William_MacAskill

• William_MacAskill 15 Dec 2018 19:59 UTC

  11 points

  on: Why You Should In­vest In Up­grad­ing Democ­racy And Give To The Cen­ter For Elec­tion Science

  As one data point: I’m very pos­i­tive about CES, and think they’re one of the best marginal uses of fund­ing right now. (Note that Aaron didn’t ask me to write this.)

  (Ties: I’ve recom­mended a grant to CES from Open Phil be­fore, and a fur­ther grant is un­der con­sid­er­a­tion at OP right now; even given this pos­si­ble grant, CES would have need for fur­ther fund­ing for the com­ing years.)

An­nounc­ing the new Forethought Foun­da­tion for Global Pri­ori­ties Research

• William_MacAskill 21 Nov 2018 17:10 UTC

  16 points

  in reply to: guzey's comment on: William MacAskill mis­rep­re­sents much of the ev­i­dence un­der­ly­ing his key ar­gu­ments in “Do­ing Good Bet­ter”

  I sec­ond Ju­lia in her apol­ogy. In hind­sight, once I’d seen that you didn’t want the post shared I should have sim­ply ig­nored it, and en­sured you knew that it had been ac­ci­den­tally shared with me.

  When it was shared with me, the dam­age had already been done, so I thought it made sense to start prep­ping a re­sponse. I didn’t think your post would change sig­nifi­cantly, and at the time I thought it would be good for me to start go­ing through your cri­tique to see if there were in­deed grave mis­takes in DGB, and offer a speedy re­sponse for a more fruit­ful dis­cus­sion. I’m sorry that I there­fore mis­rep­re­sented you. As you know, the draft you sent to Ju­lia was quite a bit more hos­tile than the pub­lished ver­sion; I can only say that as a re­sult of this I felt un­der at­tack, and that clouded my judg­ment.

• William_MacAskill 18 Nov 2018 16:59 UTC

  33 points

  in reply to: Jan_Kulveit's comment on: William MacAskill mis­rep­re­sents much of the ev­i­dence un­der­ly­ing his key ar­gu­ments in “Do­ing Good Bet­ter”

  I agree with all the points you make here, in­clud­ing on the sug­gested up­vote/​down­vote dis­tri­bu­tion, and on the na­ture of DGB. FWIW, my (cur­rent, defea­si­ble) plan for any fu­ture trade books I write is that they’d be more high­brow (and more caveated, and there­fore drier) than DGB.

  I think that’s the right ap­proach for me, at the mo­ment. But pre­sum­ably at some point the best thing to do (for some peo­ple) will be wider ad­vo­cacy (wider than DGB), which will in­evitably in­volve sim­plifi­ca­tion of ideas. So we’ll have to figure out what epistemic stan­dards are ap­pro­pri­ate in that con­text (given that GiveWell-level de­tail is off the table).

  Some pre­limi­nary thoughts on heuris­tics for this (these are sug­ges­tions only):

  Stan­dards we’d want to keep as high as ever:

  • Is the broad brush strokes pic­ture of what is be­ing con­veyed ac­cu­rate? Is there any easy way the broad brush of what is con­veyed could have been made more ac­cu­rate?

  • Are the sen­tences be­ing used to sup­port this broad brush strokes pic­ture war­ranted by the ev­i­dence?

  • Is this the way of com­mu­ni­cat­ing the core mes­sage about as caveated and de­tailed as one can rea­son­ably man­age?

  Stan­dards we’d need to re­lax:

  • Does this com­mu­ni­cate as much de­tail as pos­si­ble with re­spect to the rele­vant claims?

  • Does this com­mu­ni­cate all the strongest pos­si­ble coun­ter­ar­gu­ments to the key claim?

  • Does this in­clude ev­ery rea­son­able caveat?

  I think that a blog­post that does very well with re­spect to the above, with­out com­pro­mis­ing on the clar­ity of the core mes­sage, is Max Roser’s re­cent post: ‘The world is much bet­ter; The world is awful; The world can be much bet­ter’.

• William_MacAskill 17 Nov 2018 16:10 UTC

  72 points

  on: William MacAskill mis­rep­re­sents much of the ev­i­dence un­der­ly­ing his key ar­gu­ments in “Do­ing Good Bet­ter”

  Hi Alexey,

  I ap­pre­ci­ate that you’ve taken the time to con­sider what I’ve said in the book at such length. How­ever, I do think that there’s quite a lot that’s wrong in your post, and I’ll de­scribe some of that be­low. Though I think you have no­ticed a cou­ple of mis­takes in the book, I think that most of the alleged er­rors are not er­rors.

