Strategic High Skill Immigration

How to im­prove the global econ­omy, in­crease strate­gic sta­bil­ity, and safe­guard the far fu­ture.

Part 1: The cur­rent situ­a­tion: shift­ing bal­ance of power

When try­ing to help peo­ple at scale, it is of­ten pru­dent to go be­yond di­rect in­ter­ven­tions and to seek equil­ibria which are in the long run in­ter­est of hu­man­ity, es­pe­cially if we as­sign any sig­nifi­cant value to the far fu­ture. From this we can de­rive the in­stru­men­tal value of re­duc­ing risks which are multi-gen­er­a­tional or per­ma­nent in na­ture, and en­sur­ing the safe de­vel­op­ment of trans­for­ma­tive tech­nolo­gies which could pose im­mense risks ei­ther on their own or in the wrong hands. His­tor­i­cally, wars and state com­pe­ti­tion have had a great deal of in­fluence on these sorts of con­cerns: com­pe­ti­tions be­tween na­tions can lead to arms races that pro­duce new dan­ger­ous or benefi­cial tech­nolo­gies, the de­struc­tion of wars them­selves can have per­ma­nent effects, and the vic­tors of wars de­ter­mine the in­sti­tu­tions which gov­ern hu­man­ity and its tech­nol­ogy go­ing for­ward in time.

As China’s eco­nomic growth con­tinues to out­pace US eco­nomic growth, and its mil­i­tary ex­pen­di­ture as a per­centage of GDP con­tinues to re­main sta­ble it is plau­si­ble that the US and China may find them­selves in a differ­ent equil­ibrium from the cur­rent peace they are used to, and plau­si­bly a similar trap to the one the UK and Ger­many faced just be­fore WW1. Though the Bri­tish Em­pire spanned much of the globe and had the largest Navy in the world, in 1898 Ger­many be­gan a naval arms race with Bri­tain, just be­fore 1914 the Ger­man econ­omy caught up, and shortly af­ter World War 1 fol­lowed.

Known as the Thucy­dides Trap, in 12 out of 16 cases where a ris­ing power challenged an es­tab­lished power in the past 500 years, war was the re­sult. Though the pat­tern may seem sur­pris­ing, it is rather con­sis­tent: with no out­side en­forcer with co­er­cive power to make peace agree­ments be­tween great pow­ers trust­wor­thy, it is fairly difficult for coun­tries to cred­ibly com­mit to one of the many out­comes that leaves both par­ties bet­ter off than war, es­pe­cially since as the ris­ing power be­comes more pow­er­ful, it con­tin­u­ously gains ne­go­ti­at­ing ad­van­tage, while the formerly rul­ing power loses it. In the four cases above where war was avoided, one in­volved a prior war, and ex­ter­nal in­ter­ven­tion by the pope, two cases in­volved com­pet­ing al­lies, and the re­main­ing case was the rise of the Soviet Union, which never ac­tu­ally matched the econ­omy of the United States.

While China’s econ­omy could stag­nate due to the struc­ture of its in­sti­tu­tions, or it could fail to com­pete with the US on a mil­i­tary level due to a tech­nolog­i­cal re­search dis­ad­van­tage and lack of al­li­ances, China will likely still close in on US dom­i­nance as it builds a larger Navy, fur­ther ex­ceeds US GDP, dis­pro­por­tionately in­vests in ar­tifi­cial in­tel­li­gence, and con­tinues to be less in­hibited from ge­net­i­cally mod­ify­ing hu­mans.

Though the over­all eco­nomic in­ter­ests of the US Govern­ment and the Chi­nese Govern­ment are fairly al­igned, and China has much warmer re­la­tions with the US than the Soviet Union did, there are likely to be ar­eas where China could benefit more by fur­ther ex­pand­ing its in­ter­est in for­eign coun­tries at the ex­pense of US in­ter­ests once China is in a po­si­tion of rel­a­tive power. From the con­sid­er­a­tions above, we can see that the long term pres­sures on the US and China (es­tab­lished power vs. grow­ing eco­nomic power) will likely lead to con­flict (though not nec­es­sar­ily war given de­ter­rence, eco­nomic in­ter­de­pen­dence, and the many norms to­ward peace in the mod­ern era) and that such con­flict could fur­ther en­courage tech­nol­ogy races in ar­eas which are of great con­cern for the far fu­ture, such as AI and hu­man ge­netic en­g­ineer­ing.

