On December 18, 1944, Moe Berg — a Princeton graduate, Major League baseball star, and Office of Strategic Services operative — discreetly took his seat in a cramped conference room in Zurich. Masquerading as a studious physics graduate, the hulking Harlem-born Jew listened attentively to the keynote speaker, the Nobel-prize winning German physicist Werner Heisenberg. His instructions were straightforward. Should Heisenberg say anything indicative of the success of the covert Nazi nuclear program with which he was intimately involved, Berg was to whip out his carefully concealed .45 caliber pistol and summarily execute him in the midst of the auditorium.
As Berg sat in the audience and patiently listened to Heisenberg lecture, the American detected no note of triumphalism. Nor did he pick up on any allusion, however cryptic, to some mysterious new Wunderwaffe that might change the course of the war. Sidling up to Heisenberg after the conference, the New Yorker made small talk as they strolled through Zurich’s darkened, ice-rimed streets. As they spoke, Heisenberg voiced defeatist sentiments, glumly opining that the war was all but lost for Germany. Berg decided to spare Heisenberg’s life. This choice was proven correct: the Allies would later discover that, due to a host of bureaucratic pathologies and scientific shortcomings, Germany was nowhere close to developing a nuclear bomb.
Close to eight decades later, another leading nuclear scientist suffered a grimmer fate. In November 2020, Mohsen Fakhrizadeh, a revered figure in his native Iran and sometimes described as “Iran’s Oppenheimer,” was ambushed and slain on a country road outside Tehran. The mode of execution — via an AI-assisted, remote-controlled machine gun — was especially ingenious, and although Israeli security services made no public claim of responsibility, their involvement was in little doubt. Indeed, Fakhrizadeh was simply the latest victim in a long list of Mossad-planned assassinations of Iranian atomic scientists over the years. Israel, long surrounded by hostile states and with little strategic depth, has often relied on assassination as a means of delaying or forestalling adversarial scientific developments — particularly (but not only) with regard to nuclear weapons technology.
In both these cases, the assassination of lead scientists was considered a worthwhile and morally defensible policy option for one simple, overarching reason. Security managers in both countries understood that throughout history certain key, talented individuals have played a disproportionate role in driving technological change. Indeed, barely a few months after the Zurich conference, Washington, under the aegis of Operation Paperclip, had already begun its highly classified — and deeply morally controversial — process of feverishly exfiltrating more than 1600 of Adolf Hitler’s most accomplished technologists. This imported intellectual firepower would then go on to play a lead role in advancing U.S. space and rocketry technology, along with its chemical and biological weapons programs. The Soviet Union was equally, if not more, aggressive in its efforts to capture former Nazi talent. Thus, during one characteristically sweeping operation, notes a historian of the period, “Soviet troops rounded up about three thousand German scientists, engineers, craftsmen, and other technical specialists, along with their families and possessions, and placed them on trains heading east. No explanations were given nor were excuses or objections allowed.” By the late 1940s, hundreds of German scientists had been dragooned into working at Institut A — a secret facility near Sukhumi in the Soviet republic of Georgia — on Moscow’s nuclear program.
As we enter a new, and potentially equally consequential arms race, fueled by rapid advances in the field of AI, this troubled history serves as a useful reminder of the importance of the human dimension of military-technological rivalry. In the global competition to develop artificial general intelligence, keeping close track of compute clusters, semi-conductor export controls, and large language models will prove critical. Far more consequential, however, will be developing a carefully tailored and ruthlessly pragmatic approach toward AI talent management.
This strategy can be structured along three axes — which are not so much distinct as deeply interactive and mutually reinforcing in nature: preserving homegrown AI talent, capturing foreign AI talent, and targeting adversarial AI talent.
