G Gordon Worley III

• G Gordon Worley III 14 Jun 2019 17:56 UTC

  4 points

  on: Un­known Un­knowns in AI Alignment

  Don­ald Hob­son gives a com­ment be­low ex­plain­ing some rea­son­ing around deal­ing with un­known un­knowns, but it’s not a di­rect an­swer to the ques­tion, so I’ll offer that.

  The short an­swer is “yes”.

  The longer an­swer is that this is one of the fun­da­men­tal con­sid­er­a­tions in ap­proach­ing AI al­ign­ment and is why some or­ga­ni­za­tions, like MIRI, have taken an ap­proach that doesn’t drive straight at the ob­ject-level prob­lem and in­stead tack­les is­sues likely to be foun­da­tional to any ap­proach to al­ign­ment that could work. In fact you might say the big schism be­tween MIRI and, say, OpenAI, is that MIRI places greater em­pha­sis on ad­dress­ing the un­known whereas OpenAI ex­pects al­ign­ment to look more like an en­g­ineer­ing prob­lem with rel­a­tively small and not es­pe­cially dan­ger­ous un­known un­knowns.

  (note: I am not af­fili­ated with ei­ther or­ga­ni­za­tion so this is an in­formed opinion on their gen­eral ap­proaches, and also note that nei­ther or­ga­ni­za­tion is mono­lithic and in­di­vi­d­ual re­searches vary greatly in their as­sess­ment of these risks.)

  My own efforts ad­dress­ing AI al­ign­ment are largely about ad­dress­ing these sorts of ques­tions, be­cause I think we still poorly un­der­stand what al­ign­ment even re­ally means. In this sense I know that there is a lot we don’t know, but I don’t know all of what we don’t know that we’ll need to (so known un­known un­knowns).

• G Gordon Worley III 13 Jun 2019 23:21 UTC

  4 points

  in reply to: Hafurelus's comment on: Cry­on­ics be­fore nat­u­ral death. List of com­pa­nies?

  I’ve not heard of any­one try­ing this. I imag­ine it’s a bad idea for a cou­ple rea­sons:

  • most would-be pa­tients live in coun­tries where what you want is illegal

  • those would-be pa­tients’ ac­cess to cry­on­ics de­pends largely on cry­on­ics or­ga­ni­za­tions be­ing in good stand­ing with the lo­cal gov­ern­ment in case they die un­ex­pect­edly or don’t want to travel far

  • even if you found a ju­ris­dic­tion where you could do what you want it might have reper­cus­sions back where the or­ga­ni­za­tion is based and stores the brains/​bod­ies be­cause the home coun­try/​state/​mu­ni­ci­pal­ity might for­bid im­port due to how the brain/​body was obtained

  • the coun­tries where would-be pa­tients live might for­bid them from con­tract­ing for such a ser­vice in a lo­ca­tion where it is le­gal (com­pare the way some coun­tries re­quire their cit­i­zens fol­low na­tional laws when abroad, and that such cit­i­zens can be pros­e­cuted for ac­tions they took in for­eign na­tions)

  I think if you wanted to do this you would need to find a ju­ris­dic­tion that would be okay with this and be also oth­er­wise suit­able for bas­ing a cry­on­ics op­er­a­tion. My guess is the set of places that meet both crite­ria is empty. And this ig­nor­ing the pa­tient ac­cess is­sues I men­tioned. Since the cur­rent mar­ket for cry­on­ics is quite small, my guess is that there just isn’t enough de­mand to make this hap­pen, since I’m sure with enough de­mand you’d have the money to make a fa­vor­able ju­ris­dic­tion suit­able for bas­ing a cry­on­ics op­er­a­tion in.

• G Gordon Worley III 13 Jun 2019 19:44 UTC

  7 points

  on: Cry­on­ics be­fore nat­u­ral death. List of com­pa­nies?

  Aside from the case where you may have ac­cess to eu­thana­sia, the an­swer is no. The is­sue is that cry­on­ics, where it is legally al­lowed, is con­sid­ered a mor­tu­ary pro­ce­dure rather than a med­i­cal pro­ce­dure. The rea­sons for do­ing this are a bit in­volved, but can be summed up by say­ing it was eas­ier to get le­gal ap­proval for a novel pro­ce­dure on dead bod­ies than on live ones.

