Identifying Talent without Credentialing In EA

Em­ploy­ers usu­ally pre­fer to hire em­ploy­ees with cre­den­tials. The most com­mon ex­am­ple of this is hav­ing a uni­ver­sity de­gree: em­ploy­ees with uni­ver­sity de­grees re­ceive sig­nifi­cantly higher com­pen­sa­tion.

Univer­si­ties are very ex­pen­sive, and it would be nice to re­place them with some cheaper way of iden­ti­fy­ing tal­ent. For ex­am­ple, maybe Google could just go to high schools, give ev­ery­one some two-hour long exam, then offer those who pass the exam six months of train­ing at Google, af­ter which they be­come Google soft­ware en­g­ineers.

The fact that Ivy League uni­ver­si­ties still ex­ist seems like prima fa­cie proof that this isn’t pos­si­ble: cor­po­ra­tions could make billions of dol­lars if they could do an end run around uni­ver­si­ties and hire peo­ple right from high school, so the fact that they don’t do this must in­di­cate that it’s a hard prob­lem.

Even if it’s hard in the gen­eral case, though, there might be spe­cific sub­sets of em­ploy­ees or jobs for which we can avoid uni­ver­si­ties. Cod­ing boot camps, for ex­am­ple, seem to demon­strate a cre­den­tial­ing sys­tem for soft­ware en­g­ineer­ing jobs which is just as effec­tive as ex­ist­ing ones but mas­sively cheaper.

In this ar­ti­cle, I ex­plain the con­cerns with ex­ist­ing cre­den­tial­ing sys­tems and try to spec­u­late on which idiosyn­crasies of EA would en­able us to avoid them.

Why Cre­den­tial­ing is (Some­times) Bad

One sim­ple solu­tion to iden­ti­fy­ing tal­ent is ex­pert defer­ral: if I’m hiring a soft­ware en­g­ineer, I could just say some­thing like “if Google offers you a job then I will also offer you a job,” thereby defer­ring to Google about hiring stan­dards.

This is not a ter­rible strat­egy, but it has the down­side that we are, by defi­ni­tion, com­pet­ing with Google for ev­ery em­ployee. Similarly, if we re­cruit only from Ivy League col­leges, we are com­pet­ing against a bunch of other em­ploy­ers who re­cruit from Ivy League col­leges.

We might pre­fer to find some­one with good open-source con­tri­bu­tions who didn’t finish col­lege (as I al­most did), or grad­u­ated from a state school (as I did). Th­ese peo­ple might be bet­ter fits (in the spe­cific case of pro­gram­mers, open source con­tri­bu­tions are ar­guably more pre­dic­tive than ed­u­ca­tional pedi­gree), but even if they are less skil­led than the av­er­age Ivy League grad­u­ate, we have much less com­pe­ti­tion for re­cruit­ing them.

Se­condly, cre­den­tials don’t ex­ist for many things that EA’s con­sider im­por­tant. Some­one re­search­ing ex­is­ten­tial risks may come across in­fo­haz­ards, and hav­ing a PhD in no way cer­tifies “this per­son will avoid pub­lish­ing in­fo­haz­ards even if pub­li­ca­tion would ad­vance their ca­reer.” There­fore, we need an al­ter­na­tive cer­tifi­ca­tion pro­cess.

Fi­nally, the EA com­mu­nity cur­rently con­tains a lot of peo­ple who are early in their ca­reers and have not had the time to build a track record of suc­cess. There are le­gi­t­i­mate con­cerns with putting peo­ple into po­si­tions of au­thor­ity with­out cre­den­tials, which means that these early-ca­reer peo­ple have a lot of difficulty con­tribut­ing.

What are some ways EA could take its un­usual re­quire­ments into ac­count, in or­der to hire more effec­tively?

Hy­poth­e­sis 1: Hits-Based Hiring

Estab­lished com­pa­nies are very risk-averse. Boards will pay a large pre­mium to hire a provenly mediocre CEO ver­sus tak­ing a chance on an untested CEO who might be re­ally good but also might be re­ally awful.

Cer­tain types of EA work are pos­si­bly more hits-based, such that the or­ga­ni­za­tions do­ing that work might pre­fer the “untested, high var­i­ance” em­ployee to the “tested and mediocre” em­ployee.

As an ex­am­ple, some types of re­search might fall into this cat­e­gory: if the re­search pro­ject is done poorly, it will just go into the dust­bin as an un-cited pa­per, but if it suc­ceeds, it could define a new field.

For these po­si­tions, it seems like we can rely less on cre­den­tial­ing in fa­vor of a “spray and pray” ap­proach where we hire a lot of peo­ple with the ex­pec­ta­tion that most of them will be bad fits but the few who are suc­cess­ful will cover their costs.

Rocket In­ter­net’s model of start­ing vast num­bers of com­pa­nies and then shut­ting down the ones which aren’t suc­cess­ful is one for-profit ex­am­ple of this strat­egy. (It’s worth not­ing that Rocket has a hor­rible rep­u­ta­tion as a re­sult – it turns out that re­peat­edly lay­ing off mas­sive num­bers of peo­ple within six months of hiring them doesn’t make you pop­u­lar.)

