Carl_Shulman

• Carl_Shulman 27 Feb 2019 22:56 UTC

  21 points

  in reply to: Denis Drescher's comment on: Small an­i­mals have enor­mous brains for their size

  Re­cent large sam­ple within-fam­ily data does seem to es­tab­lish causal effects of brain size on in­tel­li­gence and ed­u­ca­tional at­tain­ment. The ge­netic cor­re­la­tion is ~0.4, so most of the ge­netic var­i­ance isn’t work­ing through over­all brain size.

  Some kinds of fea­tures that could con­tribute to ge­netic var­i­ance in hu­mans, but not scale for ar­bi­trary differ­ences across species:

  • Mu­ta­tion load (the rate at which this is trimmed back, and thus the equil­ibrium load, de­pends on the strength of se­lec­tion for cog­ni­tive abil­ities)

  • Mo­ti­va­tion: at­ten­tion to learn­ing, play, imi­ta­tion, and lan­guage comes at the ex­pense of at­ten­tion to other things

  • Pleiotropy with other se­lec­tion com­bined with evolu­tion­ary limits (se­lec­tion for lower ag­gres­sion also causes white patches in fur via changes in neu­ral crests, and re­ten­tion of a va­ri­ety of ju­ve­nile fea­tures), e.g. se­lec­tion for dis­ease re­sis­tance chang­ing path­ways so as to ac­ci­den­tally im­pair brain func­tion (with the change sur­viv­ing be­cause of its benefits)

  • Alle­les that provide re­sis­tance to dis­ease (ge­netic var­i­ance is main­tained in a Red Queen’s Race) that dam­ages the brain would be a source of ge­netic var­i­ance, like­wise var­i­ants af­fect­ing nu­tri­tion or other en­vi­ron­men­tal influences

• Carl_Shulman 27 Feb 2019 21:41 UTC

  15 points

  on: Quan­tify­ing an­thropic effects on the Fermi paradox

  Thank you for this ex­cel­lent and de­tailed post, I ex­pect to use it in the fu­ture as a go-to refer­ence for ex­plain­ing this point. You might be in­ter­ested in an old pa­per where Nick Bostrom and I went through some of this rea­son­ing (with similar con­clu­sions but much less ex­pla­na­tion) in the course of dis­cussing the im­pli­ca­tions of an­thropic the­o­ries for the pos­si­ble difficulty of evolv­ing in­tel­li­gence.

  I am not so sure about the spe­cific nu­mer­i­cal es­ti­mates you give, as op­posed to the bal­l­park be­ing within a few or­ders of mag­ni­tude for SIA and ADT+to­tal views (plus aux­iliary as­sump­tions), i.e. the vicinity of “(roughly the largest value that doesn’t make the Fermi ob­ser­va­tion too un­likely, as shown in the next two sec­tions”. But that’s com­pat­i­ble with much or most of our ex­pected on the to­tal view com­ing from sce­nar­ios where we don’t over­lap with aliens much.

  ″ How­ever, vary­ing the planet for­ma­tion rate at par­tic­u­lar times in the his­tory of the Uni­verse can make a large differ­ence.”

  We also up­date our un­cer­tainty about this sort of tem­po­ral struc­ture to some ex­tent from our ob­ser­va­tion of late ex­is­tence. Ideally we would want to let as much as pos­si­ble vary so that we don’t asym­met­ri­cally im­mu­nize some pa­ram­e­ters against up­date.

  “For this rea­son I will ig­nore sce­nar­ios where life is ex­traor­di­nar­ily un­likely to colon­ise the Uni­verse, by mak­ing fs lo­gu­niform be­tween 10−4 and 1.”

  This seems over­all too pes­simistic to me as a pre-an­thropic prior for coloniza­tion (~10% cre­dence).

• Carl_Shulman 31 Jan 2019 17:23 UTC

  7 points

  on: Cost-Effec­tive­ness of Aging Re­search

  I don’t think you can define ag­ing re­search so nar­rowly and get the same ex­pected im­pact. E.g. De Grey’s SENS in­cludes cur­ing can­cer as one of many sub­goals, and rad­i­cal ad­vances in stem cell biol­ogy and ge­netic en­g­ineer­ing, mas­sive fields that don’t fall un­der ‘ag­ing re­search.’ The more de­pen­dent progress in an area is ad­vances from out­side that field, the less re­li­able this sort of pro­jec­tion will be.

