Some personal thoughts on EA and systemic change

DavidNash re­quested that I re­post my com­ment be­low, on what to make of dis­cus­sions about EA ne­glect­ing sys­temic change, as a top-level post. Th­ese are my off-the-cuff thoughts and no one else’s. In sum­mary (to be un­packed be­low):

  • Ac­tual EA is able to do as­sess­ments of sys­temic change in­ter­ven­tions in­clud­ing elec­toral poli­tics and policy change, and has done so a num­ber of times

  • The great ma­jor­ity of crit­ics of EA in­vok­ing sys­temic change fail to pre­sent the sim­ple sort of quan­ti­ta­tive anal­y­sis given above for the in­ter­ven­tions they claim ex­cel, and fre­quently when such anal­y­sis is done the in­ter­ven­tion does not look com­pet­i­tive by EA lights

  • Nonethe­less, my view is that his­tor­i­cal data do show that the most effi­cient poli­ti­cal/​ad­vo­cacy spend­ing, par­tic­u­larly aiming at can­di­dates and is­sues se­lected with an eye to global poverty or the long term, does have higher re­turns than GiveWell top char­i­ties (even ig­nor­ing non­hu­mans and fu­ture gen­er­a­tions or fu­ture tech­nolo­gies); one can con­nect the sys­temic change cri­tique as a po­si­tion in in­tra­mu­ral de­bates among EAs about the de­gree to which one should fo­cus on highly lin­ear, giv­ing as con­sump­tion, type interventions

  • EAs who are will­ing to con­sider riskier and less lin­ear in­ter­ven­tions are mostly already pur­su­ing fairly dra­matic sys­temic change, in ar­eas with bud­gets that are small rel­a­tive to poli­ti­cal spend­ing (un­like for­eign aid)

  • As fund­ing ex­pands in fo­cused EA pri­or­ity is­sues, even­tu­ally diminish­ing re­turns there will equal­ize with re­turns for broader poli­ti­cal spend­ing, and ac­tivity in the lat­ter area could in­crease enor­mously: since broad poli­ti­cal im­pact per dol­lar is flat­ter over a large range poli­ti­cal spend­ing should ei­ther be a very small or very large por­tion of EA activity


In full:

  • Ac­tual EA is able to do as­sess­ments of sys­temic change in­ter­ven­tions in­clud­ing elec­toral poli­tics and policy change, and has done so a num­ber of times

    • Em­piri­cal data on the im­pact of votes, the effec­tive­ness of lob­by­ing and cam­paign spend­ing work out with­out any prob­lems of fancy de­ci­sion the­ory or in­creas­ing marginal returns

      • E.g. An­drew Gel­man’s data on US Pres­i­den­tial elec­tions shows that given pol­ling and fore­cast­ing un­cer­tainty a marginal vote in a swing state av­er­age some­thing like a 1 in 10 mil­lion chance of swing­ing an elec­tion over mul­ti­ple elec­tions (and one can save to make cam­paign contributions

      • 80,000 Hours has a page (there have been a num­ber of other such posts and dis­cus­sion, note that ‘worth vot­ing’ and ‘worth buy­ing a vote through cam­paign spend­ing or GOTV’ are two quite differ­ent thresh­olds) dis­cussing this data and ap­proaches to valu­ing differ­ences in poli­ti­cal out­comes be­tween can­di­dates; these sug­gest that a swing state vote might be worth tens of thou­sands of dol­lars of in­come to rich coun­try citizens

        • But if one thinks that char­i­ties like AMF do 100x or more good per dol­lar by sav­ing the lives of the global poor so cheaply, then these are com­pat­i­ble with a vote be­ing worth only a few hun­dred dollars

        • If one thinks that some other in­ter­ven­tions, such as gene drives for malaria erad­i­ca­tion, an­i­mal ad­vo­cacy, or ex­is­ten­tial risk in­ter­ven­tions are much more cost-effec­tive than AMF, that would lower the value fur­ther ex­cept in­so­far as one could iden­tify strong vari­a­tion in more highly-val­ued effects

      • Ex­per­i­men­tal data on the effects of cam­paign con­tri­bu­tions sug­gest a cost of a few hun­dred dol­lars per marginal vote (see, e.g. Ger­ber’s work on GOTV ex­per­i­ments)

      • Pre­dic­tion mar­kets and pol­ling mod­els give a good ba­sis for as­sess­ing the chance of billions of dol­lars of cam­paign funds swing­ing an election

      • If there are in­creas­ing re­turns to scale from large-scale spend­ing, small donors can con­vert their funds into a small chance of huge funds, e.g. us­ing a lot­tery, or more effi­ciently (more than 11,000 chance of more than 1000x funds) through mak­ing long­shot bets in fi­nan­cial mar­kets, so IMR are never a bar to ac­tion (also see the donor lot­tery)