  I’ll just fo­cus on what I take to be the main is­sues you high­light, and I won’t ad­dress the ‘dishon­esty’ alle­ga­tions, as I an­ti­ci­pate it wouldn’t be pro­duc­tive to do so; I’ll leave that charge for oth­ers to as­sess.

  tl;dr:

  • Of the main is­sues you re­fer to, I think you’ve iden­ti­fied two mis­takes in the book: I left out a caveat in my sum­mary of the Baird et al (2016) pa­per, and I con­flated over­heads costs and CEO pay in a way that, on the lat­ter as­pect, was un­fair to Char­ity Nav­i­ga­tor.

  • In nei­ther case are these er­rors egre­gious in the way you sug­gest. I think that: (i) claiming that the Baird et al (2016) should cause us to be­lieve that there is ‘no effect’ on wages is a mis­rep­re­sen­ta­tion of that pa­per; (ii) my core ar­gu­ment against Char­ity Nav­i­ga­tor, re­gard­ing their fo­cus on ‘fi­nan­cial effi­ciency’ met­rics like over­head costs, is both suc­cess­ful and ac­cu­rately de­picts Char­ity Nav­i­ga­tor.

  • I don’t think that the rest of the alleged ma­jor er­rors are er­rors. In par­tic­u­lar: (i) GiveWell were able to re­view the manuscript be­fore pub­li­ca­tion and were happy with how I pre­sented their re­search; the quotes you give gen­er­ally con­flate how to think about GiveWell’s es­ti­mates with how to think about DCP2’s es­ti­mates; (ii) There are many lines of ev­i­dence sup­port­ing the 100x mul­ti­plier, and I don’t rely at all on the DCP2 es­ti­mates, as you im­ply.

  (Also, caveat­ing up front: for rea­sons of time limi­ta­tions, I’m go­ing to have to pre­com­mit to this be­ing my last com­ment on this thread.)

  (Also, Alexey’s post keeps chang­ing, so if it looks like I’m re­spond­ing to some­thing that’s no longer there, that’s why.)

  1. Deworming

  Since the book came out, there has been much more de­bate about the effi­cacy of de­worm­ing. As I’ve con­tinued to learn about the state and qual­ity of the em­piri­cal ev­i­dence around de­worm­ing, I’ve be­come less happy with my pre­sen­ta­tion of the ev­i­dence around de­worm­ing in Do­ing Good Bet­ter; this fact has been re­flected on the er­rata page on my web­site for the last two years. On your par­tic­u­lar points, how­ever:

  De­worm­ing vs textbooks

  If text­books have a pos­i­tive effect, it’s via how much chil­dren learn in school, rather than an in­cen­tive for them to spend more time in school. So the fact that there doesn’t seem to be good ev­i­dence for text­books in­creas­ing test scores is pretty bad.

  If de­worm­ing has a pos­i­tive effect, it could be via a num­ber of mechanisms, in­clud­ing in­creased school at­ten­dance or via learn­ing more in school, or di­rect health im­pacts, etc. If there are big gains on any of these di­men­sions, then de­worm­ing looks promis­ing. I agree that more days in school cer­tainly aren’t good in them­selves, how­ever, so the bet­ter ev­i­dence is about the long-run effects.

  De­worm­ing’s long-run effects

  Here’s how GiveWell de­scribes the study on which I base my dis­cus­sion of the long-run effects of de­worm­ing:

  “10-year fol­low-up: Baird et al. 2016 com­pared the first two groups of schools to re­ceive de­worm­ing (as treat­ment group) to the fi­nal group (as con­trol); the treat­ment group was as­signed 2.41 ex­tra years of de­worm­ing on av­er­age. The study’s head­line effect is that as adults, those in the treat­ment group worked and earned sub­stan­tially more, with in­creased earn­ings driven largely by a shift into the man­u­fac­tur­ing sec­tor.” Then, later: “We have done a va­ri­ety of analy­ses to as­sess the ro­bust­ness of the core find­ings from Baird et al. 2016, in­clud­ing re­an­a­lyz­ing the data and code un­der­ly­ing the study, and the re­sults have held up to our scrutiny.”

