan 01 2011, 04:04:53 UTC] <moshez> rallith: formats
  12. [Jan 01 2011, 04:05:06 UTC] <Ralith> moshez: plaintext is painful? PDF is painful? ePub is painful?
  13. [Jan 01 2011, 04:05:10 UTC] <moshez> no.
  14. [Jan 01 2011, 04:05:18 UTC] <Ralith> sooooooooooo
  15. [Jan 01 2011, 04:05:19 UTC] <moshez> but converting to them is :)
  16. [Jan 01 2011, 04:05:21 UTC] <HHELLD> Also, my current information on the copyright story is that there should be some better method for monetizing informational works (*any* works, actually), but none currently known.
  17. [Jan 01 2011, 04:05:30 UTC] <Ralith> moshez: what the hell format do you put *your* text in? O.o
  18. [Jan 01 2011, 04:05:34 UTC] <HHELLD> Ralith: fb2, html?
  19. [Jan 01 2011, 04:05:37 UTC] <moshez> ralith: HTML
  20. [Jan 01 2011, 04:05:43 UTC] <moshez> always and forever
  21. [Jan 01 2011, 04:05:46 UTC] <moshez> <3 HTML
  22. [Jan 01 2011, 04:06:03 UTC] <moshez> if a man and a data format were allowed to marry, I'd marry HTML
  23. [Jan 01 2011, 04:06:26 UTC] <moshez> she has her flaws, but I'm in love :-D
  24. [Jan 01 2011, 04:06:29 UTC] <HHELLD> … Also, a scary thought: information that's really needed can be successfully monetized in any way (model) at all.
  25. [Jan 01 2011, 04:06:57 UTC] <Ralith> moshez: ebook readers can easily or do support html, conversion from HTML to all of the above formats is trivial, etc
  26. [Jan 01 2011, 04:07:24 UTC] <moshez> ralith: I don't want to convert. I want to "cp" :)
  27. [Jan 01 2011, 04:07:40 UTC] <Ralith> moshez: then read the first half of the sentence.
  28. [Jan 01 2011, 04:07:41 UTC] <Ralith> :P
  29. [Jan 01 2011, 04:07:47 UTC] <moshez> they can
  30. [Jan 01 2011, 04:07:51 UTC] <moshez> kindle can't
  31. [Jan 01 2011, 04:08:02 UTC] <moshez> because I can't put my own software on kindle
  32. [Jan 01 2011, 04:08:06 UTC] <moshez> because of DRML
  33. [Jan 01 2011, 04:08:08 UTC] <moshez> DRM
  34. [Jan 01 2011, 04:08:12 UTC] <moshez> because of copyright
  35. [Jan 01 2011, 04:08:18 UTC] <HHELLD> Didn't anyone yet install Linux on kindle yet?
  36. [Jan 01 2011, 04:08:26 UTC] <HHELLD> s/ yet / /
  37. [Jan 01 2011, 04:08:32 UTC] <moshez> ralith: I'm not saying that copyright is net loss, just that it isn't pure gain
  38. [Jan 01 2011, 04:08:51 UTC] <Ralith> moshez: copyright has *nothing to do* with the issue
  39. [Jan 01 2011, 04:08:57 UTC] <moshez> and so to calc. the net benefit, one must add positives and negatives
  40. [Jan 01 2011, 04:09:00 UTC] <Ralith> moshez: you can put your own software on the kindle, and many other ebook readers.
  41. [Jan 01 2011, 04:09:03 UTC] <moshez> ACTION shrugs
  42. [Jan 01 2011, 04:09:14 UTC] <moshez> I stopped being interested in the discussion
  43. [Jan 01 2011, 04:09:51 UTC] <Ralith> easier than supporting a point!
  44. [Jan 01 2011, 04:10:09 UTC] <moshez> indeed, it is easier.
  45. [Jan 01 2011, 04:11:19 UTC] <moshez> I know EY says that rationalists can't agree to disagree, but I disagree :) if the expected benefit of the agreement is less than the cost of coming to agreement, the rational thing to do is to agree to disagree
  46. [Jan 01 2011, 04:12:42 UTC] <moshez> here, this is a fairly minor point. it seems like convincing you won't be easy. I don't care enough about it, and the expected utility on changing my mind on the effect of copyright on ereaders is small, smaller than the utility of my time.