Preserving Homegrown AI Talent
The first overriding priority should be to preserve, strengthen, and safeguard a domestic reservoir of AI talent. While America continues to exert a strong gravitational pull on foreign talent, its domestic bench of skilled or highly specialized workers has become too shallow to properly compete with its most formidable rival, China. In international assessments, American’s K-12 education system continues to lag behind those of many other industrialized nations, especially in science, technology, engineering, and mathematics. As the U.S. National Security Commission on Artificial Intelligence observed in its 2021 report, Washington is confronted with a daunting workforce deficit, and “AI and digital talent is simply too scarce in the United States.” Although software engineers are in less high demand than two or three years ago, they are still sorely needed across the national security enterprise, with some describing the current science, technology, engineering, and mathematics shortage in the defense-industrial base as “another Sputnik moment we can’t afford to ignore.” Meanwhile, the domestic talent gap with China continues to widen, especially in the field of AI. Having invested massively in undergraduate and graduate AI programs, it is estimated that Chinese universities now churn out approximately 50 percent of the world’s top AI researchers, whereas the United States only produces about 18 percent. It is worth noting, in passing, that this phenomenon is hardly confined to the field of artificial intelligence — indeed by next year China should be generating twice as many Ph.D.s in science, technology, engineering, and mathematics as the United States.
Moreover, in a marked departure from relatively recent workflow patterns, many of these domestically educated Chinese AI scientists are now choosing to remain in China rather than seek opportunities overseas — thus further contributing to the dynamism of China’s flourishing domestic AI industry. Some of this can clearly be attributed to Beijing’s sedulous efforts to court and lure Chinese expatriate talent back to the home country, with returning scientists, colloquially known as “sea turtles,” offered hefty bonuses and research grants in exchange for establishing their labs within China.
American security managers are clearly aware of this troubling and growing imbalance. The recently issued National Defense Industrial Strategy thus emphasizes the importance of sizably investing in recruitment and upskilling programs, and of developing new forms of public-private partnership. Meanwhile, there have been intensified efforts in some quarters to launch a new nationwide science and technology talent strategy named after the National Defense Education Act of 1958, an Eisenhower administration initiative launched in response to the Sputnik satellite launch of 1957. This is all for the good, as are various calls to improve the quality of America’s K-12 education system. All of these proposed initiatives and reforms, however, will inevitably take years, if not decades, to fully bear fruit and will therefore provide no short- to medium-term solution to America’s widening AI talent gap with China.
Two immediate lines of effort should therefore be pursued.
The first consists in more aggressively shielding America’s domestic AI talent from what can best be described as adversarial predation. In the current context, this means working to prevent Chinese technology groups entrenched in Silicon Valley from poaching lead generative AI experts from companies such as OpenAI, Meta, or Google. After all, it makes little sense to restrict high-end chip exports to China, or to limit U.S. venture capital flows into Chinese firms, while turning a blind eye to Beijing’s increasingly transparent attempts to acquire an intellectual backdoor to U.S. technology through the recruitment of its best engineers. In recent years, revelations that the Chinese air force has been intensifying its efforts to recruit former Western fighter pilots to better train its airmen have generated substantial controversy, with many (rightly) questioning the moral judgment of retired U.S. and allied military personnel willing to actively enhance an adversary’s warfighting capabilities in exchange for a fat paycheck. In an era of revived Sino-American competition in which cutting edge AI research has immediate and evident military ramifications, a similar form of public opprobrium should be attached to those willing to advance Beijing’s strategy of civil-military fusion by working for Chinese technology groups such as ByteDance, Alibaba, or Meituan. Indeed, it is not inconceivable that sometime in the future, a new executive order may need to be issued — one that formally bans any U.S. citizen from working for a Chinese AI company. Given the stakes of the competition, and the degree of methodical deliberateness with which the Chinese state has been engaging in such foreign talent-poaching, this author would be in favor of such a ban. Other countries currently grappling with this challenge — such as South Korea — may eventually need to follow suit.
In addition to better monitoring and blunting these targeted recruitment efforts, U.S. government agencies may also need to consider — as we shall discuss in more depth later — preparing for bleaker scenarios, ones in which they need to guarantee the physical safety of the nation’s most brilliant minds in the field of AI. Indeed, in the not-too-distant future, much like nuclear scientists, lead AI developers may come to be seen as legitimate targets within the context of an increasingly existential struggle to attain artificial general intelligence.
