shminux

• shminux 23 Jul 2019 6:16 UTC

  2 points

  on: Where is the Mean­ing?

  In­deed, words don’t mean things on their own, peo­ple use words to mean things. But with enough shared con­text, it’s a rea­son­able ap­prox­i­ma­tion to say that words mean things. Up un­til they don’t. Scott A ex­pressed it rather well when dis­cussing whether whale is a fish. https://​​slat­estar­codex.com/​​2014/​​11/​​21/​​the-cat­e­gories-were-made-for-man-not-man-for-the-cat­e­gories/​​

• shminux 21 Jul 2019 0:40 UTC

  20 points

  on: Do you fear the rock or the hard place?

  Ac­tive listen­ing is always a good start. One does not need to agree or ex­press any opinion what­so­ever, just em­pathize with the other per­son, by restat­ing what they say in your own words, ask­ing/​nam­ing the feel­ings they might have, and mak­ing them feel un­der­stood. In your ex­am­ple:

  more gun con­trol; Y = less gun con­trol; R = peo­ple un­able to defend them­selves and hav­ing their rights taken away; H = in­creased risk of mass shoot­ings, suicides, and chil­dren shoot­ing them­selves or oth­ers.

  “I can see why gun con­trol is im­por­tant to you. Mass shoot­ings, suicides and ac­ci­den­tal deaths are ter­rible, and you are mak­ing a good point that easy availa­bil­ity of guns leads to more of these awful events. And you are say­ing that more gun con­trol would make it harder to get guns and lower the odds of some­one us­ing them, es­pe­cially ac­ci­den­tally or im­pul­sively. Is that what you are say­ing? Please cor­rect me if I missed or mis­stated some­thing.”

  “The po­ten­tial fu­ture you are de­scribing, where law-abid­ing cit­i­zens are un­able to defend them­selves from armed crim­i­nals, or worse are un­able to re­sist when their rights taken away by the gov­ern­ment agen­cies, does sound pretty scary. Looks like your point is that more gun con­trol would be a step to­ward such a fu­ture, and you find this pos­si­bil­ity ter­rify­ing. Is this a fair sum­mary?”

  Be­fore you can logic with some­one, they need to feel safe with you emo­tion­ally. This ap­plies to most peo­ple, whether as­piring ra­tio­nal­ists or not. Ac­tive listen­ing is a good way to cross this emo­tional dis­tance. Your own views and opinions can be ex­pressed af­ter, and do not have to be force­ful, more like a point to bring up and ask them to con­sider it and the ar­gu­ments they can help you eval­u­ate. There is no guaran­tee of them chang­ing their mind, or you chang­ing yours, or any con­ver­gence what­so­ever, but at least you will re­main friendly and can go for a beer to­gether af­ter.

• shminux 20 Jul 2019 2:36 UTC

  5 points

  in reply to: philh's comment on: Why ar­tifi­cial op­ti­mism?

  My point, as usual not well ar­tic­u­lated, is that the ques­tion “how to fix things?” is way down the line. First, the ap­par­ent “dis­tor­tonary dy­nam­ics” may only be an ap­pear­ance of one. The situ­a­tion de­scribed is a com­mon if metastable equil­ibrium, and it is not to me whether it is “dis­tor­tionary” or not. So, af­ter the first im­pulse to “fix” the sta­tus quo passes, it’s good to in­ves­ti­gate it first. I didn’t mean to sug­gest one *should* take ad­van­tage of the situ­a­tion, merely to in­ves­ti­gate if one *could*. Just like in one of Eliezer’s ex­am­ples, see­ing a sin­gle over­val­ued house does not help you profit on it. And if there is in­deed a way to do so, mean­ing the equil­ibrium is shal­low enough, the next step would be to model the sys­tem as it climbs out of the cur­rent state and rolls down one of many pos­si­ble other equil­ibria. Those other ones may be even worse, but the met­ric ap­plied to eval­u­ate the cur­rent state vs the imag­i­nary ideal state. A few ex­am­ples:

  • Most new busi­nesses fail within 3 years, but with­out new as­pired en­trepreneurs hav­ing a too rosy es­ti­mate of their chances for suc­cess (cf the Op­ti­mism bias men­tioned in the OP) there would be a lot fewer new busi­nesses and ev­ery­one would be worse off in the long run.