  47. [Jan 01 2011, 04:12:52 UTC] <HHELLD> Solving such argument requires bringing up common (objective, intersubjective) evidence for the information in personal (subjective) knowledge.
  48. [Jan 01 2011, 04:13:06 UTC] <HHELLD> Costly.
  49. [Jan 01 2011, 04:14:16 UTC] <HHELLD> “Agree to disagree” seems somewhat inappropriate to me even in such case, though. I suspect something like “decreasing own (both's) certainty” would be better, but I'm not yet sure.
  50. [Jan 01 2011, 04:15:58 UTC] <moshez> hhelld: why does that seem inappropriate? the hypothetical rationalist and I agree to say "we disagree, but we are not trying to convince each other further". isn't that what's commonly thought of as "agree to disagree"?
  51. [Jan 01 2011, 04:19:26 UTC] <HHELLD> moshez: (This is all unfinished-thoughs for now, but) given that both participants are sufficiently rational (quite a hypothesis, though) yet they have conflicting beliefs on some point means that both had evidence supporting those conflicting points, and combination of this evidence should (if I'm correct here) result in a belief somewhere in-between those points (and, likely, with additional orthogonal information).
  52. [Jan 01 2011, 04:19:32 UTC] ***OneWhoFrogs has quit (Quit: OneWhoFrogs)
  53. [Jan 01 2011, 04:20:22 UTC] <moshez> hhelld: yes, that's correct (let's grant the hypothetical)
  54. [Jan 01 2011, 04:20:42 UTC] <moshez> but, let's say the cost of laying out all that evidence is -10 utilons to both
  55. [Jan 01 2011, 04:21:03 UTC] <moshez> and the maximal benefit of having completely-accurate view of that issue is +5 utilons to both
  56. [Jan 01 2011, 04:21:40 UTC] <moshez> then they net at best -5 utilons from coming to agreement vs. agreeing to disagree
  57. [Jan 01 2011, 04:22:06 UTC] <HHELLD> moshez: …“agree to disagree”, as I understand, means “both keep own previous beliefs”. Which seems worse than both updating beliefs towards the middle given evidence of some rationalist stating a view opposite to their own.
  58. [Jan 01 2011, 04:22:34 UTC] <moshez> hhelld: in my example, both keep their previous beliefs, and yet end up better for it
  59. [Jan 01 2011, 04:22:39 UTC] <HHELLD> (with consideration of own subjective probability of other being a rationalist and some other bits of knowledge)
  60. [Jan 01 2011, 04:23:04 UTC] <moshez> hhelld: they can, I guess, make a tiny update towards each other
  61. [Jan 01 2011, 04:23:11 UTC] <moshez> but they'll still be far apart
  62. [Jan 01 2011, 04:23:40 UTC] <HHELLD> Well, that “tiny” part is quite more interesting and possibly valuable (I expect similar situations to be rather massive).
  63. [Jan 01 2011, 04:23:47 UTC] <moshez> granting that I think ralith is reasonable, I did slightly update my belief on "copyright causes ereaders to be crap"
  64. [Jan 01 2011, 04:24:07 UTC] <HHELLD> I.e. seeing if it's actually tiny; or if there's better updating than updating just that belief.
  65. [Jan 01 2011, 04:24:08 UTC] <moshez> but coming to complete agreement with him (her?) seems to have negative utility for me
  66. [Jan 01 2011, 04:24:35 UTC] <HHELLD> moshez: I'm certainly not suggesting complete agreement.
  67. [Jan 01 2011, 04:25:16 UTC] <moshez> ok
  68. [Jan 01 2011, 04:25:17 UTC] <HHELLD> I alraedy suggested something like “being less certain in own belief”, but I'm not sure what exactly does that mean and whether it's any way different from updating that belief towards 0.5.