• Carl_Shulman 29 Jan 2019 23:47 UTC

  3 points

  on: Ex­pected cost per life saved of the TAME trial

  Hi Emanuele,

  I saw your re­quest for com­men­tary on Face­book, so here are some off-the-cuff com­ments (about 1 hour’s worth so take with ap­pro­pri­ate grains of salt, but sum­ma­riz­ing prior think­ing):

  • My prior take on met­formin was that it seems promis­ing for its space (albeit with mixed ev­i­dence, and prior longevity drug de­vel­op­ment efforts haven’t panned out, but the re­turns would be very high for med­i­cal re­search if true), al­though over­all the space looks less promis­ing than x-risk re­duc­tion to me; the fol­low­ing com­ments will be about de­tails of the anal­y­sis where I would cur­rently differ

  • The sug­ges­tion of this trial mov­ing for­ward LEV by 3+ years through an ice­breaker effect boost­ing re­search looks wildly im­plau­si­ble to me

    • LEV is not mainly bot­tle­necked on ‘re­search on ag­ing,’ e.g. de Grey’s pro­pos­als re­quire rad­i­cal ad­vances in gen­er­ally med­i­cally ap­pli­ca­ble stem cell and ge­netic en­g­ineer­ing tech­nolo­gies that already re­ceive mas­sive fund­ing and are quite challeng­ing; the abil­ity to re­place dis­eased cells with ge­net­i­cally en­g­ineered stem cell de­rived tis­sues is already a ma­jor pri­or­ity, and cur­ing can­cer is a small sub­set of SENS

    • Much of the ex­pected gain in biomed­i­cal tech­nol­ogy is not driven by shifts within biol­ogy, and ad­vances within a par­tic­u­lar med­i­cal field are heav­ily driven by broader im­prove­ments (e.g. com­put­ers, CRISPR, genome se­quenc­ing, PCR, etc); if LEV is far off and heav­ily de­pen­dent on other ar­eas, then de­vel­op­ments in other fields will make it com­par­a­tively easy for ag­ing re­search to benefit from ‘catch up growth’ re­duc­ing the ex­pected value of im­me­di­ate speedup (al­most all of which would have washed away if LEV hap­pens in the lat­ter half of the cen­tury)

    • In par­tic­u­lar, if au­tomat­ing R&D with AI is eas­ier than LEV, and would moot prior biomed­i­cal re­search, then that adds an ad­di­tional dis­count fac­tor; I would bet that this hap­pens be­fore LEV through biomed­i­cal research

    • Get­ting ap­proval to treat ‘ag­ing’ isn’t ac­tu­ally par­tic­u­larly helpful rel­a­tive to ap­proval for ‘dis­eases of ag­ing’ since all-cause mor­tal­ity re­quires larger tri­als and we don’t have great ag­ing bio­mark­ers; and the NIH has taken steps in that di­rec­tion regardless

    • Similar sto­ries have been told about other de­vel­op­ments and ex­per­i­ments, which haven’t had mas­sive ice­breaker effects

    • Com­bined, these effects look like they cost a cou­ple or­ders of mag­ni­tude

  • From my cur­rent epistemic state the ex­pected # of years added by met­formin looks too high

  • Re the Guessti­mate model the statis­ti­cal power of the trial is tightly tied to effect size; the larger the effect size the fewer peo­ple you need to show re­sults; that raises the re­turns of small tri­als, but means you have diminish­ing re­turns for larger ones (you are spend­ing more money to de­tect smaller effects so marginal cost-effec­tive­ness goes a lot lower than av­er­age cost-effec­tive­ness, re­flect­ing high VOI of test­ing the more ex­trav­a­gant pos­si­bil­ity)

  • Like­wise the pro­por­tion us­ing met­formin con­di­tional on a pos­i­tive re­sult is also cor­re­lated with effect size (which raises av­er­age EV, but shifts marginal EV lower pro­por­tionate to av­er­age EV); also the pro­por­tion of users seems too low to me con­di­tional on success

• Carl_Shulman 29 Jan 2019 23:45 UTC

  2 points

  on: A gen­eral frame­work for eval­u­at­ing ag­ing re­search. Part 1: rea­son­ing with Longevity Es­cape Ve­loc­ity.