  In the­ory it seems likely you could get a bet­ter preser­va­tion by anes­thetiz­ing a live pa­tient, re­plac­ing their blood, and slow­ing cool­ing the body and let­ting them die slowly while freez­ing rather than dy­ing first and then start­ing the cool­ing pro­cess, but this is ex­tremely legally com­pli­cated be­cause it both in­volves a live pa­tient, so it’s a med­i­cal pro­ce­dure, and it kills the pa­tient, so it’s eu­thana­sia (or so we hope; if it wasn’t painless you definitely wouldn’t be al­lowed to do it!). This would re­quire a level of ac­cep­tance of cry­on­ics we have no rea­son to be­lieve is forth­com­ing.

  So we are left with the case where you have to die first be­fore be­ing cryo-pre­served. How­ever, it’s even a bit more com­pli­cated than that, be­cause how you die mat­ters. Mor­tu­ary pro­ce­dures can’t be­gin un­til a pa­tient has a com­pleted death cer­tifi­cate from a doc­tor in most places, and in some cases you can’t for­mally com­plete that pro­cess with­out an au­topsy to de­ter­mine cause of death, es­pe­cially in cases that look sus­pi­cious like a mur­der or suicide. In fact, with­out mod­ern as­sisted suicide laws, suicide gen­er­ally re­quires an au­topsy by law, which will of course ruin your chance of preser­va­tion.

  The only known, re­li­able way of do­ing what you pro­pose (and I know of cases in there past where it suc­cess­fully hap­pened), is that a pa­tient with a ter­mi­nal ill­ness en­tered a hos­pice near a cry­on­ics fa­cil­ity with a cry­on­ics team on standby and then re­fused all food and wa­ter. It takes sev­eral days to die this way de­pend­ing on body com­po­si­tion, and at time of death the doc­tor on staff can quickly mark that you died of nat­u­ral causes (I don’t en­tirely un­der­stand why this doesn’t count as suicide, but it ap­par­ently doesn’t) and the pro­ce­dure can be­gin within min­utes. That, to the best of my knowl­edge, is the state-of-the-art in cry­onic preser­va­tion: cry­ocide by star­va­tion/​de­hy­dra­tion.

• G Gordon Worley III 12 Jun 2019 22:05 UTC

  2 points

  in reply to: Stuart_Armstrong's comment on: How much can value learn­ing be dis­en­tan­gled?

  Right, both of these views on truth, tra­di­tional ra­tio­nal­ity and post­mod­ernism, re­sult in the­o­ries of truth that don’t quite line up with what we see in the world but in differ­ent ways. The tra­di­tional ra­tio­nal­ity view fails to ac­count for the fact that hu­mans judge truth and we have no ac­cess to the view from nowhere, so it’s right that tra­di­tional ra­tio­nal­ity is “wrong” in the sense that it in­cor­rectly as­sumes it can gain priv­ileged ac­cess to the truth of claims to know which ones are facts and which ones are false­hoods. The post­mod­ernist view makes an op­po­site and only slightly less equal mis­take by cor­rectly notic­ing that hu­mans judge truth but then failing to ad­e­quately ac­count for the ways those judge­ments are en­tan­gled with a shared re­al­ity. The way through is to see that both there is some­thing shared out there that there can in the­ory be a fact of the mat­ter of and also re­al­iz­ing that we can’t di­rectly as­cer­tain those facts be­cause we must do so across the gap of (sub­jec­tive) ex­pe­rience.

  As always, I say it comes back to the prob­lem of the crite­rion and our failure to ad­e­quately ac­cept that it de­mands we make a leap of faith, small though we may man­age to make it.

• G Gordon Worley III 12 Jun 2019 2:30 UTC

  2 points

  in reply to: rohinmshah's comment on: Con­clu­sion to the se­quence on value learning

  The stan­dard re­but­tal here is that even if a su­per­in­tel­li­gent AI sys­tem is not goal di­rected, we should be con­cerned that the AI will spon­ta­neously de­velop goal di­rected be­hav­ior be­cause it is in­stru­men­tally valuable to do­ing what­ever it is do­ing (and is not “do­ing what­ever it is do­ing” a “goal”, even if the AI does not con­ceive of it as a goal, the same way as the calcu­la­tor has a “goal” or pur­pose, even if the calcu­la­tor is un­aware of it). This is of course con­tin­gent on it be­ing “su­per­in­tel­li­gent”.