The second main answer to America’s AI deficit resides in refining a more effective strategy for talent importation — principally through the design of a tailored and bipartisan immigration policy. Indeed, the United States needs large numbers of highly skilled immigrants — and preferably large numbers of naturalized highly skilled immigrants who can take the oath of allegiance and serve the U.S. government — to compete and prevail in the 21st-century artificial general intelligence race.
Capturing Foreign AI Talent
In 1955, one of the world’s most talented experts in rocketry and jet propulsion was unceremoniously expelled from the United States. The Chinese-born Caltech prodigy had played a vital role during World War II and served his host nation with pride and distinction — taking part in the Manhattan Project, serving on the U.S. government’s Scientific Advisory Board, and then deploying to Germany to interrogate lead Nazi scientists such as Werner Von Braun. Falsely accused of espionage during the paranoia-drenched period of the Great Red Scare, the lab director sailed out of the port of Los Angeles for the country of his birth, his wife and two U.S.-born children in tow, bitterly vowing that he would never again set foot in the United States. That man was Qian Xuesen, who would then go on to become the founding father of communist China’s ballistic missile and space programs. And while his story is not, perhaps, as well-known as it should be, it has come to be seen as a textbook example of how overly exclusionary immigration policies can prove eminently counterproductive, with former Secretary of the U.S. Navy Dan Kimball later griping that deporting Qian “was the stupidest thing this country ever did.”
And indeed, at first glance, it might seem superfluous, or even unnecessary, to remind policymakers of the enormous technological, economic, and national security advantages the United States has reaped from more enlightened immigration policies in the decades since its rise to superpowerdom. From the refugee scientists who spearheaded the Manhattan Project, to the extraordinary proportion (44.8 percent) of Fortune 500 companies founded by either first- or second-generation immigrants, the lasting dividends drawn from these influxes of foreign talent appear self-evident. No other industrialized country can boast such a distinguished roster of foreign-born naturalized citizens having played such a central role in the formulation of its national security policy — from Henry Kissinger to Zbigniew Brzezinski to Madeleine Albright. This influence extends well beyond policymaking to the groves of academia. Consider, for example, one of the most historically influential volumes in the field of strategic studies, the initial edition of Makers of Modern Strategy. First published in 1943, the edited volume, which exerted a deep and lasting influence over American intellectual approaches to grand strategy, included a remarkably high proportion of German refugees among its illustrious contributors.
From the Bulgarian-born aeronautical engineering genius Assen “Jerry” Jordanoff to Hyman Rickover, the Polish-American “father of the nuclear navy,” the American military innovation enterprise has also benefited from the drive, ingenuity, and patriotism of naturalized immigrants. Indeed, freshly minted citizens — having made the deeply personal and active decision to integrate into their host nation — often approach public service with a unique brand of reverence and dedication. For one example, look to Enrico Fermi, who fled Mussolini’s Italy for America and then went on to play a lead role in the development of the Manhattan Project. His daughter wrote that it was her father’s immigrant background that had instilled in him and his fellow European refugees such a remarkable strength of purpose in the defense of their new homeland:
The determination to defend America at all costs spurred the newcomers no less than the Americans, and the European-born may have come to this determination somewhat earlier than the native-born, driven by stronger personal emotions. The picture of their country under Nazi power in the event of a German victory was something the Americans could imagine only with difficulty … And if America failed them, where would they go? It was not only gratitude to the country that had offered them asylum or pride in their new citizenship but also the fear of dictators that drove them to the limit of their physical and mental endurance.
In recognition of this longstanding tradition in U.S. innovation and statecraft, the Biden administration’s most recent executive order on AI openly advocates “attracting the world’s AI talent to our shores … not just to study, but to stay.” And indeed, on the issue of skill-based immigration reform, there may be room for bipartisan compromise and legislative action, particularly if the broader defense community works to raise awareness of its criticality to U.S. interests and security. A recent, comprehensive report aptly points out that only Congress has the authority to create a new category of lawful permanent residents specifically chosen for their advanced expertise in critical and emerging technology fields, along with the power to allocate the necessary green card quotas that provide a pathway to naturalization. Despite the commonsensical nature of implementing more effective forms of foreign talent capture, this will still prove challenging, given our current political climate. Somewhat dispiritingly, an otherwise comprehensive—and bipartisan—recent congressional report on AI chose to elide the issue of immigration entirely, notwithstanding the fact that America’s economic and societal attractiveness remains, perhaps, one of its most enduring competitive advantages.