  69. [Jan 01 2011, 04:25:26 UTC] <moshez> then I think we're saying the same thing :)
  70. [Jan 01 2011, 04:27:04 UTC] <HHELLD> moshez: I still have some uncertainty in this [meta-]topic (or, rather, unfinished computation/thinking), that's why I'm not looking for “saying the same thing” but for further ideas to make the map more elaborate.
  71. [Jan 01 2011, 04:27:26 UTC] <moshez> ah ok
  72. [Jan 01 2011, 04:28:58 UTC] <moshez> do you remember the post on lw saying that (roughly) lw sucks for instrumental rationality a few weeks ago?
  73. [Jan 01 2011, 04:29:22 UTC] <moshez> I think the crucial difference here is instrumental vs. epistemic rationality
  74. [Jan 01 2011, 04:29:23 UTC] <HHELLD> No, but I suspect I have it opened and queued for reading.
  75. [Jan 01 2011, 04:29:50 UTC] <moshez> from an epistemic rationality viewpoint, having correct beliefs about X is a terminal value
  76. [Jan 01 2011, 04:30:05 UTC] <HHELLD> http://lesswrong.com/lw/2po/selfimprovement_or_shiny_distraction_why_less/ ?
  77. [Jan 01 2011, 04:30:09 UTC] <moshez> right
  78. [Jan 01 2011, 04:30:38 UTC] <moshez> now, in general, I enjoy having correct beliefs about most Xs
  79. [Jan 01 2011, 04:31:10 UTC] <moshez> but not all (e.g., I don't care about having correct beliefs about who Paris Hilton is dating)
  80. [Jan 01 2011, 04:31:32 UTC] <moshez> I also have other things in my utility function
  81. [Jan 01 2011, 04:31:46 UTC] <HHELLD> moshez: Suppose I have only pleasure in my utility function ^_^
  82. [Jan 01 2011, 04:31:55 UTC] <moshez> for some of them, it's conceivable that knowing who PH is dating might be important
  83. [Jan 01 2011, 04:33:00 UTC] <HHELLD> Anyway, a different way to put this topic is “finding an [instrumentally] optimal course of action given sidagreement of two rationalists (given some meta-info like expected probability that other party is a rationalist unbiased in this particular topic)”.
  84. [Jan 01 2011, 04:33:24 UTC] <moshez> hrm
  85. [Jan 01 2011, 04:33:37 UTC] <moshez> I'm not sure what you mean by side-agreement
  86. [Jan 01 2011, 04:33:44 UTC] <HHELLD> s/sidagr/disagr/
  87. [Jan 01 2011, 04:33:48 UTC] <HHELLD> Mistypings :(
  88. [Jan 01 2011, 04:34:05 UTC] <moshez> no worries, just thought it might be a new notion :)
  89. [Jan 01 2011, 04:35:28 UTC] <moshez> ok, then here's some notion that I think might be important here
  90. [Jan 01 2011, 04:35:54 UTC] <moshez> let's assume there's a coin, and both our priors is that it's fair
  91. [Jan 01 2011, 04:36:14 UTC] <HHELLD> A possible (plausible!) model of the example above: given that you are generally anti-copyright (as I am, though), and that you have preference of reading anti-copyright-themed publications (like defectivebydesign) (“preaching to the converted”. It's still a quite unfinished topic to me*. An inrteresting one, too), there's significant  probability that your evidence is biased more towards problematicness of ereaders.