  • Karl Marx was call­ing for the free­dom of the work­ing class through rev­olu­tion, but any ac­tual rev­olu­tion makes things worse for ev­ery­one, in­clud­ing the work­ing class, at least in the short to medium run (years to decades). If any­thing, his­tory showed that in­cre­men­tal evolu­tion­ary ad­vances work a lot bet­ter.

  • The dis­cussed Potemkin Villages, in mod­er­a­tion, can be an emo­tional stim­u­lus for peo­ple to try harder. In fact, a lot of the fake statis­tics in the former Soviet Union served that pur­pose.

• shminux 19 Jul 2019 2:45 UTC

  5 points

  in reply to: Alexei's comment on: Nor­mal­is­ing util­ity as will­ing­ness to pay

  Hah, the auc­tion in­ter­pre­ta­tion of Quan­tum Me­chan­ics! Won­der what re­stric­tions would need to be im­posed on the bid­ders in or­der to pre­serve both the en­tan­gle­ment and rel­a­tivity.

• shminux 19 Jul 2019 2:38 UTC

  2 points

  in reply to: Dagon's comment on: Dou­ble­crux is for Build­ing Products

  Thanks, all good points. Wish we all cared to ap­ply mod­ifi­ca­tions like that.

  I also agree with off-the-shelf sup­port ad­van­tages of An­droid, though the up­date mechanism out­side of the play store seems to be noth­ing spe­cial. As for weigh­ing the di­men­sions differ­ently, there is definitely a sig­nifi­cant differ­ence: he puts pre­mium on not rock­ing the boat, while mine is on de­liv­er­ing sim­ple main­tain­able low-risk solu­tions. In gen­eral, sim­plic­ity is greatly un­der­val­ued. You prob­a­bly can re­late.

• shminux 18 Jul 2019 7:20 UTC

  2 points

  in reply to: Raemon's comment on: Dou­ble­crux is for Build­ing Products

  1. I’m just re­mov­ing an un­nec­es­sary as­sump­tion, to avoid the dis­cus­sion about what it means to be right or wrong, and whether there is a sin­gle right an­swer.

  2. I don’t have the clout to change the boss’s mind. Mak­ing sub­op­ti­mal de­ci­sions based on im­plicit un­jus­tified as­sump­tions and in­com­plete in­for­ma­tion, and then get­ting an­gry when challenged is some­thing most of hu­mans do at some point.

• shminux 17 Jul 2019 8:13 UTC

  8 points

  on: Dou­ble­crux is for Build­ing Products

  A few points. I am cur­rently in the situ­a­tion where the com­pany I work for has been wast­ing mil­lions, slip­ping sched­ule and in­creas­ing risk due to pick­ing a wrong en­vi­ron­ment for the pro­ject (An­droid in­stead of Linux. An­droid is nec­es­sary for apps, but ter­rible be­yond be­lief for IoT, where one does not need the play­store.) The owner of the com­pany drank the An­droid Kool Aid and the pro­ject man­ager re­fuses to tell him what this mar­ket­ing gim­mick end­ing up cost­ing. It would be im­poli­tic for me to challenge the two of them, and it’s not my money that is wasted. We have an oc­ca­sional “this would be so much eas­ier on Linux” mo­ment, but it goes nowhere, be­cause the de­ci­sion has been made and any change in the course, even if pro­jected to save money, would be per­ceived as risky and would ex­pose some­one’s in­com­pe­tence. So we are stuck at the “agree to dis­agree” stage, and the pro­ject man­ager makes the call, with­out any in­ter­est in dis­cussing the mer­its and get­ting an­gry at any men­tion of an al­ter­na­tive.

  Re your set of at­ti­tudes. I find that one does not need to be­lieve in any­thing like “ob­jec­tive re­al­ity is real” to use the tech­nique. So, let me mod­ify your list a bit

  Epistemic hu­mil­ity: “maybe I’m the wrong one” → “Maybe my ap­proach is not the op­ti­mal one”
  Good faith “I trust my part­ner to be co­op­er­at­ing with me”
  Belief that ob­jec­tive re­al­ity is real → Belief that bet­ter ap­proaches are pos­si­ble
  ”there’s an ac­tual right an­swer here → “there is a chance of a bet­ter an­swer, where “bet­ter” can be agreed on by all par­ties” , and it’s bet­ter for each of us if we’ve both found it”
  Earnest curiosity

• shminux 17 Jul 2019 2:05 UTC

  11 points

  on: Wolf’s Dice

  I’m con­fused about some­thing. In re­al­ity there are no perfect dice, all dice are bi­ased in some way, in­ten­tion­ally or not. Thus wouldn’t a more re­al­is­tic ap­proach be some­thing like “Given the dataset, con­struct the (mul­ti­di­men­sional) prob­a­bil­ity dis­tri­bu­tion of bi­ases.” Why priv­ilege the “un­bi­ased” hy­poth­e­sis?

