Cross-post: Think twice before talking about ‘talent gaps’ – clarifying nine misconceptions, by 80,000 Hours.

If you have a 'tool gap', it really matters which tool.

Cross-posted from the 80,000 Hours blog—see the origi­nal here. This is a par­tial fol­low-up to our pre­vi­ous post here Many EA orgs say they place a lot of fi­nan­cial value on their pre­vi­ous hire. What does that mean, if any­thing? And why aren’t they hiring faster?.

--

After push­ing the idea of ‘tal­ent gaps’ in 2015, we’ve no­ticed in­creas­ing con­fu­sion about the term.

This is partly our fault. So, here’s a quick list of com­mon mis­con­cep­tions about tal­ent gaps and how they can be fixed. This is all pretty rough and we’re still re­fin­ing our own views, but we hope this might start to clar­ify this is­sue, while we work on bet­ter ex­plain­ing the idea in our key con­tent.

1. Prob­lem ar­eas are con­strained by spe­cific skills, not ‘tal­ent’

Prob­lem ar­eas are rarely gener­i­cally ‘tal­ent con­strained’. They’re in­stead con­strained by spe­cific skills and abil­ities. It’s nearly always clearer to talk about the spe­cific needs of the field, ideally down to the level of spe­cific pro­files of peo­ple, rather than tal­ent and fund­ing in gen­eral.

For in­stance, work to pos­i­tively shape the de­vel­op­ment of AI is highly con­strained by the fol­low­ing:

  • ML re­searchers, es­pe­cially those able to do field-defin­ing work, who are in­ter­ested in and un­der­stand AI safety, the al­ign­ment prob­lem, and other is­sues rele­vant to the long-term de­vel­op­ment of AI.

  • Peo­ple skil­led in op­er­a­tions, es­pe­cially those able to run non-prof­its with un­der 50 peo­ple or aca­demic in­sti­tutes, and who are in­ter­ested in and un­der­stand is­sues re­lated to the long-term de­vel­op­ment of AI.

  • Strat­egy and policy re­searchers able to do dis­en­tan­gle­ment re­search in pre-paradig­matic fields.

  • Peo­ple with the policy ex­per­tise and ca­reer cap­i­tal to work in in­fluen­tial gov­ern­ment po­si­tions who are also knowl­edge­able about and ded­i­cated to the is­sue.

Often when peo­ple talk about a field be­ing ‘tal­ent con­strained’, they’re us­ing that as short­hand for a more spe­cific need.

Be­sides be­ing less pre­cise, talk­ing about ‘tal­ent con­straints’ is con­fus­ing for a cou­ple of rea­sons.

One rea­son is that it makes it sound like any­one who’s gen­er­ally tal­ented can con­tribute, whereas usu­ally what’s most needed is a per­son with a spe­cific set of char­ac­ter­is­tics.

In­deed, the more spe­cific the pro­file, the harder it is to hire, and so the greater the bot­tle­neck. The speci­fic­ity of the need is of­ten why there’s a ‘tal­ent gap’ in the first place. So, the ex­is­tence of a ‘tal­ent gap’ may be ev­i­dence that most peo­ple can’t eas­ily con­tribute.

Talk­ing about ‘tal­ent con­straints’ also makes it sound like peo­ple who can’t get a job in the area aren’t tal­ented, whereas it ac­tu­ally means they don’t have the spe­cific char­ac­ter­is­tics needed. They might be ex­tremely tal­ented in other ways.

Rather than fund­ing vs. tal­ent gaps, we pro­pose that peo­ple aim to iden­tify spe­cific bot­tle­necks fac­ing the field and the skills needed to re­solve them. A ‘bot­tle­neck’ is the re­source that a field most needs in or­der to make progress. In our re­cent ar­ti­cle, we pro­pose the fol­low­ing list of cat­e­gories of bot­tle­neck:

  • Fund­ing—ad­di­tional fi­nan­cial re­sources from dona­tions or fundrais­ing.

  • In­sights—new ideas about how to solve the prob­lem.

  • Aware­ness & sup­port—how many peo­ple know and care about the is­sue, and how in­fluen­tial they are.

  • Poli­ti­cal cap­i­tal—the amount of poli­ti­cal power that’s available for the is­sue.

  • Co­or­di­na­tion—the ex­tent to which ex­ist­ing re­sources effec­tively work to­gether.

  • Com­mu­nity build­ing—find­ing other peo­ple who want to work on the is­sue.

  • Lo­gis­tics and op­er­a­tions—the ex­tent to which pro­grammes can be de­liv­ered at scale.

  • Lead­er­ship and man­age­ment—the ex­tent to which con­crete plans can be formed and ex­e­cuted on us­ing the re­sources already available.

