nshepperd

• nshepperd 21 Jun 2019 3:35 UTC

  10 points

  in reply to: G Gordon Worley III's comment on: Let Values Drift

  I don’t see the usual com­mon­sense un­der­stand­ing of “val­ues” (or the un­der­stand­ing used in eco­nomics or ethics) as rely­ing on val­ues be­ing on­tolog­i­cally fun­da­men­tal in any way, though. But you’ve the fact that they’re not to make a seem­ingly un­jus­tified rhetor­i­cal leap to “val­ues are just ha­bit­u­a­tions or pat­terns of ac­tion”, which just doesn’t seem to be true.

  Most im­por­tantly, be­cause the “val­ues” that peo­ple are con­cerned with then they talk about “value drift” are ideal­ized val­ues (ala. ex­trap­o­lated vo­li­tion), not in­stan­ta­neous val­ues or opinions or ha­bit­u­a­tions.

  For in­stance, philoso­phers such as EY con­sider that chang­ing one’s mind in re­sponse to a new moral ar­gu­ment is not value drift be­cause it pre­serves one’s ideal­ized val­ues, and that it is gen­er­ally in­stru­men­tally pos­i­tive be­cause (if it brings one’s in­stan­ta­neous opinions closer to their ideal­ized val­ues) it makes one bet­ter at ac­com­plish­ing their ideal­ized val­ues. So in­deed, we should let the EAs “drift” in that sense.

  On the other hand, get­ting hit with a cos­mic ray which al­ters your brain, or get­ting hacked by a re­mote code ex­e­cu­tion ex­ploit is value drift be­cause it does not pre­serve one’s ideal­ized val­ues (and is there­fore bad, ac­cord­ing to the usual de­ci­sion the­o­retic ar­gu­ment, be­cause it makes you worse at ac­com­plish­ing them). And those are the kind of prob­lems we worry about with AI.

• nshepperd 21 Jun 2019 1:37 UTC

  12 points

  on: Let Values Drift

  When we talk of val­ues as nouns, we are talk­ing about the val­ues that peo­ple have, ex­press, find, em­brace, and so on. For ex­am­ple, a per­son might say that al­tru­ism is one of their val­ues. But what would it mean to “have” al­tru­ism as a value or for it to be one of one’s val­ues? What is the thing pos­sessed or of one in this case? Can you grab al­tru­ism and hold onto it, or find it in the mind cleanly sep­a­rated from other thoughts?

  Since this ap­pears to be a crux of your whole (fal­la­cious, in my opinion) ar­gu­ment, I’m go­ing to start by just crit­i­ciz­ing this point. This ar­gu­ment proves far too much. It proves that:

  • Peo­ple don’t have be­liefs, mem­o­ries or skills

  • Books don’t have concepts

  • Ob­jects don’t have colors

  • Shapes don’t have to­tal in­ter­nal angles

  It seems as if you’ve rhetor­i­cally de­nied the ex­is­tence of any ab­stract prop­er­ties what­so­ever, for the pur­pose of min­i­miz­ing val­ues as be­ing “merely” ha­bit­u­a­tions or pat­terns of ac­tion. But I don’t see why any­one should ac­tu­ally ac­cept that claim.

• nshepperd 10 May 2019 5:10 UTC

  4 points

  in reply to: Said Achmiz's comment on: In­ter­pre­ta­tions of “prob­a­bil­ity”

  Doesn’t it mean the same thing in ei­ther case? Either way, I don’t know which way the coin will land or has landed, and I have some odds at which I’ll be will­ing to make a bet. I don’t see the prob­lem.

  (Though my will­ing­ness to bet at all will gen­er­ally go down over time in the “already flipped” case, due to the in­creas­ing pos­si­bil­ity that who­ever is offer­ing the bet some­how looked at the coin in the in­ter­ven­ing time.)

• nshepperd 10 May 2019 4:47 UTC

  19 points

  on: In­ter­pre­ta­tions of “prob­a­bil­ity”

  The idea that “prob­a­bil­ity” is some pre­ex­ist­ing thing that needs to be “in­ter­preted” as some­thing always seemed a lit­tle bit back­wards to me. Isn’t it more straight­for­ward to say:

  1. Beliefs ex­ist, and obey the Kol­mogorov ax­ioms (at least, “cor­rect” be­liefs do, as for­mal­ized by gen­er­al­iza­tions of logic (Cox’s the­o­rem), or by pos­si­ble-world-count­ing). This is what we re­fer to as “bayesian prob­a­bil­ities”, and code into AIs when we want to them to rep­re­sent be­liefs.