  For what it’s worth this is also the ori­gin, as I re­call it, of con­cerns about pa­per­clip max­i­miz­ers: you won’t build an AI that sets out to tile the uni­verse with pa­per­clips, but through a se­ries of un­for­tu­nate mi­s­un­der­stand­ings it will, as a sub­agent or an in­stru­men­tal ac­tion, end up op­ti­miz­ing for pa­per­clips any­way be­cause it seemed like a good idea at the time.

• G Gordon Worley III 11 Jun 2019 20:15 UTC

  2 points

  on: One-step hy­po­thet­i­cal preferences

  Fi­nally, the hu­man ex­presses a judge­ment about the states of M, men­tally cat­e­goris­ing a set of states as bet­ter than an­other. This is an anti-sym­met­ric par­tial func­tion J:S×S→R, a par­tial func­tion that is non triv­ial on at least one pair of in­puts.

  I con­tinue to be un­sure if we can even claim anti-sym­me­try of the prefer­ence re­la­tion. For ex­am­ple, let $S_A$ be the state “I eat an ap­ple” and $S_O$ the state “I eat an or­ange”, and to­day $J(S_A,S_O)$ but to­mor­row $J(S_O,S_A)$, seem­ingly vi­o­lat­ing an­ti­sym­me­try. Now of course maybe I mi­s­un­der­stood my own un­der­stand­ing of $S_A$ and $S_O$ such that they ac­tu­ally in­cluded a hid­den-to-my-aware­ness prop­erty con­di­tion­ing them on time or some­thing else such that anti-sym­me­try is not vi­o­lated, but the fact that there may be some prop­erty on the states that I didn’t think about at first that sal­vages anti-sym­me­try makes my worry that this model is con­fused in this and other ways be­cause it was so eas­ily to think of and con­struct some­thing that seem­ingly vi­o­lated the prop­erty but then on fur­ther re­flec­tion seems like it doesn’t.

  That’s not a slam-dunk ar­gu­ment against this for­mal­iza­tion. This is more me shar­ing some thoughts on my reser­va­tions of us­ing this type of model. If we can so eas­ily fail to no­tice some­thing rele­vant about how we for­mal­ize some sim­ple prefer­ences, what else may we be failing to no­tice? And if so what hap­pens if we build an AI based in part on this for­mal­iza­tion? Will it also fail to ac­count for rele­vant as­pects of how hu­man prefer­ences are calcu­lated be­cause they are not eas­ily visi­ble to us in the model, or is that a failure of hu­mans to un­der­stand them­selves rather than the model? Th­ese are the things I’m wrestling with lately.

  I also have some reser­va­tions about whether we can even re­ally model hu­mans has hav­ing dis­crete prefer­ences that we can rea­son about in this way with­out get­ting our­selves into trou­ble and con­fused. Not to say that I doubt that this model of­ten works, only that I worry that it’s miss­ing some im­por­tant de­tails that are rele­vant for al­ign­ment and with­out ac­count­ing for them we will fail to pro­duce al­igned AI. I worry this be­cause there doesn’t seem to be any­thing in the hu­man mind that ac­tu­ally is a prefer­ence; prefer­ences are more like reifi­ca­tions of a pat­tern of ac­tion that ap­pears in hu­mans. Get­ting closer to un­der­stand­ing the mechanism that pro­duces the pat­tern we in­ter­pret as prefer­ences seems valuable to me in this work be­cause I worry we’re miss­ing cru­cial de­tails when we rea­son about prefer­ences at the level of de­tail you pur­sue here.