Indeed, unlike China or Russia, the United States — with its high white-collar wages, deep and liquid capital markets, world-class universities, and reliable legal system — remains a country with enormous appeal to foreign scientists, engineers, academics and entrepreneurs. America has traditionally captured 40-to-50 percent of the global inflow of college-educated migrants to nations that are members of the Organization for Economic Cooperation and Development, and more than half of all Nobel prize winners moved to the US for professional reasons. Some 42 percent of all doctoral students in science, technology, engineering, and mathematics in American universities are foreign-born, as is approximately 69 percent of Silicon Valley’s tech workforce, with the largest shares coming from India (26 percent), and China (16 percent). Meanwhile, according to a 2020 report, around 40 percent of high-skilled semiconductor workers in the United States were born abroad. It is essential to find creative ways to retain this wellspring of talent within the United States, rather than allow it to be redirected overseas. As one excellent analysis mordantly noted in these same pages only a few months ago, “locking bright minds out of the U.S. national security innovation base ecosystem because of immigration reform deadlock,” is fundamentally self-defeating, and can perhaps best be equated to a form of “strategic seppuku.”
Exempting foreign citizens with doctoral degrees in science, technology, engineering, and mathematics from annual green card limits would constitute a step in the right direction. So would the creation of a new start-up visa category for foreign-born entrepreneurs. These legislative proposals, however, still fall far short of the mark. The urgency of the situation is such that more drastic measures are needed. For example, the United States should follow the National Security Commission on AI’s recommendation to grant green cards to all students graduating with these doctoral degrees from accredited American universities. Washington should also consider expanding the Fulbright Foreign Student Program, which currently provides approximately 4,000 outstanding foreign students with the ability to study and conduct research in the United States. Establishing an effective international talent pipeline also requires facilitating the sponsorship of those foreign students whose families may not have the means to send them overseas.
As one group of national security grandees has aptly noted, however, these efforts to attract and retain foreign talent, while laudable and commonsensical, must be carefully calibrated with efforts to stem intellectual property theft and espionage. This, in turn, implies “strengthening security standards to ensure that sensitive information and intellectual property remains secure.” This is especially true when contending with the large numbers of U.S.-based talent originating from adversarial powers such as Russia or China, two nation-states with a longstanding habit of both brutalizing their diasporas, and weaponizing them as conduits for large-scale industrial espionage. At the same time, US lawmakers and security managers should ensure that any (legitimately) enhanced screening procedures of Russian or Chinese-born engineers are conducted in a legally structured and bounded fashion, rather than used as an excuse in some quarters for blanket racial profiling or discrimination. Finally, the United States should think more creatively about preemptively draining or weakening adversarial foreign talent in addition to building up its own, and about leveraging a key dimension of its soft power — its economic, cultural, and intellectual attractiveness — against its principal competitors. Conflict is a fundamentally interactive endeavor, and this includes the accelerating global competition for talent.
Targeting Adversarial AI Talent
While no contemporary U.S. policymaker would want to be confronted with the same anfractuous ethical quandaries as the architects of Operation Paperclip, one should aspire to at least a partial reintroduction of this spirit of competitive ruthlessness. For example, it is estimated that in the immediate aftermath of Moscow’s full-scale invasion of Ukraine in February 2022, over 100,000 information technology specialists left Russia. In the months following the invasion, the Biden administration proposed eliminating some visa requirements for Russians with a masters or doctoral degree in “science, technology, engineering, or mathematics, including but not limiting to degrees relevant to artificial intelligence, cybersecurity, semiconductors, and robotics.” This was shrewd thinking, but over two years later the data shows that the vast majority of these skilled emigrants chose to relocate either in Central Asia or in neighboring post-Soviet countries such as Georgia or Armenia — largely because of European and American unwillingness to provide them with visas. Since then, Russia has proven remarkably effective at enticing its wayward specialists back to the home country, where they are offered draft deferments and low mortgage rates in exchange for contributing to Moscow’s military research and development efforts.