  92. [Jan 01 2011, 04:36:28 UTC] <moshez> and there's also a third person, Foo, who has seen this coin
  93. [Jan 01 2011, 04:36:41 UTC] <HHELLD> * related to the topic of “preaching to the converted” is eagerness of local rationalists to read more of MoR. ^_^
  94. [Jan 01 2011, 04:37:00 UTC] <moshez> he saw it come up Heads once, for some reason
  95. [Jan 01 2011, 04:37:13 UTC] <moshez> now you've seen it do 50-50 heads-tails 100 times
  96. [Jan 01 2011, 04:37:24 UTC] <moshez> I've seen it do 5-5 10 times
  97. [Jan 01 2011, 04:37:41 UTC] <moshez> both our posteriors are 0.5
  98. [Jan 01 2011, 04:37:50 UTC] <HHELLD> * and yet aother note is: I think I remember some point on lw about something in-between (optimal) sharing all the evidence and sharing priors. Don't remember what it was, though :(
  99. [Jan 01 2011, 04:37:52 UTC] <moshez> Foo, however, should be more convinced by you
  100. [Jan 01 2011, 04:39:15 UTC] <moshez> I'm generally pro-weaker-copyright rather than outright anti-copyright (for the record)
  101. [Jan 01 2011, 04:39:45 UTC] <HHELLD> “should be more convinced by you” — given the possibility to state the size and completeness (or unbiasedness) of the evidence.
  102. [Jan 01 2011, 04:40:28 UTC] <moshez> but in general, I agree -- ralith should slightly update his beliefs on "ereaders are less awesome than they could be because of copyright", but not completely towards mine, because I might be biased
  103. [Jan 01 2011, 04:40:28 UTC] <HHELLD> “pro-weaker-copyright” — there should be different solution to the underlying problem but since there's none (yet), complete lack of copyright for now would be even more disastrous indeed. Or something like that.
  104. [Jan 01 2011, 04:41:44 UTC] <HHELLD> moshez: in this example, do you own a e-reader, or kindle in particular? If not, then your evidence should be expected to be less massive/certain :)
  105. [Jan 01 2011, 04:41:53 UTC] <moshez> hhelld: "copyright law was the right solution in the 1800s. it should be rewritten from the ground up to ensure we correctly balance between the need of society to produce creative works, and the limitations we put on pepole"
  106. [Jan 01 2011, 04:42:07 UTC] <moshez> I don't own one, but I have investigated buying one
  107. [Jan 01 2011, 04:42:35 UTC] <moshez> and decided against it, obviously
  108. [Jan 01 2011, 04:43:06 UTC] <moshez> my investigations included reading about it, as well as playing with friends' kindles and asking them questions about it
  109. [Jan 01 2011, 04:43:06 UTC] <HHELLD> My experience shows that investigation is often quite impricese compared to actually owning, but whatever, that's quite deep (and that experience is quite more relevant to expected convenience of interfaces rather than other points).
  110. [Jan 01 2011, 04:43:14 UTC] <moshez> I agree.
  111. [Jan 01 2011, 04:43:26 UTC] <moshez> I would probably have better conclusions if I owned one
  112. [Jan 01 2011, 04:43:56 UTC] <moshez> I don't rate the utility of the conclusions being better as more than the 150$+ it would take to own one
  113. [Jan 01 2011, 04:44:04 UTC] <HHELLD> I can also note that close-to-optimal use can be reached within month but rarely withit couple hours of playing with.
  114. [Jan 01 2011, 04:44:52 UTC] <moshez> I agree on that too
  115. [Jan 01 2011, 04:45:18 UTC] <HHELLD> Not much more to say on topic itself, I suppose; on meta-topic it could be quite useful to lay out some more formal stuff (and then put it onto lw).
  116. [Jan 01 2011, 04:45:46 UTC] <moshez> oh, so back to my example with Foo
  117. [Jan 01 2011, 04:46:02 UTC] <moshez> in advance, Foo doesn't know if he's talking to someone like me or someone like you
  118. [Jan 01 2011, 04:46:11 UTC] ***PatrickRobotham has quit (Ping timeout: 240 seconds)
  119. [Jan 01 2011, 04:47:02 UTC] <moshez> so he would need an expectation on the "quality of the average other rationalist" to properly come to a correct conclusion on how much it's worth discussing the issue with someone else
  120. [Jan 01 2011, 04:47:19 UTC] <moshez> discussing it with you is worth more than discussing it with me
  121. [Jan 01 2011, 04:47:25 UTC] <HHELLD> moshez: rather than “doesn't know” it would be better to denote some Foo's probability [distribution] regarding it (as an input value).
  122. [Jan 01 2011, 04:47:43 UTC] <moshez> isn't that what I said?