  You are cor­rect that my de­scrip­tion of the find­ings of the Baird et al pa­per was not fully ac­cu­rate. When I wrote, “More­over, when Kre­mer’s col­leagues fol­lowed up with the chil­dren ten years later, those who had been de­wormed were work­ing an ex­tra 3.4 hours per week and earn­ing an ex­tra 20 per­cent of in­come com­pared to those who had not been de­wormed,” I should have in­cluded the caveat “among non-stu­dents with wage em­ploy­ment.” I’m sorry about that, and I’m up­dat­ing my er­rata page to re­flect this.

  As for how much we should up­date on the ba­sis of the Baird et al pa­per — that’s a re­ally big dis­cus­sion, and I’m not go­ing to be able to add any­thing above what GiveWell have already writ­ten (here, here and here). I’ll just note that:

  (i) Your gloss on the pa­per seems mis­lead­ing to me. If you in­clude peo­ple with zero earn­ings, of course it’s go­ing to be harder to get a statis­ti­cally sig­nifi­cant effect. And the data from those who do have an in­come but who aren’t in wage em­ploy­ment are nois­ier, so it’s harder to get a statis­ti­cally sig­nifi­cant effect there too. In par­tic­u­lar, see here from the 2015 ver­sion of the pa­per: “The data on [non-agri­cul­tural] self-em­ploy­ment prof­its are likely mea­sured with some­what more noise. Monthly prof­its are 22% larger in the treat­ment group, but the differ­ence is not sig­nifi­cant (Table 4, Panel C), in part due to large stan­dard er­rors cre­ated by a few male out­liers re­port­ing ex­tremely high prof­its. In a ver­sion of the profit data that trims the top 5% of ob­ser­va­tions, the differ­ence is 28% (P < 0.10).”

  (ii) GiveWell finds the Baird et al pa­per to be an im­por­tant part of the ev­i­dence be­hind their sup­port of de­worm­ing. If you dis­agree with that, then you’re en­gaged in a sub­stan­tive dis­agree­ment with GiveWell’s views; it seems wrong to me to class that as a sim­ple mis­rep­re­sen­ta­tion.

  2. Cost-effec­tive­ness estimates

  Given the pre­vi­ous de­bate that had oc­curred be­tween us on how to think and talk about cost-effec­tive­ness es­ti­mates, and the mis­takes I had made in this re­gard, I wanted to be sure that I was pre­sent­ing these es­ti­mates in a way that those at GiveWell would be happy with. So I asked an em­ployee of GiveWell to look over the rele­vant parts of the manuscript of DGB be­fore it was pub­lished; in the end five em­ploy­ees did so, and they were happy with how I pre­sented GiveWell’s views and re­search.

  How can that fact be rec­on­ciled with the quotes you give in your blog post? It’s be­cause, in your dis­cus­sion, you con­flate two quite differ­ent is­sues: (i) how to rep­re­sent that cost-effec­tive­ness es­ti­mates pro­vided by DCP2, or by sin­gle stud­ies; (ii) how to rep­re­sent the (in my view much more rigor­ous) cost-effec­tive­ness es­ti­mates pro­vided by GiveWell. Al­most all the quotes from Holden that you give are about (i). But the quotes you crit­i­cise me for are about (ii). So, for ex­am­ple, when I say ‘these es­ti­mates’ are or­der of mag­ni­tude es­ti­mates that’s refer­ring to (i), not to (ii).

  There’s a re­ally big differ­ence be­tween (i) and (ii). I ac­knowl­edge that back in 2010 I was badly wrong about the re­li­a­bil­ity of DCP2 and in­di­vi­d­ual stud­ies, and that GWWC was far too slow to up­date its web pages af­ter the un­re­li­a­bil­ity of these es­ti­mates came to light. But the level of time, care and rigour that have gone into the GiveWell es­ti­mates are much greater than those that have gone into the DCP2 es­ti­mates. It’s still the case that there’s a huge amount of un­cer­tainty sur­round­ing the GiveWell es­ti­mates, but de­scribing them as “the most rigor­ous es­ti­mates” we have seems rea­son­able to me.