• G Gordon Worley III 11 Jun 2019 1:24 UTC

  2 points

  on: To first or­der, moral re­al­ism and moral anti-re­al­ism are the same thing

  I apol­o­gise for my sim­plis­tic un­der­stand­ing and defi­ni­tions of moral re­al­ism. How­ever, my par­tial ex­pe­rience in this field has been enough to con­vince me that there are many in­com­pat­i­ble defi­ni­tion of moral re­al­ism, and many ar­gu­ments about them, so it’s not clear there is a sin­gle sim­ple thing to un­der­stand. So I’ve tried to define is very roughly, enough so that the gist of this post makes sense. ↩︎

  I think this is mostly be­cause there are lots of re­al­ist and anti-re­al­ist po­si­tions and they cluster around fea­tures other than their stance on re­al­ism, i.e. whether or not moral facts ex­ist, or said less densely, whether or not moral claims can be true or false. The two camps seems to have a lot more go­ing on, though, than is cap­tured by this rather tech­ni­cal point, as you point out. In fact, most of the in­ter­est­ing de­bate is not about this point, but about things that can be func­tion­ally the same re­gard­less of your stance on re­al­ism, hence your notic­ing how re­al­ists and anti-re­al­ists can look like each other in some cases.

  (My own stance is to be skep­ti­cal, since I’m not even sure we have a great idea of what we re­ally mean when we say things are true or false. It seems like we do at first, but if we poke too hard the whole thing starts to come apart at the seams, which makes it a bit hard to worry too much about moral facts when you’re not even sure about facts in the first place!)

• G Gordon Worley III 7 Jun 2019 23:34 UTC

  4 points

  on: For the past, in some ways only, we are moral degenerates

  Re­cently Robin Han­son posted about the differ­ence be­tween fight­ing along the fron­tier vs. ex­pand­ing the fron­tier. It’s a well known point, but given I was re­cently re­minded of it it’s salient to me, and it seems quite rele­vant here.

  When we ask if hu­man val­ues have “im­proved” or “de­gen­er­ated” over time we have to have some way of judg­ing in­crease or de­crease. One way to un­der­stand this is to check if hu­mans get to re­al­ize more value, as judged by each in­di­vi­d­ual and then nor­mal­ized and ag­gre­gated, along cer­tain di­men­sions within the mul­ti­di­men­sional space of val­ues. To take your ex­am­ple of “en­gage­ment with ex­tended fam­ily”, most mod­erns have less of this than an­cients did both on av­er­age and it seems at max­i­mum, i.e. mod­ern sys­tems pre­clude as much en­gage­ment as was pos­si­ble in the past, such that a mod­ern per­son max­i­mally en­gaged with their ex­tended fam­ily is less en­gaged than was max­i­mally pos­si­ble in the past. This seems to be traded-off, though, against greater free­dom from need to en­gage with ex­tended fam­ily be­cause al­ter­na­tive sys­tems al­low a per­son to fulfill other val­ues with­out re­li­ance on ex­tended fam­ily. As a re­sult this looks much like a “fight”, i.e. a trade-off along the value fron­tier of one value to an­other.

  You give the ex­am­ple of re­duced slav­ery be­ing a gen­eral benefit, but I think we can tell a similar story that it is a trade-off. We trade-off in­di­vi­d­ual choice of labour use, liv­ing con­di­tions, etc. for the right of the pow­er­ful to make those de­ci­sions for the less pow­er­ful. In this sense the re­duc­tion in slav­ery takes away some­thing of value from some­one—the would be slave­hold­ers—to give it to some­one else—the would be slaves. We may judge this to be an ex­pan­sion or value effi­ciency im­prove­ment un­der two con­di­tions (which change slightly what we mean by ex­pan­sion):

  • (1) there is more value over­all, i.e. we traded less value away than we got back in return

  • (2) there is more value over­all along all dimensions

  I would ar­gue that case (1) is re­ally still a fight though be­cause we are still mak­ing a trade­off, we are just mov­ing to some­where more effi­cient along the fron­tier. From this per­spec­tive the end of slav­ery was not an ex­pan­sion of val­ues, but it was a trade-off for more value.

  But if we are so strict, is any­thing truly a pure ex­pan­sion? This seems quite tricky, be­cause hu­mans can value ar­bi­trary things, and so for ev­ery ac­tion that in­creases some value it would seem that we are nec­es­sar­ily de­creas­ing the abil­ity the re­al­ize some counter-value. For ex­am­ple, it might seem that some­thing like “greater availa­bil­ity of calories” would re­sult in pure value ex­pan­sion, as­sum­ing we can screen off all the com­pli­cated de­tails of how we make more calories available to hu­mans and how that pro­cess will af­fect val­ues. But sup­pose you value scarcity of calories, maybe even di­rectly, then for you this will be a fight and we must in­ter­pret an in­crease in the availa­bil­ity of calories as a trade-off rather than as a pure ex­pan­sion in val­ues.

