Racial Demographics at Longtermist Organizations

Main Findings

It’s widely un­der­stood that the Longter­mist ecosys­tem is quite racially ho­moge­nous. This anal­y­sis seeks to quan­tify and elab­o­rate on that ob­ser­va­tion, to en­able var­i­ous stake­hold­ers to make more in­formed de­ci­sions.

My re­view of 459 unique in­di­vi­d­u­als listed on the team pages of 21 Longter­mist or­ga­ni­za­tions pro­duced the fol­low­ing find­ings:

  • Peo­ple of Color (POC) make up 13-14% of staff at the 21 Longter­mist or­ga­ni­za­tions in my sam­ple. This is less di­verse, and in many cases much less di­verse, than the com­mu­ni­ties from which these or­ga­ni­za­tions tend to draw from (such as top US Com­puter Science pro­grams where POC of­ten make up a ma­jor­ity of stu­dents).

  • At about half the or­ga­ni­za­tions in my sam­ple (10 out of 21), the core staff is en­tirely white. At ev­ery or­ga­ni­za­tion ex­cept Open Philan­thropy’s AI Fel­low­ship pro­gram, POC ac­count for <30% of staff.

  • Open Philan­thropy’s AI Fel­low­ship pro­gram was an ex­treme out­lier, with 9 out of 15 fel­lows (60%) be­ing POC. This fel­low­ship ac­counted for just 6% of the core staff in my sam­ple, but 26% of the POC. One of the pro­gram’s ex­plicit goals is to im­prove di­ver­sity in the ecosys­tem, sug­gest­ing this sort of in­ten­tional effort can suc­cess­fully bring tal­ented POC into the Longter­mist com­mu­nity.

  • Or­ga­ni­za­tions/​pro­grams that are based at uni­ver­si­ties are much more racially di­verse than other types of Longter­mist or­ga­ni­za­tions (typ­i­cally in­de­pen­dent char­i­ties). At “Univer­sity” or­ga­ni­za­tions, POC make up 20% of core staff po­si­tions, com­pared to only 3% at “Non-Univer­sity” or­ga­ni­za­tions.

  • POC ap­pear to be par­tic­u­larly un­der­rep­re­sented in the most in­fluen­tial roles at Longter­mist or­ga­ni­za­tions. Only 4% of lead­er­ship po­si­tions are held by POC. How­ever, POC ac­count for 19-25% of sup­port staff.

  • Col­lec­tively, these find­ings sug­gest that the Longter­mist ecosys­tem is sig­nifi­cantly un­der­rep­re­sent­ing and un­der­uti­liz­ing POC. At the same time, Open Philan­thropy’s AI Fel­low­ship pro­gram sug­gests im­prove­ments are pos­si­ble if pri­ori­tized. Po­ten­tial strate­gies for im­prov­ing di­ver­sity are dis­cussed be­low.

Methodology

Sample

To con­duct this anal­y­sis, I looked at the “team” pages on the web­sites of 21 Longter­mist or­ga­ni­za­tions/​pro­grams, which en­com­passed 459 unique in­di­vi­d­u­als filling 558 po­si­tions (some peo­ple hold roles at mul­ti­ple or­ga­ni­za­tions). This data was col­lected in late Fe­bru­ary 2020.

I se­lected the or­ga­ni­za­tions by com­ing up with a list of promi­nent groups work­ing in the ecosys­tem, sup­ple­mented this ini­tial list with grantees of the Long-Term Fu­ture Fund, and made some ad­just­ments based on feed­back from peo­ple work­ing in the Longter­mist ecosys­tem (none of whom sug­gested adding other or­ga­ni­za­tions). This pro­cess re­quired some judg­ment calls, such as in­clud­ing CFAR (be­cause while it is not ex­plic­itly Longter­mist, it has re­ceived grants from the Long-term Fu­ture Fund and has trained many peo­ple in the Longter­mist space). Similarly, I chose to in­clude the Long-Term Fu­ture Fund it­self (be­cause de­spite be­ing a vol­un­teer-run effort rather than an or­ga­ni­za­tion, it’s role as fun­der makes it par­tic­u­larly in­fluen­tial).