To han­dle this situ­a­tion, in­ter­na­tional agree­ments are valuable for slow­ing mil­i­tary tech races but they may be hard to gen­er­ate for many rea­sons. While eco­nomic benefits of not hav­ing an arms race are of­ten win-win, some benefits are po­si­tional (e.g. be­ing a cen­tral node in a trade net­work), and coun­tries may not ac­cept an eco­nomic win-win trade that gives their ri­val a po­si­tional strate­gic ad­van­tage. Even worse, as tech­nolo­gies such as AI and hu­man en­hance­ment will im­pact eco­nomic pro­duc­tivity, there is not even nec­es­sar­ily an im­me­di­ate eco­nomic win-win from gen­er­at­ing agree­ment, just the po­ten­tial to re­duce fu­ture risks which cur­rent de­ci­sion mak­ers might dis­count. While the US and Rus­sia/​the Soviet Union were able to gen­er­ate risk re­duc­ing agree­ments (*cough cough* which Rus­sia cheats on a lot), these were par­tially caused by agree­ment on the anal­y­sis of the situ­a­tion: both sides could agree peace and sur­vival was more likely in one equil­ibrium than an­other. Un­like de­bat­ing nu­clear weapons of spe­cific speeds, ranges, and num­bers of re-en­try ve­hi­cles, AI has many more ap­pli­ca­tions which could be desta­bi­liz­ing in some man­ner and gen­er­ate a strate­gic ad­van­tage for the coun­try de­ploy­ing more ad­vanced AI sys­tems first. Fur­ther, un­like nu­clear test­ing, which is hard to con­ceal (though Rus­sia still does small yield tests de­spite the Com­pre­hen­sive Test Ban Treaty), many AI sys­tems can likely be de­ployed on­line with­out be­ing no­ticed. Be­ing a much more com­plex space, it may take sub­stan­tially longer to an­a­lyze the pos­si­bil­ities (of which there are more than hu­mans can imag­ine, as smarter than hu­man AI will be more in­tel­li­gent than hu­mans) and agree on reg­u­la­tions that would put both par­ties in a bet­ter equil­ibrium. Even if coun­tries could agree on the strate­gic im­pli­ca­tions of AI (which might be its own risk, be­cause un­der­stand­ing the ac­tual situ­a­tion in­volves hav­ing a sense of each other’s ca­pa­bil­ities, which may in­cen­tivize fur­ther rac­ing) the rate of in­no­va­tion could quickly ren­der such agree­ments not use­ful due to un­pre­dicted changes in the bal­ance of power, and the slow pace of policy change. Un­less the con­se­quences of cheat­ing on agree­ments are un­am­bigu­ously worse than co­op­er­at­ing in the minds of poli­cy­mak­ers, cheat­ing will oc­cur, and it will oc­cur dis­pro­por­tionately with gov­ern­ments that can con­ceal ac­tivi­ties from in­ves­ti­ga­tion. If these sorts of prob­lems can’t be re­solved (and we should try very hard to re­solve them!), it is valuable to have a backup plan for cre­at­ing a safe policy en­vi­ron­ment for the de­vel­op­ment and de­ploy­ment of trans­for­ma­tive tech­nolo­gies, with­out hav­ing to get ev­ery world power to agree on ex­actly what that en­vi­ron­ment will be.

While his­tor­i­cally, monopoly of force has been a mechanism by which co­or­di­na­tion prob­lems are re­solved, this does not mean war would be nec­es­sary. If in an­ti­ci­pa­tion of the nu­clear bomb, one coun­try had mo­nop­o­lized ura­nium (though un­re­al­is­tic due to its ge­o­graphic dis­tri­bu­tion), there would not have been as much po­ten­tial for nu­clear arms race con­di­tions to de­velop. The coun­try with con­trol of such re­sources would be so de­ci­sively ahead that other coun­tries pur­su­ing such re­sources would in­crease their risk via at­tract­ing at­ten­tion more than they would de­crease their risk by shrink­ing the ad­van­tage of the first mover.

A differ­ent kind of re­source, even more im­por­tant to the de­vel­op­ment of such pow­er­ful tech­nolo­gies is hu­man cap­i­tal. The abil­ity of coun­tries to gain and hold highly in­tel­li­gent peo­ple greatly af­fects their eco­nomic and strate­gic out­look. As con­crete ex­am­ples, nu­clear weapons and strat­egy in the US and China were greatly in­fluenced by the mi­gra­tion of John von Neu­mann and Quan Xue­sen re­spec­tively. To com­pete for the best and bright­est, coun­tries may en­gage in a cred­ible virtue sig­nal­ing con­test of sorts. In the 1940s the US gained hu­man cap­i­tal at the ex­pense of the USSR and Nazi Ger­many by be­ing more “vir­tu­ous” while China gained Quan Xue­sen in 1955s as the re­sult of the US’s un­fair be­hav­ior to­ward him in the pre­ced­ing years and hawk­ish stance to­ward China at the time. Such com­pe­ti­tions on virtue sig­nal­ing are prefer­able to arms races which pro­duce weapons, or other pro­jects that gen­er­ate new types of in­tel­li­gence that may gen­er­ate larger al­ign­ment risks than they are equipped to han­dle (eg. a cer­tain kind of surveillance state doesn’t have to keep the pub­lic happy to stay in power). As op­posed to cre­at­ing new abil­ities, com­pe­ti­tions for fixed re­sources mostly don’t gen­er­ate new risks. At the worst, a weapons en­g­ineer may be more pro­duc­tive in one coun­try than an­other, but the num­ber of skil­led weapon en­g­ineers is not in­creased with­out ad­di­tional in­ter­ven­tions such as es­pi­onage. As hav­ing the best minds is a key con­straint on de­vel­op­ing highly sen­si­tive tech­nolo­gies, a party that chooses not to en­gage in cred­ible virtue sig­nal­ing, in fa­vor of rac­ing in an­other di­rec­tion may lose ad­van­tage in the long run which makes this sort of com­pe­ti­tion po­ten­tially sta­ble and may par­tially offset other more harm­ful races (both by cost­ing re­sources, and by caus­ing brain drain which may pre­vent cer­tain coun­tries from be­ing ca­pa­ble of arms races in the first place).