  2. Mea­sures over imag­i­nary event classes /​ en­sem­bles also obey the Kol­mogorov ax­ioms. “Fre­quen­tist prob­a­bil­ities” fall into this cat­e­gory.

  Per­son­ally I mostly think about #1 be­cause I’m in­ter­ested in figur­ing out what I should be­lieve, not about fre­quen­cies in ar­bi­trary en­sem­bles. But the fact is that both of these obey the same “prob­a­bil­ity” ax­ioms, the Kol­mogorov ax­ioms. Deny­ing one or the other be­cause “prob­a­bil­ity” must be “in­ter­preted” as ex­clu­sively ei­ther #1 or #2 is sim­ply wrong (but that’s what fre­quen­tists effec­tively do when they loudly shout that you “can’t” ap­ply prob­a­bil­ity to be­liefs).

  Now, some­times you do need to in­ter­pret “prob­a­bil­ity” as some­thing—in the spe­cific case where some­one else makes an ut­ter­ance con­tain­ing the word “prob­a­bil­ity” and you want to figure out what they meant. But the an­swer there is prob­a­bly that in many cases peo­ple don’t even dis­t­in­guish be­tween #1 and #2, be­cause they’ll only com­mit to a spe­cific num­ber when there’s a con­ve­nient in­stance of #2 that make #1 easy to calcu­late. For in­stance, say­ing ¹⁄₆ for a roll of a “fair” die.

  Peo­ple of­ten act as though their ut­ter­ances about prob­a­bil­ity re­fer to #1 though. For in­stance when they mis­in­ter­pret p-val­ues as the post-data prob­a­bil­ity of the null hy­poth­e­sis and go around be­liev­ing that the effect is real...

• nshepperd 3 May 2019 6:07 UTC

  9 points

  in reply to: Geropy's comment on: Func­tional De­ci­sion The­ory vs Causal De­ci­sion The­ory: Ex­pand­ing on New­comb’s Problem

  No, that doesn’t work. It seems to me you’ve con­fused your­self by con­struct­ing a fake sym­me­try be­tween these prob­lems. It wouldn’t make any sense for Omega to “pre­dict” whether you choose both boxes in New­comb’s if New­comb’s were equiv­a­lent to some­thing that doesn’t in­volve choos­ing boxes.

  More ex­plic­itly:

  New­comb’s Prob­lem is “You sit in front of a pair of boxes, which are ei­ther- both filled with money if Omega pre­dicted you would take one box in this case, oth­er­wise only one is filled”. Note: de­scribing the prob­lem does not re­quire men­tion­ing “New­comb’s Prob­lem”; it can be ex­pressed as a sim­ple game tree (see here for some ex­pla­na­tion of the tree for­mat):

  .

  In com­par­i­son, your “In­verse New­comb” is “Omega gives you some money iff it pre­dicts that you take both boxes in New­comb’s Prob­lem, an en­tirely differ­ent sce­nario (ie. not this case).”

  The lat­ter is more of the form “Omega ar­bi­trar­ily re­wards agents for tak­ing cer­tain hy­po­thet­i­cal ac­tions in a differ­ent prob­lem” (of which a nearly limitless va­ri­ety can be in­vented to jus­tify any cho­sen de­ci­sion the­ory¹), rather than be­ing an ac­tual self-con­tained prob­lem which can be “solved”.

  The lat­ter also can’t be ex­pressed as any kind of game tree with­out “cheat­ing” and nam­ing “New­comb’s Prob­lem” ver­bally—or rather, you can ex­press a similar thing by em­bed­ding the New­comb game tree and refer­ring to the em­bed­ded tree, but that con­verts it into a le­gi­t­i­mate de­ci­sion prob­lem, which FDT of course gives the cor­rect an­swer to (TODO: draw an ex­am­ple ;).