  One is­sue I would add to your the­o­ret­i­cal anal­y­sis: with as­sign­ing 1000+ QALYs to let­ting some­one reach LEV is that peo­ple com­monly don’t claim lin­ear util­ity with lifes­pan, i.e. they would of­ten pre­fer to live to 80 with cer­tainty rather than die at 20 with 90% prob­a­bil­ity and live to 10,000 with 10% prob­a­bil­ity.

  I agree it’s worth keep­ing the chance that peo­ple will be able to live much longer in the fu­ture in mind when as­sess­ing benefits to ex­ist­ing peo­ple (I would also add the pos­si­bil­ity of dras­tic in­creases in qual­ity of life through tech­nol­ogy). I’d guess most of this comes from broader tech­nolog­i­cal im­prove­ments (e.g. via AI) rather than reach­ing LEV through biomed­i­cal ap­proaches), but not with ex­treme con­fi­dence.

  How­ever, I don’t think it has very rad­i­cal im­pli­ca­tions for cause pri­ori­ti­za­tion since, as you note, deaths for any rea­son (in­clude malaria and global catas­tro­phes) deny those peo­ple a chance at LEV. LEV-re­lated is­sues are also mainly a con­cern for ex­ist­ing hu­mans, so to the ex­tent one gives a boost for enor­mous im­pacts on non­hu­man an­i­mals and the ex­is­tence of fu­ture gen­er­a­tions, LEV speedup won’t reap much of those boosts.

  Within the field of biomed­i­cal re­search, ag­ing looks rel­a­tively promis­ing, and I think on av­er­age the best-tar­geted biomed­i­cal re­search does well for cur­rent peo­ple com­pared to lin­ear char­ity in sup­port of de­ploy­ment (e.g. gene drives vs bed­nets). But it’s not a slam dunk be­cause the prob­lems are so hard (in­clud­ing ones re­ceiv­ing mas­sive in­vest­ment). I don’t see it as strongly mov­ing most peo­ple who pre­fer to sup­port bed­nets over malaria gene drives, farmed an­i­mal welfare over gene drives, or GCR re­duc­tion over gene drives.

• Carl_Shulman 30 Dec 2018 17:51 UTC

  7 points

  in reply to: MichaelStJules's comment on: How High Con­tra­cep­tive Use Can Help An­i­mals?

  ″ Oh, is the con­cern that they’re look­ing at a more bi­ased sub­set of pos­si­ble effects (by fo­cus­ing pri­mar­ily on effects that seem pos­i­tive)? ”

  Yes. It doesn’t men­tion other analy­ses that have come to op­po­site con­clu­sions by con­sid­er­ing effects on wild an­i­mals and long-term de­vel­op­ment.

• Carl_Shulman 30 Dec 2018 5:10 UTC

  27 points

  on: How High Con­tra­cep­tive Use Can Help An­i­mals?

  If you’re go­ing to se­lect in­ter­ven­tions speci­fi­cally to re­duce the hu­man pop­u­la­tion and have down­stream con­se­quences, it seems ab­solutely es­sen­tial to take a broader view of the em­piri­cal con­se­quences than in the linked re­port. E.g. among oth­ers, effects on wild an­i­mals (not men­tioned but most im­me­di­ate an­i­mal effects of this change will be on wild an­i­mals), fu­ture tech­nolog­i­cal ad­vance­ment, and global catas­trophic risks have good cases for be­ing far larger and plau­si­bly of op­po­site sign to the effects dis­cussed in the re­port but are not men­tioned even as ar­eas for fur­ther in­ves­ti­ga­tion.

• Carl_Shulman 24 Dec 2018 20:23 UTC

  6 points

  in reply to: Jan_Kulveit's comment on: Should donor lot­tery win­ners write re­ports?

  What about a re­port along the lines of ‘I am donat­ing in sup­port of X, for highly illeg­ible rea­sons re­lat­ing to my in­tu­ition from look­ing at their work, and pri­vate in­for­ma­tion I have about them per­son­ally’?

• Carl_Shulman 23 Dec 2018 21:10 UTC

  6 points

  in reply to: Denise_Melchin's comment on: Should donor lot­tery win­ners write re­ports?

  This is a good point, and worth high­light­ing in dis­cus­sion of re­ports (es­pe­cially as we get more data on the effects of win­ning on dona­tion pat­terns). On the other hand, the av­er­age depth and qual­ity of in­ves­ti­ga­tion by win­ners (and the ac­cess they got) does seem higher than what they would oth­er­wise have done, whilst less than ex­pert donors.

