How can we best coordinate as a community?

I re­cently gave a talk at EAG:Bos­ton on this topic.

The video is now up.

Below is the blurb, and a rough script (it’ll be eas­ier to un­der­stand if you watch the video with boosted speed).

All these ideas are pretty spec­u­la­tive, and we’re work­ing on some more in-depth ar­ti­cles on this topic, so I’d be keen to get feed­back.

* * *

A com­mon ob­jec­tion to effec­tive al­tru­ism is that it en­courages an overly “in­di­vi­d­ual” way of think­ing, which re­duces our im­pact. Ben will ar­gue that, at least in a sense, that’s true.

When you’re part of a com­mu­nity, which ca­reers, char­i­ties and ac­tions are high­est-im­pact changes. An overly nar­row, in­di­vi­d­ual anal­y­sis, which doesn’t take ac­count of how the rest of the com­mu­nity will re­spond to your ac­tions, can lead you to have less im­pact than you could. Ben will sug­gest some bet­ter op­tions and rules of thumb for work­ing to­gether.

* * *

In­tro­duc­tion—why work to­gether?

  • Here’s one of the most pow­er­ful ways to have more im­pact: join a com­mu­nity.

  • Why’s that?

  • Well, one rea­son is it’s mo­ti­vat­ing—be­ing around other peo­ple who want to help oth­ers changes your so­cial norms, and makes you more mo­ti­vated.

  • It’s like net­work­ing on steroids—once one per­son vouches for you, they can in­tro­duce you to ev­ery­one else.

  • And also, what I want to talk about to­day, you can trade and co­or­di­nate.

    • Let’s sup­pose I want to build and sell a piece of soft­ware. One ap­proach would be to learn all the skills needed my­self—de­sign, en­g­ineer­ing, mar­ket­ing and so on.

    • A much bet­ter ap­proach is to form a team who are skil­led in each area, and then build it to­gether. Although I’ll have to share the gains with the other peo­ple, the size of the gains will be much larger, so we’ll all win.

      • [di­a­gram]

    • One thing that’s go­ing on here is spe­cial­i­sa­tion—each per­son can fo­cus on a spe­cific skill, which lets them be more effec­tive.

    • Another thing is that the team can also share fixed costs (same office, same com­pany reg­is­tra­tion, same op­er­a­tional pro­ce­dures etc.), let­ting up achieve economies of scale.

    • In to­tal, we get what’s called the “gains from trade”.

    • An im­por­tant thing about trade is that you can do it with peo­ple who don’t es­pe­cially share your val­ues, and both gain.

      • Sup­pose, hy­po­thet­i­cally, one group runs an an­i­mal char­ity, and they don’t think global poverty is that high im­pact.

      • Another group runs a global poverty char­ity, and they don’t think fac­tory farm­ing is that high im­pact.

      • But imag­ine both groups know some donors who might be in­ter­ested in the other cause. They can make mu­tual in­tro­duc­tions. Mak­ing an in­tro­duc­tion isn’t much cost, but could be a huge benefit to the other group. So, if both groups trade, they both gain.

    • This is why 100 peo­ple work­ing to­gether have the po­ten­tial to have far more im­pact than 100 peo­ple do­ing what in­di­vi­d­u­ally seems best.

Why the EA com­mu­nity?

  • What I’ve said so far ap­plies to any com­mu­nity, and there are lots of great com­mu­ni­ties out there.

  • But I know many peo­ple who think that get­ting in­volved in the EA com­mu­nity has been an es­pe­cially big boost to their im­pact.

  • Why’s that?

  • Well, as we’ve shown you can trade with in gen­eral to have a greater im­pact, even when you don’t es­pe­cially share their val­ues.

  • But if you do share val­ues you don’t even need to trade.

  • What do I mean?

  • Well, if I help some­one else in the EA com­mu­nity have more im­pact, then I’ve also had more im­pact, so we both achieve our goals.

  • This means I don’t need to worry about get­ting a favour back from the other per­son to break even. Just helping them is already valuable.

  • This un­leashes far more op­por­tu­ni­ties to work to­gether, that just wouldn’t be effi­cient in a com­mu­nity where peo­ple don’t share my aims as much.

    • (Tech­ni­cally speak­ing, trans­ac­tion costs and prin­ci­pal-agent prob­lems are dra­mat­i­cally re­duced.)

  • We don’t nor­mally think about it like this, but earn­ing to give can ac­tu­ally an ex­am­ple of this kind of co­or­di­na­tion.