  (¹): Con­sider In­verse^2 New­comb, which I con­sider the proper sym­met­ric in­verse of “In­verse New­comb”: Omega puts you in front of two boxes and says “this is not New­comb’s Prob­lem, but I have filled both boxes with money iff I pre­dicted that you take one box in stan­dard New­comb”. Ob­vi­ously here FDT takes both boxes and a tidy $1,000,1000 profit (plus the $1,000,000 from Stan­dard New­comb). Whereas CDT gets… $1000 (plus $1000 from Stan­dard New­comb).

• nshepperd 25 Apr 2019 17:19 UTC

  3 points

  in reply to: Bunthut's comment on: The Ca­cophony Hy­poth­e­sis: Si­mu­la­tion (If It is Pos­si­ble At All) Can­not Call New Con­scious­nesses Into Existence

  Yes, you need to have a the­ory of physics to write down a tran­si­tion rule for a phys­i­cal sys­tem. That is a prob­lem, but it’s not at all the same prob­lem as the “tar­get for­mat” prob­lem. The only role the tran­si­tion rule plays here is it al­lows one to ap­ply in­duc­tion to effi­ciently prove some gen­er­al­iza­tion about the sys­tem over all time steps.

  In prin­ci­ple a differ­ent more dis­t­in­guished con­cise de­scrip­tion of the sys­tem’s be­havi­our could play the a similar role (per­haps, the record­ing of the states of the sys­tem + the short­est pro­gram that out­puts the record­ing?). Or per­haps there’s some way of choos­ing a dis­t­in­guished “best” for­mal­iza­tion of physics. But that’s rather out of scope of what I wanted to sug­gest here.

  But then you are mea­sur­ing proof short­ness rel­a­tive to that sys­tem. And you could be us­ing one of countless other for­mal sys­tems which always make the same pre­dic­tions, but rel­a­tive to which differ­ent proofs are short and long.

  It would be a O(1) cost to start the proof by trans­lat­ing the ax­ioms into a more con­ve­nient for­mat. Much as Kol­mogorov com­plex­ity is “lan­guage de­pen­dent” but not asymp­tot­i­cally be­cause any par­tic­u­lar uni­ver­sal tur­ing ma­chine can be simu­lated in any other for a con­stant cost.

  The as­sump­tion (in­clud­ing that it takes in and puts out in ara­bic nu­mer­als, and uses “*” as the mul­tu­pli­ca­tion com­mand, and that but­tons must be pressed,… and all the other things you need to ac­tu­ally use it) in­cludes that.

  Th­ese are all things that can be de­rived from a phys­i­cal de­scrip­tion of the calcu­la­tor (maybe not in fewer steps than it takes to do long mul­ti­pli­ca­tion, but cer­tainly in fewer steps than less triv­ial com­pu­ta­tions one might do with a calcu­la­tor). There’s no ob­server de­pen­dency here.

• nshepperd 21 Apr 2019 22:33 UTC

  2 points

  in reply to: Bunthut's comment on: The Ca­cophony Hy­poth­e­sis: Si­mu­la­tion (If It is Pos­si­ble At All) Can­not Call New Con­scious­nesses Into Existence

  That’s not an is­sue in my for­mal­iza­tion. The “log­i­cal facts” I speak of in the for­mal­ized ver­sion would be fully speci­fied math­e­mat­i­cal state­ments, such as “if the simu­la­tion starts in state X at t=0, the state of the simu­la­tion at t=T is Y” or “given that Alice starts in state X, then <some for­mal­ized way of cat­e­goris­ing states ac­cord­ing to favourite ice cream flavour> re­turns Vanilla”. The “tar­get for­mat” is math­e­mat­i­cal proofs. Lan­guages (as in English vs Chi­nese) don’t and can’t come in to it, be­cause proof sys­tems are lan­guage-ig­no­rant.

  Note, the for­mal­ized crite­rion is broader than the in­for­mal “could you do some­thing use­ful with this simu­la­tion IRL” crite­rion, even though the lat­ter is the ‘in­spira­tion’ for it. For in­stance, it doesn’t mat­ter whether you un­der­stand the pro­gram­ming lan­guage the simu­la­tion is writ­ten in. If some­one who did un­der­stand the lan­guage could write the ap­pro­pri­ate proofs, then the proofs ex­ist.

  Similarly, if a simu­la­tion is run un­der Ho­mo­mor­phic_en­cryp­tion, it is nev­er­the­less a valid simu­la­tion, de­spite the fact that you can’t read it if you don’t have the de­cryp­tion key. Be­cause a proof ex­ists which starts by “mag­i­cally” writ­ing down the key, prov­ing that it’s the cor­rect de­cryp­tion key, then pro­ceed­ing from there.