• shminux 17 Jul 2019 0:28 UTC

  5 points

  on: Nutri­tion is Satisficing

  Good points! Also helps one avoid the Good­hart trap, op­ti­miz­ing for a wrong thing. Also ap­plies to sav­ings: Dilbert’s 9-point fi­nan­cial plan is one of satis­fic­ing, not op­ti­miz­ing:

  Dilbert cre­ator Scott Adams claims this is “ev­ery­thing you need to know about per­sonal in­vest­ing”:

  1. Make a will

  2. Pay off your credit cards

  3. Get term life in­surance if you have a fam­ily to support

  4. Fund your 401k to the max­i­mum [Or your lo­cal equiv­a­lent em­ployer con­tri­bu­tion match­ing]

  5. Fund your IRA to the max­i­mum [Or your lo­cal equiv­a­lent of tax-de­ductible in­vest­ment and/​or tax-free in­ter­est growth]

  6. Buy a house if you want to live in a house and can af­ford it

  7. Put six months worth of ex­penses in a money-mar­ket account

  8. Take what­ever money is left over and in­vest 70% in a stock in­dex fund and 30% in a bond fund through any dis­count bro­ker and never touch it un­til retirement

  9. If any of this con­fuses you, or you have some­thing spe­cial go­ing on (re­tire­ment, col­lege plan­ning, tax is­sues), hire a fee-based fi­nan­cial plan­ner, not one who charges a per­centage of your portfolio

• shminux 15 Jul 2019 21:49 UTC

  12 points

  on: Why ar­tifi­cial op­ti­mism?

  This model raises an im­por­tant ques­tion (with im­pli­ca­tions for the real world): if you’re a de­tec­tive in the king­dom of the gullible king who is at least some­what aware of the re­al­ity of the situ­a­tion and the dis­tor­tonary dy­nam­ics, and you want to fix the situ­a­tion (or at least re­duce harm), what are your op­tions?

  I sus­pect that is not the first ques­tion to ask. In the spirit of Inad­e­quate Equil­ibria, a bet­ter ini­tial ques­tion would be, “Can you take ad­van­tage of the ap­par­ent ir­ra­tional­ity of the situ­a­tion?”, and “What frac­tion of the pop­u­la­tion would have to co­op­er­ate to change things for the bet­ter?” and if there is no clear an­swer to ei­ther, then the situ­a­tion is not as ir­ra­tional as it seems, and the ar­tifi­cial op­ti­mism is, in fact, the best policy un­der the cir­cum­stances.

• shminux 13 Jul 2019 7:16 UTC

  3 points

  on: What are we pre­dict­ing for Neu­ral­ink event?

  No idea what they will show off, but, how­ever much I would like to have the in­ter­net at my nerve tips, it is un­likely to be that.

• shminux 12 Jul 2019 6:51 UTC

  6 points

  in reply to: avturchin's comment on: The AI Timelines Scam

  why is the above com­ment so badly down­voted?

• shminux 12 Jul 2019 2:50 UTC

  2 points

  in reply to: cousin_it's comment on: Reli­gion as Goodhart

  I guess my point got lost in the shuffle. It’s right there in the OP, though. The adap­ta­tion is look­ing to an ex­ter­nal higher power for an­swers. Ini­tially it would have been where to hunt, but even­tu­ally Good­harted into pray­ing and so on.

• shminux 11 Jul 2019 5:04 UTC

  17 points

  on: The AI Timelines Scam

  I see clear par­allels with the treat­ment of Sabine Hossen­felder blow­ing the whis­tle on the par­ti­cle physics com­mu­nity push­ing for a new $20B par­ti­cle ac­cel­er­a­tor. She has been go­ing through the same ad­ver­sity as any high-pro­file defec­tor from a sci­en­tific com­mu­nity, and the ar­gu­ments against her are the same ones you are list­ing.

• shminux 11 Jul 2019 5:02 UTC

  2 points

  in reply to: ioannes_shade's comment on: What’s state-of-the-art in AI un­der­stand­ing of the­ory of mind?

