George(George)

• George 30 Mar 2020 1:54 UTC

  1 point

  in reply to: maximkazhenkov's comment on: When to as­sume neu­ral net­works can solve a problem

  Yes, I was in fact. See­ing where this in­ter­net ar­gu­ment is go­ing, I think it’s best to leave it here.

  So, in that case.

  If your origi­nal chain of logic is:

  1. An RL-based al­gorithm that could play any game could pass the tur­ing test

  2. An al­gorithm that can pass the Tur­ing test is “AGI com­plete”, thus it is un­likely that (1) will hap­pen soon

  And you agree with the state­ment:

  3. An al­gorithm did pass the Tur­ing test in 2014

  You ei­ther:

  a) Have a contradiction

  b) Must have some spe­cific defi­ni­tion of the Tur­ing test un­der which 3 is un­true (and more gen­er­ally, no known al­gorithm can pass the Tur­ing test)

  I as­sume your po­si­tion here is b and I’d love to hear it.

  I’d also love to hear the causal rea­son­ing be­hind 2. (maybe ex­plained by your defi­ni­tion of the Tur­ing test ?)

  If your defi­ni­tions differ from com­monly ac­cepted defi­ni­tions and your you rely on causal­ity which is not widely im­plied, you must at least provide your ver­sions of the defi­ni­tions and some mo­ti­va­tion be­hind the causal­ity.

• George 29 Mar 2020 19:53 UTC

  1 point

  in reply to: maximkazhenkov's comment on: When to as­sume neu­ral net­works can solve a problem

  Tur­ing test, which is to say AGI-complete

  You are aware chat­bots have been “beat­ing” the origi­nal Tur­ing test since 2014, right? (And ar­guably even be­fore)
  
  Also, AGI-com­plete == fools ¹⁄₃ of hu­man judges in an x minute con­ver­sa­tion via text? Ahm, no, just no.
  

  That state­ment is mean­ingless un­less you define the Tur­ing test and keeps be­ing mean­ingless even if you define the tur­ing test, there is liter­ally no defi­ni­tion for “AGI com­plete”. AGI is more of a generic term used to mean “kinda like a hu­man”, but it’s not very con­crete.

  
  On the whole, yes, some games might prove too difficult for RL to beat… but I can’t think of any in par­tic­u­lar. I think the state­ment hold for ba­si­cally any pop­u­lar com­pet­i­tive game (e.g. one where there are cur­rently cash prizes above > 1000$ to be won). I’m sure one could de­sign an ad­ver­sar­ial game speci­fi­cally de­signed to not be beaten by RL but doable by a hu­man… but that’s an­other story. Tur­ing test, which is to say AGI-complete

• George 28 Mar 2020 11:50 UTC

  1 point

  in reply to: Daniel Kokotajlo's comment on: When to as­sume neu­ral net­works can solve a problem

  Also if you read al­most any­thing on the sub­ject, peo­ple will be con­stantly say­ing how they don’t think su­per­hu­man in­tel­li­gence is in­evitable or close

  If it’s “mean­ingfully close enough to do some­thing about it” I will take that as be­ing ’close”. I don’t think Bostrom puts a num­ber on it, or I don’t re­mem­ber him do­ing so, but he seems to ad­dress a real pos­si­bil­ity rather than a hy­po­thet­i­cal that is hun­dreds or thou­sands of years away.

  What do you mean, you’ve never seen a con­sis­tent top-to-bot­tom rea­son­ing for it? This is not a rhetor­i­cal ques­tion, I am just not sure what you mean here. If you are ac­cus­ing e.g. Bostrom of in­con­sis­tency, I am pretty sure you are wrong about that.

  I mean, I don’t see a chain of con­clu­sions that leads to the the­ory be­ing “cor­rect” , Vaniver men­tioned bel­low how this is not the cor­rect per­spec­tive to adopt and I agree with that.… or I would, as­sum­ing that the hy­poth­e­sis would be Pop­pe­rian (i.e. that one could do some­thing to dis­prove AI be­ing a large risk in the rel­a­tive near fu­ture).

  If you are just say­ing he hasn’t got an ar­gu­ment in premise-con­clu­sion form, well, that seems true but not very rele­vant or im­por­tant. I could make one for you if you like.

  If you could make such a premise-con­clu­sion case I’d be more then wel­come to hear it out.

  
  

  ease of data col­lec­tion? Cost of com­put­ing power? Use­ful­ness of in­tel­li­gence? -- but all three of these things seem like things that peo­ple have ar­gued about at length, not assumed

  Well, I am yet to see the arguments

  Also the case for AI safety doesn’t de­pend on these things be­ing prob­a­ble, only on them be­ing not ex­tremely un­likely.

  It de­pends on you be­ing able to put num­ber on those prob­a­bil­ities though, oth­er­wise you are in a Pas­cal wa­ger’s sce­nario, where any event that is not al­most cer­tainly ruled out should be taken into ac­count with an amount of se­ri­ous­ness pro­por­tional to it’s fic­tive im­pact.

• George 28 Mar 2020 9:56 UTC

  1 point

  in reply to: Daniel Kokotajlo's comment on: When to as­sume neu­ral net­works can solve a problem

  more­over I think Stu­art Rus­sell is too

  Yes, I guess I should have made the clar­ifi­ca­tion about that, I don’t think Stu­art Rus­sell is nec­es­sar­ily much di­ver­gent from Bostrom in his views. Rather, he’s most poniente ar­gu­ments seem not to be very re­lated to that view, so I think his books is a good guide for what I la­beled as the sec­ond view in the ar­ti­cle.

  But he cer­tainly tries to up­hold both.

  How­ever the ar­ti­cle was already too long and go­ing into that would have made it even longer.… in hind­sight I’ve de­cided to just split it into two, but the ver­sion here I shall leave as is.