    • In the early days of 80k, we needed one per­son to run the org and we needed fund­ing. Me and an­other guy called Matt con­sid­ered the po­si­tion. We re­al­ised Matt had higher earn­ing po­ten­tial than me, while I was bet­ter suited to run­ning 80,000 Hours, hope­fully.

    • There were other fac­tors at play, but in part, this is why I be­came the CEO, and Matt earned to give and be­came one of our largest donors, as well as seed fund­ing sev­eral other orgs.

    • The al­ter­na­tive would have been to for us both to earn to give, in which case, no 80k; or for us both to work at 80k, in which case it would have taken us much longer to fundraise (and the other orgs wouldn’t have benefited).

    • With the com­mu­nity as a whole, some peo­ple are like Matt, rel­a­tively bet­ter suited to earn­ing money, and oth­ers to run­ning non-prof­its. We can achieve more if the peo­ple best suited to earn­ing money earn to give and fund ev­ery­one else.

  • In sum, by work­ing to­gether effec­tively, we have the po­ten­tial to achieve far more.

How can we work to­gether bet­ter?

  • How­ever, I don’t think we work to­gether as a com­mu­nity as well as we could.

  • Effec­tive al­tru­ism en­courages us to ask which in­di­vi­d­ual ac­tions lead to the most im­pact.

  • Some cri­tiques of the com­mu­nity have sug­gested this ques­tion could bias us so that we don’t ac­tu­ally do the high­est im­pact things.

  • Here is per­haps the most well-known crit­i­cism in this vein, in the Lon­don Re­view of Books.

    • [read out]

    • I agree with Jeff McMa­han’s re­sponse to this ar­ti­cle. He points out that ul­ti­mately what we can con­trol are our in­di­vi­d­ual ac­tions, so they are the most rele­vant. We can’t di­rect the whole of so­ciety.

  • How­ever, I think there’s also truth in Amia’s view: when we think about what’s best from an in­di­vi­d­ual per­spec­tive, we have to be wary of the bi­ases of this way of think­ing, oth­er­wise we might not ac­tu­ally find the high­est-im­pact in­di­vi­d­ual ac­tions.

  • I of­ten see peo­ple in the com­mu­nity tak­ing what I call a “nar­row, sin­gle player” per­spec­tive to figur­ing out what’s best—not fully fac­tor­ing in the rele­vant coun­ter­fac­tu­als and how the com­mu­nity will ad­just to their ac­tions.

  • This might have worked be­fore we had a com­mu­nity, but these days it doesn’t.

  • In­stead we need to move to what I call a “mul­ti­player per­spec­tive”,

  • In par­tic­u­lar:

    • First, we need to take a differ­ent ap­proach to choos­ing be­tween our op­tions.

    • Se­cond, new op­tions be­come promis­ing.

    • I’ll cover both in turn.

1) How to choose be­tween our options

    • Let’s con­sider this ques­tion: should I take a job at a char­ity in the com­mu­nity? Like GiveWell, or AMF or CEA.

    • Amy is con­sid­er­ing work­ing at a char­ity. What’s her im­pact?

    • The naive view is that the job is high im­pact, so if I take it, I’ll have a big im­pact.

    • But then you hear about EA, and some­one points out: if you don’t take the job, some­one else will take it, so ac­tu­ally your im­pact is small. The job is only worth tak­ing if you’d be much bet­ter than the per­son who re­places you.

    • We call this anal­y­sis “sim­ple re­place­abil­ity” and it’s an ex­am­ple of sin­gle player think­ing.

    • This leads to lots of peo­ple not want­ing to do di­rect work, and think­ing it’s bet­ter to earn to give in­stead.

    • But this is wrong. And I apol­o­gise, be­cause it’s partly our fault for talk­ing loosely about the sim­ple re­place­abil­ity view in the past. But to­day I want to help stamp it out.

  • The first prob­lem is that you won’t always be re­placed. There’s a chance that the char­ity just won’t hire any­one oth­er­wise.

    • In fact this seems to be of­ten the case. When we talk to the or­gani­sa­tions, there are roles they’ve been try­ing to fill for a while, but haven’t been able to.

    • One rea­son this is that, be­cause there are donors with money on the sidelines, if the or­gani­sa­tions were able to find some­one with a good level of fit, they could fundraise enough money to pay for their salaries.

    • This means the or­gani­sa­tions do what’s called “thresh­old hiring”—hire any­one above that level of fit.

    • And there ways you can end up be­ing not re­place­able, such as sup­ply-de­mand effects, which we cover el­se­where.

    • Either way, the sim­ple anal­y­sis of re­place­abil­ity ends up un­der­es­ti­mat­ing the im­pact.

    • In­stead, you can end up be­ing pretty valuable to the or­gani­sa­tion you work at.