  123. [Jan 01 2011, 04:48:04 UTC] <HHELLD> Yes, it is what you said while I was saying what I said ^_^
  124. [Jan 01 2011, 04:48:04 UTC] <moshez> oh, I meant "expectation" in the technical probability theory sense of the word
  125. [Jan 01 2011, 04:48:10 UTC] <moshez> ok :)
  126. [Jan 01 2011, 04:49:04 UTC] <moshez> so that's something that should belong in a formal treatment of this issue
  127. [Jan 01 2011, 04:49:41 UTC] <moshez> hhelld: if I were writing this, I'd call it "Rationally Agreeing to Disagree", just to tweak EY's nose
  128. [Jan 01 2011, 04:49:50 UTC] <HHELLD> Furthermore, we mentioned such things as “probability that other party is a rationalist”, “… is unbiased regarding this particular evidence”, “size of other party's evidence”.
  129. [Jan 01 2011, 04:50:00 UTC] <moshez> right
  130. [Jan 01 2011, 04:50:10 UTC] <HHELLD> "Rationally Agreeing to Disagree" — I'll note that ^_^
  131. [Jan 01 2011, 04:50:32 UTC] ***Adelene has quit (Quit: Leaving.)
  132. [Jan 01 2011, 04:50:57 UTC] <HHELLD> moshez: … what of those things I listed above come into which Foo's expectations, and which of them can be easily updated, is yet undecided question.
  133. [Jan 01 2011, 04:51:34 UTC] <moshez> another thing that should be considered is "How important is it to have correct beliefs"
  134. [Jan 01 2011, 04:52:03 UTC] <HHELLD> given a honest rationalist, it's probably easy to ask size of evidence; bit more/harder to find any reasons for thinking about biased/unbiased.
  135. [Jan 01 2011, 04:52:20 UTC] <moshez> e.g., it's important to have correct beliefs on the temperature of the stove (or you might get burned) but not that important to have correct beliefs on PH's beau
  136. [Jan 01 2011, 04:52:35 UTC] <moshez> unless you're a gossip column writer, I guess
  137. [Jan 01 2011, 04:53:19 UTC] <HHELLD> "How important is it to have correct beliefs" — moreover, value of time/mental resources to spend on it. Easiest for a general topic is to note some easiest ways and leave out anything that's noticeably hard :)
  138. [Jan 01 2011, 04:54:00 UTC] <moshez> weird synchronicity
  139. [Jan 01 2011, 04:54:18 UTC] <moshez> just happened to misclick in my browser, and found http://lesswrong.com/lw/go/why_truth_and/
  140. [Jan 01 2011, 04:55:04 UTC] <HHELLD> (i.e. I'm talking about some unformalized level of “general interest” which denotes some similar level of value of knowing …)
  141. [Jan 01 2011, 04:56:59 UTC] <gwern> moshez: but don't you care about being correctly calibrated if you were to make predictions about who hilton is dating?
  142. [Jan 01 2011, 04:57:16 UTC] <moshez> gwern: I do care about calibration
  143. [Jan 01 2011, 04:57:17 UTC] <HHELLD> “expected value of information / expected value of time(mental resources)”
  144. [Jan 01 2011, 04:57:23 UTC] <gwern> (just finished _Summer Wars_. Not as good as I hoped it'd be.)
  145. [Jan 01 2011, 04:57:29 UTC] <moshez> but if there is a mountain of evidence on who she's dating
  146. [Jan 01 2011, 04:57:36 UTC] <gwern> moshez: so there is a sense in which you care about correct beliefs about her
  147. [Jan 01 2011, 04:57:44 UTC] <HHELLD> gwern: why did you expect it to be good?
  148. [Jan 01 2011, 04:57:48 UTC] <moshez> yes, there is one
  149. [Jan 01 2011, 04:58:02 UTC] <gwern> HHELLD: well, everyone and the reviewers were saying it was good
  150. [Jan 01 2011, 04:58:02 UTC] <moshez> but I don't care about improving my discrimination
  151. [Jan 01 2011, 04:58:13 UTC] <moshez> so if you disagreed with me about who she's dating
  152. [Jan 01 2011, 04:58:49 UTC] <moshez> and it seemed like you had some evidence, but it would take me a long time to ferret out the strength of evidence on which you base your assumptions
  153. [Jan 01 2011, 04:58:54 UTC] <HHELLD> If I had some belief on such an irrelevant topic and see someone disagree I'd just drop my belief and go to almost complete uncertainty.
