Halstead

• Halstead 11 Oct 2019 9:18 UTC

  3 points

  on: Shap­ley val­ues: Bet­ter than counterfactuals

  Thanks for this in­ter­est­ing post. As I ar­gued in the post that you cite and as Ge­orge Bridg­wa­ter notes be­low, I don’t think you have iden­ti­fied a prob­lem in the idea of coun­ter­fac­tual im­pact here, but have in­stead shown that you some­times can­not ag­gre­gate coun­ter­fac­tual im­pact across agents. As you say, Coun­ter­fac­tu­alIm­pact(Agent) = Value(World with agent) - Value(World with­out agent).

  Sup­pose Karen and An­drew have a one night stand which leads to Karen hav­ing a baby Ge­orge (and Karen and An­drew oth­er­wise have no effect on any­thing). In this case, An­drew’s coun­ter­fac­tual im­pact is:

  Value (world with one night stand) - Value (world with­out one night stand)

  The same is true for Karen. Thus, the coun­ter­fac­tual im­pact of each of them taken in­di­vi­d­u­ally is an ad­di­tional baby Ge­orge. This doesn’t mean that the coun­ter­fac­tual im­pact of An­drew and Karen com­bined is two ad­di­tional baby Ge­orges. In fact, the coun­ter­fac­tual im­pact of Karen and An­drew com­bined is also given by:

  Value (world with one night stand) - Value (world with­out one night stand)

  Thus, the coun­ter­fac­tual im­pact of Karen and An­drew com­bined is an ad­di­tional baby Ge­orge. There is noth­ing in the defi­ni­tion of coun­ter­fac­tual im­pact which im­plies it can be always be ag­gre­gated across agents.

  This is the differ­ence be­tween “if me and Karen hadn’t ex­isted, nei­ther would Ge­orge” and “If I hadn’t ex­isted, nei­ther would Ge­orge, and if Karen hadn’t ex­isted nei­ther would Ge­orge, there­fore if me and Karen hadn’t ex­isted, nei­ther would two Ge­orges.” This last state­ment is con­fused, be­cause the ba­bies referred to in the an­tecedent are the same.

  I dis­cuss other ex­am­ples in the com­ments to Joey’s post.

  **

  The coun­ter­fac­tual un­der­stand­ing of im­pact is how al­most all vot­ing the­o­rists analyse the ex­pected value of vot­ing. EAs tends to think that vot­ing is some­times al­tru­is­ti­cally ra­tio­nal be­cause of the small chance of be­ing the one pivotal voter and mak­ing a large coun­ter­fac­tual differ­ence. On the Shapely value ap­proach, the large coun­ter­fac­tual differ­ence would be di­vided by the num­ber of win­ning vot­ers. Firstly, to my knowl­edge al­most no-one in vot­ing the­ory as­sesses the im­pact of vot­ing in this way. Se­condly, this would I think im­ply that vot­ing is never ra­tio­nal since in any large elec­tion the prospec­tive pay-off of vot­ing would be di­vided by the po­ten­tial set of win­ning vot­ers and so would be >100,000x smaller than on the coun­ter­fac­tual approach

• Halstead 7 Oct 2019 15:20 UTC

  4 points

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: Au­gust 2019 grant recommendations

  I dis­agree that 80k should tran­si­tion to­wards a £3k re­treat + no on­line con­tent model, but it doesn’t seem worth get­ting into why here.

  On premises, here is the top defi­ni­tion I have found from googling… “a pre­vi­ous state­ment or propo­si­tion from which an­other is in­ferred or fol­lows as a con­clu­sion”. This fits with my (and CFAR’s) char­ac­ter­i­sa­tion of dou­ble crux­ing. I think we’re agreed that the ques­tion is which premises you dis­agree on cause your dis­agree­ment. It is log­i­cally im­pos­si­ble that dou­ble crux­ing ex­tends this char­ac­ter­i­sa­tion.