  More broadly: Do I re­ally think that you do as much good or more in ex­pec­ta­tion from donat­ing $3500 to AMF as sav­ing a child’s life? Yes. GiveWell’s es­ti­mate of the di­rect benefits might be op­ti­mistic or pes­simistic (though it has stayed rel­a­tively sta­ble over many years now — the me­dian GiveWell es­ti­mate for ‘cost for out­come as good as avert­ing the death of an in­di­vi­d­ual un­der 5’ is cur­rently $1932), but I re­ally don’t have a view on which is more likely. And, what’s more im­por­tant, the biggest con­sid­er­a­tion that’s miss­ing from GiveWell’s anal­y­sis is the long-run effects of sav­ing a life. While of course it’s a thorny is­sue, I per­son­ally find it plau­si­ble that the long-run ex­pected benefits from a dona­tion to AMF are con­sid­er­ably larger than the short-run benefits — you speed up eco­nomic progress just a lit­tle bit, in ex­pec­ta­tion mak­ing those in the fu­ture just a lit­tle bit bet­ter off than they would have oth­er­wise been. Be­cause the fu­ture is so vast in ex­pec­ta­tion, that effect is very large. (There’s *plenty* more to dis­cuss on this is­sue of long-run effects — Might those effects be nega­tive? How should you dis­count fu­ture con­sump­tion? etc — but that would take us too far afield.)

  3. Char­ity Navigator

  Let’s dis­t­in­guish: (i) the use of over­head ra­tio as a met­ric in as­sess­ing char­i­ties; (ii) the use of CEO pay as a met­ric in as­sess­ing char­i­ties. The idea of eval­u­at­ing char­i­ties on over­heads and on the ba­sis of CEO pay are of­ten run to­gether in pub­lic dis­cus­sion, and are both wrong for similar rea­sons, so I bun­dled them to­gether in my dis­cus­sion.

  Re­gard­ing (ii): CN-of-2014 did talk a lot about CEO pay: they fea­tured CEO pay, in both ab­solute terms and as a pro­por­tion of ex­pen­di­ture, promi­nently on their char­ity eval­u­a­tion pages (see, e.g. their page on Books for Africa), they had top-ten lists like, “10 highly-rated char­i­ties with low paid CEOs”, and “10 highly paid CEOs at low-rated char­i­ties” (and no lists of “10 highly-rated char­i­ties with high paid CEOs” or “10 low-rated char­i­ties with low paid CEOs”). How­ever, it is true that CEO pay was not a part of CN’s rat­ing sys­tem. And, reread­ing the rele­vant pas­sages of DGB, I can see how the reader would have come away with the wrong im­pres­sion on that score. So I’m sorry about that. (Per­haps I was sub­con­sciously still ornery from their spec­tac­u­larly hos­tile hit piece on EA that came out while I was writ­ing DGB, and was there­fore less care­ful than I should have been.) I’ve up­dated my er­rata page to make that clear.

  Re­gard­ing (i): CN’s two key met­rics for char­i­ties are (a) fi­nan­cial health and (b) ac­countabil­ity and trans­parency. (a) is in very sig­nifi­cant part about the char­i­ties’ over­heads ra­tios (in sev­eral differ­ent forms), where they give a char­ity a higher score the lower its over­heads are, break­ing the scores into five broad buck­ets: see here for more de­tail. The dough­nuts for po­lice officers ex­am­ple shows that a re­ally bad char­ity could score ex­tremely highly on CN’s met­rics, which shows that CN’s met­rics must be wrong. Similarly for Books for Africa, which gets a near-perfect score from CN, and fea­tures in its ‘ten top-notch char­i­ties’ list, in sig­nifi­cant part be­cause of its very low over­heads, de­spite hav­ing no good ev­i­dence to sup­port its pro­gram.

  I rep­re­sent CN fairly, and make a fair crit­i­cism of its ap­proach to as­sess­ing char­i­ties. In the ex­tended quote you give, they caveat that very low over­heads are not make-or-break for a char­ity. But, on their char­ity rat­ing method­ol­ogy, all other things be­ing equal they give a char­ity a higher score the lower the char­ity’s over­heads. If that scor­ing method is a bad one, which it is, then my crit­i­cism is jus­tified.

  4. Life satis­fac­tion and in­come and the hun­dred­fold multiplier

  The hun­dred­fold multiplier

  You make two ob­jec­tions to my 100x mul­ti­plier claim: that the DCP2 de­worm­ing es­ti­mate was off by 100x, and that the Steven­son and Wolfers pa­per does not sup­port it.