In al­most ev­ery case, I in­cluded all team mem­bers listed. The only ex­cep­tion was Open Philan­thropy, where I in­cluded the lead­er­ship and peo­ple whose job ti­tles/​de­scrip­tions ex­plic­itly sug­gest they work full-time on Longter­mist causes. I did this be­cause Open Philan­thropy plays a crit­i­cal part in the Longter­mist ecosys­tem, but has a scope that ex­ceeds Longter­mism. This method­ol­ogy ex­cluded some Open Philan­thropy em­ploy­ees, in­clud­ing some POC, who work part-time on Longter­mist causes. (CEA and 80,000 Hours also do work be­sides Longter­mism, but I in­cluded all of their em­ploy­ees as it was not clear which of them do Longter­mist work.)

Ra­cial Classification

For each per­son in my sam­ple, I tried to de­ter­mine whether they are white or a per­son of color (POC) based on the pic­tures dis­played on the rele­vant team page (in some cases, I performed a quick google search to try to find ad­di­tional in­for­ma­tion). I in­cluded peo­ple I be­lieved to be mul­ti­ra­cial in the POC cat­e­gory. This is ad­mit­tedly an im­perfect ap­proach, and I ex­pect I mis­clas­sified some peo­ple. I sincerely apol­o­gize to any­one I mis­clas­sified. I be­lieve it is un­likely that this im­pre­ci­sion ma­te­ri­ally changes the find­ings of this anal­y­sis, as there gen­er­ally didn’t seem to be much am­bi­guity.

Note: While there are of course many di­men­sions of di­ver­sity, for pur­poses of brevity my use of that term through­out this anal­y­sis will re­fer to racial di­ver­sity un­less oth­er­wise stated.

Role Classification

Differ­ent or­ga­ni­za­tions have differ­ent crite­ria for list­ing peo­ple on their team pages. Some or­ga­ni­za­tions use loose crite­ria, and in­clude in­terns, vol­un­teers, and even past em­ploy­ees.

Us­ing job de­scrip­tions listed on the team pages, I dis­t­in­guished be­tween differ­ent types of team mem­bers in two ways. First, I tried to de­ter­mine whether or not each per­son was part of an or­ga­ni­za­tion’s “core staff”. I ex­cluded from the core staff des­ig­na­tion roles such as board mem­bers, free­lancers, in­terns, and re­search af­fili­ates. If some­one was listed on a team page in any ca­pac­ity, I in­clude them in the “all staff” cat­e­gory.

I also tried to cat­e­go­rize the type of work each per­son did, dis­t­in­guish­ing be­tween “Lead­er­ship” roles (e.g. CEO, COO, board mem­bers), “Pro­fes­sional Staff” roles (e.g. an­a­lysts and re­searchers), “Sup­port Staff” roles (e.g. op­er­a­tions and ad­minis­tra­tion), and “As­so­ci­ate” roles (e.g. alumni and stu­dents).

As an ex­am­ple of how these clas­sifi­ca­tions play out, a board mem­ber would be clas­sified as a lead­er­ship role, but would not be con­sid­ered a mem­ber of an or­ga­ni­za­tion’s “core staff”.

Shortcomings

My method­ol­ogy in­cludes sev­eral short­com­ings, in­clud­ing rely­ing on sub­jec­tive judge­ments in as­sess­ing peo­ples’ race, who con­sti­tutes an or­ga­ni­za­tion’s “core team”, and which peo­ple are in which type of roles.

To miti­gate these short­com­ings, I have tried to provide mul­ti­ple per­spec­tives when pos­si­ble. For in­stance, look­ing only at core team mem­bers re­lies on my as­sess­ment of who falls into that cat­e­gory. By con­trast, look­ing at ev­ery­one listed on an or­ga­ni­za­tion’s team page does not re­quire that as­sess­ment, but places equal weight on peo­ple who are es­sen­tial to an or­ga­ni­za­tion’s work and those who are some­what tan­gen­tial. Each ap­proach has its weak­nesses, but look­ing at both per­spec­tives helps miti­gate those weak­nesses. In the same vein, I sup­ple­ment the over­all num­bers I provide (which effec­tively weight or­ga­ni­za­tions by their staff sizes) with the av­er­age and me­dian figures across or­ga­ni­za­tions (which give or­ga­ni­za­tions equal weight).