Meanwhile, there was a time, only a few years ago, when a growing number of China’s most highly educated scientists were feverishly seeking ways to escape their increasingly stagnant and dystopic society to find opportunity — and freedom — overseas. Over the past year, however, as mentioned earlier, a “reverse brain drain” has been in effect, with many returning to China with now little intention of leaving. In both cases, there was perhaps a missed opportunity for Washington to durably affect the trajectory of U.S. technological competition with its most formidable foes.
Obviously, any strategy for long-term talent capture should also be fused with one that maximizes operational safety, along with the defense of U.S. intellectual property. No Chinese or Russian-born AI scientist should be allowed to remain in the United States, work for the U.S. government, or work on U.S. government-adjacent contracts without undergoing a draconian naturalization and security clearance process. Moreover, given that most revisionist nations have developed the unsavory habit of targeting the families of emigrants — and especially those of expatriated political opponents — a generous policy of family reunification would have to be offered to a select number of high-value foreign AI luminaries. This would blunt Chinese or Russian intelligence services’ efforts to extort said scientists by threatening their extended families.
Of course, there is also the possibility that as the United States appears to move toward artificial general intelligence, an increasingly desperate foreign adversary such as China might seek to abduct or harm star American or U.S.-based AI scientists, regardless of the exact nature of their work. After all, the arguments regarding the dual use of AI research could also be advanced for a number of assassinated Iranian scientists whose work had civilian as well as military applications. And if lead American AI scientists and developers come to be viewed as “fair game” by authoritarian powers, what can be done to both identify those few key individuals most vital to the artificial general intelligence enterprise, and then ensure their protection from physical harm or extortion? Somewhat ominously, Beijing has already begun developing a cyclopean AI-powered platform called “Supermind” to continuously track millions of information technology specialists and AI scientists around the world.
Conversely, if it appears that a foreign adversarial power is clearly taking the lead in developing a technology as hugely consequential as artificial general intelligence, U.S. security managers may need to reluctantly reacquaint themselves with some of the more shadowy and morally fraught chapters of their nation’s past — from the plot against Heisenberg to Operation Paperclip. Indeed, the stakes of the competition may demand no less. As one particularly sophisticated treatment of the ethics of civilian assassinations in times of war reminds us,
Contemporary theories of just war allow for the partial extension of combatant status to civilians who are either threatening or responsible for unjust threats. Weapons manufacturers, their factories and employees are accorded less than absolute protection within just war theory, and even less under international law … . The various moral arguments against assassinations on the one hand and the complex status of munitions workers on the other suggest that scientists involved in weapon manufacturing may in some cases be morally liable to direct harm, as well as being legally liable to proportionate collateral damage.
In short, this is something of a miasmic moral issue — and as the lines between AI’s civilian and military applications continue to fade, so will governments’ perceptions of the physical inviolability of lead AI scientists.
In 1946, during the vanishingly brief period of U.S. nuclear monopoly, Winston Churchill famously cautioned that it would be “criminal madness” to cast the “secret knowledge or experience of the atomic bomb … adrift in this still agitated and un-united world.” Close to eight decades later, as we now find ourselves potentially on the cusp of unleashing another, even more transformative technology into our own fractured world, it would be just as foolhardy for Washington to cede its increasingly wafer-thin advantage in the field of AI. In order to prevent such a dread outcome, however, America’s leaders will need to come to terms with one overarching truth: For all its algorithmic complexity and technological intricacy, the race toward artificial general intelligence remains, above all, part of an age-old series of battles for human talent and ingenuity. It is time, perhaps, to begin planning accordingly.
Iskander Rehman is a senior political scientist at the RAND Corporation and can be followed on X @IskanderRehman. He would like to thank his RAND colleague Nathan Waechter for pointing him toward the sorry tale of Qian Xuesen.
Image: Greg Gerken via Department of Defense
Jesus Christ is King