Hy­poth­e­sis 2: Altru­is­tic Internships

Sup­pose a cor­po­ra­tion feels that some­one with no rele­vant back­ground could be brought up to speed af­ter com­plet­ing a two-year train­ing pro­gram. The cor­po­ra­tion might think of some strat­egy like: train some­one for two years (dur­ing which the com­pany loses money on them), but then have them em­ployed at be­low-mar­ket salaries for three years (dur­ing which the com­pany makes money on them). This would seem to benefit both the cor­po­ra­tion and the em­ployee.

In prac­tice, this doesn’t usu­ally hap­pen be­cause there’s noth­ing to pre­vent the em­ployee from leav­ing af­ter they com­plete the train­ing pro­gram and go­ing to a com­peti­tor which pays mar­ket rates. Cor­po­ra­tions some­times try to put a “poi­son pill” in their con­tracts, forc­ing em­ploy­ees who leave be­fore a cer­tain date to pay money back to the cor­po­ra­tion. But it’s hard to legally en­force these con­tracts.

Altru­is­tic em­ploy­ers might not have this prob­lem: if I train some­one to be­come awe­some at cor­po­rate out­reach cam­paigns, and then they go do cor­po­rate out­reach cam­paigns some­where else, that’s kind of okay?

More speci­fi­cally: al­tru­is­tic fun­ders would be okay with this. If The Hu­mane League has un­usu­ally high HR costs be­cause their skil­led em­ploy­ees leave to go work at Mercy For An­i­mals, donors to THL might be happy to sub­si­dize that cost be­cause the donor is al­most in­differ­ent be­tween work at THL ver­sus MFA. Or a fun­der might sup­port a grant for some­one to work on a prob­lem apart from ei­ther or­ga­ni­za­tion.

Hy­poth­e­sis 3: Avoid­ing Nega­tive Salaries

The US has a cul­tural and le­gal norm that em­ploy­ees must be paid if they provide value to their em­ployer. Univer­sity stu­dents, for ex­am­ple, do not need to be paid for their stud­ies (and in­deed usu­ally pay for the priv­ilege of study­ing) be­cause their stud­ies do not benefit the Univer­sity.

This pre­vents a lot of mu­tu­ally benefi­cial ap­pren­tice­ships: young ap­pren­tices might be will­ing to work for free or even pay to help more skil­led work­ers, and the skil­led work­ers might be happy to have free la­bor. But this is cul­turally and of­ten legally pro­hibited, e.g. by min­i­mum wage laws. For ex­am­ple, short­ages of sur­geons are some­times blamed on the fact that hos­pi­tals are re­quired to pay sur­gi­cal res­i­dents, even though the mar­ket equil­ibrium would be for the res­i­dents to pay the hos­pi­tals.

Non­prof­its are not as shielded from these re­quire­ments as you might ex­pect – there are laws pre­vent­ing vol­un­teers from do­ing the work that em­ploy­ees would oth­er­wise do. But, as in the pre­vi­ous sec­tion, non­prof­its can co­or­di­nate with fun­ders in ways for-prof­its can’t. For ex­am­ple, a donor could sub­si­dize some­one’s em­ploy­ment through a dona­tion to their em­ployer, let­ting the non­profit avoid tak­ing a risk on their new em­ployee while still com­ply­ing with min­i­mum wage laws. The donor is thus effec­tively in­vest­ing in the ap­pren­tice’s em­ploy­ment, with the hope that the ap­pren­tice will provide pos­i­tive value later on.

Another ap­proach is en­trepreneur­ship: the self-em­ployed are gen­er­ally ex­empt from min­i­mum wage laws and of­ten lose money on their com­pa­nies – if EA has valuable pro­jects which can be worked on in­de­pen­dently, peo­ple could do that in­stead of more ex­pen­sive cre­den­tials.

Conclusion

It seems like there is some mo­men­tum in EA be­hind these ap­proaches. BERI’s In­di­vi­d­ual Level-Up Grants is one rele­vant pro­gram; some EA Grants re­cip­i­ents have been funded to gain im­por­tant skills through in­de­pen­dent study. Eliezer Yud­kowsky, the founder of MIRI, is an ex­treme ex­am­ple of suc­cess with­out cre­den­tials, hav­ing at­tended nei­ther high school or col­lege.

Given that EA has a lot of peo­ple with raw ca­pa­bil­ities but rel­a­tively few cre­den­tials, I would be ex­cited for more peo­ple to ex­plore this.

I would like to thank Aaron Gertler for re­view­ing a draft of this. Any re­main­ing mis­takes are mine. Views ex­pressed are my own, and do not rep­re­sent my em­ployer’s. I have not been in­volved in CEA’s hiring, nor in any rounds of EA or Com­mu­nity Build­ing Grants. This post is re­lated to sev­eral re­cent ones such as EA is Vet­ting Con­strained but was writ­ten sev­eral months ago and not in­fluenced by them (it just took me a while to get around to pub­lish­ing).