  154. [Jan 01 2011, 04:59:28 UTC] <moshez> I'd update slightly towards you and agree to disagree, although you think she's dating Arnold Schwarzenneger with 0.99 probability
  155. [Jan 01 2011, 04:59:47 UTC] <moshez> I'd update to 0.001, and leave it that (maybe)
  156. [Jan 01 2011, 04:59:58 UTC] <HHELLD> (Hm. Not totally complete uncertainty because I'd have at least two noticeably likely options, not none)
  157. [Jan 01 2011, 05:00:05 UTC] <moshez> hhelld: I'm pretty sure she's not dating arnold
  158. [Jan 01 2011, 05:00:41 UTC] <moshez> if you offered me a bet on that, with a nice payoff, I'd take your dollar in return for giving you 1k$ if she is
  159. [Jan 01 2011, 05:01:24 UTC] <moshez> (assume that after you offer me the bet, I don't have time to seek out more evidence)
  160. [Jan 01 2011, 05:01:29 UTC] <HHELLD> … more interestingly, if I could drop even that information for the sake of optimization (and go to *complete* uncertainty), I'd do that ^_^
  161. [Jan 01 2011, 05:01:55 UTC] <moshez> but right now, the utility of me knowing who she's dating is close to 0, so I wouldn't have the conversation
  162. [Jan 01 2011, 05:02:07 UTC] <moshez> except as an example of the meta-principle in question
  163. [Jan 01 2011, 05:03:04 UTC] <HHELLD> That meta-principle, if overlayed with optimization (i.e. some reasons to prefer one topic over another), will likely result in indifference too.
  164. [Jan 01 2011, 05:03:49 UTC] <HHELLD> If you're a super-brain that can limitlessly spawn thinkers to find out anything, then ofc there would be no reason for such an optimization‘…
  165. [Jan 01 2011, 05:04:07 UTC] <moshez> right
  166. [Jan 01 2011, 05:04:45 UTC] <moshez> but if you're maximizing your pleasure, and have a choice about whether to discuss the evidence of PH's love life or reading an interesting paper
  167. [Jan 01 2011, 05:04:54 UTC] <HHELLD> But generally, curiosity makes more sense for things that are quite out of your map, not things that are in area of your map but are just details.
  168. [Jan 01 2011, 05:05:03 UTC] <moshez> well
  169. [Jan 01 2011, 05:05:09 UTC] <moshez> I'd disagree on that
  170. [Jan 01 2011, 05:05:19 UTC] <moshez> there are some details I'm passionate about :)
  171. [Jan 01 2011, 05:05:50 UTC] <HHELLD> Maybe, but that's irrelevant to what even EY would call a “virtue”.
  172. [Jan 01 2011, 05:05:57 UTC] <moshez> hrm
  173. [Jan 01 2011, 05:06:03 UTC] <moshez> I'm not even sure about that
  174. [Jan 01 2011, 05:06:27 UTC] <HHELLD> It's somewhat unimportant anyway :)
  175. [Jan 01 2011, 05:06:27 UTC] <moshez> EY thinks it's virtuous to care about the tiny details of arguments about FAI, I assume
  176. [Jan 01 2011, 05:06:57 UTC] <moshez> ok :)
  177. [Jan 01 2011, 05:07:38 UTC] <HHELLD> I'd suggest to go to the point before you referencig that post and continuing from it.