• Halstead 7 Oct 2019 15:10 UTC

  6 points

  in reply to: Max_Daniel's comment on: Long-Term Fu­ture Fund: Au­gust 2019 grant recommendations

  Yes I don’t fully un­der­stand why they’re not leg­ible. A 4 day work­shop seems pretty well-placed for a care­fully done im­pact eval­u­a­tion.

• Halstead 7 Oct 2019 2:28 UTC

  3 points

  in reply to: Max_Daniel's comment on: [link] An­dreas Mo­gensen’s “Max­i­mal Clue­less­ness”

  On the bit­ing the bul­let an­swer, that doesn’t seem plau­si­ble to me. The prefer­ence we have are a product of the be­liefs we have about what will make our lives bet­ter over the long-run. My prefer­ence not to smoke is en­tirely a product of the fact that I be­lieve that it will in­crease my risk of pre­ma­ture death. Per pro­po­nents of clue­less­ness, I could ar­gue “maybe it will make me look cool to smoke, and that will in­crease my chances of get­ting a de­sir­able part­ner” or some­thing like that. In that sense the sign of the effect of smok­ing on my own in­ter­ests is not cer­tain. Nev­er­the­less, I think it is ir­ra­tional to smoke. I don’t think a Parfi­tian un­der­stand­ing of iden­tity would help here be­cause then my re­fusal to smoke would be al­tru­is­tic—I would be helping out my fu­ture self.

  The dodge the bul­let an­swer is more plau­si­ble, and I may fol­low up with more later.

• Halstead 7 Oct 2019 2:18 UTC

  3 points

  in reply to: Max_Daniel's comment on: [link] An­dreas Mo­gensen’s “Max­i­mal Clue­less­ness”

  On the lat­ter, yes that is a good point—there are gen­eral fea­tures at play here, so I re­tract my pre­vi­ous com­ment. How­ever, it still seems true that your ra­tio­nal credal state will always de­pend to a very sig­nifi­cant ex­tent on the par­tic­u­lar facts.

  I find the use of the long-ter­mist point of view a bit weird as ap­plied to the AMF ex­am­ple. AMF is not usu­ally jus­tified from a long-ter­mist point of view, so it is not re­ally sur­pris­ing that its benefits seem less ob­vi­ous when you con­sider it from that point of view.

• Halstead 7 Oct 2019 2:14 UTC

  3 points

  in reply to: Max_Daniel's comment on: [link] An­dreas Mo­gensen’s “Max­i­mal Clue­less­ness”

  Hello,

  Here is a good pa­per on this - https://​​www.prince­ton.edu/​​~adame/​​pa­pers/​​sharp/​​elga-sub­jec­tive-prob­a­bil­ities-should-be-sharp.pdf

• Halstead 7 Oct 2019 2:03 UTC

  30 points

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: Au­gust 2019 grant recommendations

  thanks for this.

  If the re­treats are valuable, one would ex­pect them to com­mu­ni­cate gen­uinely use­ful con­cepts and ideas. Which ideas that CFAR teaches do you think are most use­ful?

  On the pay­ment model, imag­ine that in­stead of putting their ma­te­rial on choos­ing a high im­pact ca­reer on­line, 80k charged peo­ple £3000 to have 4 day coach­ing and net­work­ing re­treats in a large man­sion, af­ter­wards giv­ing them ac­cess to the rele­vant writ­ten ma­te­rial. I think this would shave off ~100% of the value of 80k. The differ­ences be­tween the two or­gani­sa­tions don’t seem to me to be large enough to make a rele­vant differ­ence to this anal­y­sis when ap­plied to CFAR. Do you think there is a case for 80k to move to­wards the CFAR £3k re­treat model?