  But there are very many lines of ev­i­dence in favour of the 100x mul­ti­plier, which I refer­ence in Do­ing Good Bet­ter. I men­tion that there are many in­de­pen­dent jus­tifi­ca­tions for think­ing that there is a log­a­r­ith­mic (or even more con­cave) re­la­tion­ship be­tween in­come and hap­piness on p.25, and in the end­notes on p.261-2 (all refer­ences are to the Bri­tish pa­per­back edi­tion—yel­low cover). In ad­di­tion to the Steven­son and Wolfers life­time satis­fac­tion ap­proach (which I dis­cuss later), here are some rea­sons for think­ing that the hun­dred­fold mul­ti­plier ob­tains:

  • The ex­pe­ri­en­tial sam­pling method of as­sess­ing hap­piness. I men­tion this in the end­note on p.262, point­ing out that, on this method, my ar­gu­ment would be stronger, be­cause on this method the re­la­tion­ship be­tween in­come and wellbe­ing is more con­cave than log­a­r­ith­mic, and is in fact bounded above.

  • Im­puted util­ity func­tions from the mar­ket be­havi­our of pri­vate in­di­vi­d­u­als and the ac­tions of gov­ern­ment. It’s ab­solutely main­stream eco­nomic thought that util­ity varies with log of in­come (that is, eta=1 in an isoe­las­tic util­ity func­tion) or some­thing more con­cave (eta>1). I refer­ence a pa­per that takes this ap­proach on p.261, the Groom and Mad­di­son (2013). They es­ti­mate eta to be 1.5.

  • Es­ti­mates of cost to save a life. I dis­cuss this in ch.2; I note that this is an­other strand of sup­port­ing ev­i­dence prior to my dis­cus­sion of Steven­son and Wolfers on p.25: “It’s a ba­sic rule of eco­nomics that money is less valuable to you the more you have of it. We should there­fore ex­pect $1 to provide a larger benefit for an ex­tremely poor In­dian farmer than it would for you or me. But how much larger? Economists have sought to an­swer this ques­tion through a va­ri­ety of meth­ods. We’ll look at some of these in the next chap­ter, but for now I’ll just dis­cuss one [the Steven­son and Wolfers ap­proach].” Again, you find 100x or more dis­crep­ancy in the cost to save a life in rich or poor coun­tries.

  • Es­ti­mate of cost to provide one QALY. As with the pre­vi­ous bul­let point.

  Note, cru­cially, that the de­vel­op­ing world es­ti­mates for cost to provide one QALY or cost to save a life come from GiveWell, not — as you im­ply — from DCP2 or any in­di­vi­d­ual study.

  Is there a causal re­la­tion­ship from in­come to wellbe­ing?

  It’s true that there Steven­son and Wolfers only shows the cor­re­la­tion is be­tween in­come and wellbe­ing. But that there is a causal re­la­tion­ship, from in­come to wellbe­ing, is be­yond doubt. It’s perfectly ob­vi­ous that, over the scales we’re talk­ing, higher in­come en­ables you to have more wellbe­ing (you can buy anal­gesics, health­care, shelter, eat more and bet­ter food, etc).

  It’s true that we don’t know ex­actly the strength of the causal re­la­tion­ship. Un­der­stand­ing this could make my ar­gu­ment stronger or weaker. To illus­trate, here’s a quote from an­other Steven­son and Wolfers pa­per, with the nu­mer­als in square brack­ets added in by me:

  “Although our anal­y­sis pro­vides a use­ful mea­sure­ment of the bi­vari­ate re­la­tion­ship be­tween in­come and well-be­ing both within and be­tween coun­tries, there are good rea­sons to doubt that this cor­re­sponds to the causal effect of in­come on well-be­ing. It seems plau­si­ble (per­haps even likely) that [i] the within-coun­try well-be­ing-in­come gra­di­ent may be bi­ased up­ward by re­verse cau­sa­tion, as hap­piness may well be a pro­duc­tive trait in some oc­cu­pa­tions, rais­ing in­come. A differ­ent per­spec­tive, from offered by Kah­ne­man, et al. (2006), sug­gests that [ii] within-coun­try com­par­i­sons over­state the true re­la­tion­ship be­tween sub­jec­tive well-be­ing and in­come be­cause of a “fo­cus­ing illu­sion”: the very na­ture of ask­ing about life satis­fac­tion leads peo­ple to as­sess their life rel­a­tive to oth­ers, and they thus fo­cus on where they fall rel­a­tive to oth­ers in re­gard to con­crete mea­sures such as in­come. Although these spe­cific bi­ases may have a more im­por­tant im­pact on within-coun­try com­par­i­sons, it seems likely that [iii] the bi­vari­ate well-be­ing-GDP re­la­tion­ship may also re­flect the in­fluence of third fac­tors, such as democ­racy, the qual­ity of na­tional laws or gov­ern­ment, health, or even fa­vor­able weather con­di­tions, and many of these fac­tors raise both GDP per cap­ita and well-be­ing (Kenny, 1999).29 [iv] Other fac­tors, such as in­creased sav­ings, re­duced leisure, or even in­creas­ingly ma­te­ri­al­ist val­ues may raise GDP per cap­ita at the ex­pense of sub­jec­tive well-be­ing. At this stage we can­not ad­dress these short­com­ings in any de­tail, al­though, given our re­assess­ment of the stylized facts, we would sug­gest an ur­gent need for re­search iden­ti­fy­ing these causal pa­ram­e­ters.”