Th­ese bot­tle­necks can be made much more spe­cific again, such as with the AI ex­am­ples above. The ideal is to be as pre­cise as pos­si­ble about the char­ac­ter­is­tics needed.

There is a defi­ni­tion of ‘tal­ent con­strained’ that we and other or­gani­sa­tions some­times use:

An or­gani­sa­tion is tal­ent con­strained when, for some­one who could take (a rea­son­ably im­por­tant) job at that or­gani­sa­tion, they would typ­i­cally con­tribute more to that or­gani­sa­tion by tak­ing the job than earn­ing to give.[fn 1]

This defi­ni­tion can be use­ful, but we sug­gest that when this more pre­cise claim is what’s meant, peo­ple should just say that.

2. Skill bot­tle­necks are a mat­ter of degree

As the above list of bot­tle­necks also shows, there are many ways to con­tribute to a field. We try to high­light the key bot­tle­necks fac­ing a field be­cause these are the most ur­gent to ad­dress, but most fields are some­what con­strained by many differ­ent in­puts.

In par­tic­u­lar, even in fields that are most con­strained by a cer­tain skill set, ad­di­tional dona­tions are usu­ally still valuable, as well as other con­tri­bu­tions.

This pre­sents a difficult com­mu­ni­ca­tion challenge, which we have not yet solved. If there’s a valuable di­rect work op­por­tu­nity, we want to high­light that those who might be able to take it should strongly con­sider pur­su­ing it over earn­ing to give. But that doesn’t mean that earn­ing to give is low im­pact, or that earn­ing to give won’t be the top op­tion for many peo­ple.

In gen­eral, we’re keen to see peo­ple am­bi­tiously pur­su­ing top di­rect work roles, but only if they might be a good fit, and have a good backup plan.

3. Skill and fund­ing con­straints are not opposites

All else equal, in­creas­ing fund­ing gen­er­ally in­creases the ex­tent to which hiring is the bot­tle­neck, and vice versa. So, there is some in­verse re­la­tion­ship be­tween fund­ing and skill bot­tle­necks.

How­ever, it’s pos­si­ble for an or­gani­sa­tion to be nei­ther fund­ing con­strained nor skill con­strained, and sev­eral or­gani­sa­tions in our 2018 sur­vey re­ported this.

This is be­cause they might in­stead be mainly con­strained by man­age­ment ca­pac­ity, co­or­di­na­tion over­heads, in­sights into their plan, etc.

An or­gani­sa­tion can also be both con­strained by fund­ing and by mak­ing cer­tain types of hires.

For ex­am­ple, imag­ine an or­gani­sa­tion that has iden­ti­fied a sim­ple, scal­able model that re­quires lit­tle man­age­ment over­head but lots of ex­pen­sive staff who are hard to hire be­cause they have a rare and difficult-to-mea­sure skill set. This or­gani­sa­tion might be con­strained by both fund­ing and ‘tal­ent.’ By con­trast, man­age­ment ca­pac­ity and in­sights do not seem like im­por­tant bot­tle­necks.

Over­all, it’s pos­si­ble for an or­gani­sa­tion to be mainly con­strained by a cer­tain skill, or fund­ing, or both, or nei­ther.

4. Skill bot­tle­necks vary by prob­lem area

Each area has differ­ent needs, so faces differ­ent bot­tle­necks.

This is an­other rea­son why it can be mis­lead­ing to talk about ‘the com­mu­nity be­ing skill con­strained’. It’s bet­ter to talk about spe­cific prob­lem ar­eas or pro­jects.

We try to high­light how our views de­pend on prob­lem se­lec­tion in our re­cent ar­ti­cle and the sur­vey. For in­stance, global health is sig­nifi­cantly more fund­ing con­strained than global catas­trophic risks, so earn­ing to give is a rel­a­tively more at­trac­tive path if you’re fo­cused on health—though as per point 2, ad­di­tional fund­ing is use­ful in both.

5. Skill bot­tle­necks aren’t only about effec­tive al­tru­ism organisations

Peo­ple some­times act as if the main al­ter­na­tive to earn­ing to give is work­ing at an ‘effec­tive al­tru­ism non-profit’. How­ever, this misses many types of high im­pact roles in­clud­ing those in academia, policy and rele­vant com­pa­nies, which could ab­sorb far more peo­ple. Our re­cent sur­vey showed that roles in policy are highly val­ued, as are re­search po­si­tions that could be done within academia.

When peo­ple talk about how the com­mu­nity is ‘tal­ent con­strained’, they don’t just mean that the top pri­or­ity is more peo­ple work­ing at EA or­gani­sa­tions. They of­ten have these other roles in mind too.