  178. [Jan 01 2011, 05:10:17 UTC] <moshez> so here's another thing that we should think about in formalizing
  179. [Jan 01 2011, 05:10:34 UTC] <moshez> there's something, Foo, we both care about
  180. [Jan 01 2011, 05:10:49 UTC] <moshez> you think that Foo is true with probability 0.8001
  181. [Jan 01 2011, 05:10:57 UTC] <HHELLD> “Foo” is already used identifier :)
  182. [Jan 01 2011, 05:11:02 UTC] <moshez> sorry
  183. [Jan 01 2011, 05:11:03 UTC] <moshez> Bar
  184. [Jan 01 2011, 05:11:28 UTC] <moshez> you think Bar is true with 0.8001, I think it's true with 0.7999
  185. [Jan 01 2011, 05:11:56 UTC] <HHELLD> And, for the sake of topic, I suggest starting point as “there something, ‘Bar’, which we don't mind spending some minutes talking about”.
  186. [Jan 01 2011, 05:12:03 UTC] <moshez> ok
  187. [Jan 01 2011, 05:12:28 UTC] <HHELLD> (formalizing the value at least into “some minutes”)
  188. [Jan 01 2011, 05:12:35 UTC] <moshez> it's obvious that if we'll update towards each other, we'll be somewhere near 0.8
  189. [Jan 01 2011, 05:13:08 UTC] <HHELLD> With such numbers you're already somewhere near 0.8 >.>
  190. [Jan 01 2011, 05:13:12 UTC] <moshez> it might actually not be that important to come to agreement, even if we do care about Bar
  191. [Jan 01 2011, 05:13:14 UTC] <moshez> exactly
  192. [Jan 01 2011, 05:14:25 UTC] <HHELLD> Indeed it's not important; what is interesting is a way to most efficiently spend few minutes updating each other's probabilities :)
  193. [Jan 01 2011, 05:14:40 UTC] <HHELLD> (“updating each other's probabilities” sounds… intimate ^_^)
  194. [Jan 01 2011, 05:15:21 UTC] <moshez> it would be hard to be efficient, though. You have your set A of evidence, I have B. |A symmetric difference B| << |A intersect B|
  195. [Jan 01 2011, 05:15:47 UTC] <moshez> so you sending me A and me sending you B would be horribly redundant
  196. [Jan 01 2011, 05:16:28 UTC] <HHELLD> Uhh… It's hard to be “very efficient”, ofc. But I'm talking about “being much closer to most efficient possible” (by knowing some methods/ideas/howto).
  197. [Jan 01 2011, 05:16:59 UTC] <moshez> I think I meant "it's hard to avoid being horribly inefficient" (in a typical case)
  198. [Jan 01 2011, 05:17:37 UTC] <HHELLD> Yup; but anything better than “not updating probabilities at all” would be good already.
  199. [Jan 01 2011, 05:18:41 UTC] <moshez> well, depends. what if the reason we're this close is because we spent several hours discussing the evidence?
  200. [Jan 01 2011, 05:19:11 UTC] <moshez> in that case, we can agree to "slightly" disagree -- "OK, it looks like we're close enough on this issue."
  201. [Jan 01 2011, 05:19:45 UTC] <moshez> so my central thesis here is that it's *always* rational to agree to disagree -- the only question is "how much probability updating is enough?"
  202. [Jan 01 2011, 05:20:07 UTC] <HHELLD> “spent several hours” is already out of the topic :). Or, it's a different question on “how to efficiently spend several hours” (and decide that you don't value further updating enough to continue).
  203. [Jan 01 2011, 05:20:25 UTC] <moshez> let's assume we spent those hours as efficiently as possible
  204. [Jan 01 2011, 05:20:47 UTC] <HHELLD> I'm still proposing to concentrate on that vast amount of cases where it's not (likely) valuable to spend more than few minutes.
  205. [Jan 01 2011, 05:21:09 UTC] <moshez> that's reasonable
  206. [Jan 01 2011, 05:22:05 UTC] <moshez> so you know the story about "lady, we already figured out what you are, now we're just quibbling about the price?"
  207. [Jan 01 2011, 05:22:21 UTC] <HHELLD> … Somewhere up there I suggested points about stating the probability, amount of evidence and possible facts that's relevant to biasedness of the evidence.
  208. [Jan 01 2011, 05:22:31 UTC] <HHELLD> Nope, I probably don't.