  To the ex­tent to which (i), (ii) or (iv) are true, the case for the 100x mul­ti­plier be­comes stronger. To the ex­tent to which (iii) is true, the case for the 100x mul­ti­plier be­comes weaker. We don’t know, at the mo­ment, which of these are the most im­por­tant fac­tors. But, given that the wide va­ri­ety of differ­ent strands of ev­i­dence listed in the pre­vi­ous sec­tion all point in the same di­rec­tion, I think that es­ti­mat­ing a 100x mul­ti­plier as a causal mat­ter is rea­son­able. (Fi­nal point: not­ing again that all these es­ti­mates do not fac­tor in the long-run benefits of dona­tions, which would in­crease the ra­tio of benefits oth­ers to benefits to your­self even fur­ther in the di­rec­tion of benefits to oth­ers.)

  On the Steven­son and Wolfers data, is the re­la­tion­ship be­tween in­come and hap­piness weaker for poor coun­tries than for rich coun­tries?

  If it were the case that money does less to buy hap­piness (for any given in­come level) in poor coun­tries than in rich coun­tries, then that would be one coun­ter­ar­gu­ment to mine.

  How­ever, it doesn’t seem to me that this is true of the Steven­son and Wolfers data. In par­tic­u­lar, it’s highly cherry-picked to com­pare Nige­ria and the USA as you do, be­cause Nige­ria is a clear out­lier in terms of how flat the slope is. I’m only eye­bal­ling the graph, but it seems to me that, of the poor­est coun­tries rep­re­sented (PHL, BGD, EGY, CHN, IND, PAK, NGA, ZAF, IDN), only NGA and ZAF have flat­ter slopes than USA (and even for ZAF, that’s only true for in­comes less than $6000 or so); all the rest have slopes that are similar to or steeper than that of USA (IND, PAK, BGD, CHN, EGY, IDN all seem steeper than USA to me). Given that Nige­ria is such an out­lier, I’m in­clined not to give it too much weight. The av­er­age trend across coun­tries, rich and poor, is pretty clear.

• William_MacAskill 13 Jun 2017 17:34 UTC

  6 points

  in reply to: Owen_Cotton-Barratt's comment on: Pro­jects I’d like to see

  Thanks Owen!

  Re Etg buy-out—yes, you’re right. For peo­ple who think that CEA is a top dona­tion tar­get, hope­fully we could just come to agree­ment as a trade wouldn’t be pos­si­ble, or would be pro­hibitively costly (if there were only slight differ­ences in our views on which places were best to fund).

  Re lo­cal group ac­tivi­ties: Th­ese are just ex­am­ples of some of the things I’d be ex­cited about lo­cal groups do­ing, and I know that at least some lo­cal groups are fund­ing con­strained (e.g. some­one is run­ning them part-time, un­paid, and will oth­er­wise need to get a job).

  Re AI safety fel­low­ship at ASI—as I un­der­stand it, that is cur­rently fund­ing con­strained (they had great ap­pli­cants who wanted to take the fel­low­ship but ASI couldn’t fund it). For other ap­pli­ca­tions (e.g. Google Brain) it could in­volve, say, spend­ing some amount of time dur­ing or af­ter a physics or math PhD in or­der to learn some ma­chine learn­ing and be more com­pet­i­tive.

  Re an­thro­pogenic ex­is­ten­tial risks—ah, I had thought that it was only in pre­sen­ta­tion form. In which case: that pa­per is ex­actly the sort of thing I’d love to see more of.