So far the frame­work above im­plies a model of states as unified ra­tio­nal ac­tors, while in re­al­ity they con­sist of coal­i­tions which are not always al­igned to­ward the cen­tral goals of a state, but rather to­ward those of the smaller groups and in­di­vi­d­u­als. In democ­ra­cies, broad dis­tri­bu­tion of power helps keep the gov­ern­ment more al­igned with its pop­u­la­tion, but re­duces the abil­ity to co­or­di­nate on con­tentious is­sues and to be­have strate­gi­cally in the long run, hence there may be low hang­ing fruit where democ­ra­cies can be­have more strate­gi­cally at low cost. On the other hand, in less demo­cratic sys­tems, top down force can re­solve in­ter­nal co­or­di­na­tion prob­lems and get things done but, this power is not nec­es­sar­ily al­igned with the pop­u­la­tion, and can be pre­car­i­ous since when some loy­alties switch, a new dom­i­nant coal­i­tion may de­velop ab­solute power. Over­all this im­plies both that high skill im­mi­grants are likely to pre­fer to live in demo­cratic sys­tems, and that such sys­tems are less able to abuse any ad­van­tages over other coun­tries that they gain from such im­mi­grants.

Part 2: The arguments

In bul­let point form be­low, we dis­cuss the po­ten­tial benefits of fur­ther liber­al­iz­ing im­mi­gra­tion policy for in­di­vi­d­u­als of high skill or po­ten­tial, and ways to ad­dress the po­ten­tial risks of such policy changes.

Strate­gic benefits of high skill im­mi­gra­tion for the far fu­ture:

(click each sec­tion be­low for jus­tify­ing bul­let points)

With high skill im­mi­gra­tion less re­strained, the nat­u­ral co­a­lesc­ing of top tal­ent in­creases differ­en­tials in progress be­tween lead­ers and fol­low­ers: some coun­tries would get more of an ad­van­tage in AI and other trans­for­ma­tive tech­nolo­gies than oth­ers, which re­duces com­pet­i­tive pres­sure, al­low­ing more re­sources to be al­lo­cated to safe de­sign and safe de­ploy­ment of new tech­nolo­gies rather than to just stay­ing ahead in tech races.

  • Peo­ple vote with their feet: the places with the sys­tems that most cred­ibly sig­nal re­spect for their cit­i­zens in­ter­ests dis­pro­por­tionately gain in an equil­ibrium with in­creased high skill/​high po­ten­tial im­mi­gra­tion be­tween coun­tries.

    • Dur­ing the Man­hat­tan pro­ject, many physi­cists de­liber­ately chose to work for the United States since they did not want the Nazis to get the atomic bomb. (e.g. Klaus Fuchs, John von Neu­mann, and var­i­ous other sci­en­tists.) Coun­tries that offer the great­est re­spect for the rights of their cit­i­zens, can cred­ibly offer greater se­cu­rity for in­tel­lec­tu­als and re­searchers than those which hold poli­ti­cal pris­on­ers.

Higher mo­bil­ity for the highly skil­led al­lows them to move to the sys­tems that are most in their in­ter­ests: the al­tru­is­tic can move to sys­tems where they have greater lev­er­age, while the less al­tru­is­tic can go where they can ex­pect to per­son­ally benefit the most. As the best run sys­tems will tend to both have more eco­nomic op­por­tu­ni­ties and larger gov­ern­ment bud­gets, both groups are likely con­cen­trate in the same coun­tries, leav­ing gov­ern­ments where ra­tio­nal al­tru­ists can gain in­fluence more re­sources at their dis­posal due to higher tax rev­enues from im­mi­gra­tion driven in­creases in eco­nomic growth.

  • While peo­ple have many in­ter­ests, and lo­cal in­cen­tives can de­ter­mine the types of be­hav­ior oth­er­wise self­ish or oth­er­wise al­tru­is­tic peo­ple would take, com­pet­i­tive pro­cesses will most re­ward effec­tive op­ti­miz­ers, so it is im­por­tant to ac­count for the in­cen­tives that cause peo­ple to seek the high­est lev­er­age po­si­tions in gov­er­nance struc­tures.

  • In­tel­li­gent lo­cal op­ti­miz­ers (peo­ple who’d like to be­come per­son­ally rich, fa­mous, (lo­cally) pow­er­ful, etc.) are more likely to head to the pri­vate sec­tor in the US, while in coun­tries with less in­clu­sive in­sti­tu­tions they com­pete for con­trol of the trea­sury, nat­u­ral re­sources, and aid from in­side the gov­ern­ment (e.g. re­source curse) dis­plac­ing global op­ti­miz­ers. Even as China’s econ­omy de­vel­ops, cre­at­ing more op­por­tu­ni­ties to grow wealth out­side of gov­ern­ment, there is still a stronger in­cen­tive for lo­cal op­ti­miz­ers to head to gov­ern­ment since the dom­i­nant strat­egy for lo­cal op­ti­miz­ers in less in­clu­sive sys­tems is to aim more di­rectly for power.