      • The main thing needed to im­prove pre­ci­sion for such es­ti­ma­tion of elec­toral poli­tics spend­ing is care­fully cat­a­loging and valu­ing differ­ent chan­nels of im­pact (cost per vote and elec­toral im­pact per vote are well-un­der­stood)

        • More broadly there are also likely higher re­turns than cam­paign spend­ing in some ar­eas such as think tanks, lob­by­ing, and grass­roots move­ment-build­ing; bal­lot ini­ti­a­tive cam­paign spend­ing is one ex­am­ple that seems like it may have bet­ter re­turns than spend­ing on can­di­dates (and EAs have sup­ported sev­eral bal­lot ini­ti­a­tives fi­nan­cially, such as restora­tion of vot­ing rights to con­victs in Florida, cage bans, and in­creased for­eign spend­ing)

    • A re­cent blog post by the Open Philan­thropy Pro­ject de­scribes their cost-effec­tive­ness es­ti­mates from policy search in hu­man-ori­ented US do­mes­tic policy, in­clud­ing crim­i­nal jus­tice re­form, hous­ing re­form, and others

      • It states that thus far even ex ante es­ti­mates of effect there seem to have only rarely out­performed GiveWell style charities

      • How­ever it says: “One hy­poth­e­sis we’re in­ter­ested in ex­plor­ing is the idea of com­bin­ing mul­ti­ple sources of lev­er­age for philan­thropic im­pact (e.g., ad­vo­cacy, sci­en­tific re­search, helping the global poor) to get more hu­man­i­tar­ian im­pact per dol­lar (for in­stance via ad­vo­cacy around sci­en­tific re­search fund­ing or poli­cies, or sci­en­tific re­search around global health in­ter­ven­tions, or policy around global health and de­vel­op­ment). Ad­di­tion­ally, on the ad­vo­cacy side, we’re in­ter­ested in ex­plor­ing op­por­tu­ni­ties out­side the U.S.; we ini­tially fo­cused on U.S. policy for epistemic rather than moral rea­sons, and ex­pect most of the most promis­ing op­por­tu­ni­ties to be el­se­where.

    • Let’s Fund’s fundrais­ing for cli­mate policy work similarly made an at­tempt to es­ti­mate the im­pacts of their pro­posed in­ter­ven­tion in this sort of fash­ion; with­out en­dors­ing all the de­tails of their anal­y­sis, I think it is an ex­am­ple of EA method­olo­gies be­ing quite ca­pa­ble of mod­el­ing sys­temic interventions

    • An­i­mal ad­vo­cates in EA have ob­vi­ously pur­sued cor­po­rate cam­paigns and bal­lot ini­ti­a­tives which look like sys­temic change to me, in­clud­ing quan­ti­ta­tive es­ti­mates of the im­pact of the changes and the effects of the campaigns

  • The great ma­jor­ity of crit­ics of EA in­vok­ing sys­temic change fail to pre­sent the sim­ple sort of quan­ti­ta­tive anal­y­sis given above for the in­ter­ven­tions they claim ex­cel, and fre­quently when such anal­y­sis is done the in­ter­ven­tion does not look com­pet­i­tive by EA lights

    • A com­mon rea­son for this is EAs tak­ing into ac­count the welfare of for­eign­ers, non­hu­man an­i­mals and fu­ture gen­er­a­tions; crit­ics may pro­pose to get lev­er­age by work­ing through the poli­ti­cal sys­tem but give up on lev­er­age from con­cern for ne­glected benefi­cia­ries, and in other cases the com­pe­ti­tion is in­ter­ven­tions that get lev­er­age from ad­vo­cacy or sci­ence com­bined with a fo­cus on ne­glected beneficiaries

    • Some­times sys­temic change cri­tiques come from a Marx­ist per­spec­tive that as­sumes Marx­ist rev­olu­tion will pro­duce a utopia, whereas em­piri­cally such rev­olu­tion has been re­spon­si­ble for im­pov­er­ish­ing billions of peo­ple, mass kil­ling, the Cold War, (with risk of nu­clear war) and in­creased ten­sions be­tween China and democ­ra­cies, cre­at­ing large ob­ject-level dis­agree­ments with many EAs who want to ac­cu­rately fore­cast the re­sults of poli­ti­cal action

  • Nonethe­less, my view is that his­tor­i­cal data do show that the most effi­cient poli­ti­cal/​ad­vo­cacy spend­ing, par­tic­u­larly aiming at can­di­dates and is­sues se­lected with an eye to global poverty or the long term, does have higher re­turns than GiveWell top char­i­ties (even ig­nor­ing non­hu­mans and fu­ture gen­er­a­tions or fu­ture tech­nolo­gies); one can con­nect the sys­temic change cri­tique as a po­si­tion in in­tra­mu­ral de­bates among EAs about the de­gree to which one should fo­cus on highly lin­ear, giv­ing as con­sump­tion, type in­ter­ven­tions