  **

  On dou­ble crux­ing, here is how CFAR defines dou­ble cruxing

  “Let’s say you have a be­lief, which we can la­bel A (for in­stance, “mid­dle school stu­dents should wear uniforms”), and that you’re in dis­agree­ment with some­one who be­lieves some form of ¬A. Dou­ble crux­ing with that per­son means that you’re both in search of a sec­ond state­ment B, with the fol­low­ing prop­er­ties:

  1. You and your part­ner both dis­agree about B as well (you think B, your part­ner thinks ¬B)

  2. The be­lief B is cru­cial for your be­lief in A; it is one of the cruxes of the ar­gu­ment. If it turned out that B was not true, that would be suffi­cient to make you think A was false, too.

  3. The be­lief ¬B is cru­cial for your part­ner’s be­lief in ¬A, in a similar fash­ion.”

  So, if I were to dou­ble crux with you, we would both es­tab­lish which were the premises we dis­agree on that cause our dis­agree­ment. B is a premise in the ar­gu­ment for A. This is dou­ble crux­ing, right?

  You say:

  “if you ask me “what are my premises for the be­lief that Na­ture is the most pres­ti­gious sci­ence jour­nal?” then I definitely won’t have a nice list of premises I can re­spond with, but if you ask me “what would change my mind about Na­ture be­ing the most pres­ti­gious sci­ence jour­nal?” I might be able to give a rea­son­ably good an­swer and start hav­ing a pro­duc­tive con­ver­sa­tion”

  Your an­swer could be ex­pressed in the form of premises right? Premises are just propo­si­tions that bear on the like­li­hood of the conclusion

• Halstead 7 Oct 2019 1:07 UTC

  5 points

  on: Defend­ing Philan­thropy Against Democracy

  This is a great post, which I think will be use­ful for the com­mu­nity!

• Halstead 6 Oct 2019 14:43 UTC

  29 points

  on: Long-Term Fu­ture Fund: Au­gust 2019 grant recommendations

  I’m in­ter­ested in the recom­men­da­tion of CFAR (though I ap­pre­ci­ate it is not funded by the LTFF). What do you think are the top ideas re­gard­ing epistemics that CFAR has come up with that have helped EA/​the world?

  You men­tion dou­ble crux­ing in the other post dis­cussing CFAR. Rather than an in­no­va­tion, isn’t this merely agree­ing on which premise you dis­agree on? Similarly, isn’t mur­phyjitsu just the pre-mortem, which was defined by Kah­ne­man more than a decade ago?

  I also won­der why CFAR has to charge peo­ple for their ad­vice. Why don’t they write down all of their in­sights and put it on­line for free?

• Halstead 6 Oct 2019 5:50 UTC

  8 points

  on: [link] An­dreas Mo­gensen’s “Max­i­mal Clue­less­ness”

  I’m pretty scep­ti­cal of ar­gu­ments for clue­less­ness. Some thoughts:

  • Knigh­tian un­cer­tainty seems to me never ra­tio­nal. There are strong ar­gu­ments that cre­dence func­tions should be sharp. Even if you can bound your cre­dences very broadly with in­ter­vals, it seems like you would never be un­der knigh­tian un­cer­tainty given your in­for­ma­tion—your credal state is always some­where be­tween 0 and 1, and surely your mean es­ti­mate will differ be­tween differ­ent prob­lems.

  • Similar ar­gu­ments for com­plex clue­less­ness also seems to ap­ply to my own de­ci­sions about what would be in my ra­tio­nal self-in­ter­est to do. Nev­er­the­less, I will not be wan­der­ing blindly into the road out­side my ho­tel room in 10 min­utes.

  • I don’t see how you could make a gen­eral ar­gu­ment for clue­less­ness with re­spect to all de­ci­sions made by the com­mu­nity. You could make an ar­gu­ment that the sign of the ex­pected benefits of EA ac­tions is much more un­cer­tain than has been ac­knowl­edged. I don’t see how this could ever gen­er­al­ise to an ar­gu­ment that all of our de­ci­sions are clue­less, since the level of un­cer­tainty will always be al­most en­tirely de­pen­dent on the facts about the par­tic­u­lar case. Why would un­cer­tainty about the effects of AMF have any bear­ing on un­cer­tainty about the effects of MIRI or the Clean Air Task Force?