  An in­for­mal crite­rion which maybe cap­tures this bet­ter would be: If you and your friend both have (view) ac­cess to a gen­uine com­pu­ta­tion of some log­i­cal facts X, it should be pos­si­ble to con­vince your friend of X in fewer words by refer­ring to the alleged com­pu­ta­tion (but you are per­mit­ted un­limited time to think first, so you can re­verse en­g­ineer the simu­la­tion, brute­force some en­cryp­tion keys, learn Chi­nese, what­ever you like, be­fore talk­ing). A bit like how it’s more effi­cient to con­vince your friend that 637265729567*37265974 = 23748328109134853258 by punch­ing the num­bers into a calcu­la­tor and say­ing “see?” than by hand­ing over a pa­per with a com­plete long mul­ti­pli­ca­tion deriva­tion (as­sum­ing you are fa­mil­iar with the calcu­la­tor and can con­vince your friend that it calcu­lates cor­rectly).

• nshepperd 16 Apr 2019 7:41 UTC

  25 points

  on: The Ca­cophony Hy­poth­e­sis: Si­mu­la­tion (If It is Pos­si­ble At All) Can­not Call New Con­scious­nesses Into Existence

  This idea is, as oth­ers have com­mented, pretty much Dust the­ory.

  The solu­tion, in my opinion, is the same as the an­swer to Dust the­ory: namely, it is not ac­tu­ally the case that any­thing is a simu­la­tion of any­thing. Yes, you can claim that (for in­stance) the mo­tion of the atoms in a peb­ble can be in­ter­preted as a simu­la­tion of Alice, in the sense that any­thing can be mapped to any­thing… but in a cer­tain more real sense, you can’t.

  And that sense is this: an ac­tual simu­la­tion of Alice run­ning on a com­puter grants you cer­tain pow­ers—you can step through the simu­la­tion, ex­am­ine what Alice does, and de­ter­mine cer­tain facts such as Alice’s favourite ice cream flavour (these are log­i­cal facts, given the simu­la­tion’s ini­tial state). If the simu­la­tion is an up­load of your friend Alice, then by do­ing so you learn mean­ingful new facts about your friend.

  In com­par­i­son, a peb­ble “in­ter­preted” as a simu­la­tion of Alice af­fords you no such pow­ers, be­cause the in­ter­pre­ta­tion (map­ping from peb­ble states to simu­la­tion data) is en­tirely post-hoc. The only way to pin down the map­ping—such that you could, for in­stance, ex­plic­itly write it down, or take the peb­ble’s state and map it to an an­swer about Alice’s favourite ice cream—is to already have car­ried out the ac­tual simu­la­tion, sep­a­rately, and already know these things about Alice.

  In gen­eral, “le­gi­t­i­mate” com­pu­ta­tions of cer­tain log­i­cal facts (such as the an­swers one might ask about simu­la­tions of peo­ple) should, in a cer­tain sense, make it eas­ier to calcu­late those log­i­cal facts then do­ing so from scratch.

  A spe­cific for­mal­iza­tion of this idea would be that a proof sys­tem equipped with an or­a­cle (ax­iom schema) de­scribing the states of the phys­i­cal sys­tem which allegedly com­puted these facts, as well as its tran­si­tion rule, should be able to find proofs for those log­i­cal facts in less steps than one with­out such ax­ioms.

  Such proofs will in­volve first com­ing up with a map­ping (such as in­ter­pret­ing cer­tain elec­tri­cal junc­tions as nand gates), prov­ing them valid us­ing the tran­si­tion rules, then us­ing in­duc­tion to jump to “the phys­i­cal state at timestep t is X there­fore Alice’s favourite ice cream colour is Y”. Note that the re­quire­ment that these proofs be short nat­u­rally re­sults in these “in­ter­pre­ta­tions” be­ing sim­ple.

  As far as I know, this spe­cific for­mal­iza­tion of the anti-Dust idea is origi­nal to me, though the idea that “in­ter­pre­ta­tions” of things as com­pu­ta­tions ought to be “sim­ple” is not par­tic­u­larly new.