    • While plenty of low level pub­lic ser­vants are in­com­pe­tent and many of the al­tru­is­ti­cally ori­ented may not be ra­tio­nal, nonethe­less it is those more ra­tio­nal and al­tru­is­tic that dis­pro­por­tionately make it to the high­est lev­er­age civil ser­vice po­si­tions in coun­tries that have bet­ter eco­nomic op­por­tu­ni­ties for the highly skil­led out­side gov­ern­ment.

      • This may seem hard to jus­tify, but broadly there are many peo­ple with high earn­ing ca­pac­ity who nev­er­the­less chose to go into lower paid policy po­si­tions to im­prove the qual­ity of de­ci­sions that benefit oth­ers.

        • Peo­ple with pres­ti­gious quan­ti­ta­tive PhDs work­ing as run­ning gov­ern­ment agen­cies/​ac­tivi­ties, etc.

        • Yale/​Har­vard JD’s work­ing as se­nior ca­reer con­gres­sional staffers on com­mit­tees such as the Emerg­ing Threats and Ca­pa­bil­ities sub­com­mit­tee, for 150k/​year in­stead of in pri­vate prac­tice for $600K to $1+ mil­lion/​year.

      • Ba­si­cally, the pri­vate sec­tor is more likely to pay peo­ple a what they can ne­go­ti­ate for, while the pub­lic sec­tor is more likely to over­pay low skill work and un­der pay high skill work.

  • In con­trast, due to cor­rup­tion, the high­est lev­er­age po­si­tions in non-demo­cratic coun­tries typ­i­cally are the most com­peted for, as that lev­er­age can be mon­e­tized. Even for dic­ta­tors who pre­fer for there not to be cor­rup­tion, their own grasp of power may de­pend on it.

    • This is why it is im­por­tant to main­tain strong norms against nepo­tism and cor­rup­tion, so al­tru­is­tic ra­tio­nal ac­tors gain ad­van­tage and pre­vent lo­cal op­ti­miz­ers from be­ing in­cen­tivized to go into gov­ern­ment.

          • With im­perfect in­for­ma­tion, it’s much more difficult for lo­cal op­ti­miz­ers/​defec­tors to co­or­di­nate. In the non­tech­ni­cal world, ev­ery­thing is some­what am­bigu­ous so groups with lots of defec­tors are dis­ad­van­taged.

          • All else be­ing equal, this forces gov­ern­ments to take ac­tions to in­crease the ad­van­tage of global op­ti­miz­ers over lo­cal op­ti­miz­ers in­side their gov­er­nance sys­tems in or­der to re­main com­pet­i­tive: ei­ther bait­ing lo­cal op­ti­miz­ers away with bet­ter pri­vate sec­tor pay, or by dis­pro­por­tionately pun­ish­ing defec­tion and lack of virtue in the pub­lic ser­vices.

In­creased free­dom of move­ment for skil­led work­ers in­creases their lev­er­age vis-a-vis states: re­sult­ing in net brain gain for those states which most cred­ibly sig­nal pro-so­cial poli­cies.

  • For ex­am­ple Rus­sian billion­aires store their money in the US since the US has more cred­ibly sig­naled that wealth will not be un­justly con­fis­cated and poli­ti­cal dis­si­dents will not be jailed.

  • This also forces states to com­pete at in­creas­ing eco­nomic op­por­tu­nity for high skill im­mi­grants and sig­nal­ing the pres­ence of in­sti­tu­tions in which global op­ti­miz­ers can gain in­fluence.

      • In the case sci­en­tists and en­g­ineers do want to make weapons, it makes sense for coun­tries to avoid de­port­ing en­g­ineers, and to be cau­tious about de­vel­op­ing peo­ple’s tal­ents for strate­gi­cally rele­vant tech­nolo­gies when there is es­pi­onage risk.

        • A par­tic­u­larly sad ex­am­ple is the story of Quan Xue­sen. Not only did the US deny him cit­i­zen­ship af­ter his many achieve­ments in rock­etry, but also ac­cused him of be­ing a com­mu­nist and try­ing to steal clas­sified in­for­ma­tion (even though the doc­u­ments he was found with were his own work). After re­vok­ing his se­cu­rity clear­ance and putting him un­der 5 years of house ar­rest, the US sent him back to China, where he be­came the di­rec­tor of their nu­clear weapon and bal­lis­tic mis­sile de­vel­op­ment pro­gram, con­tribut­ing to the world record for fis­sion to fu­sion de­vel­op­ment speed.

          • Via such poor treat­ment, and US defense pos­ture at the time, is seems likely that Quan was con­vinced he needed to defend his own coun­try: this stresses the im­por­tance of not be­ing overly bel­liger­ent when tak­ing defen­sive mea­sures against po­ten­tial es­pi­onage.