  209. [Jan 01 2011, 05:22:54 UTC] <moshez> someone asks a woman "would you sleep with me for a million dollars?" "yes"
  210. [Jan 01 2011, 05:23:04 UTC] <HHELLD> * http://lesswrong.com/lw/wo/the_mechanics_of_disagreement/ (possibly relevant)
  211. [Jan 01 2011, 05:23:13 UTC] <moshez> "would you sleep with me for a hundred?" "what do you think I am, a prostitute?"
  212. [Jan 01 2011, 05:23:41 UTC] <moshez> in the same vein, my point is that it's always rational to agree to disagree at some point
  213. [Jan 01 2011, 05:23:49 UTC] <moshez> so the analysis should be of the cost/benefit :)
  214. [Jan 01 2011, 05:24:59 UTC] <HHELLD> Again, ofc the result would be difference in probabilities; I'm just arguing against “keeping the probabilities same as before the discussion” and looking for formalized ideas on how can they be properly updated.
  215. [Jan 01 2011, 05:26:29 UTC] <moshez> I see
  216. [Jan 01 2011, 05:27:32 UTC] <HHELLD> Also, I still fail to find that reference I remember about some more correct way than sharing the priors (which might just lead to the wrong direction) yet more optimal than sharing all the evidence.
  217. [Jan 01 2011, 05:28:11 UTC] <gwern> log?
  218. [Jan 01 2011, 05:28:22 UTC] <HHELLD> gwern: ?
  219. [Jan 01 2011, 05:29:15 UTC] <gwern> likelihood ratio
  220. [Jan 01 2011, 05:30:01 UTC] ***PatrickRobotham has joined #lesswrong
  221. [Jan 01 2011, 05:30:02 UTC] <HHELLD> … And, simplest but still interesting point: given disagreement, is it more appropriate to shift probability [slightly] towards the other's distribution (assigning more to other party's most probable expectation), or more towards the unitary distribution (less certainty in anything)?
  222. [Jan 01 2011, 05:30:28 UTC] <HHELLD> gwern: yep, right, seems like it.
  223. [Jan 01 2011, 05:33:27 UTC] <moshez> well
  224. [Jan 01 2011, 05:34:04 UTC] <moshez> it would be silly if we disagreed 0.7 (me) to 0.8 (you), and I shifted towards 0.5
  225. [Jan 01 2011, 05:34:44 UTC] <HHELLD> not necessarily; or, rather, AFAIU that depends on the underlying semantics.
  226. [Jan 01 2011, 05:37:18 UTC] <moshez> hhelld: our priors for the coin were "there's a .5 chance it's fixed". You saw it land heads-up X time, and shifted to 0.8. I saw it land heads-up Y(<X) and shifted to 0.7
  227. [Jan 01 2011, 05:37:40 UTC] <moshez> maybe some of the experiments were common and some weren't
  228. [Jan 01 2011, 05:38:04 UTC] <moshez> it's hard to tell which are which, so knowing the exact Bayes optimum is hard
  229. [Jan 01 2011, 05:38:24 UTC] <moshez> but shifting back towards 0.5 is clearly away from optimum
  230. [Jan 01 2011, 05:39:08 UTC] <HHELLD> back towards 0.5 from 0.8, too, you mean?
  231. [Jan 01 2011, 05:39:48 UTC] <moshez> no
  232. [Jan 01 2011, 05:39:58 UTC] <moshez> I mean for me, it would be silly
  233. [Jan 01 2011, 05:40:24 UTC] <moshez> clearly, if both of us update, at least one of us will update towards 0.5
  234. [Jan 01 2011, 05:46:22 UTC] <HHELLD> “… Overconfidence is probably a large part of the reason for persistent disagreement, each party ranking his own knowledge above that of the other. …” [q]
  235. [Jan 01 2011, 05:46:38 UTC] <HHELLD> (relevant but very far from complete comclusion)
  236. [Jan 01 2011, 05:58:07 UTC] <HHELLD> http://lesswrong.com/lw/gr/the_modesty_argument/
RAW Paste Data
Top