• Carl_Shulman 23 Dec 2018 21:05 UTC

  4 points

  in reply to: DavidNash's comment on: Should donor lot­tery win­ners write re­ports?

  I don’t think this is true. The prob­a­bil­ities and pay­outs are the same for any given par­ti­ci­pant, re­gard­less of what oth­ers do, so peo­ple who are un­likely to write up a re­port don’t re­duce the av­er­age num­ber of re­ports pro­duced by those who would.

• Carl_Shulman 23 Dec 2018 21:00 UTC

  4 points

  in reply to: rk's comment on: Should donor lot­tery win­ners write re­ports?

  Ex­cept that the pot size isn’t con­strained by the par­ti­ci­pa­tion of small donors: the CEA donor lot­tery has fixed pot sizes guaran­teed by large donors, and the largest donors could be ~risk-neu­tral over lot­ter­ies with pots of many mil­lions of donors. So there is no effect of this kind, and there is un­likely to ever be one ex­cept at lu­dicrously large scales (where one could use deriva­tives or the like to get similar effects).

• Carl_Shulman 23 Dec 2018 20:50 UTC

  4 points

  in reply to: Lukas_Finnveden's comment on: Should donor lot­tery win­ners write re­ports?

  Yes, the main effect bal­ances out like that.

  But in­so­far as the lot­tery en­hances the effec­tive­ness of donors (by let­ting them in­vest more in re­search if they win, amor­tized against a larger dona­tion), then you want donors do­ing good to be en­hanced and donors do­ing bad not to be en­hanced. So you might want to try to avoid boost­ing pot size available to bad donors, and en­sure good donors have large pots available. The CEA lot­tery is struc­tured so that ques­tion doesn’t arise.

  There is also the minor is­sue of cor­re­la­tion with other donors in the same block men­tioned in the above com­ment, al­though you could ask CEA for a sep­a­rate block if some un­usual situ­a­tion meant your dona­tion plans would change a lot if you found out an­other block par­ti­ci­pant had won.

• Carl_Shulman 23 Dec 2018 4:46 UTC

  8 points

  in reply to: Ben Pace's comment on: Should donor lot­tery win­ners write re­ports?

  > but also in the other 80% of wor­lds you have a prefer­ence for your money be­ing al­lo­cated by peo­ple who are more thought­ful.

  For the CEA donor lot­tery, the pot size is fixed in­de­pen­dent of one’s en­try as the guaran­tor (Paul Chris­ti­ano last year, the re­grant­ing pool I am ad­minis­ter­ing this year) puts in funds for any un­claimed tick­ets. So the dis­tri­bu­tion of fund­ing amounts for each en­trant is un­af­fected by other en­trants. It’s set up this way speci­fi­cally so that peo­ple don’t even have to think about the sort of effect you dis­cuss (the back­stop fund has ~lin­ear value of funds over the rele­vant range, so that isn’t an im­pact ei­ther).

  The only thing that par­ti­ci­pat­ing in the same lot­tery block as some­one else mat­ters for is cor­re­la­tions be­tween your dona­tions and theirs. E.g. if you would wind up choos­ing a differ­ent char­ity to give to de­pend­ing on whether an­other par­ti­ci­pant won the lot­tery. But nor­mally the be­hav­ior of one other donor wouldn’t change what you think is the best op­por­tu­nity.

• Carl_Shulman 22 Dec 2018 18:19 UTC

  7 points

  in reply to: Julia_Wise's comment on: How Effec­tive Altru­ists Can Be Wel­com­ing To Conservatives

  What hap­pened in those cases?

• Carl_Shulman 10 Dec 2018 0:30 UTC

  5 points

  in reply to: Milan_Griffes's comment on: Open Thread #43

  I would love to see a canon­i­cal post mak­ing this ar­gu­ment, con­flat­ing EA with the benefits of max­ing out per­sonal warm fuzzies is one of my pet peeves.

• Carl_Shulman 9 Dec 2018 4:00 UTC

  2 points

  in reply to: Arepo's comment on: Why we have over-rated Cool Earth

  I ac­tu­ally hap­pen to think that the re­port was too dis­mis­sive of more lev­er­aged cli­mate change in­ter­ven­tions that I ex­pected could be a lot bet­ter than the es­ti­mates for Cool Earth (es­pe­cially effi­cient an­gles on sci­en­tific re­search and poli­ti­cal ac­tivity in the cli­mate space), but the OP is sug­gest­ing that the origi­nal Cool Earth num­bers (which in­di­cate much lower cost-effec­tive­ness than char­i­ties recom­mended by EAs in other ar­eas with more ro­bust data) were over­stated, not un­der­stated (as the origi­nal re­port would sug­gest due to re­gres­sion to the mean and mea­sure­ment er­ror).