• nshepperd 16 Apr 2019 6:31 UTC

  2 points

  in reply to: habryka's comment on: High­lights from “In­te­gral Spiritu­al­ity”

  We can (and should) have that dis­cus­sion, we should just have it on a sep­a­rate post

  Can you point to the spe­cific lo­ca­tion that dis­cus­sion “should” hap­pen at?

• nshepperd 15 Apr 2019 17:59 UTC

  5 points

  in reply to: mr-hire's comment on: The Hard Work of Trans­la­tion (Bud­dhism)

  The two parts I men­tioned are sim­ply the most ob­vi­ously spec­u­la­tive and un­jus­tified ex­am­ples. I also don’t have any real rea­son to be­lieve the va­guer pop psy­chol­ogy claims about build­ing sto­ries, back­logs, etc.

  The post would prob­a­bly have been a bit cleaner to not men­tion the few wild spec­u­la­tions he men­tions, but get­ting caught up on the tiny de­tails seems to miss the for­est from the trees.

  It seems to me LW has a big epistemic hy­giene prob­lem, of late. We need to col­lec­tively stop make ex­cuses for post­ing wild spec­u­la­tions as if they were fact, just be­cause the same post also con­tains some in­ter­est­ing ‘in­sight’. We should be down­vot­ing such posts and say­ing to the au­thor “go away and write this again, with the parts that you can’t jus­tify re­moved”.

  Do­ing so may re­veal that the ‘in­sight’ ei­ther does or does not suc­cess­fully stand alone when the sup­port­ing ‘spec­u­la­tion’ is re­moved. Either way, we learn some­thing valuable about the alleged in­sight; and we benefit di­rectly by not spread­ing claims that lack ev­i­dence.

• nshepperd 12 Apr 2019 21:09 UTC

  17 points

  on: The Hard Work of Trans­la­tion (Bud­dhism)

  For a post that claims to be a “trans­la­tion” of Bud­dhism, this seems to con­tain:

  • No Pali text;

  • No spe­cific refer­ences to Pali text, or any sources at all;

  • No ac­tual trans­la­tion work of any kind.

  On the other hand, it does con­tain quite a bit of un­jus­tified spec­u­la­tion. “Literal elec­tri­cal re­sis­tance in the CNS”, re­ally? “Rewiring your CNS”? Why should I be­lieve any of this?

  Why are peo­ple up­vot­ing this?

• nshepperd 2 Feb 2019 4:47 UTC

  8 points

  in reply to: Elo's comment on: Boundaries—A map and ter­ri­tory ex­per­i­ment. [post-ra­tio­nal­ity]

  “Above the map”? “Out­side the ter­ri­tory”? This is ut­ter non­sense. Ra­tion­al­ity in­sists no such thing. Ex­plic­itly the op­po­site, in fact.

  Given things like this too:

  Ex­ist­ing map-less is very hard. The hu­man brain re­ally likes to put maps around things.

  At this point I have to won­der if you’re just round­ing off ra­tio­nal­ity to the near­est thing to which you can ap­ply new-age plat­i­tudes. Frankly, this is in­sult­ing.

• nshepperd 31 Jan 2019 17:54 UTC

  8 points

  in reply to: Jacobian's comment on: Mas­culine Virtues

  You don’t need to es­ti­mate this.

  A McGill Univer­sity study found that more than 60 per­cent of col­lege-level soc­cer play­ers re­ported symp­toms of con­cus­sion dur­ing a sin­gle sea­son. Although the per­centage at other lev­els of play may be differ­ent, these data in­di­cate that head in­juries in soc­cer are more fre­quent than most pre­sume.

  A 60% chance of con­cus­sion is more than enough for me to stay far away.

• nshepperd 8 Jan 2019 17:40 UTC

  6 points

  in reply to: Raemon's comment on: No op­tion to re­port spam

  Preven­tion over re­moval. Old LW re­quired a cer­tain amount of karma in or­der to cre­ate posts, and we cor­re­spond­ingly didn’t have a post spam prob­lem that I re­mem­ber. I strongly be­lieve that this re­quire­ment should be re-in­tro­duced (with or with­out a mod­er­a­tor ap­proval op­tion for users with­out suffi­cient karma).