6. High value of re­cent em­ploy­ees doesn’t nec­es­sar­ily en­tail that hiring is the top priority

Our re­cent sur­vey showed that or­gani­sa­tions were will­ing to pay a lot to re­tain re­cent hires. It’s nat­u­ral to think that this im­plies the or­gani­sa­tions are con­strained by that type of tal­ent, and that their key pri­or­ity should be to hire more peo­ple. But this doesn’t fol­low.

Run­ning a hiring round re­quires large costs, de­lays and risk, so the net benefit of hiring could be lower than the benefit of re­tain­ing ex­ist­ing staff.

It’s easy to con­flate ‘skill bot­tle­necks’, ‘re­cent hires were worth a lot’ and ‘hiring is the top pri­or­ity’, and these con­cepts are all closely con­nected, but they can di­verge, so it’s clearer to sep­a­rate them.

Read more.

7. The ex­is­tence of skill bot­tle­necks doesn’t mean that it’s easy to get a job in the area

Even if hiring is the key bot­tle­neck fac­ing an or­gani­sa­tion, it may still not be easy to get a job there. Skill con­straints usu­ally ex­ist pre­cisely be­cause it’s very difficult for an or­gani­sa­tion to find peo­ple to fill the skill set it needs. One rea­son this might hap­pen is that they’re look­ing for a very spe­cific skill set. This makes it un­likely that any given per­son will be a good fit. If peo­ple with the nec­es­sary skills were com­mon, the or­gani­sa­tion would prob­a­bly be hav­ing an eas­ier time hiring.

For in­stance, de­vel­op­ment of clean meat seems to be bot­tle­necked by out­stand­ing sci­en­tists, but very few peo­ple can re­al­is­ti­cally take those jobs.

Again, it would be helpful if peo­ple com­mu­ni­cated more spe­cific skill pro­files and in­di­ca­tors of fit that po­ten­tial hires can use to gauge their chances.

Another way to see the prob­lem is that a typ­i­cal job ap­pli­ca­tion pro­cess only ac­cepts 1-10% of ap­pli­cants. This means that even if an or­gani­sa­tion is 3-times keener to hire than av­er­age, its ac­cep­tance rate would still only be 3-30%, and most ap­pli­cants will still not get the job.

In gen­eral, fit is both very im­por­tant but hard to pre­dict, so any given hiring pro­cess has a low chance of work­ing out.

To help com­mu­ni­cate this, it may be helpful if or­gani­sa­tions pub­lished typ­i­cal ra­tios of ap­pli­cants to hires to let peo­ple plan ac­cord­ingly.

How­ever, we also of­ten see peo­ple rule them­selves out of jobs pre­ma­turely, and fail to ap­ply even when they were ac­tu­ally a good fit. It’s un­clear whether we and or­gani­sa­tions in the com­mu­nity should be more or less en­courag­ing of ap­pli­ca­tions.

One way to solve this is to do more to en­courage peo­ple to test out their fit. If a role might have a lot of im­pact, it’s of­ten worth in­ves­ti­gat­ing more to see if it might work out for you. This is es­pe­cially true if you can do it cheaply, such as by talk­ing to a few peo­ple or mak­ing an ap­pli­ca­tion.

How­ever, you should always do this while main­tain­ing a good backup plan, or ac­cu­mu­lat­ing widely use­ful ca­reer cap­i­tal. Be­cause the most likely sce­nario is that it doesn’t work out, prob­a­bly don’t quit your job or bet the farm.

8. Skill bot­tle­necks are not always solv­able by rais­ing salaries

A low salary is one rea­son why you might strug­gle to hire, but there are many other is­sues you can face, such as search costs or a long train­ing pro­cess.

Rais­ing salaries also some­times has a limited abil­ity to at­tract new staff, since peo­ple are mo­ti­vated by many differ­ent things, of which salary is just one. If your po­ten­tial em­ploy­ees are not highly money-mo­ti­vated, you can in­crease salaries a lot for a small gain.

What’s more, the costs of rais­ing salaries to at­tract new staff are of­ten large be­cause in or­der to be fair, you may also need to raise salaries for ex­ist­ing staff. For in­stance, if you have 10 equally-paid staff, and raise salaries 10% to at­tract one ex­tra per­son, that fi­nal staff mem­ber effec­tively costs dou­ble the av­er­age pre­vi­ous salary.

Higher salaries can also at­tract peo­ple who do the job for the wrong rea­sons, can cre­ate PR is­sues, and so on.

This point is a par­tic­u­larly com­pli­cated de­bate. In gen­eral, if an or­gani­sa­tion thinks that it’s con­strained by a par­tic­u­lar role, we think they should se­ri­ously con­sider rais­ing salaries for that role. How­ever, we don’t think this will fully solve the prob­lem, and it isn’t always the right thing to do.