    • Lastly, one can eas­ily imag­ine that top Rus­sian and Chi­nese stu­dents if pro­vided the op­por­tu­nity, would prob­a­bly have hap­pily moved to the west in 1940. If they had, ev­ery­one would have been bet­ter off. Bloody wars all around the de­vel­op­ing world could have been avoided. This in turn may have sped up eco­nomic de­vel­op­ment in Asia, and Latin Amer­ica.

In­creased ease of reg­u­lat­ing risky technologies

  • In­ter­na­tional col­lab­o­ra­tion is hard since each party can sim­ply act in its own in­ter­est and defect. In­side one state, the state owns a monopoly of force and can im­pose re­straints and reg­u­la­tions that are oth­er­wise im­pos­si­ble. It is easy to imag­ine that miti­gat­ing cli­mate change would be eas­ier if coun­tries didn’t lose var­i­ous eco­nomic and strate­gic ad­van­tages by co­op­er­at­ing.

    • While we think this ar­gu­ment is mostly rele­vant with re­spect to the reg­u­la­tion of ar­tifi­cial in­tel­li­gence, it may also be eas­ier to ne­go­ti­ate and en­force guidelines for ge­netic en­g­ineer­ing within a coun­try, or be­tween two coun­tries (such as the US and China) than among all states. For more on the poli­tics of ge­netic en­g­ineer­ing, this pod­cast with the Fu­ture Strate­gist is use­ful in lay­ing out some po­ten­tial sce­nar­ios of in­cen­tives and out­comes in the fu­ture.

      • It also seems rea­son­able to ex­pect that par­ents would be more in­clined to chose more pro-so­cial traits for their offspring if they live in well off coun­tries with (rel­a­tively) strong so­cial safety nets than if they don’t.

        • That is, par­ents in rich coun­tries would prob­a­bly be much more in­clined to give their offspring pro-so­cial traits that in­crease per­sonal hap­piness and so­cietal well-be­ing while those in more malthu­sian en­vi­ron­ments would prob­a­bly pre­fer traits that more purely in­crease offspring’s fit­ness, since with­out so­cial safety nets there’s no limit to how far you can fall. If those with greater means move to coun­tries with more pro-so­cial norms, the odds of hu­man ge­netic en­g­ineer­ing go­ing well im­prove.

        • Scan­d­i­na­vian coun­tries may have more pro-so­cial norms than the US, but over­all the US gen­er­ally seems to have more pro-so­cial norms than the global av­er­age.

          • More­over, for high skill im­mi­grants who are go­ing to be par­ti­ci­pat­ing in high sen­si­tivity parts of the econ­omy, elite so­cial norms are more likely to be rele­vant. In the US, the cit­i­zens of cities like Bos­ton can out­com­pete even Copen­hagen in co­op­er­a­tion game ex­per­i­ments, though this ad­van­tage goes away when the abil­ity to pun­ish non-co­op­er­a­tion is re­moved.

        • The above ar­gu­ment ap­plies less in situ­a­tions where hu­man ge­netic en­g­ineer­ing is broadly available rather than to just a nar­row eco­nomic elite. How­ever even if hu­man ge­netic en­g­ineer­ing be­comes broadly available, early adopters will likely be elites. Given the in­fluence of start­ing con­di­tions on long-run out­comes, we sus­pect that in­fluence on elites could have an out­size im­pact on the long run fu­ture: for ex­am­ple if ge­netic en­g­ineer­ing were to yield nu­mer­ous peo­ple with Von Neu­mann lev­els of in­tel­li­gence.

Higher salaries for white col­lar work in the US dis­cour­age riskier but po­ten­tially high pay­ing shady black hat(ish) in­dus­tries.

  • De­creas­ing bar­ri­ers to im­mi­gra­tion for Is­raelis in par­tic­u­lar seems valuable with re­gards to miti­gat­ing the spread of cy­ber-weapons.

    • Wages for skil­led pro­gram­mers are lower in Is­rael than in Sili­con valley, so re­mote work such as black hat(ish) in­dus­tries are rel­a­tively more ap­peal­ing when in­di­vi­d­u­als are ge­o­graph­i­cally con­strained.

In­creased al­ign­ment be­tween the elites of differ­ent coun­tries via over­lap­ping and in­ter­laced so­cial networks

  • By mak­ing it eas­ier for the best from other coun­tries to live in the coun­try which has the most to offer them, a larger por­tion of their so­cial ties and in­vest­ments will be in the same place, which fur­ther re­duces the odds of war be­tween both ris­ing pow­ers and ex­ist­ing great pow­ers, as well as re­duc­ing the odds of re­gional con­flict. (eg. Taiwan and China, In­dia and Pak­istan, etc.)

Pos­si­ble ob­jec­tions:

(click each sec­tion be­low for sup­port­ing de­tails and our counter ar­gu­ments)

In­creased high skill im­mi­gra­tion may speed up the de­vel­op­ment of high risk technologies

  • As con­cen­tra­tion of hu­man cap­i­tal al­lows faster in­for­ma­tion flow, and there­fore faster in­tel­lec­tual progress, it is pos­si­ble that un­safe tech­nolo­gies will progress faster than our abil­ity to cre­ate defen­sive tech­nolo­gies or cre­ate safe­guards.