• nshepperd 21 Nov 2018 1:16 UTC

  24 points

  AF

  on: Topolog­i­cal Fixed Point Exercises

  Proof of #4, but with un­nec­es­sary calcu­lus:

  Not only is there an odd num­ber of tri­color tri­an­gles, but they come in pairs ac­cord­ing to their ori­en­ta­tion (RGB clock­wise/​an­ti­clock­wise). Proof: define a con­tin­u­ously differ­en­tiable vec­tor field on the plane, by let­ting the field at each ver­tex be 0, and the field in the cen­ter of each edge be a vec­tor of mag­ni­tude 1 point­ing in the di­rec­tion R->G->B->R (or 0 if the two ad­ja­cent ver­tices are the same color). Ex­tend the field to the com­plete edges, then the in­te­ri­ors of the tri­an­gles by some in­ter­po­la­tion method with con­tin­u­ous deriva­tive (eg. co­sine in­ter­po­la­tion).

  As­sume the line in­te­gral along one unit edge in the di­rec­tion R->G or G->B or B->R to be ¹⁄₃. (Without loss of gen­er­al­ity since we can rescale the graph/​vec­tors to make this true). Then a similar par­ity ar­gu­ment to Sperner’s 1d lemma (or the FTC) shows that the clock­wise line in­te­gral along each large edge is ¹⁄₃, hence the line in­te­gral around the large tri­an­gle is 1/​3+1/​3+1/​3=1.

  By green’s the­o­rem, this is equal to the in­te­grated curl of the field in the in­te­rior of the large tri­an­gle, and hence equal (by an­other in­vo­ca­tion of green’s the­o­rem) to the summed clock­wise line in­te­grals around each small tri­an­gle. The in­te­grals around a uni­color or bi­color tri­an­gle are 0 and −1/​3 + ¹⁄₃ + 0 = 0 re­spec­tively, leav­ing only tri­color tri­an­gles, whose in­te­gral is again 1 de­pend­ing on ori­en­ta­tion. Thus: (tri­color clock­wise) - (tri­color an­ti­clock­wise) = 1. QED.

• nshepperd 8 Nov 2018 3:02 UTC

  2 points

  in reply to: Said Achmiz's comment on: Beta-Test: Strong Votes

  Your in­ter­pre­ta­tion of the bolded part is cor­rect.

• nshepperd 8 Nov 2018 2:32 UTC

  24 points

  on: Beta-Test: Strong Votes

  We got to dis­cussing this on #less­wrong re­cently. I don’t see any­one here point­ing this out yet di­rectly, so:

  Can you tech­ni­cally Strong Upvote ev­ery­thing? Well, we can’t stop you. But we’re hop­ing a com­bi­na­tion of mostly-good-faith + triv­ial in­con­ve­niences will re­sult in peo­ple us­ing Strong Upvotes when they feel it’s ac­tu­ally im­por­tant.

  This ap­proach, hop­ing that good faith will pre­vent peo­ple from us­ing Strong votes “too much”, is a good ex­am­ple of an Ass­hole Filter (linkposted on LW last year). You’ve set some (un­clear) bound­aries, then due to not en­forc­ing them, re­ward those who vi­o­late them with in­creased con­trol over the site con­ver­sa­tion. Chris_Leong ges­tures to­wards this with­out di­rectly nam­ing it in a sibling com­ment.

  In my opinion “maybe put limits on strong up­votes if this seems to be a prob­lem” is not the cor­rect re­sponse to this prob­lem, nor would be ban­ning or oth­er­wise ‘dis­ci­plin­ing’ users who use strong votes “too much”. The cor­rect re­sponse is to re­move the ass­hole filter by al­ter­ing the in­cen­tives to match what you want to hap­pen. Op­tions in­clude:

  1. Mak­ing votes nor­mal by de­fault but en­courag­ing users to use strong votes freely, up to 100% of the time, so that good faith users are not dis­ad­van­taged. (Note: still dis­en­fran­chises users who don’t no­tice that this fea­ture ex­ists, but maybe that’s ok.)

  2. Mak­ing votes strong by de­fault so that it’s mak­ing a “weak” vote that takes ex­tra effort. (Note: this gives users who care­fully make weak votes when they have weak opinions less weight, but at least they do this with eyes open and in the ab­sence of per­verse in­cen­tives.)

  3. #2 but with some al­gorith­mic ad­just­ment to give care­ful users more weight in­stead of less. This seems ex­tremely difficult to get right (cf. slash­dot meta­mod­er­a­tion). Prob­a­bly the cor­rect an­swer there is some form of col­lab­o­ra­tive fil­ter­ing.