• William_MacAskill 13 Jun 2017 11:19 UTC

  4 points

  in reply to: riceissa's comment on: An­nounc­ing Effec­tive Altru­ism Grants

  It is a suc­ces­sor to EA Ven­tures, though EA Grants already has fund­ing, and is more fo­cused on in­di­vi­d­u­als than start-up pro­jects.

• William_MacAskill 13 Jun 2017 11:18 UTC

  7 points

  in reply to: joshjacobson's comment on: Pro­jects I’d like to see

  Yes, the money is raised; we have a pot of £500,000 in the first in­stance.

  It is a suc­ces­sor to EA Ven­tures, though EA Grants already has fund­ing, and is more fo­cused on in­di­vi­d­u­als than start-up pro­jects.

Pro­jects I’d like to see

• William_MacAskill 11 Apr 2017 20:18 UTC

  7 points

  in reply to: the_jaded_one's comment on: How ac­cu­rately does any­one know the global dis­tri­bu­tion of in­come?

  “How­ever, $700/​year (= $1.91/​day, =€1.80/​day, =£1.53 /​day) (with­out gifts or hand­outs) is not a suffi­cient amount of money to be al­ive in the west. You would be home­less. You would starve to death. In many places, you would die of ex­po­sure in the win­ter with­out shelter.”

  One could live on that amount of money per day in the West. You’d live in a sec­ond-hand tent, you’d scav­enge food from bins (which would count to­wards your ‘ex­pen­di­ture’, be­cause we’re talk­ing about con­sump­tion ex­pen­di­ture, but wouldn’t count that much). Your life ex­pec­tancy would be con­sid­er­ably lower than oth­ers in the West, but prob­a­bly not lower than the 55 years which is the life ex­pec­tancy in Burk­ina Faso (as an ex­am­ple com­par­i­son, bear in mind that num­ber in­cludes in­fant mor­tal­ity). Your life would suck very badly, but you wouldn’t die, and it wouldn’t be that dis­similar to the lives of the mil­lions of peo­ple who live in makeshift slums or shanty towns and scav­enge from dumps to make a liv­ing. (Such peo­ple aren’t rep­re­sen­ta­tive of all ex­tremely poor peo­ple, but they are a no­table frac­tion.)

• William_MacAskill 4 Apr 2017 19:00 UTC

  0 points

  in reply to: RobBensinger's comment on: In­tu­ition Joust­ing: What It Is And Why It Should Stop

  “counts as an xrisk (and there­fore as a GCR)”

  My un­der­stand­ing: GCR = (some­thing like) risk of ma­jor catas­tro­phe that kills 100mn+ people

  (I think the GCR book defines it as risk of 10mn+ deaths, but that seemed too low to me).

  So, as I was us­ing the term, some­thing be­ing an x-risk does not en­tail it be­ing a GCR. I’d count ‘Hu­man­ity’s moral progress stag­nates or we oth­er­wise end up with the wrong val­ues’ as an x-risk but not a GCR.

  In­ter­est­ing (/​wor­ry­ing!) how we’re un­der­stand­ing widely-used terms so differ­ently.

• William_MacAskill 4 Apr 2017 18:55 UTC

  6 points

  in reply to: Carl_Shulman's comment on: In­tu­ition Joust­ing: What It Is And Why It Should Stop

  Mea culpa that I switched from “im­pact on be­ings al­ive to­day” to “benefits over the next 50 years” with­out notic­ing.

• William_MacAskill 31 Mar 2017 17:13 UTC

  1 point

  in reply to: Robert_Wiblin's comment on: In­tu­ition Joust­ing: What It Is And Why It Should Stop

  That’s rea­son­able, though if the aim is just “benefits over the next 50 years” I think that cam­paigns against fac­tory farm­ing seem like the stronger com­par­i­son:

  “We’ve es­ti­mated that cor­po­rate cam­paigns can spare over 200 hens from cage con­fine­ment for each dol­lar spent. If we roughly imag­ine that each hen gains two years of 25%-im­proved life, this is equiv­a­lent to one hen-life-year for ev­ery $0.01 spent.” “One could, of course, value chick­ens while valu­ing hu­mans more. If one val­ues hu­mans 10-100x as much, this still im­plies that cor­po­rate cam­paigns are a far bet­ter use of funds (100-1,000x) [So $30-ish per equiv­a­lent life saved].” http://​www.open­philan­thropy.org/​blog/​wor­ld­view-di­ver­sifi­ca­tion

  And to clar­ify my first com­ment, “un­likely to be op­ti­mal” = I think it’s a con­tender, but the base rate for “X is an op­ti­mal in­ter­ven­tion” is re­ally low.