• George 27 Mar 2020 22:37 UTC

  5 points

  in reply to: Vaniver's comment on: When to as­sume neu­ral net­works can solve a problem

  I will prob­a­bly be steal­ing the per­spec­tive of the view be­ing dis­junc­tive as a way to look at why it’s hard to pin down.

  And thus, just like the state of neu­ral net­works in 2010 was only weakly in­for­ma­tive about what would be pos­si­ble in 2020, it seems rea­son­able to ex­pect the state of things in 2020 will be only weakly in­for­ma­tive and about will be pos­si­ble in 2030.

  This state­ment I would par­tially dis­agree with.

  I think the idea of train­ing on a GPU was com­ing to the fore­front by 2010 and also the idea of CNNs for image recog­ni­tion: https://​​hal.in­ria.fr/​​in­ria-00112631/​​doc­u­ment (see both in that 2006 pa­per)y K. et al. (2006)

  I’d ar­gue it’s fairly easy to look at to­day’ land­scape and claim that by 2030 the things that are likely to hap­pen in­clude:

  • ML play­ing any pos­si­ble game bet­ter than hu­mans as­sum­ing a team ac­tu­ally works on that spe­cific game (maybe even if one doesn’t), with huma-like in­puts and hu­man-like limi­ta­tions in terms of gran­u­lar­ity of tak­ing in­puts and giv­ing out­puts.

  • ML achiev­ing all the things we can do with 2d images right now for 3d images and short (e.g. < 5 minute) videos.

  • Al­gorithms be­ing able to write e.g. ar­ti­cles sum­ma­riz­ing var­i­ous knowl­edge it gath­ers from given sources and pos­si­bly even find rele­vant sources via search­ing based on key­words (so you could just say “Write an ar­ti­cle about Peru’s eco­nomic cli­mate in 2028, rather than feed a bunch of ar­ti­cles about Peru’s econ­omy in 2028)… the sec­ond part is already doable, but I’m men­tion­ing them to­gether since I as­sume peo­ple will be more im­pressed with the fi­nal product

  • Al­gorithms be­ing able to trans­late from and to al­most any lan­guage about as well as hu­man, but still not well enough to trans­late sources which re­quire a lot of in­ter­pre­ta­tion (e.g. yes for trans­lat­ing a biol­ogy pa­per from en­glish to hindi or vice versa, no for trans­lat­ing a phe­nomenol­ogy pa­per from en­glish to hindi or vice versa)

  • Con­trol­ling me­chan­i­cal sys­tems (e.g. robotic arms) via net­works trained us­ing RL.

  • Gen­er­ally speak­ing, al­gorithms be­ing used in ar­eas where they already out-perform hu­mans but where reg­u­la­tions and sys­tem­atic in­effi­cien­cies com­bined with is­sues of stake don’t cur­rently al­low them to be used (e.g. ac­count­ing, risk anal­y­sis, set­ting in­surance poli­cies, di­ag­no­sis, treat­ment plan­ning). Al­gorithms be­ing jointly used to help in var­i­ous sci­en­tific fields by re­plac­ing the need for hu­mans to use clas­si­cal statis­tics and or man­u­ally fit­ting equa­tions in or­der to model cer­tain pro­cesses.

  I’d wa­ger points 1 to 4 are ba­si­cally a given, point 5 is de­bat­able since it de­pends on hu­man reg­u­la­tors and cul­tural ac­cep­tance for the most part.

  I’d also wa­ger than, other than au­dio pro­cess­ing, there won’t be much in­no­va­tion be­yond those 5 points that will cre­ate load of hype by 2030. You might have en­sem­bles of those 4 things build­ing up to some­thing big­ger, but those 5 things will be at the core of it.

  But that’s just my in­tu­ition, par­tially based on the kind of heuris­tics above about what is eas­ily doable and what isn’t. But alas, the point of the ar­ti­cle was to talk about what’s doable in the pre­sent, rather than what to ex­pect from the fu­ture, so it’s not re­ally that re­lated.

When to as­sume neu­ral net­works can solve a problem

• George 27 Mar 2020 8:53 UTC

  3 points

  on: Ge­orge’s Shortform

  I find it in­ter­est­ing what kind of be­liefs one needs to ques­tion and in which ways in or­der to get peo­ple an­gry/​up­set/​touchy.

  Or, to put it in more pop­u­lar terms, what kind of ar­gu­ments make you seem like a smart-ass when ar­gu­ing with some­one.

  For ex­am­ple, read­ing Eliezer yud­kowsky’s Ra­tion­al­ity from AI to Zom­bies, I found my­self gen­er­ally speak­ing lik­ing the writ­ing style and to a karge ex­tent the book was just re­in­forc­ing the bi­ases I already had. Other then some of his poorly thought out meta­physics based on which he bases his ethics ar­gu­ment… I hon­estly can’t think of a sin­gle thing from that book I dis­agree with. Same goes for Inad­e­quate Equil­ibria.

  Yet, I can re­mem­ber a cer­tain feel­ing pop­ping up in my head fairly of­ten when read­ing it, one that can be best de­scribed in an image: https://​​i.kym-cdn.com/​​en­tries/​​icons/​​face­book/​​000/​​021/​​665/​​DpQ9YJl.jpg

  ***

  One seem­ing pat­tern for this is some­thing like:

  • Ar­gu­ing about a spe­cific belief

  • Go­ing a level down and challeng­ing a pillar of the op­po­nent’s be­lief that was not be­ing con­sid­ered as part of the dis­cus­sion.