  This is po­ten­tially trou­bling be­cause it means there’s no uni­ver­sal way to judge moral progress if there can be no ex­pan­sion with­out some con­trac­tion some­where. It would seem that there must always be con­trac­tion of some­thing, even if it is an effi­cient con­trac­tion that gen­er­ates more value than it gives up.

  So in the end I guess I am forced to (mostly) agree with your as­sess­ment even though you frame it in a way that seem for­eign to me. It feels for­eign to me be­cause it seems ev­ery im­prove­ment is also a de­gen­er­a­tion and vice versa, and the rele­vant ques­tion of im­prove­ment is mostly whether or not we are gen­er­at­ing more value in ag­gre­gate (an effi­ciency im­prove­ment) if we want to be neu­tral on which value di­men­sions to op­ti­mize along.

  I ac­tu­ally don’t love the idea of mak­ing ag­gre­gate value some­thing we op­ti­mize for, though, be­cause I worry about de­gen­er­ate cases like highly op­ti­miz­ing along a sin­gle value di­men­sion at the ex­pense of all oth­ers such that it re­sults in an over­all in­crease in value but in a way we wouldn’t want, even though ar­guably if we were mea­sur­ing value cor­rectly in this sys­tem such a situ­a­tion would be im­pos­si­ble be­cause it would be fac­tored in by a de­crease in what­ever value we had that was be­ing traded off against that made us dis­like the “op­ti­miza­tion”.

  I in­stead con­tinue to think that value is a con­fused con­cept that we need to break apart and re­un­der­stand, but I’m still work­ing on de­con­fus­ing my­self on this, so I have noth­ing ad­di­tional to re­port in that di­rec­tion for now.

• G Gordon Worley III 6 Jun 2019 19:27 UTC

  22 points

  on: Steel­man­ning Divination

  Some years ago I got in­ter­ested in the Yi Jing af­ter read­ing Philip K. Dick’s The Man in the High Cas­tle, which fea­tures the Yi Jing promi­nently where the book within the book (which is the al­ter­nate di­men­sion/​his­tory ver­sion of The Man in the High Cas­tle) is writ­ten by us­ing the Yi Jing to make plot de­ci­sions and one of the char­ac­ters re­lies on it heav­ily to nav­i­gate life. I went on to write a We­bOS Yi Jing phone app so I could more eas­ily con­sult it from my phone and played around with it my­self.

  My ex­pe­rience of it was mostly that it offered me noth­ing I wasn’t already do­ing on my own, but I could see how it would have been helpful to oth­ers who lack my par­tic­u­lar nat­u­ral dis­po­si­tion to let­ting my mind go quiet and see­ing what it has to tell me. As you note, it seems a good way to be able to step back and con­sider some­thing from a differ­ent an­gle, and to con­sider differ­ent as­pects of some­thing you may be cur­rently ig­nor­ing. The com­men­tary on the Yi Jing is care­fully worded such that it’s more about the de­ci­sion gen­er­a­tion pro­cess than the de­ci­sion it­self, and when used well I think can re­sult in the sort of sud­den re­al­iza­tion of the ac­tion you will take the same way my sit­ting quietly and wait­ing for in­sight does.

  I also know a de­cent num­ber of ra­tio­nal­ists who en­joy play­ing with Tarot cards for seem­ingly this same rea­son. Tarot works a bit differ­ent be­cause it more tells a story than high­lights a virtue, but I think like you much of the value comes from plac­ing an ran­dom fram­ing on events, in­ject­ing noise into an oth­er­wise too sta­ble al­gorithm, and helping peo­ple get out of lo­cal max­ima/​min­ima traps.

  I’d also in­clude rub­ber duck­ing as a mod­ern div­ina­tion method. I think it does some­thing similar, but by us­ing a differ­ent method to get you to see things more clearly and find out what you already im­plic­itly knew but weren’t mak­ing ex­plicit enough to let it have an im­pact on your ac­tions. My spec­u­la­tion at a pos­si­ble mechanism of ac­tion here is some­thing like what hap­pens when I sit quietly with a de­ci­sion and wait for an an­swer: you let the es­tab­lished pat­terns of thought get out of the way and let other things come through so you can con­sider them, in part be­cause you can gen­er­ate your own in­ter­nal noise if you stop try­ing to di­rect your thought. But not ev­ery­one finds this easy or pos­si­ble, in which case more tra­di­tional div­ina­tion meth­ods with ex­ter­nal noise in­jec­tion are likely use­ful.