    • The gen­eral counter ar­gu­ment to this po­si­tion is that in­creased in­for­ma­tion flow also makes it eas­ier to spread ideas of safety, and to ac­com­plish dis­pro­por­tionate work on defen­sive tech­nolo­gies, so it is difficult to say that this will be de­ci­sive in one di­rec­tion or an­other. For ar­tifi­cial in­tel­li­gence, both Deep­Mind and OpenAI have safety teams, and broad con­nec­tions to the effec­tive al­tru­ism com­mu­nity, while it is still un­known how other com­pa­nies and coun­tries would com­pete if they be­came more dom­i­nant in AI re­search.

In­creased high skill im­mi­gra­tion may in­crease risks from espionage

  • With mod­ern es­pi­onage, there is a lot of data which can be of value to states, so in ad­di­tion to hav­ing small num­bers of high fidelity spies, send­ing coun­tries can try to en­courage large num­bers of their de­part­ing cit­i­zens to make plau­si­bly de­ni­able efforts and small leaks to benefit their home coun­tries, es­pe­cially with grad stu­dents. With such in­com­plete in­for­ma­tion, marginal leaks can some­times have in­creas­ing rather than de­creas­ing marginal value to spy­ing states, so the suc­cess of es­pi­onage may ac­tu­ally in­crease with in­creased im­mi­gra­tion.

  • Es­pi­onage is likely one of the great­est risks from high skill im­mi­gra­tion, as it has been one of the best means for coun­tries be­hind in an arms race to catch up di­rectly, and may also lead to un­der­stand­ing that in­cen­tivizes tech rac­ing.

    • One counter is that the sub­set of mu­tual es­pi­onage which can’t re­sult in the de­ploy­ment of new sys­tems would be good as it would re­duce strate­gic un­cer­tainty, and there­fore the odds of a mis­calcu­la­tion lead­ing to war. When coun­tries un­der­stand each other’s ca­pa­bil­ities and mo­ti­va­tions, they can’t gain ne­go­ti­at­ing ad­van­tage by ly­ing about their abil­ities or will­ing­ness to fight, which clar­ifies the bar­gain­ing range for any given dis­pute, al­low­ing more op­tions to avoid war.

  • One op­tion for han­dling es­pi­onage is to have rather ex­treme in­for­ma­tion se­cu­rity mea­sures. Air gaps are a blunt way to in­crease cy­ber-se­cu­rity and his­tor­i­cally se­cret cities played a role in var­i­ous coun­tries’ weapon pro­grams.

  • Other forms of cy­ber se­cu­rity via en­cryp­tion, se­cure com­mu­ni­ca­tion lines, etc. would also be nec­es­sary as cut­ting edge re­search can be difficult with­out ac­cess to the large vol­umes of re­search that can be found on the in­ter­net and with differ­ent re­search groups.

    • For ex­am­ple, though many pro­grams are clas­sified, DARPA and IARPA use a great deal of open col­lab­o­ra­tion with pub­lic and pri­vate com­pa­nies to ac­cel­er­ate the rate at which they can do re­search: open­ness fa­cil­i­tates in­for­ma­tion flow which is in­stru­men­tal in the speed of progress in re­search. Pro­grams which have to be fully shrouded in se­crecy, all else be­ing equal, effec­tively get less feed­back and eval­u­a­tion, so progress is slower.

  • To de­crease the odds of es­pi­onage risk, it is likely bet­ter to seek younger im­mi­grants who are of high promise so they can go though many of their for­ma­tive ex­pe­riences and de­velop many of their life long friends and so­cial net­works dur­ing un­der­grad­u­ate ed­u­ca­tion. With more so­cial ties in their new coun­tries of res­i­dence, there may be in­creased af­finity for their new coun­tries of res­i­dence which would not oc­cur dur­ing more iso­lated grad school pro­grams.

Devel­op­ing top re­searchers who then re­turn home speeds up arms races by re­duc­ing the gap be­tween tech lead­ers and followers

“Western cul­ture” would be “de­graded” (na­tivist back­lash)

  • Highly tal­ented im­mi­grants typ­i­cally end up in ur­ban melt­ing pots so it seems un­likely cul­tural con­ser­va­tives will ac­tu­ally no­tice much of a differ­ence (com­pet­ing for differ­ent job types, liv­ing in differ­ent places, defy­ing stereo­types that cul­tural con­ser­va­tives fit to lower skill im­mi­grants, etc.)

  • If you are se­lect­ing dis­pro­por­tionately for in­tel­li­gent im­mi­grants, it seems es­pe­cially un­likely that cul­ture would get worse.

Too many “eggheads” in one basket

  • Though a lot of the world’s in­tel­lec­tual cap­i­tal may fur­ther con­cen­trate due to high skill im­mi­gra­tion, this does not nec­es­sar­ily in­crease the risk of los­ing that in­tel­lec­tual cap­i­tal due to the bet­ter in­cen­tives to co­op­er­ate. Many peo­ple will be in fewer places, but those places will in­evitably be­come bet­ter defended.

  • How­ever there would be risk that if a dic­ta­tor took over the US (or what­ever coun­try ends up with the great­est con­cen­tra­tion of hu­man cap­i­tal), this could cre­ate a long term sta­ble to­tal­i­tar­ian world gov­ern­ment.