  E.g: “Ar­gu­ing about whether or not cli­mate change is a threat, go­ing one level down and ar­gu­ing that there’s not enough proof cli­mate change is hap­pen­ing to be­ing with”

  You can make this pat­tern even more an­noy­ing by do­ing some­thing like:

  • Ar­gu­ing about a spe­cific belief

  • Go­ing a level down and challeng­ing a pillar of the op­po­nent’s be­lief that was not be­ing con­sid­ered as part of the dis­cus­sion.

  • Not en­ter­tain­ing an op­po­site ar­gu­ment about one of your own pillars be­ing shaky.

  E.g.: After the pre­vi­ous cli­mate change ar­gu­ment, not en­ter­tain­ing the idea that “Maybe act­ing upon cli­mate change as if it were real and as if it were a threat, would ac­tu­ally re­sult in pos­i­tive con­se­quences even if those two things were un­ture”

  You can make this pat­tern even more an­noy­ing by do­ing some­thing like:

  • Ar­gu­ing about a spe­cific belief

  • Go­ing a level down and challeng­ing a pillar of the op­po­nent’s be­lief that was not be­ing con­sid­ered as part of the dis­cus­sion.

  • Do­ing so with some ev­i­dence that the other party is un­aware or can­not understand

  E.g.: After the pre­vi­ous cli­mate change ar­gu­ment, back up your point about cli­mate change not be­ing real by cit­ing var­i­ous stud­ies that would take hours to fact check and might be out of reach knowl­edge-wise for ei­ther of you.

  ***

  I think there’s other things that come into ac­count.

  For ex­am­ple there’s some spe­cific fields which are con­sid­ered more sacro­sanct then oth­ers, try­ing to ar­gue against a stan­dard po­si­tion in that field as part of your ar­gu­ment seems to much more eas­ily put you into the “smar­tass” camp.

  For ex­am­ple, ar­gu­ing against com­monly held re­li­gious or med­i­cal knowl­edge, seems to be al­most im­pos­si­ble, un­less you are tak­ing an already-ap­proved side of the de­bate.

  E.g. You can ar­gue ibupro­fen against parac­eta­mol as the go to for com­mon cold since there’s au­thor­i­ta­tive claims for each, you can’t ar­gue for a 3rd lesser backed NSAID or for us­ing cor­ti­cos­teroids or no treat­ment in­stead of NSAIDs.

  Other fields such as ethics or physics or com­puter sci­ence seem to be fair game and no­body re­ally minds peo­ple try­ing to ar­gue for an un­sanc­tioned view­point.

  ***

  There’s ob­vi­ously the idea of poli­tics be­ing over­all bad, and the more poli­ti­cized a cer­tain sub­ject is the less you can change peo­ple’s minds about it.

  But to some ex­tent I don’t feel like poli­tics re­ally comes into play.

  It seems that peo­ple are fairly open to hav­ing their minds changed about eco­nomic policy but not about iden­tity policy.… no mat­ter which side of the spec­trum you are on. Which seem counter in­tu­itive, since the is­sue of “should coun­tries have open bor­ders and free health­care” seems like one much more deeply em­bed­ded in ex­ist­ing poli­ti­cal agen­das and of much more im­port than “What gen­der should trans­gen­der peo­ple be counted in when par­ti­ci­pat­ing in the olympics”.

  ***

  One in­ter­est­ing thing that I ob­served: I’ve per­son­ally been able to an­noy a lot of peo­ple when talk­ing with them on­line. How­ever, IRL, in the last 4 years or so (since I ac­tu­ally be­gun ex­plic­itly learn­ing how to com­mu­ni­cate), I can’t think of a sin­gle per­son that I’ve offended.

  Even though I’m more ver­bose when I talk. Even though the ideas I talk about over coffee are usu­ally much more niche and ques­tion­able in their ver­ity then the ones I write about on­line.

  I won­der if there’s some sort of “magic or­a­tory skill” I’ve come closer to at­tain­ing IRL that ei­ther can’t be at­tained on the in­ter­net or is very differ­ent… granted, it’s more likely it’s the in­her­ent bias of the peo­ple I’m usu­ally dis­cussing with.

SARS-CoV-2, 19 times less likely to in­fect peo­ple un­der 15

• George 23 Mar 2020 11:22 UTC

  1 point

  in reply to: Dustin's comment on: The ques­tions one needs not address

  Well, not re­ally, since the way they get talked about is es­sen­tially search­ing for a “bet­ter” defi­ni­tion or try­ing to make all defi­ni­tions co­in­cide.

  Even more so, some of the terms al­low for defi­ni­tions but those defi­ni­tions in them­selves run into the same prob­lem. For ex­am­ple, could you try to come up with one or mul­ti­ple defi­ni­tions for the mean­ing of “free will” ? In my ex­pe­rience it ei­ther leads to very bor­ing ones (in which case the sub­ject would be moot) or, more likely, to a defi­ni­tion that is just as prob­le­matic as ‘free will’ it­self.

• George 22 Mar 2020 18:59 UTC

  3 points

  in reply to: Donald Hobson's comment on: The ques­tions one needs not address

  We can now say that try­ing to an­swer ques­tions like “what is the true na­ture of god” isn’t go­ing to work

  I mean, I don’t think and I’m not ar­gu­ing we can do that. I just think that the ques­tion in it­self is mis­tak­enly for­mu­late, the same way “How do we han­dle AI risk ?” is a mis­taken for­mu­la­tion (see Jau Mols­tad’s an­swer to the post which seems to ad­dress this).

  All that I am claiming is that cer­tain ill-defined ques­tion on which no progress can be made ex­ist and that they can be to some ex­tent eas­ily spot­ted be­cause they would make no sense if de-con­structed or if an out­side ob­serve were to judge your progress on them.