• G Gordon Worley III 6 Jun 2019 19:05 UTC

  3 points

  in reply to: Chris Leong's comment on: All knowl­edge is cir­cu­larly justified

  Ac­tu­ally, good thing you asked, be­cause I gave wrong in­for­ma­tion in my origi­nal com­ment. Chisholm is an ex­pert on the prob­lem of the crite­rion, but I was ac­tu­ally think­ing of William Als­ton in my com­ment. Here’s two pa­pers, one by Als­ton and one by an­other au­thor that I’ve refer­enced in the past and found use­ful:

  William P. Als­ton. Epistemic Cir­cu­lar­ity. Philos­o­phy and Phenomenolog­i­cal Re­search, 47(1):1, sep 1986.

  Jonathan Dancy. Eth­i­cal Par­tic­u­larism and Mo­rally Rele­vant Prop­er­ties. Mind, XCII(368):530– 547, 1983.

• G Gordon Worley III 6 Jun 2019 0:47 UTC

  3 points

  on: All knowl­edge is cir­cu­larly justified

  You might like the work of Rod­er­ick Chisholm on this topic. He spent a good deal of effort on ad­dress­ing the is­sue of epistemic cir­cu­lar­ity (the is­sue cre­ated by the prob­lem of the crite­rion) and gives what is, in my opinion, one of the bet­ter and more tech­ni­cal treat­ments of the topic. His work also lets us make a dis­tinc­tion be­tween par­tic­u­larism (mak­ing min­i­mal leaps of faith) and prag­ma­tism (mak­ing any leaps of faith), which I find use­ful be­cause in prac­tice most peo­ple seem to be prag­ma­tists (they have other things to do than wres­tle with episte­mol­ogy) while think­ing they are par­tic­u­larists be­cause their par­tic­u­lar leaps of faith (the facts they as­sume with­out jus­tifi­ca­tion) are in­tu­itive to them and they can’t think of a way to make them smaller.

• G Gordon Worley III 6 Jun 2019 0:07 UTC

  12 points

  on: “But It Doesn’t Mat­ter”

  First, let me start by say­ing this com­ment is ul­ti­mately a nit­pick. I agree with the thrust of your po­si­tion and think in most cases your point stands. How­ever, there’s no fun and noth­ing to say if I leave it at that, so grab your tweez­ers and let’s get that nit.

  Even if Hy­poth­e­sis H is true, it doesn’t have any de­ci­sion-rele­vant im­pli­ca­tions,

  So to me there seems to be a spe­cial case of this that is not ra­tio­nal­iza­tion, and that’s in cases where one fact dom­i­nates an­other.

  By “dom­i­nates” I here mean that for the pur­pose for which the fact is be­ing con­sid­ered, i.e. the de­ci­sion about which the truth value of H may have rele­vant im­pli­ca­tions, there may be an­other fact about an­other hy­poth­e­sis, H’, such that if H’ is true or H’ is false then whether or not H is true or false will have no im­pact on the out­come be­cause H’ is rel­a­tively so much more im­por­tant than H.

  To make this con­crete, con­sider the case of the sin­gle-is­sue voter. They will vote for a can­di­date pri­mar­ily based on whether or not that can­di­date sup­ports their fa­vored po­si­tion on the sin­gle is­sue they care about. So let’s say Can­di­date Brain Slug is run­ning for Pres­i­dent of the World on a plat­form whose main plank is im­plant­ing brain slugs on all peo­ple. You ar­gue with your sin­gle-is­sue voter friend they should not vote for Brain Slug be­cause it will put a brain slug on them, but they say even if that’s true, it’s not rele­vant to their de­ci­sion, be­cause Brain Slug also sup­ports a ban on trol­ley switches, which is your friend’s sin­gle is­sue.

  Now maybe you think your friend is be­ing stupid, but in this case they’re ar­guably not ra­tio­nal­iz­ing. In­stead they’re mak­ing a de­ci­sion based on their val­ues that place such a pre­mium on the is­sue of trol­ley switch bans that they rea­son­ably don’t care about any­thing else, even if it means vot­ing for Pres­i­dent Brain Slug and its brain slug im­plant­ing agenda.