We asked or­gani­sa­tions about why they don’t just raise salaries in our 2017 tal­ent gaps sur­vey.

9. Skill bot­tle­necks can change rapidly

We re­cently saw some po­si­tions that would have been hard to fill a cou­ple of years ago get so many good ap­pli­cants they couldn’t differ­en­ti­ate be­tween the top ones. This sug­gests we may be mak­ing progress on find­ing the right peo­ple.

Or­gani­sa­tions work­ing on build­ing effec­tive al­tru­ism and re­duc­ing GCRs are also in­creas­ing their ca­pac­ity to ab­sorb new staff as they bet­ter work out their plans and es­tab­lish man­age­ment teams.

Th­ese fac­tors could lead to an eas­ing up of skill con­straints in some key ar­eas.

Th­ese re­cent changes also show how it can be difficult to make these kinds of as­sess­ments.

For in­stance, or­gani­sa­tions of­ten talk about their im­me­di­ate con­straints over the next year or so. But if you’re con­sid­er­ing in­vest­ing in your ca­reer cap­i­tal, then your time hori­zon is prob­a­bly more like 5-20 years. So, the key ques­tion is which bot­tle­necks will be most press­ing over that en­tire pe­riod, which could be sig­nifi­cantly differ­ent.

To help with this, in our last sur­vey we also asked re­spon­dents to as­sess skill needs over 5 years.

The fact that key bot­tle­necks can change is also an­other rea­son why it’s im­por­tant to always have a backup plan.

Conclusion

While we haven’t suc­cess­fully com­mu­ni­cated it to this point, this more pre­cise no­tion of skill con­straints is a big part of the rea­son­ing be­hind our cur­rent core ca­reer ad­vice.

We be­lieve many of our top prob­lem ar­eas are highly con­strained by spe­cific skill sets, as we out­lined for AI safety ear­lier. What’s more, there was an im­por­tant shift in this di­rec­tion from 2014 as ad­di­tional large fun­ders en­tered these ar­eas, es­pe­cially the Open Philan­thropy Pro­ject. This in­crease in fund­ing cre­ated a spike in de­mand for cer­tain key po­si­tions that couldn’t be quickly matched by an equiv­a­lent in­crease in peo­ple able to fill them. This led to a bot­tle­neck of peo­ple with these types of skills, which per­sists to­day.

To­day we usu­ally recom­mend that peo­ple who are a good fit for filling these bot­tle­necks treat them as their first pri­or­ity. This usu­ally means ini­tially con­sid­er­ing rele­vant jobs in re­search, top non-prof­its and policy, and if you’re will­ing to con­sider some­thing es­pe­cially com­pet­i­tive, our list of pri­or­ity paths.

In con­trast, we rarely think that earn­ing to give should be the top pri­or­ity for peo­ple who could be a good fit for these other roles. This is an­other idea we hoped to high­light by talk­ing about ‘tal­ent con­straints’.

How­ever, we also rec­og­nize that our pri­or­ity prob­lems aren’t ‘tal­ent con­strained’ in gen­eral, and our pri­or­ity paths re­quire a fairly nar­row set of skills. So, we con­tinue to recom­mend build­ing ca­reer cap­i­tal and earn­ing to give as a high im­pact op­tion for peo­ple whose skills don’t match the par­tic­u­lar con­straints cur­rently faced by our pri­or­ity prob­lems.

In the fu­ture, we’ll aim to re­place ‘tal­ent con­straints’ with a more spe­cific phrase, which will usu­ally be a claim about which spe­cific skills are most needed.

What should peo­ple look­ing to choose a ca­reer ac­tu­ally do based on this? Here’s a quick sum­mary of our pro­cess in prac­ti­cal terms:

  1. Make a list of po­ten­tially high-im­pact op­tions: jobs that help re­lieve key bot­tle­necks in top prob­lem ar­eas (but not just at ‘EA orgs’).

  2. Work out if any spe­cific jobs in these paths have the po­ten­tial to be a good fit. If not, ex­pand your op­tions by con­sid­er­ing differ­ent prob­lem ar­eas, differ­ent bot­tle­necks, or earn­ing to give.

  3. Do cheap tests to learn more about fit. Even when a skill bot­tle­neck ex­ists, fit is both very im­por­tant and hard to pre­dict.

  4. If, af­ter these tests, an op­tion seems promis­ing, try it out for a num­ber of years.

  5. If it goes well, keep go­ing. Other­wise try some­thing else.

  6. Each op­tion could eas­ily fail to work out, so do all of this while main­tain­ing a good backup plan.