  Cel­e­brat­ing the peo­ple who ded­i­cated their lives to build­ing the first steam en­g­ine, while mock­ing peo­ple who tried to build per­pet­ual mo­tion ma­chines be­fore con­ser­va­tion of en­ergy was un­der­stood, is just pure hindsight

  Ahm, I mean, Epicu­rus and Thales would have had pretty strong in­tu­itions against this, and con­ser­va­tion of en­ergy has been pos­tu­lated in physics since Is­sac New­ton and even be­fore him, when the whole thing wasn’t even called “physics”.

  Nor is there a way to “prove” con­ser­va­tion of en­ergy other than purely philo­soph­i­cally, or in an em­piri­cal way by say­ing: “All our for­mu­las make sense if this is a thing, so let’s as­sume the world works this way, and if there is some part of the world that doesn’t we’ll get to it when we find it”.

  Also, build­ing a per­pet­ual mo­tion ma­chine (or try­ing to) is not work­ing on an un­san­swer­able prob­lem/​ques­tion of the sort I re­fer to.

  As in, work­ing on one will pre­sum­ably lead you to build bet­ter and bet­ter en­g­ines, and/​or see your failure and give up. There is a “failure state”, and there’s no ob­vi­ous way of get­ting into “meta­physics” from try­ing to re­search per­pet­ual mo­tion.

  In­deed, “Can we build a per­pet­ual mo­tion ma­chine ?” is a ques­tion I see as en­tirely valid, not worth pur­su­ing, but it’s at worst harm-neu­tral and it has proven so in the last 2,000+ years of peo­ple try­ing to an­swer it.

The ques­tions one needs not address

• George 28 Feb 2020 13:36 UTC

  5 points

  on: Ge­orge’s Shortform

  Walk­ing into a new coun­try where peo­ple speak very lit­tle English re­minds me of the dan­gers of over com­mu­ni­ca­tion.

  Go­ing into a restau­rant and say­ing: “Could I get the turk­ish coffee and an omelette with a.… crois­sant, oh, and a glass of wa­ter, no ice and, I know this is a bit weird, but I like cin­na­mon in my turk­ish coffee, could you add a bit of cin­na­mon to it ? Oh, ac­tu­ally, could you scratch the omelette and do poached eggs in­stead”

  Is a recipe for failure, at best the waiter looks at you con­fused and you can be ashamed of your poor com­mu­ni­ca­tion skills and start over.

  At worst you’re get­ting an omelette, with a cin­na­mon bun in­stead of a crois­sant, two cups of turk­ish coffee, with some ad­di­tional poached eggs and a room-tem­per­a­ture bot­tle of wa­ter.

  Maybe a far fetched ex­am­ple, but the point is: The more in­struc­tions you give, the flour­ishes you put into your re­quest, the higher the like­li­hood that the core of the re­quests gets lost.

  If you can point at the items on the menu and hold a num­ber of fingers in the air to in­di­cate the quan­tity, that’s an ideal way to or­der.

  But it’s cu­rios that this sort of over com­mu­ni­ca­tion never hap­pens in, say, Ja­pan. In places where peo­ple know very lit­tle to no English and where they don’t mind tel­ling you that what you just said made no sense (or at least they get very visi­bly em­bar­rassed, more so than their stan­dard over-the-top anx­iety, and the fact that it made no sense is in­stantly ob­vi­ous to any­one).

  It hap­pens in the coun­tries where peo­ple kinda-know English and where they con­sider it rude to ad­mit to not un­der­stand­ing you.

  Ja­panese and Taiwanese clerks, ran­dom pedes­tri­ans I ask for di­rec­tions and servers, know about as much English as I know Ja­panese or Chi­nese. But we can com­mu­ni­cate just fine via grunts, smiles, point­ing, shak­ing of heads and tak­ing out a phone to google trans­late if the in­ter­ac­tions is bar­ing close to the 30s mark with no re­s­olu­tion in sight.

  The same archtypes in In­dia and Le­banon speak close to cur­sive English though, give them 6-12 months in the UK or US plus a panache for learn­ing and they’d be a na­tive speaker (I guess it could be ar­gued that many peo­ple in In­dia speak 100% perfect English, but their own di­alect, but for the in­tents and pur­poses of this post I’m refer­ring to English as UK/​US city English).

  Yet it’s always in the sec­ond kind of coun­try where I find my over com­mu­nica­tive style fails me. Par­tially be­cause I’m more in­clined to use it, par­tially be­cause peo­ple are less in­clined to ad­mit I’m not mak­ing any sense.

  I’m pretty sure it’s this phe­nomenon is a very good metaphor or in­stan­ti­a­tion of a prin­ci­ple that ap­plies in many other situ­a­tions, es­pe­cially in ex­pert com­mu­ni­ca­tion. Or rather, in how ex­pert-lay­man vs ex­pert-ex­pert vs ex­pert-{al­most ex­pert} com­mu­ni­ca­tion works.

• George 28 Feb 2020 13:15 UTC

  1 point

  in reply to: romeostevensit's comment on: Ge­orge’s Shortform

  This just boils down to “show­ing off” though. But this makes lit­tle sense con­sid­er­ing:

  a) both gen­ders en­gage in bad prac­tices. As in, I’d ex­pect to see a lot of men do­ing cross fit, but it doesn’t make sense when you con­sider there’s a pretty even gen­der split. “Show­ing off health” in a way that’s harm­ful to health is not evolu­tion­ary adap­tive for women (where it ar­guably pays off to live for a long time, evolu­tion­ar­ily speak­ing). This is backed up by other high-risk be­hav­iors be­ing mainly a men’s thing

  b) sports are a very bad way to show off, es­pe­cially the sports that come with high risk of in­jury and per­ma­nent degra­da­tion when prac­ticed in their cur­rent ex­treme (e.g. weight lift­ing, climb­ing, gym­nas­tics, rugby, hokey). The high­est pay-off sports I can think of (in terms of so­cial sig­nal­ing) are foot­ball, amer­i­can foot­ball, bas­ket­ball and base­ball… since they are pop­u­lar and thus the com­pe­ti­tion is both in­tense and achiev­ing high rank is re­ward­ing. Other than amer­i­can foot­ball they are all pretty phys­i­cally safe as far as sports go… when there are risks, they come from other play­ers (e.g. get­ting a ball to the head) not from over-train­ing or over-perform­ing.