• William_MacAskill 30 Mar 2017 23:34 UTC

  9 points

  in reply to: RobBensinger's comment on: In­tu­ition Joust­ing: What It Is And Why It Should Stop

  Agree that GCRs are a within-our-life­time prob­lem. But in my view miti­gat­ing GCRs is un­likely to be the op­ti­mal dona­tion tar­get if you are only con­sid­er­ing the im­pact on be­ings al­ive to­day. Do you know of any sources that make the op­po­site case?

  And it’s framed as long-run fu­ture be­cause we think that there are po­ten­tially lots of things that could have a huge pos­i­tive on the value of the long-run fu­ture which aren’t GCRs—like hu­man­ity hav­ing the right val­ues, for ex­am­ple.

• William_MacAskill 8 Mar 2017 1:59 UTC

  12 points

  in reply to: jpaddison's comment on: In­tro­duc­ing CEA’s Guid­ing Principles

  In my pre­vi­ous post I wrote: “The ex­is­tence of this would bring us into al­ign­ment with other so­cieties, which usu­ally have some doc­u­ment that de­scribes the prin­ci­ples that the so­ciety stands for, and has some mechanism for en­sur­ing that those who choose to rep­re­sent them­selves as part of that so­ciety abides by those prin­ci­ples.” I now think that’s an in­cor­rect state­ment. EA, cur­rently, is all of the fol­low­ing: an idea/​move­ment, a com­mu­nity, and a small group of or­gani­sa­tions. On the ‘move­ment’ un­der­stand­ing of EA, analo­gies of EA don’t have a com­mu­nity panel similar to what I sug­gested, and only some have ‘guid­ing prin­ci­ples’. (Though com­mu­ni­ties and or­gani­sa­tions, or groups of or­gani­sa­tions, of­ten do.)

  Ju­lia cre­ated a list of po­ten­tial analo­gies here:

  [https://​docs.google.com/​doc­u­ment/​d/​1aXQp_9pGauMK9rKES9W3Uk3soW6c1oSx68bhDmY73p4/​edit?usp=shar­ing].

  The clos­est anal­ogy to what we want to do is given by the open source com­mu­nity: many but not all of the or­gani­sa­tions within the open source com­mu­nity cre­ated their own codes of con­duct, many of them very similar to each other.

In­tro­duc­ing CEA’s Guid­ing Principles

[CEA Up­date] Up­dates from Jan­uary 2017

• William_MacAskill 11 Feb 2017 0:07 UTC

  4 points

  in reply to: jpaddison's comment on: In­tro­duc­ing the EA Funds

  One thing to note, re di­ver­sifi­ca­tion (which I do think is an im­por­tant point in gen­eral) is that it’s easy to think of Open Phil as a sin­gle agent, rather than a col­lec­tion of agents; and be­cause Open Phil is a col­lec­tive en­tity, there are gains from di­ver­sifi­ca­tion even with the funds.

  For ex­am­ple, there might be a grant that a pro­gram officer wants to make, but there’s in­ter­nal dis­agree­ment about it, and the pro­gram officer doesn’t have time (given op­por­tu­nity cost) to con­vince oth­ers at Open Phil why it’s a good idea. (This has been his­tor­i­cally true for, say, the EA Giv­ing Fund). Hav­ing a sep­a­rate pool of money would al­low them to fund things like that.

• William_MacAskill 10 Feb 2017 23:45 UTC

  10 points

  in reply to: LukeDing's comment on: In­tro­duc­ing the EA Funds

  Thanks so much for this, Luke! If some­one who spends half their work­ing time ded­i­cat­ing to philan­thropy, as you do, says “There is a limit to how much high qual­ity due dili­gence one could do. It takes time to build re­la­tion­ships, analyse op­por­tu­ni­ties and mon­i­tor them”—that’s pretty use­ful in­for­ma­tion!

• William_MacAskill 10 Feb 2017 7:45 UTC

  2 points

  in reply to: Mac-'s comment on: In­tro­duc­ing the EA Funds

  Thanks! That’s re­ally helpful to know. The Funds are po­ten­tially solv­ing a num­ber of prob­lems at once, and we know there’s some de­mand for each of these prob­lems to be solved, but not how much de­mand, so com­ments like this are very use­ful.