  Hu­mans are triv­ial to kill. Phys­i­cally, chem­i­cally, biolog­i­cally or psy­cholog­i­cally. And a com­bi­na­tion of those would be even more effec­tive in col­laps­ing the hu­man pop­u­la­tion. I will not go here into the de­tails, to avoid ar­gu­ments and nega­tive at­ten­tion. And if your ar­gu­ment is that hu­mans are tough to kill, then look into the his­toric data of pop­u­la­tion col­lapse, and that was with­out any ad­ver­sar­ial pres­sure. Or with, if you con­sider the in­dige­nous pop­u­la­tion of the Amer­i­can con­ti­nent.

• shminux 10 Jul 2019 20:48 UTC

  4 points

  in reply to: dxu's comment on: Ex­per­i­men­tal Open Thread April 2019: So­cratic method

  t seems, based on what you’re say­ing, that you’re tak­ing “re­al­ity” to mean some preferred set of mod­els.

  Depend­ing on the mean­ing of the word preferred. I tend to use “use­ful” in­stead.

  my be­lief in an ex­ter­nal re­al­ity, if we phrase it in the same terms we’ve been us­ing (namely, the lan­guage of mod­els and pre­dic­tions), can be sum­ma­rized as the be­lief that there is some (reach­able) model within our hy­poth­e­sis space that can perfectly pre­dict fur­ther in­puts.

  It’s a com­mon be­lief, but it ap­pears to me quite un­founded, since it hasn’t hap­pened in mil­len­nia of try­ing. So, a di­rect ob­ser­va­tion speaks against this model.

  I ex­pect that (bar­ring an ex­is­ten­tial catas­tro­phe that erases us en­tirely) there will even­tu­ally come a point when we have the “full pic­ture” of physics, such that no ex­per­i­ment we perform will pro­duce a re­sult we find sur­pris­ing.

  It’s an­other com­mon be­lief, though sep­a­rate from the be­lief of re­al­ity. It is a be­lief that this re­al­ity is effi­ciently know­able, a bold pre­dic­tion that is not sup­ported by ev­i­dence and has hints to the con­trary from the com­plex­ity the­ory.

  If we ar­rive at such a model, I would be com­fortable refer­ring to that model as “true”, and the phe­nom­ena it de­scribes as “re­al­ity”.

  Yes, in this highly hy­po­thet­i­cal case I would agree.

  Ini­tially, I took you to be as­sert­ing the nega­tion of the above state­ment—namely, that we will never stop be­ing sur­prised by the uni­verse, and that our mod­els, though they might asymp­tot­i­cally ap­proach a rate of 100% pre­dic­tive suc­cess, will never quite get there.

  I make no claims one way or the other. We tend to get bet­ter at pre­dict­ing ob­ser­va­tions in cer­tain limited ar­eas, though it tends to come at a cost. In high-en­ergy physics the progress has slowed to a stand­still, no in­ter­est­ing ob­ser­va­tions has been pre­dicted since last mil­len­nium. Gen­eral Rel­a­tivity plus the stan­dard model of the par­ti­cle physics have stood un­changed and un­challenged for decades, the magic num­bers they re­quire re­main­ing un­ex­plained since the Higgs mass was pre­dicted a long time ago. While this sug­gests that, yes, we will prob­a­bly never stop be­ing sur­prised by the -uni­verse- (no strike through markup here?) ob­ser­va­tions, I make no such claims.

  It is this claim that I find im­plau­si­ble, since it seems to im­ply that there is no model in our hy­poth­e­sis space ca­pa­ble of pre­dict­ing fur­ther in­puts with 100% ac­cu­racy—but if that is the case, why do we cur­rently have a model with >99% pre­dic­tive ac­cu­racy?

  Yes we do have a good han­dle on many iso­lated sets of ob­ser­va­tions, though what you mean by 99% is not clear to me. Similarly, I don’t know what you mean by 100% ac­cu­racy here. I can imag­ine that in some limited ar­eas 100% ac­cu­racy can be achiev­able, though we of­ten get sur­prised even there. Say, in math the Hilbert Pro­gram had a sur­pris­ing twist. Feel free to give ex­am­ples of 100% pre­dictabil­ity, and we can dis­cuss them. I find this model (of no uni­ver­sal perfect pre­dictabil­ity) very plau­si­ble and con­firmed by ob­ser­va­tions. I am still un­sure what you mean by co­in­ci­dence here. The dic­tio­nary defines it as “A re­mark­able con­cur­rence of events or cir­cum­stances with­out ap­par­ent causal con­nec­tion.” and that open a whole new can of worms about what “ap­par­ent” and “causal” mean in the situ­a­tion we are de­scribing, and we soon will be back to a cir­cu­lar ar­gu­ment of im­ply­ing some un­der­ly­ing re­al­ity to ex­plain why we need to pos­tu­late re­al­ity.