  So ba­si­cally, if it’s ge­netic miss-firing then I’d ex­pect to see it mis­fire al­most only in men, and this is un­true.

  If it’s “ra­tio­nal” be­hav­ior (as in, ra­tio­nal from the per­spec­tive of our pri­mate an­ces­tor) then I’d ex­pect to see the more dan­ger­ous forms of show­ing off bring the most so­cial gains rather than vice-versa.

  Granted, I do think hand­i­cap prin­ci­ple can be par­tially to blame for “start­ing” the thing, but I think it con­tinues be­cause of higher level memes that have lit­tle to do with so­cial sig­nal­ing or ge­net­ics.

• George 22 Feb 2020 22:37 UTC

  1 point

  on: Ge­orge’s Shortform

  Should dis­com­fort be a re­quire­ment for im­por­tant ex­pe­riences ?

  A while ago I was dis­cussing with a friend ma­lign­ing about the fact that there doesn’t ex­ist some sort of sub­lin­gual DMT, with an ab­sorp­tion pro­file similar to smok­ing DMT, but with­out the ran­cid taste.

  (Side note, there are some ways to get sub­lin­gual DMT: https://​​www.dmt-nexus.me/​​fo­rum/​​de­fault.aspx?g=posts&t=10240 , but you prob­a­bly won’t find it for sale at your lo­cal drug dealer and effects will differ a lot from smok­ing. In most ex­pe­riences I’ve read about I’m not even con­vinced that the peo­ple are ex­pe­rienc­ing sub­lin­gual ab­sorp­tion rather than just slowly swal­low­ing DMT with MAOIs and see­ing the effects that way)

  My point where some­thing along the way of:

  I wish there was a way to get high on DMT with­out go­ing through the un­pleas­ant ex­pe­rience of smok­ing it, I’m pretty sure that ex­pe­rience serves to “prime” your mind to some ex­tent and leads to a worst trip.

  My friend’s point was:

  We are talk­ing about one of the most re­al­ity-shat­ter­ing ex­pe­riences ever pos­si­ble to a hu­man brain that doesn’t in­volve death or per­ma­nent dam­age, surely hav­ing a small cost of en­try for that in terms of the un­pleas­ant taste is ac­tu­ally a de­sir­able side-effect.

  I kind of ended up agree­ing with my friend and I think most peo­ple would find that view­point appealing

  But

  You could make the same ar­gu­ment for some­thing like knee surgery (or any life-chang­ing surgery, which is most of them).

  You are elect­ing to do some­thing that will al­ter your life for­ever and will re­sult in you ex­pe­rienc­ing se­vere side-effects for years to come… but the step be­tween “de­cide to do it” and “sup­port ma­jor con­se­quences” has 0 dis­com­fort as­so­ci­ate to it.

  That’s not to say knee surgery is good, much like a DMT trip, I have a lot of prior of it be­ing good for peo­ple (well, in this case as­sum­ing that doc­tor recom­mends you to do it).

  But I do find it a bit strange that this is the case with most surgery, even if it’s life al­ter­ing, when I think of it in light of the DMT ex­am­ple.

  But

  If you’ve vis­ited South Korea and seen the pro­gres­sive note mu­tila­tion go­ing on in their so­ciety (I’m pretty sure this has a fancier name… see some term they use in the study of su­per-stim­uli, seag­ulls sit­ting on gi­gan­tic painted balls kinda king), I’m pretty sure the surgery ex­am­ple can be­come blur­rier.

  As in, I think it’s pretty easy to ar­gue peo­ple are do­ing a lot of un­nec­es­sary plas­tic surgery, and I’m pretty sure some cost of en­try (e.g. you must feel mild dis­com­fort for 3 hours to get this done… equiv­a­lent to say, get­ting a tat­too on your arm), would re­duce that num­ber a lot and in­tu­itively that seem like a good thing.

  It’s not like you could do that though, as in, in prac­tice you can’t re­ally do “anes­the­sia with con­trol­led pain level” it’s ei­ther zero or op­er­at­ing within a huge er­ror range (see peo­ple’s sub­jec­tive re­ports of pain af­ter den­tal anes­the­sia with similar quan­tities of li­do­caine).

• George 22 Feb 2020 21:09 UTC

  1 point

  in reply to: Dagon's comment on: Ge­orge’s Shortform

  Hmh, I ac­tu­ally did not think of that one all-im­por­tant bit. Yeap, what I de­scribed as a “meta model for Dave’s mind” is in­deed a “meta model for hu­man minds” or at least a “meta model for Amer­i­can minds” in which I plugged in some Dave-spe­cific ob­ser­va­tions.

  I’ll have to re-work this at some point with this in mind, un­less there’s already some­thing much bet­ter on the sub­ject out there.

  But again, I’ll ex­cuse this with hav­ing been so tried when I wrote this that I didn’t even re­mem­ber I did un­til your com­ment re­minded me about it.

• George 21 Feb 2020 2:13 UTC

  5 points

  on: Ge­orge’s Shortform

  90% cer­tainty that this is bs be­cause I’m wait­ing for a flight and I’m sleep de­prive, but:

  For most peo­ple there’s not a very clear way or in­cen­tive to have a meta model of them­selves in a cer­tain situ­a­tion.

  By meta model, I mean one that is mod­el­ing “high level gen­er­a­tors of ac­tion”.