• Carl_Shulman 9 Dec 2018 3:56 UTC

  3 points

  in reply to: Taymon's comment on: Why we have over-rated Cool Earth

  One thing to em­pha­size more than that writeup did, is that in EA terms donat­ing to such a lightly re­searched in­ter­ven­tion (a few months work) is very likely dom­i­nated by dona­tions to bet­ter re­search the area, find­ing higher ex­pected value op­tions and in­fluenc­ing oth­ers.

  On the other hand, the point es­ti­mates in that re­port fa­vored other char­i­ties like AMF over Cool Earth any­way, a con­clu­sion strength­ened by the OP cri­tique (not that it ex­cludes some­thing else or­ders of mag­ni­tude bet­ter be­ing found like un­usual en­ergy re­search, very effec­tive poli­ti­cal lob­by­ing, geo­eng­ineer­ing, etc; Open Philan­thropy has made a few cli­mate grants that look rel­a­tively lev­er­aged).

  And I agree with John Maxwell about it be­ing over­sold in some cases.

• Carl_Shulman 14 Nov 2018 20:51 UTC

  21 points

  on: Is Ne­glect­ed­ness a Strong Pre­dic­tor of Marginal Im­pact?

  ″ I’d like to hear if there’s been any rele­vant work done on this topic (ei­ther within EA or­ga­ni­za­tions or within gen­eral academia). In­creas­ing re­turns is a fairly com­mon topic within eco­nomics, so I figure there is plenty of rele­vant re­search out there on this. ”

  Th­ese are my key rea­sons (with links to aca­demic EA and other dis­cus­sions) for see­ing diminish­ing re­turns as the rele­vant situ­a­tion on av­er­age for EA as a whole, and in par­tic­u­lar the most effec­tive causes:

  • If prob­lems can be solved, and vary in difficulty over mul­ti­ple or­ders of mag­ni­tude (in re­quired in­puts), you will tend to see diminish­ing re­turns as you plot the num­ber of prob­lems solved with in­creas­ing re­sources; see this se­ries of posts by Owen Cot­ton-Bar­ratt and others

  • Em­piri­cally, we do see sys­tem­atic diminish­ing re­turns to R&D in­puts across fields of sci­en­tific and tech­nolog­i­cal in­no­va­tion, and for global to­tal fac­tor pro­duc­tivity; but his­tor­i­cally the great­est suc­cesses of philan­thropy, re­duc­tions in poverty, and in­creased pros­per­ity have stemmed from in­no­va­tion, and many EA pri­ori­ties in­volve re­search and development

  • In poli­tics and pub­lic policy the liter­a­tures on lob­by­ing and cam­paign fi­nance sug­gest diminish­ing returns

  • In grow­ing new move­ments, there is an el­e­ment of com­pound­ing re­turns, as new par­ti­ci­pants carry for­ward work (in­clud­ing fur­ther growth), and so in­fluenc­ing; this topic has been the sub­ject of a fair amount of EA attention

  • When there are varied pos­si­ble ex­pen­di­tures with widely vary­ing cost-effec­tive­ness and some limits on room for more fund­ing (even­tu­ally, there may be in­creas­ing re­turns be­fore that), then work­ing one’s way from the most effec­tive op­tions to the least pro­duces a diminish­ing re­turns curve at a scale large enough to en­com­pass mul­ti­ple in­ter­ven­tions; Toby Ord dis­cusses the land­scape of global health in­ter­ven­tions hav­ing this property

  • Elab­o­rat­ing on the idea of limits to fund­ing and scal­ing: an ex­tremely cost-effec­tive in­ter­ven­tion with lin­ear or in­creas­ing re­turns that scaled to very large ex­pen­di­tures would of­ten im­ply im­pos­si­bly large effects; there can be cheap op­por­tu­ni­ties to save a hu­man life to­day for $100 un­der spe­cial cir­cum­stances, but there can’t be trillions of dol­lars worth of such op­por­tu­ni­ties, since you would be sav­ing more than the world pop­u­la­tion; like­wise the prob­a­bil­ity of pre­ma­ture ex­tinc­tion can­not fall be­low 0, etc