Other weak­nesses are harder to miti­gate. My anal­y­sis re­lies on the team pages it is based upon be­ing (rea­son­ably) ac­cu­rate. It also re­lies on these team pages as they stood in Fe­bru­ary 2020, as it was not prac­ti­cal to con­tin­u­ously up­date my sam­ple. In some cases teams have shifted since then (e.g. I’m aware that 80,000 Hours has added a POC to their core team, and the Long-term Fu­ture Fund has re­placed the sole POC on its man­age­ment com­mit­tee).

Staff Demographics

Of the 459 unique in­di­vi­d­u­als in this sam­ple, 65 (14%) are POC. POC make up a slightly lower per­centage of to­tal roles (13%) and core staff po­si­tions (13%).

It’s not ob­vi­ous which refer­ence point(s) should be used to con­tex­tu­al­ize these figures. The fol­low­ing table lists a va­ri­ety of pos­si­bil­ities.

Sources: EA Com­mu­nity, UK stu­dents inc. Oxford, UK fac­ulty, US Doc­toral Univer­si­ties (over­all and high re­search), All US Doc­torates and APA mem­bers, Stan­ford un­der­grad, Stan­ford, Har­vard, US Fac­ulty, Philos­o­phy de­grees, CS De­grees, Har­vard CS, Stan­ford CS, Eco­nomics, Google, Face­book.

Longter­mist or­ga­ni­za­tions ap­pear less di­verse than even the most ho­moge­nous of these bench­marks: UK uni­ver­sity fac­ulty (15%), the mem­ber­ship of the Amer­i­can Philo­soph­i­cal As­so­ci­a­tion (17%), Oxford’s un­der­grad­u­ate pop­u­la­tion (18%), and the EA com­mu­nity (19.5%).

It’s plau­si­ble that Longter­mist di­ver­sity is very roughly on par with di­ver­sity at top UK philos­o­phy de­part­ments (while I could not find de­mo­graphic data for the lat­ter, it seems rea­son­able to ex­pect that philos­o­phy pro­grams are less di­verse than uni­ver­sity pop­u­la­tions as a whole in the UK, as is the case in the US). But while UK philoso­phers are cer­tainly well rep­re­sented at Longter­mist or­ga­ni­za­tions, it seems a stretch to ar­gue that they’re the ap­pro­pri­ate bench­mark for the ecosys­tem as a whole. Roughly a quar­ter of the or­ga­ni­za­tions in my sam­ple have an ex­plicit fo­cus on AI and are based in the Bay Area; for these or­ga­ni­za­tions one might rea­son­ably ex­pect half the staff or more to be POC. More­over, there’s sig­nifi­cant vari­a­tion in job func­tion both within and across or­ga­ni­za­tions in my sam­ple, which ar­gues for con­sid­er­ing a wider range of dis­ci­plines as a refer­ence point.

The sig­nifi­cant gap in racial di­ver­sity be­tween the US and the UK (which col­lec­tively are the home of all the or­ga­ni­za­tions in my sam­ple) ad­mit­tedly com­pli­cates the bench­mark­ing pro­cess; later, I take a closer look at the re­la­tion­ship be­tween ge­og­ra­phy and di­ver­sity.

De­mo­graphic differ­ences across organizations

Ra­cial di­ver­sity varies sig­nifi­cantly across the 21 Longter­mist or­ga­ni­za­tions in this sam­ple.

At about half the or­ga­ni­za­tions, core staff is en­tirely white. At some of the most di­verse or­ga­ni­za­tions, a quar­ter of core staff are POC. The Open Phil AI Fel­low­ship is a huge out­lier, with 60% of fel­lows be­ing POC (twice as much as any other or­ga­ni­za­tion.)

There ap­pears to be a strong pat­tern whereby uni­ver­sity-based re­search cen­ters/​pro­grams have much more di­verse teams than other types of Longter­mist or­ga­ni­za­tions (which are gen­er­ally stan­dalone reg­istered char­i­ties). At “Univer­sity” or­ga­ni­za­tions, POC make up 20% of core staff po­si­tions, com­pared to only 3% at “Non-Univer­sity” or­ga­ni­za­tions.