  So, say that I know Dave:

  • Likes peanut-but­ter-jelly on thin cracker

  • Dis­likes peanut-but­ter-jelly in sandwiches

  • Likes but­ter fingers candy

  A com­pletely non-meta model of Dave would be:

  • If I give Dave a but­ter fingers candy box as a gift, he will en­joy it

  Another non-meta model of Dave would be:

  • If I give Dave a box of Reese’s as a gift, he will en­joy it, since I thing they are kind of a com­bi­na­tion be­tween peantu-but­ter-jelly and but­ter fingers

  A meta model of Dave would be:

  • Based on the 3 items above, I can de­duce Dave likes things which are sweet, fatty, smooth with a touch of bit­ter (let’s as­sume peanut but­ter has some bit­ter to it) and crunchy but he doesn’t like them be­ing too starchy (hence why he dis­likes sand­wiches).

  • So, if I give Dave a cup of sweet milk ice cream with bits of crunchy dark choco­late on top as a gift, he will love it.

  Now, I’m not say­ing this meta-model is a good one (and Dave is imag­i­nary, so we’ll never know). But my point is, it seems highly use­ful for us to have very good meta-mod­els of other peo­ple, since that’s how we can pre­dict their ac­tions in ex­treme situ­a­tions, sur­prise them, im­press them, make them laugh… etc

  On the other hand, we don’t need to con­struct meta-mod­els of our­selves, be­cause we can just query our “high level gen­er­a­tors of ac­tion” di­rectly, we can think “Does a cup of milk ice cream with crunchy dark choco­late on top sound tasty ?” and our high level gen­er­a­tors of ac­tion will strive to give us an es­ti­mate which will usu­ally seem “good enough to us”.

  So in some way, it’s eas­ier for us to get meta mod­els of other peo­ple, out of sim­ple ne­ces­sity and we might have bet­ter meta mod­els of other peo­ple than we have of our own self… not be­cause we couldn’t con­struct a bet­ter one, but be­cause there’s no need for it. Or at least, based on the fal­lacy of know­ing your own mind, there’s no need for it.

• George 20 Feb 2020 2:19 UTC

  7 points

  on: Ge­orge’s Shortform

  Phys­i­cal perfor­mance is one thing that isn’t re­ally “needed” in any sense of the word for most peo­ple.

  For most peo­ple, the need for phys­i­cal ac­tivity seems to boil down to the fact that you just feel bet­ter, live longer and over­all get less health re­lated is­sues if you do it.

  But on the whole, I’ve seen very lit­tle proof that ex­cel­ling in phys­i­cal ac­tivity can help you with any­thing (other than be­ing a pro­fes­sional ath­lete or trainer, that is). In­deed, it seems that the whole re­la­tion to mor­tal­ity ba­si­cally breaks down if you look at top perform. Go­ing from things like strong­man com­pe­ti­tions and amer­i­can foot­ball where life ex­pec­tancy is lower, to things like run­ning and cy­cling where some would ar­gue but ev­i­dence is lack­ing, to foot­ball and ten­nis where it’s a bit above av­er­age.

  If the sub­ject in­ter­ests you, I’ve per­son­ally looked into it a lot, and I think this is the defini­tive re­view: https://​​yorkspace.library.yorku.ca/​​xm­lui/​​bit­stream/​​han­dle/​​10315/​​32723/​​Le­mez_Srd­jan_2016_PhD.pdf

  But it’s ba­si­cally a bloody book, I per­son­ally haven’t read all of it, but I of­ten go back to it for refer­ences.

  Also, there’s the much more ob­vi­ous prob­lem with push­ing your­self to the limits, in­jury. I think this is hard to quan­tify and there’s few stud­ies look­ing at it. In my ex­pe­rience I know a sur­pris­ing amount of “ac­tive” peo­ple that got in­jured in life-al­ter­ing ways from things like skat­ing, sky­ing, snow­board­ing and even foot­ball (not in the para­plegic sense, more in the “I have a bar of tita­nium go­ing through my spine and I can’t lift more than 15kg safely” sort of way). Con­versely, 100% of my couch-dwelling bud­dies in av­er­age phys­i­cal shape doesn’t seem to suffer from any chronic pain.

  To some ex­tent, this an­noys me, though I won­der if poor stud­ies and anec­do­tal ev­i­dence is enough to war­rant that an­noy­ance.

  For ex­am­ple, I fre­quent a climb­ing gym. Now, if you look at climb­ing, it’s rel­a­tively safe, there’s two things peo­ple com­plain about most sci­at­ica and “climbers back” (ba­si­cally a very weird look­ing but not that harm­ful form of kypho­sis).

  I hon­estly found the idea rather weird… since one of the main rea­son I climb (be­sides the fact that it’s fun) is that it helps and helped me cor­rect my kypho­sis and ba­si­cally got rid of any back/​neck dis­com­fort I felt from sit­ting too much at a com­puter.

  I think this boils down to how peo­ple climb, es­pe­cially how they do boulder­ing.

  A refer­ence as to how the ex­treme kind of boulder­ing looks like: https://​​www.youtube.com/​​watch?v=7brSdnHWBko

  The two is­sues I see here is:

  • Hurl­ing limbs at tremen­dous speeds to try and crab onto some­thing tiny.

  • Fal­ling on the mat, of­ten and from large heights. Climb­ing goes two ways up and down, most peo­ple do­ing boulder­ing only care about up

  In­deed, a typ­i­cal boulder­ing run might look some­thing like: “Climb care­fully and skil­lfully as much as pos­si­ble, hurl your­self with the last bit of effort you have hop­ing you reach the top, fall on the mat rinse and re­peat”.