• G Gordon Worley III 31 May 2019 23:27 UTC

  9 points

  in reply to: moridinamael's comment on: Book Sum­mary: Con­scious­ness and the Brain

  To my read­ing, all of this seems to pretty well match a (part of) the Bud­dhist no­tion of de­pen­dent origi­na­tion, speci­fi­cally the way senses beget sense con­tact (ex­pe­rience) begets feel­ing begets crav­ing (prefer­ences) begets cling­ing (be­liefs/​val­ues) begets be­ing (for­mal on­tol­ogy). There the fo­cus is a bit differ­ent and is ori­ented around ad­dress­ing a differ­ent ques­tion, but I think it’s tack­ling some of the same is­sues via differ­ent meth­ods.

• G Gordon Worley III 31 May 2019 17:38 UTC

  3 points

  in reply to: toonalfrink's comment on: Ego­ism In Disguise

  1) The bedrock of our val­ues are prob­a­bly the same for any hu­man be­ing, and any differ­ence be­tween con­scious val­ues is ei­ther due to hav­ing seen differ­ent data, but more likely due to differ­ent peo­ple situ­a­tion­ally benefit­ting more un­der differ­ent moral­ities. For ex­am­ple a strong per­son will have “val­ues” that are more ac­cept­ing of com­pe­ti­tion, but that will change once they be­come weaker.

  I con­tinue to find min­i­miza­tion of con­fu­sion while main­tain­ing home­osta­sis around biolog­i­cally de­ter­mined set points a rea­son­able ex­pla­na­tion for the bedrock of our val­ues. Hope­fully these ideas will co­a­lesce well enough in me soon to be able to write some­thing more about this than that head­line.

• G Gordon Worley III 30 May 2019 1:46 UTC

  4 points

  in reply to: Wei_Dai's comment on: Boo votes, Yay NPS

  I agree there are other pos­si­ble in­ter­pre­ta­tions; mainly wanted to doc­u­ment for my­self in case I wanted to refer­ence it later, and it seems po­ten­tially rele­vant, es­pe­cially if we wanted to go back and in­ter­view the vot­ers or an­a­lyze the com­ments.

• G Gordon Worley III 30 May 2019 1:43 UTC

  4 points

  on: De­bate AI and the De­ci­sion to Re­lease an AI

  This is a pretty in­ter­est­ing idea. I can imag­ine this be­ing part of a safety-in-depth ap­proach: not a sin­gle method we would rely on but one of many fail-safes along with sand­box­ing and ac­tu­ally try­ing to di­rectly ad­dress al­ign­ment.

• G Gordon Worley III 29 May 2019 18:00 UTC

  6 points

  on: Boo votes, Yay NPS

  Ad­di­tional ev­i­dence of boo/​yay vot­ing cul­ture: Ben Hoff­man’s “Down­ing chil­dren are rare” post (on EAF, on LW, my com­ment about votes)

• G Gordon Worley III 28 May 2019 20:52 UTC

  5 points

  on: Micro feed­back loops and learning

  To give an ad­di­tional ex­am­ple of tight feed­back loops be­ing helpful, I’ve been tak­ing Alexan­der les­sons for nearly a year. Each les­son con­sists of 30 min­utes of me do­ing move­ments (al­though some­time the “move­ment” is hold­ing a pos­ture, like sit­ting, stand­ing, stand­ing on toes, or crouch­ing) and 30 min­utes of “table time”, i.e. I lay on a mas­sage table while my teacher users her hands to very sub­tly sug­gest changes to my pos­ture. Although I could go on about how great this has been and how much value I get from it, what I mostly want to say about it here is that it de­pends very much on tight feed­back loops to perform a kind of re­in­force­ment learn­ing. As I make a move­ment she uses her hands and some taught jar­gon (part of the tech­nique in­volves as­so­ci­at­ing jar­gon with pos­tures and move­ments so you can eas­ily call them up on com­mand by say­ing or think­ing the jar­gon) to ad­just what I do, giv­ing me rapid feed­back on how I’m do­ing. The re­sult was that within the first 10 hours of train­ing I dra­mat­i­cally im­proved my pos­ture and re­duced pos­ture and move­ment re­lated pain.