This pat­tern is quite con­sis­tent across or­ga­ni­za­tions. None of the 14 “Non-Univer­sity” or­ga­ni­za­tions has more than one POC on its core staff, and 10 (71%) have none. By con­trast, each of the 7 Univer­sity or­ga­ni­za­tions has mul­ti­ple POC on their core staff, and POC make up 24% of the core staff at the av­er­age or­ga­ni­za­tion.

De­mo­graphic differ­ences by role

Over­all, POC ac­count for ~13% of all roles at the Longter­mist or­ga­ni­za­tions in my sam­ple. Cer­tain types of roles, how­ever, have much differ­ent lev­els of di­ver­sity. There ap­pears to be a strong pat­tern in which the more in­fluen­tial a role is within an or­ga­ni­za­tion, the less likely it is to be held by a POC.

Only 4% of lead­er­ship roles are held by POC. Diver­sity among pro­fes­sional staff (14%) is in line with over­all lev­els. But for “Sup­port staff” roles (op­er­a­tions, ad­minis­tra­tion, free­lancers, part-timers, in­terns, and vol­un­teers), POC ac­count for 25% of po­si­tions.

In­ter­est­ingly, POC make up 17% of staff alumni. Since this rate is higher than the rep­re­sen­ta­tion of POC across all roles (13%) and much higher than the rep­re­sen­ta­tion at or­ga­ni­za­tions ex­clud­ing the Open Phil AI Fel­low­ship (10%), this raises the ques­tions of whether POC are more likely to leave Longter­mist or­ga­ni­za­tions than their white coun­ter­parts, and if so, why.

Note: Within the “Pro­fes­sional Staff” cat­e­gory, differ­ences in di­ver­sity be­tween “Staff” and “Re­searcher/​Fel­low” roles are in large part a func­tion of the Open Philan­thropy AI Fel­low­ship’s high per­centage of POC. Ex­clud­ing that pro­gram, POC make up 11% of Re­searcher/​Fel­low roles.

De­mo­graphic differ­ences by geography

As dis­cussed above, the differ­ence in di­ver­sity be­tween US and UK hiring pools com­pli­cates the as­sess­ment of di­ver­sity in the Longter­mist ecosys­tem. And since these two coun­tries house a differ­ent mix of Univer­sity-based and Non-Univer­sity-based or­ga­ni­za­tions in my sam­ple, it is im­por­tant to tease apart this con­found­ing effect. The fol­low­ing table does just that (look­ing only at Core Staff since non-Core Staff num­bers in­clude many re­search af­fili­ates who are not lo­cated where the or­ga­ni­za­tions are).

As ex­pected, US-based or­ga­ni­za­tions have a higher per­centage of POC than UK-based or­ga­ni­za­tions. A large part of this is be­cause of the out­sized in­fluence of the AI Fel­low­ship pro­gram. Ex­clud­ing those fel­lows, UK or­ga­ni­za­tions are ac­tu­ally more di­verse than US or­ga­ni­za­tions over­all, but only be­cause UK or­ga­ni­za­tions are more likely to be based at uni­ver­si­ties. Mak­ing an ap­ples-to-ap­ples com­par­i­son (e.g. uni­ver­sity-based or­ga­ni­za­tions in the US vs. UK), we again see US or­ga­ni­za­tions are more di­verse as ex­pected given their re­spec­tive hiring pools.

There are just two Univer­sity-based or­ga­ni­za­tions based in the US, the AI Fel­low­ship and CHAI. To­gether, they have a similar per­centage of POC (42%) to en­rol­l­ment at US Doc­toral Univer­si­ties with the high­est re­search ac­tivity (40%), though that is largely driven by the high rate of di­ver­sity within the AI Fel­low­ship. Rep­re­sen­ta­tion at CHAI alone (25%) is 38% be­low the same bench­mark. This short­fall is roughly similar to what we find at Univer­sity-based or­ga­ni­za­tions in the UK, where POC’s 15% rep­re­sen­ta­tion is 31% be­low the 22% POC en­rol­l­ment rate at lead­ing UK uni­ver­si­ties.