  This is prob­a­bly one of the stupi­dest things I’ve seen from a health per­spec­tive. You’re es­sen­tially pray­ing for ar­tic­u­la­tion dam­age, dis­lo­cat­ing a shoulder/​knee, tear­ing a mus­cle (doesn’t look pretty, I as­sume doesn’t feel nice, re­cov­ery times are long and some­times fully re­cov­er­ing is a mat­ter of years) and spine dam­age (or­tho­pe­dics don’t agree on much, but I think all would agree the worst thing you can do for your spine is fall from a con­sid­er­able height… re­peat­edly, like, dozens of time ev­ery day).

  But the thing is, you can pretty much do boulder­ing with­out this, as in, you can be “de­cent” at it with­out do­ing any of this. Per­son­ally I ap­proach boulder­ing as slowly and steadily climb­ing… to the top, with enough en­ergy to also climb down + climb­ing down when­ever I feel that I’m to ex­hausted to con­tinue. Some­how, this ap­proach to the sport is the one that give you strange looks. The peo­ple push­ing them­selves above the limits risk­ing in­jury and get­ting per­sis­tent spine dam­age from fal­ling… are the stan­dard.

  Another things I en­joy is weight lift­ing, I es­pe­cially en­joy weighted squats. Weighted squats are fun, they wake you up in the morn­ing, they are a lazy per­son ex­er­cise when you’ve got noth­ing else in dur­ing that day.

  I’ve heard peo­ple claim you can get lower back pain and in­jury from weighted squats, again, this seems con­fus­ing to me. I ac­tu­ally used to have minor lower back pain on oc­ca­sions (again, from sit­ting), the one ex­er­cise that seemed to have per­ma­nently fixed that is a squat. A squat is what I do when I feel that my back is a bit stiff and I need some help.

  But I think, again, this is be­cause I am “get­ting squats wrong”, my ap­proach to a squat is “Let me load a 5kg er­gonomic bar with 25kg, do a squat like 8 times, check my pos­ture on the last 2, if I’m able to hold it and don’t feel tired, do 5-10 more, if I still feel nice and en­er­getic af­ter a 1 minute break, rinse and re­peat”.

  But the cor­rect squat, I be­lieve, looks some­thing like this: https://​​www.youtube.com/​​watch?v=nLVJTBZtiuw

  Load­ing a bar with a few hun­dred kg, at least 2.5x your body weight, putting on a belt so that your in­testines don’t fall out and low­er­ing it “ONCE”, be­cause fuck me you’re not go­ing to be able to do that twice in a day. You should at least get some nose­bleed ev­ery 2 or 3 tries if you’re do­ing this stuff cor­rectly.

  I’ve seen this in gyms, I’ve seen this in what peo­ple recom­mend, if I google “how much weight should I squat”, the first thing I get is: https://​​www.live­strong.com/​​ar­ti­cle/​​286849-nor­mal-squat-weight/​​

  If you weigh 165 pounds and have one of the fol­low­ing fit­ness lev­els, the stan­dard for your squat one-rep max is:

  Un­trained: 110 pounds

  Novice: 205 pounds

  … etc

  To say this seems in­sane is far fetched, ba­si­cally the ad­vice around the in­ter­net seems to be “If you’ve never done this be­fore, aim for 40-60kg, if you’ve been to the gym a few times, go for 100+”

  Again, it’s hard to find data on this, but as some­one that’s pretty bloody tall who has been us­ing weight to train for years, the idea of start­ing with 50kg for a squat as an av­er­age per­son seem in­sane. I do 45kg from time to time to change things up, I’d never squat any­thing over 70kg even if you paid me… I can feel my body dur­ing the move, I can feel the ten­ta­tive pres­sure on my lower back if my pos­ture slips for a bit… that’s fine if you’re lift­ing 30kg, that seems dan­ger­ous as heck if you’re lift­ing more than your body weight, it even feels dan­ger­ous at 60kg.

  But again, I’m not do­ing squats cor­rectly, I am in the wrong here as far as peo­ple do­ing weight train­ing are con­cerned.

  I’m also wrong when it comes to ev­ery sport. I’m a bad run­ner be­cause I give up once my lungs are burn­ing for 5 min­utes straight. I’m a hor­rible swim­mer be­cause I al­ter styles and stick with low-speed ones that are over­all bet­ter for ton­ing all mus­cles and have less risk of in­jury… etc

  Granted, I don’t think that peo­ple are too pushy about go­ing to ex­tremes. The few times peo­ple tell me some ver­sion of “try harder” phrased as a friendly en­courage­ment. I finish what I’m do­ing, say thanks and lie to them that I have a slight in­jury and I’d rather not push it.

  But deep in­side I have a very strong sus­pi­cion that I’m not wrong on this thing. That some­how we’ve got our­selves into a very un­healthy memetic loop around sports, where push­ing your­self is seen as the nat­u­ral thing to do, as the thing you should be do­ing ev­ery day.

  A very dan­ger­ous memetic loop, dan­ger­ous to some ex­tent in that it causes in­jury, but much more dan­ger­ous be­cause it might be dis­cour­ag­ing peo­ple from sports. Both in that they try once, get an in­jury and quit. Or in that they see it, they think it’s too hard (and, I think it is, the way most peo­ple do it) and they never re­ally bother.

  I’m hon­estly not sure why it might have started…

  The ob­vi­ous rea­son is that it phys­i­cally feels good to do it, lift­ing a lot of run­ning more than your body tells you that you should is “nice”. But it’s nice in the same way that smok­ing a tiny bit of hero­ine be­fore go­ing about your day is nice (as in, quite liter­ally, it seems to me the feel­ings are re­lated and I think there’s some phar­ma­colog­i­cal ev­i­dence to back that up). It’s nice to do it once to see how it is, maybe I’ll do it ev­ery few months if I get the oc­ca­sion and I feel I need a men­tal boost… but I wouldn’t nec­es­sar­ily ad­vise it or struc­ture my life around it.