  For com­par­i­son, over­lap­ping with learn­ing Alexan­der tech­nique I’ve been more deeply prac­tic­ing for­mal med­i­ta­tion, and learn­ing for­mal med­i­ta­tion has very long feed­back loops and re­quires months to make sig­nifi­cant progress. Now, maybe the long feed­back cy­cles are not why it takes months to make progress, and I can think of rea­son­able sto­ries as to why that would be, I can also imag­ine find­ing ways to shorten feed­back cy­cles would have made progress much faster. For ex­am­ple, when I’ve done biofeed­back stuff in the past it only took 4 or 5 hours of ses­sions be­fore I could make my­self fall asleep at will (sadly I’ve for­got­ten how to do this), and I think it’s quite likely that it was helped a lot by hav­ing a com­puter tel­ling me when I got a lit­tle closer to what I needed to do to make that hap­pen and when I got a lit­tle farther away, such that I didn’t have to spend as much time guess­ing and wait­ing for strong ev­i­dence that I was do­ing the right thing be­fore I could re­li­ably train that abil­ity and then go on to the next step.

• G Gordon Worley III 28 May 2019 8:55 UTC

  17 points

  on: What should ra­tio­nal­ists think about the re­cent claims that air force pi­lots ob­served UFOs?

  While these ob­jects may be uniden­ti­fied, the idea that they are the prod­ucts of aliens, a simu­la­tion, AI, or some­thing else seems un­likely given the low qual­ity of the ev­i­dence. In all cases I’m aware of ev­i­dence for some­thing like this be­ing the true ori­gin of a UFO would have to over­come the more likely al­ter­na­tives of

  • se­cret, ex­per­i­men­tal, or stealth air­craft, prob­a­bly mil­i­tary, with ad­vanced ca­pa­bil­ities undis­closed to the pub­lic;

  • ob­ser­va­tional er­rors and in­stru­men­ta­tion glitches;

  • mis­re­mem­ber­ing, em­bel­lish­ment, and out­right ly­ing.

  For a com­par­i­son, the liter­a­ture on cryp­tids (claimed to be real but un­ob­served by sci­ence an­i­mals like bigfoot, the Loch Ness mon­ster, and the chu­pacabra) is full of cases where the ev­i­dence looks pretty com­pel­ling...so long as we only look for ev­i­dence that con­firms the hope that a cryp­tid ex­ists. Per­haps sadly, there are no cryp­tid hu­manoids or sea mon­sters that we know of, and all ev­i­dence of them thus far col­lected is ei­ther best cat­e­go­rized as hoaxes, misi­den­ti­fi­ca­tions, and hope­ful mis­in­ter­pre­ta­tions or turned out to be ev­i­dence of real, undis­cov­ered, and not fan­tas­ti­cal an­i­mals.

• G Gordon Worley III 24 May 2019 22:30 UTC

  5 points

  on: Sepa­ra­tion of Concerns

  The nat­u­ral ar­gu­ment against this is of course that sep­a­ra­tion is an illu­sion. I don’t say that to sound mys­te­ri­ous, I mean that just in the sim­ple way that ev­ery­thing is tan­gled up to­gether, de­pen­dent on each other for its ex­is­tence, and its only in our mod­els that clean sep­a­ra­tion can ex­ist, and then only by ig­nore some parts of re­al­ity in or­der to keep our mod­els clean.

  As a work­ing pro­gram­mer, I’m very fa­mil­iar with the origi­nal con­text of the idea of sep­a­ra­tion of con­cerns, and I can also tell you that even there is never to­tally works. It’s a tool we use to help us poor hu­mans who can’t fathom the to­tal, com­plete, awe­some com­plex­ity of the world to get along well enough any­way to col­lect a pay­check. Or some­thing like that.

  Re­lat­edly, ev­ery ab­strac­tion is leaky, and if you think it isn’t you just haven’t looked hard enough.

  None of that is to say we shouldn’t re­spect the sep­a­ra­tion of con­cerns when use­ful, but also that we shouldn’t ele­vate it more than it de­serves to be, be­cause the sep­a­ra­tion is a con­struc­tion of our minds, not a nat­u­ral fea­ture of the world.