  • Clue­less­ness seems to im­ply that al­tru­ists should be in­differ­ent be­tween all pos­si­ble ac­tions that they can take. Is this im­pli­ca­tion of the view em­braced?

  • Re­lated to the above, in the AMF vs make a wish foun­da­tion ex­am­ple, I don’t ac­tu­ally agree that we are as un­cer­tain as sug­gested. e.g. you list stud­ies cit­ing differ­ent effects of life sav­ing on fer­til­ity say­ing “Un­for­tu­nately, the stud­ies just noted are of differ­ent kinds (cross-coun­try com­par­i­sons, panel stud­ies, quasi-ex­per­i­ments, large-sam­ple micro-stud­ies), with differ­ent strengths and weak­nesses, mak­ing it difficult to draw firm con­clu­sions”. This seems to be ask­ing for the re­ac­tion “what are we to do in the face of all this method­olog­i­cal com­plex­ity?” But an economist would ac­tu­ally have an an­swer to this—cross-coun­try com­par­i­sons with cross-sec­tional data are out of fash­ion for ex­am­ple.

  • Over­all, ar­gu­ments about clue­less­ness seem to merely re­assert that the world is com­plex and we should think care­fully be­fore act­ing. I don’t see how it points to some deep per­ma­nent fea­ture of our epistemic situ­a­tion.

The ITN frame­work, cost-effec­tive­ness, and cause prioritisation

• Halstead 6 Oct 2019 1:35 UTC

  6 points

  in reply to: Habryka's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  ok cheers. I dis­agree with that but feel we have reached the end of pro­duc­tive argument

• Halstead 5 Oct 2019 7:10 UTC

  4 points

  in reply to: Habryka's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  What do you make of my ‘offen­sive be­liefs’ poll idea and ques­tions?

• Halstead 5 Oct 2019 2:40 UTC

  5 points

  in reply to: Ben Pace's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  There are two is­sues here. The less im­por­tant one is - (1) are peo­ple’s be­liefs that many of these opinions are taboo ra­tio­nal? I think not, and have dis­cussed the rea­sons why above.

  The more im­por­tant one is (2) - this poll is a blunt in­stru­ment that en­courages peo­ple to en­ter offen­sive opinions that threaten the rep­u­ta­tion of the move­ment. If there were a way to do this with those opinions laun­dered out, then I wouldn’t have a prob­lem.

  This has been done in a very care­less way with­out due thought to the very ob­vi­ous risks

• Halstead 5 Oct 2019 2:29 UTC

  11 points

  in reply to: Habryka's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  They have a sec­tion on ‘why do this?’ and don’t dis­cuss any of the ob­vi­ous risks which sug­gests they haven’t thought prop­erly about the is­sue. I think a good norm to prop­a­gate would be—peo­ple put a lot of thought into whether they should pub­lish posts that could po­ten­tially dam­age the move­ment. Do you agree?

  Sup­pose I am go­ing to run a poll on ‘what’s the most offen­sive thing you be­lieve—anony­mous pub­lic poll for effec­tive al­tru­ists’. (1) do you think I should have to pub­li­cly ex­plain why I am do­ing this? (2) do you think I should run this poll and pub­lish the re­sults?

• Halstead 3 Oct 2019 9:57 UTC

  35 points

  in reply to: Ben Pace's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  Hi Ben,

  Thanks for this, this is use­ful (up­voted)

  1. I think we dis­agree on the em­piri­cal facts here. EA seems to me un­usu­ally open to con­sid­er­ing ra­tio­nal ar­gu­ments for un­fash­ion­able po­si­tions. Peo­ple in my ex­pe­rience lose points for bad ar­gu­ments, not for weird con­clu­sions. I’d be very per­plexed if some­one were not will­ing to dis­cuss whether or not util­i­tar­i­anism is false (or whether re­mote work­ing is bad etc) in front of EAs, and would think some­one was over­come by ir­ra­tional fear if they de­clined to do so. Michael Plant be­lieves one of the allegedly taboo opinions here (men­tal health should be a pri­or­ity) and is cur­rently on a speak­ing tour of EA events across the Far East.