    • One way to es­ti­mate the effect is to ask the org how much you’d need to donate to them to be in­differ­ent be­tween you tak­ing the job and them get­ting dona­tions. This helps to mea­sure the size of the benefit to the or­gani­sa­tion. [show Q on slide]

    • We ac­tu­ally did this with 12 or­gani­sa­tions in the com­mu­nity, and for their most re­cent hire, they gave figures of

      • [slide: av­er­age of $126,000 – $505,000 and a me­dian of $77,000 – $307,000 per year.]

      • The or­gani­sa­tions may well be bi­ased up­wards, but since it’s sig­nifi­cantly more than most peo­ple who could work at EA orgs could donate, it at least sug­gests they’re hav­ing much more im­pact than they would through earn­ing to give.

      • We also asked the orgs sim­ply how fund­ing vs tal­ent con­strained they are, and you can see a clear bias to­wards tal­ent con­straint rather than fund­ing con­straint.

        • In­ter­est­ingly, the an­i­mal fo­cused orgs were more fund­ing con­strained, so if you re­move them, the figures were higher.

  • So, the first prob­lem with the sim­ple anal­y­sis of re­place­abil­ity is that you might not ac­tu­ally be re­placed. The sec­ond prob­lem is where the com­mu­nity comes in.

  • Even if you take the job, and some­one else would have taken it any­way, that per­son is freed up to go and do some­thing else that’s valuable. So there’s a spillover benefit to the rest of the com­mu­nity.

  • If you were con­sid­er­ing a job that would be filled by some­one who didn’t care about so­cial im­pact oth­er­wise, like a ran­dom job in the cor­po­rate world, then it’s fine, you can mostly ig­nore these spillovers. The “sin­gle player” view would be fine.

  • But in the cur­rent com­mu­nity, how­ever, that per­son will prob­a­bly go and do some­thing else you think is high-im­pact, so it’s a sig­nifi­cant benefit.

    • This is not hy­po­thet­i­cal—I’ve seen real cases where some­one didn’t take a job be­cause they thought they’d be re­place­able, which then meant some­one else had to be taken from a high-im­pact role.

  • So, there’s a sec­ond benefit that’s ig­nored by the sim­ple anal­y­sis of re­place­abil­ity.

    • And, for both rea­sons, the im­pact of tak­ing the job is higher.

  • How valuable? I think this is still an un­solved prob­lem, but here is a sketch of our think­ing.

  • Ba­si­cally, you cause a chain re­ac­tion of re­place­ments. The chain can end if ei­ther (i) it hits a thresh­old job where the per­son isn’t re­place­able, or you (ii) hit the marginal op­por­tu­nity in the com­mu­nity—the best role that has not yet been taken.

    • So, at the worst, you’re adding some­one to the marginal po­si­tion in the com­mu­nity, which is still a sig­nifi­cant im­pact.

    • Plus, by trig­ger­ing the chain, you’re hope­fully helping peo­ple switch into roles that bet­ter play to their rel­a­tive strengths. So we also get to a more effi­cient al­lo­ca­tion.

  • Zoom­ing out, rather than iden­tify the sin­gle high­est-im­pact job in gen­eral, and try to take that, in­stead we have sev­eral thou­sand roles to fill, and want to achieve the op­ti­mal al­lo­ca­tion over them. The ques­tion is: what can you do to take the com­mu­nity to­wards the op­ti­mal al­lo­ca­tion over the best roles?

  • I think the key con­cept here is your com­par­a­tive ad­van­tage, com­pared to the rest of the com­mu­nity.

  • Com­par­a­tive ad­van­tage can be a lit­tle coun­ter­in­tu­itive, and is not always the same as per­sonal fit, so let’s ex­plore a lit­tle more.

  • Let’s imag­ine there are two roles that need filling, re­search and out­reach; and two peo­ple.

    • [di­a­gram]

    • Char­lie is 1 at out­reach and 2 at re­search.

    • Dora is 2 at out­reach and 10 at re­search.

    • What’s best?

    • It de­pends on ex­actly how we in­ter­pret these num­bers, but prob­a­bly Char­lize should do out­reach and Dex­ter should do re­search, be­cause 1 + 10 > 2 + 2

  • This is sur­pris­ing be­cause, in a sense, Char­lie is ac­tu­ally worse at out­reach than Dora, and worse at out­reach than re­search, so he in no sense has an ab­solute ad­van­tage in out­reach, but it turns out he does have a com­par­a­tive ad­van­tage in it. This is be­cause he’s rel­a­tively less bad at out­reach com­pared to Dora.

  • I have a sus­pi­cion this might be a real ex­am­ple.