    • E.g. I would rather see $1000 go to some­thing like the Cen­ter for Global Devel­op­ment, Tar­get Malaria’s gene drive effort, or the Swiss effec­tive for­eign aid bal­lot ini­ti­a­tive than the Against Malaria Foundation

    • I do think it is true that well-tar­geted elec­toral poli­tics spend­ing has higher re­turns than AMF, be­cause of the im­pacts of elec­tions on things such as sci­ence, for­eign aid, great power war, AI policy, etc, pro­vided that one ac­tu­ally di­rects one’s efforts based on the ne­glected considerations

  • EAs who are will­ing to con­sider riskier and less lin­ear in­ter­ven­tions are mostly already pur­su­ing fairly dra­matic sys­temic change, in ar­eas with bud­gets that are small rel­a­tive to poli­ti­cal spend­ing (un­like for­eign aid)

    • Global catas­trophic risks work is fo­cused on re­search and ad­vo­cacy to shift the di­rec­tion of so­ciety as a whole on crit­i­cal is­sues, and the col­lapse of hu­man civ­i­liza­tion or its re­place­ment by an un­de­sir­able suc­ces­sor would cer­tainly be a sys­temic change

    • As men­tioned pre­vi­ously, short-term an­i­mal EA work is over­whelm­ingly fo­cused on sys­temic changes, through chang­ing norms and laws, or pro­duc­ing tech­nolo­gies that would re­place and elimi­nate the fac­tory farm­ing system

    • A num­ber of EA global poverty fo­cused donors do give to or­ga­ni­za­tions like CGD, meta in­ter­ven­tions to grow the EA move­ment (which can even­tu­ally be cashed in for larger sys­temic change), and groups like GiveWell or the Poverty Ac­tion Lab

      • Although there is a rel­a­tive gap in longter­mist and high-risk global poverty work com­pared to other cause ar­eas, that does make sense in terms of ceiling effects, ar­gu­ments for the im­por­tance of tra­jec­tory changes from a longter­mist per­spec­tive, and the role of GiveWell as a re­spected char­ity eval­u­a­tor pro­vid­ing a ser­vice lack­ing for other ar­eas

    • Is­sue-spe­cific fo­cus in ad­vo­cacy makes sense for these ar­eas given the view that they are much more im­por­tant than the av­er­age is­sue and with very low spending

  • As fund­ing ex­pands in fo­cused EA pri­or­ity is­sues, even­tu­ally diminish­ing re­turns there will equal­ize with re­turns for broader poli­ti­cal spend­ing, and ac­tivity in the lat­ter area could in­crease enor­mously: since broad poli­ti­cal im­pact per dol­lar is flat­ter over a large range poli­ti­cal spend­ing should ei­ther be a very small or very large por­tion of EA activity

    • Essen­tially, the cost per vote achieved through things like cam­paign spend­ing is cur­rently set by the broader poli­ti­cal cul­ture and has the ca­pac­ity to ab­sorb billions of dol­lars at similar cost-effec­tive­ness to the cur­rent level, so it should ei­ther be the case that EA funds very lit­tle of it or enor­mous amounts of it

      • There is a com­pli­ca­tion in that close elec­tions or other op­por­tu­ni­ties can vary the effec­tive­ness of poli­ti­cal spend­ing over time, which would sug­gest sav­ing most funds for those

    • The con­sid­er­a­tions are similar to GiveDirectly: since cash trans­fers could ab­sorb all EA funds many times over at similar cost-effec­tive­ness (with con­tinued rapid scal­ing), it should take in ei­ther very lit­tle or al­most all EA fund­ing; in a forced choice it should ei­ther be the case that most fund­ing goes to cash trans­fers, whereas for other in­ter­ven­tions with diminish­ing re­turns on the rele­vant scale as mixed port­fo­lio will yield more impact

    • For now ar­eas like an­i­mal ad­vo­cacy and AI safety with bud­gets of only tens of mil­lions of dol­lars are very small rel­a­tive to poli­ti­cal spend­ing, and the im­pact of the fo­cused work (in­clud­ing rele­vant move­ment build­ing) makes more of a differ­ence to those ar­eas than a typ­i­cal differ­ence be­tween poli­ti­cal can­di­dates; but if billions of dol­lars were be­ing spent in those ar­eas it would seem that poli­ti­cal ac­tivity could be a com­pet­i­tive use (e.g. sup­port­ing pro-an­i­mal can­di­dates for office)