    • While hu­man cap­i­tal is likely not enough for this sce­nario to oc­cur on its own, ad­di­tional fac­tors such as the rise of new rad­i­cal ide­olo­gies, grad­ual in­crease of ex­ec­u­tive power, and/​or the de­vel­op­ment of de­ci­sive strate­gic ad­van­tage via new weapons ad­vances, surveillance, and/​or ar­tifi­cial in­tel­li­gence, could be enough to com­plete the tran­si­tion to a less plu­ral­is­tic gov­ern­ment with sig­nifi­cant power over the rest of the world. Even within democ­racy, the later tech­nolo­gies could en­able bet­ter con­trol of pub­lic choice with tar­geted pro­pa­ganda and in­fluence.

    • For­tu­nately not all the best/​most strate­gi­cally rele­vant jobs are con­cen­trated in one coun­try, e.g. Google Brain and Deep­mind are in the US and UK re­spec­tively. The best jobs tend to be con­cen­trated in poli­ti­cal sys­tems that best re­spect hu­man prefer­ences/​have the most in­clu­sive in­sti­tu­tions. If to­tal­i­tar­ian regimes arise, they would have to re­strain la­bor mo­bil­ity, and keep multi­na­tional com­pa­nies happy to not suffer cap­i­tal flight, which par­tially de­ters the worst poli­cies.

    • Gen­er­ally speak­ing, it is eas­ier to copy than to in­no­vate: even with much hu­man cap­i­tal con­cen­trated, the eco­nomic growth rate a coun­try with a huge hu­man cap­i­tal ad­van­tage might not ex­ceed the growth rates of less de­vel­oped coun­tries, so it may be hard for a to­tal­i­tar­ian regime to stay ahead of other coun­tries in a long term sta­ble man­ner.

Ero­sion of trust and good so­cial norms

Peace prospects are not nec­es­sar­ily in­creased by in­ter-socialization

  • Though oc­ca­sion­ally in­te­gra­tion can in­crease con­flict, this pri­mar­ily seems to be the case more so in Malthu­sian en­vi­ron­ments, where groups benefit at the ex­pense of each other.

    • This is ul­ti­mately one more ar­gu­ment for fa­cil­i­tat­ing in­ter-so­cial­iza­tion while we ex­ist in a time of plenty.

In an equil­ibrium with more mo­bil­ity for the elites of differ­ent coun­tries, the gov­er­nance of some coun­tries may worsen, lead­ing to con­flict.

  • Greater mo­bil­ity also fa­cil­i­tates bet­ter in­for­ma­tion flow, which can spread bet­ter ideas for gov­er­nance be­tween coun­tries.

  • Per­mit­ting elites to spread their as­sets across mul­ti­ple coun­tries, re­sults in those elites hav­ing a vested in­ter­est in the coun­tries of their in­vest­ments not fight­ing amongst them­selves.

Re­ac­tion to eas­ier high skill im­mi­grant cit­i­zen­ship in democ­ra­cies could cause more au­thor­i­tar­ian coun­tries to fur­ther re­strict their own cit­i­zens from leav­ing.

  • In the cur­rent equil­ibrium, smart stu­dents can leave their coun­tries, and of­ten come back due to the difficulty of be­com­ing a cit­i­zen in coun­tries such as the US which benefits send­ing coun­tries.

  • In the equil­ibrium where peo­ple stay in the US, there may be benefits from re­mit­tances, but send­ing gov­ern­ments mostly won’t de­rive the strate­gic benefit of get­ting back their best and bright­est with more ed­u­ca­tion.

  • His­tor­i­cally, effec­tively re­strict­ing the abil­ity to leave a coun­try is a very hos­tile move by gov­ern­ments, which sig­nals ill in­tent. Coun­tries that do this seem likely to have an dis­ad­van­tage at at­tract­ing im­mi­grants since im­mi­grants would likely not want to be­come per­ma­nently trapped in their new coun­try.

As­sum­ing long term AI al­ign­ment prob­lems are solved, but that de­ployed AI sys­tems aren’t al­igned with all hu­mans, a world with mul­ti­po­lar out­comes may be prefer­able to unipo­lar ones, since hu­man co­op­er­a­tion and mu­tual benefit will no longer be the way to wield the most power.

  • When hu­man in­tel­li­gence no longer offers a lot of marginal value with re­spect to solv­ing prob­lems or gain­ing ad­van­tage, then there is less in­cen­tive forc­ing so­ciety to be more in­clu­sive and the few might be able to benefit at the ex­pense of the many in a long term sus­tain­able way: un­der­min­ing long term trends to­ward larger more in­clu­sive so­cieties be­ing stronger.

    • This ar­gu­ment fails be­cause this is true re­gard­less of the po­lar­ity of the out­come. In a sin­gle­ton situ­a­tion, those with power are more able to do what they wish and make the world how they think it should be, in a mul­ti­po­lar situ­a­tion there is less room for ma­neu­ver and more in­cen­tive to put all re­sources to­ward out­com­pet­ing oth­ers to en­sure sur­vival or dom­i­nance: leav­ing far less po­ten­tial sur­plus value for other hu­mans. In the mul­ti­po­lar situ­a­tion with ad­vanced AI, tak­ing care of peo­ple who do not have high strate­gic value could be a li­a­bil­ity, so com­pet­ing par­ties might be less al­tru­is­tic.