At non-Univer­sity-based or­ga­ni­za­tions, POC are ex­tremely rare (5%) in the US, but are com­pletely ab­sent in the UK in my sam­ple. (As men­tioned in the method­ol­ogy sec­tion, I’m aware of one POC who was hired by a UK or­ga­ni­za­tion af­ter my sam­ple was col­lected, which would bring the ag­gre­gate rate of POC at non-uni­ver­sity-based UK or­ga­ni­za­tions to 2%.) One could ar­gue that the lack of rep­re­sen­ta­tion in the US is ac­tu­ally worse than in the UK, given the US’s more di­verse hiring pool (es­pe­cially when one con­sid­ers that many of the US-based or­ga­ni­za­tions are fo­cused on AI and likely fre­quently hire from rel­a­tively di­verse CS pro­grams).

Sugges­tions for im­prov­ing racial diversity

  • Proac­tively seek out di­verse can­di­dates. Steps could in­clude sur­vey­ing net­works to iden­tify tal­ented POC, en­courag­ing POC to ap­ply for open po­si­tions or start their own or­ga­ni­za­tions, invit­ing POC to board roles, and ad­ver­tis­ing job open­ings in com­mu­ni­ties with strong POC rep­re­sen­ta­tion (e.g. the EA Global meetup for POC).

  • Sur­vey POC about their ex­pe­riences in the Longter­mist com­mu­nity. This sur­vey (which should per­mit anony­mous an­swers) could cap­ture valuable in­for­ma­tion on whether and why POC may feel ex­cluded, and ideas for im­prov­ing di­ver­sity.

  • Look for les­sons from the Open Philan­thropy AI Fel­low­ship pro­gram. This pro­gram, which ex­plic­itly sought to in­crease di­ver­sity, had more than twice as many POC (in per­centage terms) as any other or­ga­ni­za­tion in my sam­ple. There are likely as­pects of its op­er­a­tions, in­clud­ing how it sources and se­lects can­di­dates, that could help other or­ga­ni­za­tions be­come more di­verse. Scal­ing this pro­gram could also help im­prove the broader ecosys­tem’s di­ver­sity over time. That said, it’s im­por­tant to rec­og­nize that the AI Fel­low­ship pro­gram is not ex­cep­tion­ally di­verse; it’s di­ver­sity is sim­ply in line with that at Com­puter Science pro­grams at top US schools.

  • Pro­mote for­mal di­ver­sity and in­clu­sion pro­grams. My anal­y­sis doesn’t provide a defini­tive ex­pla­na­tion of why be­ing based at a uni­ver­sity ap­pears to have such a siz­able effect on an or­ga­ni­za­tion’s di­ver­sity. But it seems rea­son­able to as­sume that the for­mal poli­cies uni­ver­si­ties typ­i­cally have in place may play an im­por­tant role. Wider adop­tion of such poli­cies could im­prove di­ver­sity through­out the Longter­mist ecosys­tem.

  • Shift fund­ing, on the mar­gins, to more di­verse or­ga­ni­za­tions. On the mar­gins, some donors may want to sup­port uni­ver­sity-based Longter­mist or­ga­ni­za­tions, which gen­er­ally have much more racial di­ver­sity than other or­ga­ni­za­tions. Similarly, donors may want to sup­port US-based or­ga­ni­za­tions rather than their less-di­verse UK coun­ter­parts. Fun­ders who value di­ver­sity should com­mu­ni­cate this pri­or­ity to the or­ga­ni­za­tions they sup­port or are con­sid­er­ing sup­port­ing, par­tic­u­larly large fun­ders who can in­fluence or­ga­ni­za­tional di­rec­tion.

  • Con­sider POC who are already work­ing with Longter­mist or­ga­ni­za­tions for core staff po­si­tions. Nearly 30% of vol­un­teers, in­terns, free­lancers, and con­trac­tors (past or cur­rent) are POC. Ad­ding top perform­ers from this co­hort to core staff po­si­tions would be a sim­ple way to im­prove di­ver­sity.