  Per­son­ally I favour solu­tion #1.

  I’ll add that this is not just a hy­po­thet­i­cal troll-con­trol is­sue. This is also a UX is­sue. Forc­ing users to nav­i­gate an un­clear eth­i­cal ques­tion and pris­oner’s dilemma—how much strong vot­ing is “too much”—in or­der to use the site is un­pleas­ant and a bad user ex­pe­rience. There should not be a “wrong” ac­tion available in the user in­ter­face.

  PS. I’ll con­cede that mak­ing strong votes an ac­tu­ally limited re­source that is en­forced by the site eco­nom­i­cally (eg. with To­ken Bucket quota) would in a way also work, due to elimi­nat­ing the per­ceived need for strong votes to be limited by “good faith”. But IMO the need is only per­ceived, and not real. Vot­ing is for ex­press­ing prefer­ences, and prefer­ences are un­limited.

• nshepperd 23 Oct 2018 4:21 UTC

  3 points

  on: Kal­man Filter for Bayesians

  Good post!

  Is it com­mon to use Kal­man filters for things that have non­lin­ear trans­for­ma­tions, by ap­prox­i­mat­ing the pos­te­rior with a Gaus­sian (eg. calcu­lat­ing the clos­est Gaus­sian dis­tri­bu­tion to the true pos­te­rior by JS-di­ver­gence or the like)? How well would that work?

  Gram­mar com­ment—you seem to have ac­ci­den­tally a few words at

  Mea­sur­ing mul­ti­ple quan­tities: what if we want to mea­sure two or more quan­tities, such as tem­per­a­ture and hu­midity? Fur­ther­more, we might know that these are [miss­ing words?] Then we now have mul­ti­vari­ate nor­mal dis­tri­bu­tions.

• nshepperd 21 Oct 2018 22:57 UTC

  4 points

  in reply to: Elo's comment on: Out­line of Me­ta­ra­tional­ity, or much less than you wanted to know about postrationality

  How big was your mir­ror, and how much of your face did you see in it?

• nshepperd 20 Oct 2018 2:43 UTC

  10 points

  in reply to: G Gordon Worley III's comment on: Out­line of Me­ta­ra­tional­ity, or much less than you wanted to know about postrationality

  C is ba­si­cally a state­ment that, if in­cluded in a valid ar­gu­ment about the truth of P, causes the ar­gu­ment to tell us ei­ther P or ~P. That’s defi­ni­tion­ally what it means to be able to know the crite­rion of truth.

  That’s not how al­gorithms work and seems… in­co­her­ent.

  That you want to deny C is great,

  I did not say that ei­ther.

  be­cause I think (as I’m find­ing with Said), that we already agree, and any dis­agree­ment is the con­se­quence of mi­s­un­der­stand­ing, prob­a­bly be­cause it comes too close to sound­ing to you like a po­si­tion that I would also re­ject, and the rest of the fun­da­men­tal dis­agree­ment is one of sen­ti­ment, per­spec­tive, hav­ing worked out the de­tails, and em­pha­sis.

  No, I don’t think we do agree. It seems to me you’re deeply con­fused about all of this stuff.

  Here’s an ex­er­cise: Say that we re­place “C” by a spe­cific con­crete al­gorithm. For in­stance the el­e­men­tary long mul­ti­pli­ca­tion al­gorithm used by pri­mary school chil­dren to mul­ti­ply num­bers.

  Does any­thing what­so­ever about your ar­gu­ment change with this sub­sti­tu­tion? Have we proved that we can ex­plain mul­ti­pli­ca­tion to a rock? Or per­haps we’ve proved that this al­gorithm doesn’t ex­ist, and nei­ther do schools?

  Another ex­er­cise: sup­pose, as a coun­ter­fac­tual, that Laplace’s de­mon ex­ists, and fur­ther­more likes an­swer­ing ques­tions. Now we can take a spe­cific al­gorithm C: “ask the de­mon your ques­tion, and await the an­swer, which will be re­ceived within the minute”. By con­struc­tion this al­gorithm always re­turns the cor­rect an­swer. Now, your task is to give the al­gorithm, given only these premises, that I can fol­low to con­vince a rock that Eu­clid’s the­o­rem is true.