  The other ob­vi­ous rea­son is that it’s a sta­tus thing, the whole “I can do this thing bet­ter than you thus my rank in the hi­er­ar­chy is higher”. But then… why is it so com­mon with both gen­ders, I’d see some rea­son for men to do this, be­cause his­tor­i­cally we’ve been do­ing it, but women com­pet­ing in sports is a re­cent things, hardly “built into our na­ture” and most of the ones I know that prac­tice things like climb­ing are among the most chilled out dudes I’ve ever meet.

  The last rea­son might be that it’s about break­ing a psy­cholog­i­cal bar­rier, the “Oh, I to­tally thought I couldn’t do that, but ap­par­ently I can”. But it seems to me like a very very bad way of do­ing that. I can think of many other safer bet­ter ways from solv­ing a hard calcu­lus prob­lem to learn­ing a for­eign lan­guage in a month to forc­ing your­self to write an ar­ti­cle ev­ery day… you know, things that have zero risks of paral­y­sis and long term dam­age in­volved.

  But I think at this point imi­ta­tion alone is enough to keep it go­ing.

  The “real” rea­son if I take the out­side view is prob­a­bly that that’s how sports are sup­posed to be done and I just got stuck with a weird per­spec­tive be­cause “I play things safe”.

• George 19 Feb 2020 23:23 UTC

  1 point

  in reply to: Dagon's comment on: Does donat­ing to EA make sense in light of the mere ad­di­tion para­dox ?

  To the ex­tent that you’re pur­su­ing top­ics that EA or­ga­ni­za­tions are also pur­su­ing, you should prob­a­bly donate to their recom­mended char­i­ties rather than try­ing to do it your­self or go­ing through less-mea­sured char­i­ties.

  Well yes, this is ba­si­cally the crux of my ques­tion.

  As in, I ob­vi­ously agree with the E and I tend do agree with the A , buy my is­sue is why how A seems to be defined in EA (as in, mainly around im­prov­ing the lives of peo­ple that you will never in­ter­act with or ‘care’ about on a per­sonal level).

  So I agree with: I should donate to some of my fa­vorite writ­ers/​video-mak­ers that are less pop­u­lar and thus might be kept in busi­ness by 20$ monthly on pa­teron is an­other hun­dred peo­ple think like me. (effi­cient as op­posed, to, say, donat­ing to an org that helps all artists or donat­ing to well-off cre­ators).

  I also agree with: It’s effi­cient to save a life halfway across the globe for x,000$ as op­posed to one in the EU where it would cost x00,000$ to achieve a similar ad­di­tion in healthy life years.

  Where I don’t un­der­stand how the in­tu­ition re­ally works is “Why is it bet­ter to save the life of a per­son you will never know/​meet than to help 20 artists that you love” (or some such equiv­alence).

  As in, I get there some in­tu­ition about it be­ing “bet­ter” and I agree that might be strong enough in some peo­ple that it’s just “ob­vi­ous”, but my think­ing was that there might be some sort of bet­ter ethic-rooted ar­gu­ment for it.

• George 19 Feb 2020 18:58 UTC

  4 points

  in reply to: Kaj_Sotala's comment on: Does donat­ing to EA make sense in light of the mere ad­di­tion para­dox ?

  No wor­ries, I wasn’t as­sum­ing you were a speaker for the EA com­mu­nity here, I just wanted to bet­ter un­der­stand pos­si­ble mo­ti­va­tions for donat­ing to EA given my cur­rent per­spec­tive on ethics. I think the an­swer you gave out­line on such line of rea­son­ing quite well.

• George 19 Feb 2020 17:00 UTC

  3 points

  in reply to: Kaj_Sotala's comment on: Does donat­ing to EA make sense in light of the mere ad­di­tion para­dox ?

  Utili­tar­i­anism is not the only sys­tem that be­comes prob­le­matic if you try to for­mal­ize it enough; the prob­lem is that there is no com­pre­hen­sive moral sys­tem that wouldn’t ei­ther run into para­dox­i­cal an­swers, or be so vague that you’d need to fill in the miss­ing gaps with in­tu­ition any­way.

  Agree, I wasn’t try­ing to im­ply otherwise

  Any de­ci­sion that you make, ul­ti­mately comes down to your in­tu­ition (that is: de­ci­sion-weight­ing sys­tems that make use of in­for­ma­tion in your con­scious­ness but which are not them­selves con­sciously ac­cessible) fa­vor­ing one de­ci­sion or the other. You can try to for­mu­late ex­plicit prin­ci­ples (such as util­i­tar­i­anism) which ex­plain the prin­ci­ples be­hind those in­tu­itions, but those ex­plicit prin­ci­ples are always go­ing to only cap­ture a part of the story, be­cause the full de­ci­sion crite­ria are too com­plex to de­scribe.

  Also agree, as in, this is how I usu­ally for­mu­late my moral de­ci­sion and it’s ba­si­cally a prag­matic view on ethics, which is one I gen­er­ally agree with.

  is just “the kinds where donat­ing to EA char­i­ties makes more in­tu­itive sense than not donat­ing”; of­ten peo­ple de­scribe these kinds of moral in­tu­itions as “util­i­tar­ian”, but few peo­ple would ac­tu­ally en­dorse all of the con­clu­sions of purely util­i­tar­ian rea­son­ing.

  So ba­si­cally, the idea here is that it ac­tu­ally makes in­tu­itive moral sense for most EA donors to donate to EA causes ? As in, it might be that they par­tially jus­tify it with one moral sys­tem or an­other, but at the end of the day it seems “in­tu­itively right” to them to do so.