    • [slide]

    • Lots of peo­ple in the com­mu­nity are good at an­a­lyt­i­cal things com­pared to peo­ple in gen­eral, so they figure they should do re­search.

    • But this doesn’t fol­low. What ac­tu­ally mat­ters is how good they are at re­search rel­a­tive to oth­ers in the com­mu­nity.

    • If we have lots of an­a­lyt­i­cal peo­ple and few out­reach peo­ple, then even though you’re good at re­search com­pared to peo­ple in gen­eral, you might have a com­par­a­tive ad­van­tage in out­reach.

    • Some­thing similar seems true with op­er­a­tions roles.

    • It may also be true with earn­ing to give. Peo­ple of­ten rea­son that be­cause they have high earn­ing po­ten­tial, they should earn to give. But this doesn’t quite fol­low. What ac­tu­ally mat­ters is their earn­ing po­ten­tial rel­a­tive to oth­ers who could do di­rect work. If the other di­rect work peo­ple also have high earn­ing po­ten­tial, then they might have a com­par­a­tive ad­van­tage in di­rect work.

    • Un­less you’re chuck nor­ris, who has a com­par­a­tive ad­van­tage in ev­ery­thing.

  • How can you figure out your com­par­a­tive ad­van­tage in real cases? Ask peo­ple in charge of hiring at the orgs—what would the next best hire do any­way, and what would their dona­tion po­ten­tial be com­pared to yours?

  • So, let’s sum up. How to work out your im­pact by tak­ing a job in the com­mu­nity?

    • First, you prob­a­bly cause some boost to the org it­self, and you’re not fully re­place­able. You can roughly es­ti­mate the size of this boost by ask­ing the org to make trade­offs, or mak­ing your own es­ti­mates.

    • Se­cond, you cause some spillover benefit to the rest of the com­mu­nity, be­cause you free up some­one else to go and work el­se­where.

      • The key ques­tion then is whether the role is the above the bar for the com­mu­nity as a whole, and whether it plays to your com­par­a­tive ad­van­tage com­pared to the other peo­ple who might take the role?

      • Are you get­ting the com­mu­nity to­wards a bet­ter over­all al­lo­ca­tion?

  • I think ba­si­cally the same anal­y­sis ap­plies to donat­ing as well.

    • Some­times peo­ple don’t want to donate be­cause they think some­one else will fund the char­ity any­way.

    • But they ig­nore the fact that if even if their dona­tion were 100% re­placed, at worst they’d be free­ing up an­other donor, sooner, to go and donate to some­thing else.

    • There’s lots more to say here, I go into a bit more de­tail in this ar­ti­cle.

2) The new op­tions that come available

  • Be­ing part of a com­mu­nity also changes your ca­reer by open­ing up new op­tions that wouldn’t be on the table if you were just in a sin­gle player game.

  • The EA mind­set, with a fo­cus on in­di­vi­d­ual ac­tions, can lead us to ne­glect paths to im­pact through helping oth­ers achieve more, be­cause the im­pact is less salient. How­ever, now that there are 1000s of other effec­tive al­tru­ists, the high­est im­pact op­tion for you could well be one that in­volves “boost­ing” oth­ers in the com­mu­nity.

    • Here are five ex­am­ples. They aren’t ex­haus­tive or ex­clu­sive, but are just some ideas.

    • First, five minute favours.

  • We all have differ­ent strengths and weak­nesses, knowl­edge and re­sources. Now the com­mu­nity is so big, there are prob­a­bly lots of small ways we can help oth­ers in the com­mu­nity have far more im­pact, that would be very lit­tle cost to our­selves- 5 minute favours. (a term I took from Adam Grant). Th­ese are re­ally worth look­ing for.

      • Do you know a job that needs filling? There’s a good chance some­one in this room would be a good can­di­date. If you could in­tro­duce them to the job, it might only take you un­der an hour, but it would have a ma­jor im­pact for years.

      • There’s prob­a­bly some­one in this room who has a prob­lem you could eas­ily solve, be­cause you’ve already solved it, or you know a book that con­tains the an­swer, and so on.

    • Here’s a sec­ond, more in­volved ex­am­ple of helping oth­ers be more effec­tive: Oper­a­tions roles in gen­eral.

    • Kyle went to Oxford, and ended up be­com­ing Nick Bostrom’s as­sis­tant. He rea­soned that if he could save Bostrom time, then he could let more re­search and out­reach get done. I think this could be re­ally high im­pact.

    • Peo­ple feel like these roles are eas­ily re­place­able by peo­ple out­side the com­mu­nity, but be­cause you have to make lots of lit­tle de­ci­sions that re­quire a good un­der­stand­ing of the aims of the or­gani­sa­tion, they’re ac­tu­ally of­ten very hard to out­source.