  • Draw ad­vice from peo­ple and or­ga­ni­za­tions that ex­cel in di­ver­sity. As men­tioned above, the Open Philan­thropy AI Fel­low­ship likely has les­sons that can be ap­plied through­out the Longter­mist ecosys­tem, and the same can be said for peo­ple and or­ga­ni­za­tions from out­side that ecosys­tem. Re­search into best prac­tices (e.g. here and here) can also help this effort.

  • Provide men­tor­ship op­por­tu­ni­ties for POC. WANBAM pro­vides men­tor­ship for women and non-bi­nary mem­bers of the EA com­mu­nity. An analo­gous pro­gram for POC could provide im­por­tant sup­port and train­ing, while also sig­nal­ing that im­prov­ing racial di­ver­sity is a pri­or­ity. Men­tor­ship that could help POC take on lead­er­ship po­si­tions could be par­tic­u­larly valuable. I reached out to WANBAM’s founder, Kathryn Me­crow-Flynn, who would be will­ing to help a men­tor­ship pro­gram for POC get started: “I would love sug­ges­tions [for] women of color who would like to men­tor with us go­ing for­ward. I would ad­di­tion­ally wel­come a men­tor­ship pro­gram for women of color and peo­ple of color who are in­ter­ested in pur­su­ing high im­pact ca­reer paths and I would per­son­ally sup­port and lend re­sources and les­sons learned to such a pro­gram. You can reach me at <EAMen­tor­ship­pro­gram@gmail.com>.”

Conclusion

This anal­y­sis has found that POC are sig­nifi­cantly un­der­rep­re­sented at Longter­mist or­ga­ni­za­tions and pro­jects rel­a­tive to the pop­u­la­tions they typ­i­cally draw from. At many or­ga­ni­za­tions, and at most of the or­ga­ni­za­tions that are not based at a uni­ver­sity, core staff is en­tirely white. Where POC are em­ployed, it is dis­pro­por­tionately in sup­port roles and very rarely in lead­er­ship roles.

I be­lieve this situ­a­tion un­der­mines the effi­cacy of Longter­mist or­ga­ni­za­tions, which de­pends on the tal­ent they em­ploy. Their high de­gree of racial ho­mo­gene­ity sug­gests these or­ga­ni­za­tions have been miss­ing out on tal­ented em­ploy­ees and al­ter­na­tive per­spec­tives, and may con­tinue to strug­gle to at­tract strong POC can­di­dates in the ab­sence of changes (see here for re­lated dis­cus­sion).

I also be­lieve this lack of racial di­ver­sity un­der­mines the le­gi­t­i­macy of Longter­mist efforts. If you were a policy maker in the Global South, how cred­ible would you view “global pri­ori­ties re­search” from an ecosys­tem that’s as white as South Dakota? If you were an AI strate­gist in China, and knew that Asi­ans out­num­bered whites in many top CS pro­grams in the US, would you be skep­ti­cal about an over­whelm­ingly white ecosys­tem pro­duc­ing re­search on top­ics like “dis­tribut­ing the benefits of AI for the com­mon good”?

To be clear, I don’t think the racial ho­mo­gene­ity of Longter­mist or­ga­ni­za­tions is in any way due to ex­plicit racism or dis­crim­i­na­tion on the part of those or­ga­ni­za­tions or their em­ploy­ees. Rather, as CEA’s stance on di­ver­sity and in­clu­sion states, “Without con­scious effort, groups tend to main­tain a similar de­mo­graphic makeup over time. Coun­ter­act­ing that ten­dency to­ward nar­row­ness and ho­mo­gene­ity takes at­ten­tion and effort.”

My anal­y­sis also found that when at­ten­tion and effort is ap­plied, it can lead to pos­i­tive out­comes. The ex­am­ple of Open Philan­thropy’s AI Fel­lows clearly shows that in­ten­tion­ally pri­ori­tiz­ing di­ver­sity works. On a similar note, the efforts of CEA’s Com­mu­nity Health Team have paid off in terms of im­prov­ing speaker di­ver­sity at EA Global. I hope this anal­y­sis helps mo­ti­vate fur­ther at­ten­tion and effort to­ward im­prov­ing di­ver­sity, and ul­ti­mately effi­cacy, in the Longter­mist ecosys­tem.