  • So far EA is still small and un­usual rel­a­tive to the world, and much of its ac­tivity is har­vest­ing low-hang­ing fruit from ar­eas with diminish­ing re­turns (a con­se­quence of those fruit) that couldn’t be scaled to ex­tremes (this is least true for lin­ear aid in­ter­ven­tions added to the already large global aid and lo­cal pro­grams, and in par­tic­u­lar GiveDirectly, but holds for what I would con­sider more promis­ing, in cost-effec­tive­ness, EA global health in­ter­ven­tions such as gene drive R&D for malaria erad­i­ca­tion); as EA ac­tivity ex­pands more cur­rently un­der­funded ar­eas will see re­turns diminish to the point of fal­ling be­hind in­ter­ven­tions with more lin­ear or in­creas­ing re­turns but worse cur­rent cost-effectiveness

  • Ex­pe­rience with suc­cesses us­ing ne­glect­ed­ness (which in pri­ori­ti­za­tion prac­tice does in­volve look­ing at the rea­sons for ne­glect) thus far, at least on di­men­sions for which feed­back has yet arrived

  ″ Ideally, we would like to not sim­ply se­lect causes that are ne­glected, but to se­lect causes that are ne­glected for rea­sons other than their im­pact. ”

  Agreed.

• Carl_Shulman 14 Nov 2018 19:59 UTC

  7 points

  on: Tiny Prob­a­bil­ities of Vast Utilities: A Prob­lem for Long-Ter­mism?

  The ex­am­ples in the post have ex­pected util­ities as­signed us­ing in­con­sis­tent method­olo­gies. If it’s pos­si­ble to have long-run effects on fu­ture gen­er­a­tions, then many ac­tions will have such effects (elec­tions can cause hu­man ex­tinc­tion some­times, an ad­di­tional per­son saved from malaria could go on to cause or pre­vent ex­tinc­tion). If lu­dicrously vast uni­verses and in­fluence over them are sub­jec­tively pos­si­ble, then we should like­wise con­sider be­ing less likely to get lu­dicrous re­turns if we are ex­tinct or badly-gov­erned (see ‘em­piri­cal sta­bi­liza­tion as­sump­tions’ in Nick Bostrom’s in­finite ethics pa­per). We might have in­finite im­pact (un­der cer­tain de­ci­sion the­o­ries) when we make a de­ci­sion to eat a sand­wich if there are in­finite phys­i­cally iden­ti­cal be­ings in the uni­verse who will make the same de­ci­sion as us.

  Any ar­gu­ment of the form “con­sider type of con­se­quence X, which is larger than con­se­quences you had pre­vi­ously con­sid­ered, as it ap­plies to op­tion A” calls for ap­pli­ca­tion of X to an­a­lyz­ing other op­tions. When you do that you don’t get any 10^100 differ­ences in ex­pected util­ity of this sort, with­out an over­whelming amount of ev­i­dence to in­di­cate that A has 10^100+ times the im­pact on X as op­tion B or C (or your prior over other and un­known al­ter­na­tives you may find later).

• Carl_Shulman 4 Nov 2018 4:06 UTC

  1 point

  in reply to: stijnbruers's comment on: Re­duc­ing ex­is­ten­tial risks or wild an­i­mal suffer­ing?

  A strictly pos­i­tive crit­i­cal level that is low enough such that it would not re­sult in the choice for that counter-in­tu­itive situ­a­tion, is still posiible.

  As a mat­ter of math­e­mat­ics this ap­pears im­pos­si­ble. For any crit­i­cal level c that you pick where c>0, there is some level of pos­i­tive welfare w where c>w>0, with rel­a­tive util­ity u, 0>u, u=c-w.

  There will then be some quan­tity of ex­pected nega­tive util­ity and rel­a­tive util­ity peo­ple with rel­a­tive util­ity be­tween 0 and u that vari­able CLU would pre­fer to the ex­is­tence of you with c and w. You can use gam­bles (with ar­bi­trar­ily di­visi­ble prob­a­bil­ities) or ag­gre­ga­tion across similar peo­ple to get ar­bi­trar­ily close to zero. So ei­ther c<=0 or CLU will recom­mend cre­ation of nega­tive util­ity and rel­a­tive util­ity peo­ple to pre­vent your ex­is­tence for some pos­i­tive welfare lev­els.