  Now, per­haps you ac­tu­ally do hold the po­si­tion de­scribed in the above para­graph. (If you do, please let me know.) But based on what you wrote, it doesn’t seem nec­es­sary for me to as­sume that you do. Rather, you seem to be say­ing some­thing along the lines of, “It may be tempt­ing to take our cur­rent set of mod­els as de­scribing how re­al­ity ul­ti­mately is, but in fact we have no way of know­ing this for sure, so it’s best not to as­sume any­thing.”

  I don’t dis­agree with the quoted part, it’s a de­cent de­scrip­tion.

  If that’s all you’re say­ing, it doesn’t nec­es­sar­ily con­flict with my view (al­though I’d sug­gest that “re­al­ity doesn’t ex­ist” is a rather poor way to go about ex­press­ing this sen­ti­ment). Nonethe­less, if I’m cor­rect about your po­si­tion, then I’m cu­ri­ous as to what you think it’s use­ful for? Pre­sum­ably it doesn’t help make any pre­dic­tions (al­most by defi­ni­tion), so I as­sume you’d say it’s use­ful for dis­solv­ing cer­tain kinds of con­fu­sion. Any ex­am­ples, if so?

  “re­al­ity doesn’t ex­ist” was not my origi­nal state­ment, it was “mod­els all the way down”, a suc­cinct way to ex­press the cur­rent state of knowl­edge, where all we get is ob­ser­va­tions and lay­ers of mod­els based on them pre­dict­ing fu­ture ob­ser­va­tions. It is use­ful to avoid get­ting astray with ques­tions about ex­is­tence or non-ex­is­tence of some­thing, like num­bers, mul­ti­verse or qualia. If you stick to mod­els, these ques­tions are dis­solved as mean­ingless (not use­ful for pre­dict­ing fu­ture ob­ser­va­tions). Just like the ques­tion of count­ing an­gels on the head of a pin. Teg­mark Level X, the hard prob­lem of con­scious­ness, MWI vs Copen­hagen, none of these are worth ar­gu­ing over un­til and un­less you sug­gest some­thing that can be po­ten­tially ob­serv­able.

• shminux 10 Jul 2019 19:36 UTC

  2 points

  in reply to: dxu's comment on: If physics is many-wor­lds, does ethics mat­ter?

  As I men­tioned there, Jes­sica was ap­par­ently pissed and un­char­ac­ter­is­ti­cally un­char­i­ta­ble in her re­ply. The up­vote count in this case seems to re­flect tribal af­fili­a­tions more than any­thing.

• shminux 10 Jul 2019 19:34 UTC

  2 points

  in reply to: AlexMennen's comment on: If physics is many-wor­lds, does ethics mat­ter?

  I said “a” not “the”. Yes, you could also quote Teg­mark and Deutsch. I tend to fa­vor a prag­matic ap­proach to sci­ence, same as Sabine. You don’t have to, but it helps to re­al­ize that untestable mod­els still “add up to nor­mal­ity”, to quote The Founder, and so have no bear­ing on your ethics.

• shminux 10 Jul 2019 15:50 UTC

  −19 points

  on: If physics is many-wor­lds, does ethics mat­ter?

  Con­sider read­ing a real physi­cist’s take on the is­sue: Why the mul­ti­verse is re­li­gion, not sci­ence.

• shminux 10 Jul 2019 15:16 UTC

  3 points

  in reply to: cousin_it's comment on: Reli­gion as Goodhart

  You got it back­wards. Faith never recom­mends ran­dom­iza­tion, it jus­tifies it. Like trust­ing the tea leaves to pre­dict the fu­ture.

  ran­dom­ness in physics is cheap, and na­ture uses ran­dom­iza­tion in many rock-pa­per-scis­sors games with­out re­quiring re­li­gion or even brains.

  Yes, ran­dom­ness in physics is cheap, but I have a hard time find­ing ex­am­ples of, say, a uniform or ex­po­nen­tial dis­tri­bu­tion in the be­hav­iors of higher an­i­mals. Just be­cause some­thing is cheap at a lower lev­els (e.g. quan­tum pro­cesses), it does not mean that it is cheap at the higher lev­els. I wel­come ex­am­ples of higher-lev­els rock-pa­per-scis­sors type of be­hav­ior.