  2. This is a good point and up­dates me to­wards the use­ful­ness of the sur­vey, but I won­der whether there is a bet­ter way to achieve this that doesn’t carry such clear rep­u­ta­tional risks for EA.

  3. The is­sue is not whether my col­leagues have suffi­cient pub­lic ac­cessible rea­son to be­lieve that EA is full of good peo­ple act­ing in good faith (which they do), but whether this sur­vey weighs heav­ily or not in the ev­i­dence that they will ac­tu­ally con­sider. i.e. this might lead them not to con­sider the rest of the ev­i­dence that EA is mostly full of good peo­ple work­ing in good faith. I think there is a se­ri­ous risk of that.

  4. As men­tioned el­se­where in the thread, I’m not say­ing that EA should em­brace poli­ti­cal level self-re­straint. What I am say­ing is that there are some­times rea­sons to self-cen­sor hold­ing forth on all of your opinions in pub­lic when you rep­re­sent a com­mu­nity of peo­ple try­ing to achieve some­thing im­por­tant. The re­spon­dents to this poll im­plic­itly agree with that given that they want to re­main anony­mous. For some of these state­ments, the rep­u­ta­tional risk of airing them anony­mously does not trans­fer from them to the EA move­ment as a whole. For other state­ments, the rep­u­ta­tional risk does trans­fer from them to the com­mu­nity as a whole.

  Do you think any­one in the com­mu­nity should ever self-cen­sor for the sake of the rep­u­ta­tion of the move­ment? Do you think sci­en­tists should ever self-cen­sor their views?

• Halstead 3 Oct 2019 9:40 UTC

  10 points

  in reply to: Khorton's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  This post ac­tively en­courages peo­ple to post their least ac­cept­able views on­line, so seems bad by this ar­gu­ment.

• Halstead 3 Oct 2019 9:39 UTC

  8 points

  in reply to: Habryka's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  Hi, you start with a straw man here—I’m not re­quest­ing that they write a whole es­say, I’m just re­quest­ing that they put some thought into the po­ten­tial down­sides, rather than zero thought (as oc­curred here). As I un­der­stand your view, you think the per­son has no obli­ga­tion to put any thought into whether pub­lish­ing this post is a good idea or not. I have to say I find this an im­plau­si­ble and strange po­si­tion.

• Halstead 3 Oct 2019 4:18 UTC

  17 points

  in reply to: Habryka's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  I re­spect your view Oli, but I don’t think the per­son or­ganis­ing it put suffi­cient thought into the down­sides of do­ing a poll such as this. They didn’t dis­cuss any of the ob­vi­ous risks in the ‘why this is a valuable ex­er­cise’ sec­tion.

• Halstead 3 Oct 2019 0:18 UTC

  13 points

  in reply to: Ben Pace's comment on: [Link] What opinions do you hold that you would be re­luc­tant to ex­press in front of a group of effec­tive al­tru­ists? Anony­mous form.

  The poli­ti­cal anal­ogy was an ex­am­ple; it was not meant to say that stan­dard poli­ti­cal con­straints should ap­ply to EA. The thought ap­plies to any so­cial move­ment, e.g. for peo­ple in­volved in en­vi­ron­men­tal­ism, rad­i­cal ex­change or liber­tar­i­anism. If I were a liber­tar­ian and some­one came to me say­ing “why don’t we run a poll of liber­tar­i­ans on opinions they are scared to air pub­li­cly and then pub­lish those opinions on­line for the world to see”, I think it would be pretty ob­vi­ous that this would be an ex­tremely bad idea.