    • A third cat­e­gory is that build­ing com­mu­nity in­fras­truc­ture be­comes much more valuable.

      • Hav­ing a job board isn’t re­ally needed when there are only a few hun­dred peo­ple in the com­mu­nity, but now there are 1000s it can play a use­ful role, so we re­cently made one.

        • [show screen­shot]

      • In­fras­truc­ture is any­thing that helps to make the com­mu­nity co­or­di­nate more effi­ciently, such as this event, or set­ting up good norms, that make it eas­ier to work to­gether….like stat­ing the ev­i­dence for your views, or be­ing nice.

      • If you can help 1000 peo­ple be 1% more effec­tive, then that’s like hav­ing the im­pact of 10 peo­ple.

      • On the other hand, if you do some­thing de­struc­tive, then it ru­ins it for ev­ery­one else

    • A fourth cat­e­gory, is knowl­edge shar­ing.

    • The more peo­ple there are in the com­mu­nity, the more valuable it is to do re­search into what the com­mu­nity should do and share it, be­cause there are more peo­ple who might act on the find­ings.

      • One ex­am­ple is writ­ing up re­ports on ar­eas we have spe­cial knowl­edge of

        • [give ex­am­ples from EA fo­rum]

      • This can mean it’s some­times worth go­ing and learn­ing about ar­eas that don’t seem like the high­est pri­or­ity but might turn out to be. In a smaller com­mu­nity, this ex­plo­ra­tion wouldn’t be worth the time, but as we be­come larger, it is.

  • A fifth ex­am­ple, spe­cial­i­sa­tion be­comes more worth do­ing.

      • If the com­mu­nity were just a cou­ple of peo­ple, we’d all need to be­come gen­er­al­ists. But in a com­mu­nity of, say, 1000 peo­ple, we can all be­come ex­perts in our in­di­vi­d­ual ar­eas, and be more than 1000 times as pro­duc­tive as an in­di­vi­d­ual. This is just di­vi­sion of labour like we men­tioned at the very start, with the soft­ware ex­am­ple.

      • For in­stance, Dr. Greg Lewis did the re­search on 80,000 Hours into how many lives doc­tors save, and this con­vinced him that he wouldn’t have much so­cial im­pact as a doc­tor through clini­cal prac­tice.

      • In­stead he de­cided to do a mas­ters in pub­lic health.

      • Part of the rea­son was be­cause it’s an im­por­tant area for the com­mu­nity, es­pe­cially around pan­demics, but there’s a lack of peo­ple with the skil­lset.

      • Greg ac­tu­ally thinks AI risk might be higher pri­or­ity in gen­eral, but as a doc­tor, he has a com­par­a­tive ad­van­tage in pub­lic health.

      • Right now, I, and many oth­ers, think that one of the great­est weak­nesses of the com­mu­nity is a lack of spe­cial­ist ex­per­tise. We’re gen­er­ally pretty young and in­ex­pe­rienced.

      • Some par­tic­u­lar gaps in­clude the fol­low­ing, which I’m not go­ing to read out:

        • Policy ex­perts—go and work in poli­tics or at a think tank

        • Bio­eng­ineer­ing PhDs

        • Ma­chine learn­ing PhDs

        • Eco­nomics PhDs

        • Other un­der-rep­re­sented ar­eas—his­tory, an­thro­pol­ogy,

        • En­trepreneurial man­agers and op­er­a­tional people

        • Mar­ket­ing and out­reach experts

Summary

  • When choos­ing whether to take a job, or donate some­where, don’t as­sume you’re re­place­able.

    • Rather, ask the or­gani­sa­tion, or oth­ers who are aware of the al­ter­na­tive op­tions, about your rel­a­tive strengths and weak­nesses.

    • You might also trig­ger a chain of peo­ple go­ing into other roles. Con­sider whether the role plays to your com­par­a­tive ad­van­tage.

  • Look for ways to boost the im­pact of oth­ers in the com­mu­nity.

    • 5 minute favours

    • Oper­a­tions roles

    • Com­mu­nity infrastructure

    • Shar­ing knowl­edge.

    • Spe­cial­i­sa­tion.

End

  • We still have a lot to learn about how to best work to­gether, and there’s a lot more we could do. But I re­ally be­lieve that if we do work to­gether effec­tively, then, in our life­times, the com­mu­nity can make ma­jor progress on re­duc­ing catas­trophic risks, elimi­nat­ing fac­tory farm­ing, end­ing global poverty, and many other is­sues.