      • Though it may gen­er­ally be benefi­cial to have good in­sti­tu­tion de­sign and many ra­tio­nal al­tru­ists in gov­ern­ment, these ar­gu­ments im­ply it may be es­pe­cially more im­por­tant than usual in the fu­ture, as tech­nol­ogy grad­u­ally re­duces the la­bor value of most peo­ple, and there­fore their di­rect value to gov­er­nance struc­tures.

Most democ­ra­cies already have fa­vor­able im­mi­gra­tion poli­cies for high skill im­mi­grants, and even in the US there’s mul­ti­ple visas high skil­led per­sons can take ad­van­tage of.

  • The prob­lem with US high skill im­mi­gra­tion isn’t that peo­ple can’t go to the US at all, nor is it that per­ma­nent res­i­dence sta­tus is im­pos­si­ble to ob­tain for de­ter­mined in­di­vi­d­u­als. It’s that there are nu­mer­ous bar­ri­ers and un­cer­tain­ties to per­ma­nent res­i­dency, while the to­tal is limited, which keeps lev­els of net high skill im­mi­gra­tion be­low the global op­ti­mum.

High skill im­mi­gra­tion is low lev­er­age: His­tor­i­cally, the im­mi­gra­tion shock af­ter the col­lapse of the Soviet Union didn’t no­tice­ably ac­cel­er­ate sci­ence in the US, so the im­pact is un­clear.

  • The lack of ac­cel­er­a­tion of sci­ence fol­low­ing the high skill im­mi­gra­tion shock to the US is not nec­es­sar­ily bad news: it may also im­ply that fu­ture shocks won’t ac­cel­er­ate risky tech­nolo­gies, that re­search fund­ing is a more fun­da­men­tal con­straint, or that other sec­tors of the econ­omy are bet­ter at ab­sorb­ing high skill im­mi­grants.

    • Fur­ther em­i­gra­tion likely de­cel­er­ated progress for po­ten­tially risky tech­nolo­gies in the former USSR, which is a net re­duc­tion of risk: there is less in­cen­tive for the US gov­ern­ment to en­gage in an arms race if there is no one to race.

Con­clu­sion:

To sum­ma­rize, we think the strongest fu­ture fo­cused rea­sons to make cit­i­zen­ship eas­ier for high skill/​high po­ten­tial im­mi­grants to ob­tain are that it would likely in­crease the strate­gic ad­van­tage of sys­tems that bet­ter ap­prox­i­mate hu­man val­ues, give ra­tio­nal al­tru­ists in­creased ad­van­tage in gov­er­nance, re­duce the odds of catas­trophic wars, and re­duce the odds of un­reg­u­lated work and arms races on dan­ger­ous tech­nolo­gies. The best po­ten­tial coun­ters seem to be that we should miti­gate the risk of a to­tal­i­tar­ian global regime, that the in­creased in­for­ma­tion flow that comes from col­lect­ing the best and bright­est could lead to faster de­vel­op­ment of dan­ger­ous tech­nolo­gies if not con­trol­led, and that via es­pi­onage and re­turn­ing home, such progress on dan­ger­ous tech­nolo­gies could pro­lifer­ate to the rest of the world. We think that these coun­ters could po­ten­tially be ad­dressed by in­creased efforts at strength­en­ing in­sti­tu­tions, fund­ing differ­en­tial tech­nolog­i­cal progress in defen­sive tech­nolo­gies such as cy­ber­se­cu­rity, and spread­ing ideas of safety en­g­ineer­ing among those work­ing in po­ten­tially dan­ger­ous fields. Since the benefits of in­creased high skill im­mi­gra­tion ap­pear difficult to gain with­out ac­tu­ally hav­ing it, and the down­sides can be coun­tered, a well calcu­lated liber­al­iza­tion of high skill im­mi­gra­tion is likely the best path for­ward.

There are many meth­ods by which one could at­tempt to miti­gate arms/​tech race type dy­nam­ics which could be threat­en­ing to the far fu­ture of hu­man­ity. While ex­plor­ing ways to achieve agree­ment and spread knowl­edge of risks, we should also seek ways to make the best of the situ­a­tions we find our­selves in should global co­or­di­na­tion fail. We provide im­prov­ing high skill im­mi­gra­tion policy as an ex­am­ple of a way in which a state ac­tor could unilat­er­ally make an arms race type of situ­a­tion less harm­ful, while also act­ing in their own in­ter­est. There are likely other in­ter­ven­tions which fall into a similar cat­e­gory that are wor­thy of in­ves­ti­ga­tion.

In a fu­ture post, we will dis­cuss the cur­rent bar­ri­ers to a bet­ter high skill im­mi­gra­tion policy for the US, where it cur­rently suc­ceeds, how much room there is for im­prove­ment, and why such im­prove­ments may be tractable across party lines.


Notes: The gen­eral out­look of this post was par­tially in­formed by the fol­low­ing books: