Keeping Top AI Talent in the United States

Keeping Top

AI Talent in the

United States

FINDINGS AND

POLICY OPTIONS

FOR INTERNATIONAL

GRADUATE STUDENT

RETENTION

DECEMBER 2019

LEAD AUTHOR

Remco Zwetsloot

CO-AUTHORS

James Dunham

Zachary Arnold

Tina Huang

Center for Security and Emerging Technology2

Established in January 2019, the Center for Security and

Emerging Technology (CSET) at Georgetown’s Walsh

School of Foreign Service is a research organization fo-

cused on studying the security impacts of emerging tech-

nologies, supporting academic work in security and tech-

nology studies, and delivering nonpartisan analysis to the

policy community. CSET aims to prepare a generation of

policymakers, analysts, and diplomats to address the chal-

lenges and opportunities of emerging technologies. During

its ﬁrst two years, CSET will focus on the effects of progress

in artiﬁcial intelligence and advanced computing.

CSET.GEORGETOWN.EDU | CSET@GEORGETOWN.EDU

FINDINGS AND POLICY OPTIONS FOR

INTERNATIONAL GRADUATE STUDENT RETENTION

Keeping Top AI Talent

in the United States

DECEMBER 2019

LEAD AUTHOR

Remco Zwetsloot

CO-AUTHORS

James Dunham

Zachary Arnold

Tina Huang

ACKNOWLEDGEMENTS

For comments and conversations that inf

ormed the content of this pa-

per, we thank Tarun Chhabra, Teddy Collins, Richard Danzig, Jeff Ding,

Melissa Flagg, Joy Ma, Jason Matheny, Doug Rand, Igor Mikolic-Tor-

reira, Josh Trapani, Allie Vreeman, and others we talked to who prefer to

remain unnamed. Thanks also to Maura McCarthy and Lynne Weil for

editorial support.

For help with original data collection (see Appendix A), we are

grateful to our dedicated research assistants Kiren Chaudry, Christina

Huntzinger, Jonathan Murdick, Santiago Mutis, Daniel Zhang, and Kath-

erine Zhuo. Data collection was also supported by Ben Murphy, Philippe

Loustaunau, Jennifer Melot, and Dewey Murdick. Roxanne Heston and

Will Hunt provided valuable further research assistance.

For sharing their data and insights with us, we thank the Computing

Research Association (especially Betsy Bizot and Burcin Tamer) and the

National Science Foundation (especially Darius Singpurwalla). The use

of NSF data does not imply NSF endorsement of the research, research

methods, or conclusions contained in this report. Any opinions, ﬁndings,

and conclusions or recommendations expressed in this material are

those of the author and do not necessarily reflect the views of NSF or

CRA.

is licensed under a Creative Commons Attribution-Non Commercial 4.0

International License.

To view a copy of this license, visit https://creativecommons.org/

licenses/by-nc/4.0/.

Document Identifier:10.51593/20190007

Cover photo: Trombax/Adobe Stock.

RELATED CSET REPORTS

Strengthening the U.S. AI Workforce: A Policy and Research

Agenda, by Remco Zwetsloot, Roxanne Heston, and

Zachary Arnold

Immigration Policy and the U.S. AI Sector: A Preliminary

Assessment, by Zachary Arnold, Remco Zwetsloot, Roxanne

Heston, and Tina Huang

China’s Access to U.S. AI Technology: An Assessment, by

William Hannas and Huey-meei Chang

Center for Security and Emerging Technology i

EXECUTIVE SUMMARY

INTRODUCTION: THE IMPORTANCE OF INTERNATIONAL

GRADUATE STUDENTS TO U.S. AI COMPETITIVENESS

1 | UNDERSTANDING STUDENT RETENTION

2 | THE POLICY CONTEXT

3 | PRIORITIES AND OPTIONS FOR U.S. POLICYMAKERS

CONCLUSION

FUTURE WORK

APPENDIX

ENDNOTES

iii

vii

Contents

Center for Security and Emerging Technologyiv

Center for Security and Emerging Technology iii

alent is core to U.S. competitiveness in artiﬁcial intelligence, and

international graduate students are a large source of AI talent for

the United States. More than half of the AI workforce in the United

States was born abroad, as were around two-thirds of current graduate

students in AI-related ﬁelds. Tens of thousands of international students

get AI-related degrees at U.S. universities every year. Retaining them,

and ensuring a steady future talent inﬂow, is among the most important

things the United States can do to address persistent domestic AI work-

force shortages and to remain the global leader in AI.

This paper holds both good news and bad news for the United States.

The good news is that student retention has historically been a core

U.S. strength, with well over 80 percent of international U.S.-trained AI

PhDs staying in the country, including those from AI competitors such as

China. By contrast, other studies have found that the vast majority of Chi-

na-trained AI talent currently lives outside China. Moreover, contrary to

popular perception and anecdotal reports, there is no evidence of recent

declines in U.S. retention rates.

The bad news is that two trends are placing this U.S. strength

in student retention at risk. The immigration obstacles international

graduates face have grown steadily in the past two decades and have

worsened in recent years. At the same time, other countries are investing

heavily in AI talent attraction and retention, pumping money into their do-

mestic AI ecosystems and opening up their immigration systems to foreign

AI talent. In the past, the United States could rely on its status as the world’s

sole science and technology superpower to compensate for the ﬂaws of

Executive Summary

Center for Security and Emerging Technologyiv

its immigration system, but in today’s more competitive world, complacency is likely

to come at a higher cost. Without serious immigration policy changes, the United

States stands to lose a vital asset in the international competition for AI leadership.

Results presented below are based on CSET-collected comprehensive career

data on 2,000 recent AI PhD graduates from U.S. universities, as well as original

analysis of 43,000 immigration records of AI professionals and multiple AI-related

survey instruments. Key ﬁndings include the following:

• International students are a key source for graduate-level U.S. talent

in AI.

• Two-thirds of graduate students in AI-related programs are inter-

national students, and the number of domestic graduate students in

these programs has not increased since 1990. Currently, U.S. uni-

versities graduate around 50,000 international graduate students

(44,000 master’s, 3,000 PhDs) in AI-related ﬁelds per year.

• About 70 percent of immigrants sponsored by AI companies for per-

manent residency studied at U.S. universities, as did more than half of

all international AI workers entering the U.S. labor market each year.

• International graduates ﬁll critical AI talent gaps in the U.S. labor

market. Objective labor market indicators and expert assessments

suggest demand for AI talent will far outstrip supply for the foresee-

able future.

• Stay rates among international graduates in AI are persistently high.

• Around 90 percent of international AI PhD students take a job in the

United States after graduating, and more than 80 percent stay in the

country for at least ﬁve years. Past studies strongly suggest stay rates

are likely to be high beyond the ﬁve-year window for which there is

hard AI-speciﬁc data.

• Multiple data sources indicate retention rates have not fallen in re-

cent years, contrary to popular perception and anecdotal reports.

• Stay rates are highest—exceeding 90 percent—among students from

Taiwan, India, Iran, and China, and lower—around 75 percent—

among students from European countries.

• Among the few graduates who leave the United States, the large

majority go to U.S. allies and partners in Europe and Asia, such as

the U.K., Canada, Singapore, and South Korea. Less than 20 percent

of those leaving go to China.

Center for Security and Emerging Technology v

• Professional considerations are the main reasons for international

talent to stay in the United States, while immigration difﬁculties and

cultural factors are the most important issues pushing away talent.

• The U.S. private sector is especially attractive to graduates; around

60 percent go on to work for companies after completing their de-

gree, with most of the remainder going into academia.

• Graduates with ambitions to launch or work at startups are particularly

hampered by immigration obstacles. Whereas more than 40 percent

of domestic graduates who go into the private sector work at small

companies, less than 20 percent of international graduates do so.

On the policy front, research highlights two important trends that, together,

could erode the U.S. AI talent advantage:

• Domestically, international graduates who want to stay are faced

with signiﬁcant obstacles in the U.S. immigration system, and these

problems are getting worse.

• Green card wait times have increased signiﬁcantly in recent years.

One study estimates that an Indian AI PhD graduate sponsored for a

green card today would face a wait time of around 50 years in the

absence of immigration reforms.

• Optional Practical Training, a program used by tens of thousands of

international graduates from AI-related programs every year, is cur-

rently facing signiﬁcant legal and policy challenges. Given the lack

of available alternative visas for these graduates, many would likely

be forced to leave the United States if OPT were eliminated.

• There is no suitable U.S. entrepreneur visa for international graduates

who want to start AI companies. Sponsoring employees for visas is

often too costly for startups, in large part due to inﬂexible and long

application timelines.

• Internationally, the United States faces increasing competition for top

AI talent.

• The United States has lost its historical near-monopoly on AI R&D

and commercial activity. In 2013, the United States accounted for

more than 70 percent of funding deals for AI startups. By 2018, this

number had dropped to 40 percent.

• Other countries are opening their immigration systems and aggres-

sively recruiting U.S.-trained AI talent. Nearly two dozen countries

Center for Security and Emerging Technologyvi

have recently launched startup visa programs marketed mainly to

tech entrepreneurs.

• Other countries are also investing heavily in their education systems.

The number of U.S. universities reporting international students de-

clining admission offers because they preferred to study at home or in

third countries increased three-fold between 2016 and 2018.

Based on these ﬁndings, the report lays out two priorities and several concrete

options for U.S. policymakers.

• First, policymakers need to reform high-skill immigration rules in

order to maintain and improve U.S. international AI talent retention.

Options for achieving this include:

• Reforming student visa regulations and procedures, for example by

codifying OPT in statute and eliminating processing backlogs.

• Streamlining post-graduation transitions into the U.S. labor force, as

could be done through the creation of a statutory student-to-work

pathway and a dedicated visa program for entrepreneurs.

• Shortening the path to permanent residency and citizenship, for ex-

ample by removing numerical caps for in-demand graduate talent or

creating accelerated citizenship-through-service programs.

• Second, policymakers should address legitimate security concerns

around foreign AI talent while avoiding broad and potentially coun-

terproductive restrictions. This can be done by:

• Improving policy coordination domestically and internationally by

creating a new interagency task force and increasing engagement

with allies, without whom counter-transfer efforts for diffuse technolo-

gies such as AI would almost certainly be ineffective.

• Raising awareness of transfer practices through open-source collec-

tion and dissemination, for example by allocating more resources to

open-source intelligence activities or adopting FARA-like legislation

for foreign talent recruitment activities.

• Collecting more and better data about student retention trends,

including among master’s students, for whom there is no govern-

ment survey or other data source that tracks post-graduation career

choices.

Center for Security and Emerging Technology vii

Introduction

The Importance of International Graduate

Students to U.S. AI Competitiveness

here is widespread consensus among U.S. policymakers that AI is

central to the future of U.S. security and economic competitive-

ness.

Improving the retention of international AI graduate students

should be a key part of U.S. AI policy because talent is core to compet-

itiveness in AI and because international graduate students are a key

source of U.S. AI talent.

Talent is crucial for building and deploying the different parts of AI

systems—algorithms, hardware, and data. Much of the knowledge for how

to work with AI systems is tacit and acquired through experience, and con-

tinuous progress in the ﬁeld means that today’s cutting-edge models could

be outdated tomorrow. Countries and companies thus require adaptable

skilled individuals who can continuously learn by doing and keep up with

rapid changes. Xi Jinping has called talent “the ﬁrst resource” in China’s

push for “independent innovation.”

As analyst Elsa Kania has put it, in a

sentiment echoed by many industry observers, “the real ‘arms race’ in [AI]

is not military competition but the battle for talent.”

Yet domestic AI talent in the United States is—and will for the foresee-

able future remain—inadequate to ﬁll rapidly rising demand. More than

half of the AI workforce in both academia and the private sector was born

abroad, and U.S. companies are increasingly setting up AI labs abroad

because they cannot ﬁnd enough talent at home. For example, on job site

Indeed, the number of AI job postings in the United States more than dou-

bled between 2015 and 2018, while the number of job searches increased

only marginally.

Expert consensus about workforce shortages in AI sets

the ﬁeld apart from other ﬁelds with many international students, including

STEM ﬁelds, where labor shortages claims are heavily debated.

Talent remains the

most important

driver of progress

in all facets of AI.”

—U.S. National

Security Commission

for Artiﬁcial

Intelligence (2019)

Center for Security and Emerging Technologyviii

International students are an especially valuable source of talent for the U.S. AI

workforce. Most top AI programs in the world are at U.S. universities, and the stu-

dents accepted into and trained in those programs represent the world’s top talent.

Employers also prefer students with a U.S. education because employers are better

able to assess the reliability of prospective employees’ qualiﬁcations and evaluate

them through internships.

Lastly, workers who come into a country as students tend

to remain longer and integrate better than those who initially enter through a tempo-

rary employment visa.

One study showed that international students co-founded

21 of the 87 “unicorns” the United States had in 2016, cumulatively worth $60

billion and responsible for nearly 20,000 U.S. jobs.

It should come as no surprise, then, that the majority of foreign-born workers

hired today by American AI companies are former international students. In abso-

lute terms, the number of international students graduating from U.S. universities with

AI-related degrees stands at more than 50,000 per year, and nearly all of them

are graduate students. By comparison, the annual number of domestic graduates

with AI-related graduate degrees is around 23,000. In short, international graduate

students are a main source of AI talent for the United States.

This paper asks what can be done to retain this source of AI talent, and its ﬁnd-

ings highlight the need for urgent action. The United States has historically excelled

at international graduate recruitment and retention. But there are warning signs in

the form of mounting domestic immigration difﬁculties and increased international

competition. While it is conceivable that the strength of the U.S. AI ecosystem will

continue to draw international talent despite these trends, complacency carries

signiﬁcant risks. It is much easier to maintain than to recover an advantage. And

because talent attracts talent and ecosystems grow in self-reinforcing ways, any

short-term increase in other states’ relative attractiveness—even if counteracted after

the fact—can have long-term and potentially irreversible consequences. U.S. poli-

cymakers have a window of opportunity for reform that they should not let pass.

To help policymakers bolster U.S. competitiveness in AI, this paper proceeds in

three steps. First, it draws on evidence from a wide range of sources to show how

the United States performs when it comes to graduate student retention in AI (Chap-

ter 1: “Understanding Student Retention”). Second, it examines relevant policy

trends, focusing on student-related U.S. immigration policies and recent measures

adopted by other countries competing for U.S.-trained AI talent (Chapter 2: “The

Policy Context”). Third, it lays out targeted policy options for improving graduate

student retention and recruitment while also addressing security concerns around

foreign talent (Chapter 3: “Priorities and Options for U.S. Policymakers”).

Center for Security and Emerging Technology 1

his chapter provides data and other evidence on the number of

international AI students in the United States, how many and who

among them stay in the country after graduating, why they decide

to stay or leave, and what work they end up doing after graduating.

Many of the ﬁndings and ﬁgures are based on original data collection

and analysis that CSET conducted. Data sources and methodology are

outlined in Box 1 and discussed in more detail in Appendix A.

Understanding

Student Retention

Data sources analyzed in this report

• Data newly collected by CSET on the career and educational histories of

1,999 AI PhDs who graduated from top U.S. universities between 2014

and 2019.

• Data from the National Science Foundation’s Survey of Earned Doctor-

ates on CS PhD students’ countries of origin and post-graduation profes-

sional plans.

• Data from the National Science Foundation and Department of Educa-

tion on national enrollment trends in AI-related programs.

• Data from the Computing Research Association’s Taulbee and Data

Buddies surveys on computing students’ ﬁelds of specialization and

post-graduation professional plans.

• Data from the Department of Labor’s PERM labor certiﬁcation process on

the educational and professional backgrounds of nearly 900,000 green

card applicants (43,000 of whom we classify as AI-related).

BOX 1

Center for Security and Emerging Technology2

WHAT DEGREES AND FIELDS ARE MOST IMPORTANT FOR AI?

Because AI is a new ﬁeld, it is not clear which students should be counted as

doing AI-relevant work or having AI-relevant skills, or what the most important

degrees are. Yet answering those questions, even provisionally, is necessary for

the analysis this paper sets out to do. Our research led us to focus on graduate

students and look primarily at the ﬁelds of computer science and computer engi-

neering.

We focus on graduate students because they represent around 85 percent of

all U.S.-based international students in AI-relevant disciplines (see Table 1) and be-

cause most international workers hired by AI companies hold graduate degrees. For

example, our analysis of data on AI companies’ sponsorship of green cards shows

that roughly 70 percent of individuals sponsored for technical jobs at these compa-

nies hold graduate degrees.

We focus primarily on computer science and computer engineering because

those seem to be the main feeder ﬁelds into AI jobs and thus the most representative

measure for AI talent pipelines.

Among workers sponsored for green cards for

technical jobs at AI companies, about two thirds have computer science, computer

engineering, or electrical engineering degrees.

(This does not mean, of course,

that everyone in CS, CE, or EE does AI-focused work; in fact, even in these ﬁelds AI

appears to be the focus of only about a quarter of all students, though data on this

question is sparse.

)

HOW MANY INTERNATIONAL STUDENTS ARE THERE?

International students ﬁrst accounted for more than 50 percent of total CS/EE

graduate students in the United States after 2000, with a rapid rise starting in

2013 bringing them to approximately 65 percent of the 150,000 total today

(Figure 1).

NUMBER OF CS/

EE GRADUATES

Bachelor’s

103,541

Masters

PhD

PERCENT CS/EE

GRADUATES

INTERNATIONAL

NUMBER OF

DOMESTIC CS/EE

GRADUATES

NUMBER OF

FOREIGN CS/EE

GRADUATES

Number of domestic and international gr

aduates

in AI-relevant fields,

AY2016-2017.

TABLE 1

65,943

4,713

67%

64%

94,624

21,665

1,682

8,917

44,278

3,031

Source: Depar

tment of Education Integrated Postsecondary Education Data System (IPEDS).

Center for Security and Emerging Technology 3

Looking at students’ countries of origin (Figure 2), for which NSF PhD data is

currently available to us only for CS students, Chinese and Indian nationals made

up a majority of international CS PhD graduates in 2016. Together, they slightly

outnumbered domestic U.S. graduates: 36 percent for China and India versus 35

percent for Americans. They are distantly followed by Iran (4 percent), South Korea

(4 percent), Bangladesh (2 percent), Taiwan (2 percent), and Turkey (2 percent).

What do these percentages translate into in terms of absolute numbers? Table 1

shows that U.S. universities graduate roughly 45,000 international master’s students

and 3,000 international PhD students in CS/EE per year (67 and 64 percent of

total graduates respectively). At the bachelor’s level, there is a much lower share of

international students (9 percent) and about 9,000 international graduates per year.

(Note that these numbers refer to annual graduates; the number of enrolled students is

signiﬁcantly higher at the undergraduate and doctoral levels since those degrees take

multiple years to complete.)

There is also a signiﬁcant number of U.S.-based international post-doctoral

researchers in CS/EE. There were a total of 2,100 CS/EE postdocs in 2016, for

example, of which roughly 70 percent were international.

1990 1995 2000 2005 2010 2015

20k

30k

40k

50k

60k

70k

80k

90k

100k

Number of CS/EE graduate students enrolled at U.S. universities,

1990-2016.

Foreign United States

FIGURE 1

Source: NSF Survey of Graduate Students and Postdoctorates in Science and Engineering (see Appendix A).

YEAR

Center for Security and Emerging Technology4

HOW MANY GRADUATES STAY, AND FOR HOW LONG?

Calculating stay rates is complicated because of the different ways in which one

can deﬁne and measure what it means to “stay” in the United States. One common

approach is to ask students if they intend to stay or have plans to stay in the United

States after completing their degree. Another is to track where graduates end up

working and to see if they actually stay based on publicly available career data

(e.g., from CVs).

Both measures have advantages and disadvantages. One advantage of us-

ing CV data is that it provides reliable longitudinal data on the same individual.

Intention-to-stay data, on the other hand, has the advantage of reﬂecting students’

underlying preferences more closely than their behavior does (which reﬂects legal

restrictions as well as their preferences). Intentions are also prospective, while stay

behavior is historical and thus a lagging indicator of changes in retention trends. This

report therefore presents data on both measures.

Looking ﬁrst at data on intentions, the vast majority of international PhD

students want to stay.* This ﬁnding is consistent across different data sources. In

a survey by the National Science Foundation (NSF) of CS PhD graduates, roughly

75 percent said they intend to stay (Figure 3a). A survey by the Computing Research

*No data source that we know of tracks stay rates among master’s students, so we report results

speciﬁc to PhD students in this report. In Chapter 3, we recommend that the National Science

Foundation ﬁll this informational gap by launching a survey of graduating master’s students similar

to the survey it runs of graduating PhD students.

Country of origin among CS PhD gr

aduates in the United States,

1986-2016.

FIGURE 2

1990 1995 2000 2005 2010 2015

0%

20%

40%

60%

YEAR

PERCENT OF U.S. CS PHDS

United States Other

India

China OECD

Source: NSF Survey of Earned Doctorates (see Appendix A).

Center for Security and Emerging Technology 5

Percentage of international CS PhD students intending to stay in the

United States after graduating, 1998-2016.

FIGURE 3A

Intend to stay

Unknown

Intend to leave

Source: NSF Survey of Earned Doctorates (see Appendix A).

1998 2002 2006 2010 20142000 2004 2008 2012

20%

40%

60%

80%

100%

2006 2008 2010 2012 2014 2016 2018

2016

20%

40%

60%

80%

100%

PERCENT OF PHD GRADUATES

Intend to stay UnknownIntend to leave

Source: CRA Taulbee Survey (see Appendix A).

Percentage of international AI PhD students intending to stay in the

United States after graduating, 2005-2018.

FIGURE 3B

YEAR

PERCENT OF PHD GRADUATES

YEAR

Center for Security and Emerging Technology6

Association (CRA), which collects information on students’ subﬁelds and thus allows

us to look speciﬁcally at doctoral graduates doing AI research, also ﬁnds inten-

tion-to-stay rates around 80 percent (Figure 3b). About half of the remaining stu-

dents surveyed by NSF and CRA had not yet made up their mind about post-grad-

uation plans when asked (around 10 percent of total), while the other half intended

to leave the United States (also around 10 percent of total).

To study stay rates, CSET also undertook a months-long data collection effort

on the pre- and post-PhD educational and professional histories of 1,999 PhDs

who completed an AI-related dissertation at a U.S. university ranked in the top 20

nationally for AI between 2014 and 2019 (described in more detail in Appendix

A). Looking at this group, there are also very high stay rates when it comes to

actual behavior, with more than 90 percent staying in the United States initially

and more than 80 percent remaining in the United States ﬁve years after graduat-

ing (Figure 4).

Percentage of top international U.S. AI PhD graduates still in the

United States, by years since graduation.

FIGURE 4

92%

88%

86%

85%

81%

82%

12345

20%

40%

60%

80%

100%

YEARS SINCE DOCTORATE

Source: CSET U.S. AI PhD Career Data (see Appendix A).

INTERNATIONAL STUDENTS

WORKING IN THE UNITED STATES

Center for Security and Emerging Technology 7Center for Security and Emerging Technology

While CSET’s AI-speciﬁc data cannot speak to retention beyond a ﬁve-year pe-

riod, prior research on PhD stay rates more broadly suggests that if graduates stay

for ﬁve years, they are also very likely to stay for a much longer period. For exam-

ple, Michael Finn at the Department of Energy, who has studied retention for years,

ﬁnds that a large majority of attrition has historically happened in the ﬁrst few years

after graduating; while 30 percent of all international PhD students leave within two

years, in the subsequent ten years only another 10 percent leave.

Other studies

conﬁrm this.

For example, a survey by Nature ﬁnds that younger researchers are

more open to moving “because their career paths were not settled and they were

less likely to be tied down by relationships and families.”

BOX 2

Experts and media outlets have claimed that an increasing number of internation-

al AI graduates are leaving the United States, especially Chinese students due to

recent tensions and a booming domestic tech sector and Indian students due to

the incredibly long green card queues.

However, the multiple datasets examined in this report show no evidence of

downward retention trends in either the overall PhD graduate population or for

these speciﬁc countries of origin. For example, neither NSF or CRA surveys on

intention-to-stay data (Figure 3) nor CSET-collected career data broken down

by graduation cohort (Figure 5) show any signs of recent decline. And NSF data

shows that there aren’t notable differences in retention trends across students from

different countries (Figure 6).

Still, we do not interpret our ﬁndings as entirely disproving claims about declining

stay rates. First, there is good retention data only on PhD students, and it could be

that stay rate patterns among bachelor’s or master’s students are different. Sec-

ond, most data sources lag by one or two years, while many of the events cited

as decreasing students’ desire or ability to stay—such as rising feelings of discrim-

ination among Chinese students and perceived upticks in visa processing times,

denials, and cancellations—have occurred recently.

Third, it could be that the same percentage of graduates stay in the United States

immediately after graduation but that the duration of (intended) stay is declining

Have stay rates been declining recently?

Center for Security and Emerging Technology8

Percentage of international AI PhD students who remain in the

United States directly after graduating, by year of graduation.

FIGURE 5

Source: CSET U.S. AI PhD Career Data (see Appendix A).

89%

91% 91%

92%

91%

2014 2015 2016 2017 2018

20%

40%

60%

80%

100%

COMPLETION YEAR

INTERNATIONAL STUDENTS

WORKING IN THE UNITED STATES

BOX 2 CONTINUED

for recent cohorts, which would not show up in the data for a while.

For exam-

ple, while it is possible to know what the ﬁve-year stay rate for the 2013 cohort

is, we won’t know that stay rate for the 2018 cohort is until 2023. Given how fast

the ﬁeld of AI is changing, it’s important to be cautious about extrapolating past

trends into the future.

Center for Security and Emerging Technology 9

WHO STAYS AND WHO LEAVES?

One of the most consistent predictors of a student’s decision to stay or leave is

their country of origin. As Figure 6 shows, there are a lot of differences across

nationality in how many students want to stay. The highest intention-to-stay rates

are among Chinese and Indian students, with lower rates among citizens of highly

developed Organization for Economic Cooperation and Development (OECD)

member countries. Unsurprisingly, U.S. citizens also tend to remain in the Unit-

ed States at very high rates, but there have been years when more Indians than

Americans intended to stay in the country.

This same cross-country stay rate pattern is also reﬂected in CSET’s top U.S.-

trained AI PhD career history data, where we estimate a student’s nationality by

the county where they did their undergraduate studies (Figure 7).

For example,

more than 90 percent of Indian and Iranian students are still in the United States ﬁve

years after obtaining their PhD, compared to around 75 percent for many European

countries.

Interestingly, there are lower stay rates among students from traditional U.S.

allies than among students from countries with which the United States has less

friendly relations. The most common explanation for this pattern is wealth: as coun-

tries become richer, they tend to have more professional opportunities and higher

Percentage of CS PhD students intending to stay in the United States

after graduating by nationality, 1998-2016.

FIGURE 6

YEAR

PERCENT OF PHDS INTENDING

TO STAY IN THE UNITED STATES

United States OtherChina India OECD

Source: NSF Survey of Earned Doctorates (see Appendix A).

1998 2000 2002 2004 2006 2008 2010 2012 2014 2016

0%

20%

40%

60%

80%

100%

Center for Security and Emerging Technology10

quality of life, and so they will see more top talent returning home. From the per-

spective of the United States losing valuable talent, then, its allies have been a much

more persistent challenge than potential adversaries (as is underscored in Figure 9).

Past research has also identiﬁed other factors that predict leaving. For example,

older international students or those who receive home government funding tend to

return at much greater rates.

Studies of other ﬁelds such as economics indicate that

the highest-quality international students generally have the highest likelihood of

remaining in the United States,

but there is no evidence available on whether this

is also true for AI.

WHY STAY OR LEAVE?

Unfortunately, no systematic data addresses the “why” question for the AI-rele-

vant graduate student population speciﬁcally.

However, research on other ﬁelds

points to factors that affect whether international graduate students—among the

Percentage of top AI PhD students still in the United States five years

after graduating, by country of undergraduate degree.

FIGURE 7

Source: CSET U.S. AI PhD Career Data (see Appendix A).

ESTIMATED NATIONALITY

RETENTION

95%

93%

92%

91%

89%

82%

81%

77%

75%

74%

50%

75%

Iran

Taiwan

United States

China

India

South Korea

Turkey

Germany

United Kingdom

Canada

Singapore

Other

20%

40%

60%

80%

100%

Center for Security and Emerging Technology 11

most mobile migrant populations in the world—decide to stay in their country of ed-

ucation. The most important factors are professional opportunities, immigration rules,

and personal and cultural considerations.

(Appendix C contains a more in-depth

review of evidence on return decisions among Chinese and Indian graduates.)

Professional opportunities. Several types of evidence point to professional

considerations as being decisive for many graduate students. For example:

• Job opportunities and the ability to stay and work after graduation are often

among the main reasons that students decide to study abroad in the ﬁrst

place.

When the U.K. closed its “post-study work route” immigration cat-

egory in 2010, international student applications fell by an unprecedented

30 percent.

• Historically, return rates have increased when domestic labor markets

improve.

For example, return rates among Taiwanese and South Korean

graduates rose sharply after their home countries underwent rapid industri-

alization in the 1980s

but brieﬂy fell again during the ﬁnancial crises of the

1990s.

• In surveys, most graduating students cite career prospects as their main

reason for staying.

This pattern also holds for more senior migrants; for

example, a plurality of international scientists who initially stayed abroad

after obtaining their PhD report “job opportunities” as the main determinant

of whether they’ll eventually return to their home country.

Immigration rules. Immigration restrictions, including those that do not bar

graduates from staying outright, have also been found to reduce stay rates

• In interviews and surveys, one of the most common reasons graduates cite

for not attempting to stay in the United States is “uncertainties about obtain-

ing green cards following graduation.”

Even if graduates can ﬁnd ways to

stay in the short term, long-term uncertainty and unpredictability are strong

deterrents.

• Due to numerical caps in the U.S. system, students from certain countries,

especially India and China, face years- or even (for Indians) decades-long

waitlists for permanent residency (see Chapter 2). One study estimates that

the number of Indian and Chinese graduates staying in the United States

drops by several percentage points for each year of extra delay due to

green card waitlists.

Another study focused speciﬁcally on Indian high-skill

immigrants found 94 percent concerned about green card wait times and

70 percent actively considering emigrating to a more visa-friendly country.

Center for Security and Emerging Technology12

• Restrictions on the type of work international graduates are allowed to do

also affects stay rates. For example, the lack of a U.S. visa category for

those wishing to start their own business has driven away many graduates

with entrepreneurial ambitions.

Lack of work authorization for spouses can

have similar effects.

Personal and cultural considerations. Personal and cultural considerations

often both pull students back to their home country and push them away from their

host country:

• Social ties are important pull factors; family is usually an especially

important consideration.

Research ﬁnds that “professional factors were

generally cited as encouraging students to stay in the United States, while

societal and personal factors were more likely to draw them back to their

home countries.”

• Social and cultural concerns can also serve as push factors. In one recent

survey, roughly 60 percent of international STEM graduate students report

experiencing cultural and/or social challenges.

Amid geopolitical ten-

sions and security measures, many international students, especially from

China and Iran, have felt less welcome in the United States due to a sense

of discrimination.

In summary, an accommodating immigration system is a necessary but not a

sufﬁcient reason for students to remain in the United States, with professional rea-

sons typically being decisive in motivating people to stay. Even when immigration

barriers are absent, people may still leave for professional or personal reasons. For

example, in one survey of 1,203 Indian and Chinese returnees, around 25 percent

of Indian and 35 percent of Chinese respondents held either permanent residency

or citizenship in the United States at the time of their departure.

WHAT DO GRADUATES DO IF THEY STAY?

Because professional considerations play such a large role in graduates’ deci-

sions about whether to stay or leave, it is useful to understand their career choices

in more detail. Looking at data on career trajectories can also uncover evidence

of immigration rules’ labor-displacing effects and help assess the security risks

that departing students pose to the United States, topics discussed in more detail

in Chapters 2 and 3.

After graduating, doctorate holders could go on to work in the private sector,

academia, nonproﬁts, or the public sector. Based on CSET-collected data, the

private sector is the most popular sector among top AI PhD graduates who stay in

Center for Security and Emerging Technology 13

the United States (with 60 percent of graduates taking a job there), followed by

academia (about 35 percent), with public sector and nonproﬁt roles far behind (5

percent combined). There do not appear to be any changes in the relative populari-

ty of the different sectors in the past ﬁve years (Table 2).

Of course, not everyone stays in their ﬁrst job, or even in the same sector.

Looking at where graduates work four years after obtaining their degrees—data

available only for the 2014 and 2015 graduating cohorts—nearly 20 percent have

switched sectors. Of the graduates who start in government or nonproﬁt jobs, nearly

75 percent leave for either industry or academia within four years. Around 20 per-

cent of the graduates who started off in academia moved to the private sector, and

10 percent of those who started off in private sector traveled the opposite path.

WHAT DO GRADUATES DO IF THEY LEAVE?

While a large majority of U.S.-trained AI PhDs stay in the United States, 205 stu-

dents (out of the 1,881 for which CSET has complete career data) left the country

after graduating. These graduates face two important choices. Like those who stay,

they have to decide what kind of job to take. But since they are in demand across

the world, they also have to decide on the place where they pursue that path.

Europe and Asia are the most popular destinations for those who leave, with

the United Kingdom and China taking the top spots (Figure 9). In total, about 39

percent go to Europe (18 percent to the United Kingdom, 7 percent to Germany,

4 percent to Switzerland, and 10 percent to other European countries), and about

Academic

Government/

Nonproﬁt

Private Sector

Sector of first post-graduation job among U.S. AI PhD

graduates staying in the United States, by year of

graduation.

TABLE 2

16 (7%)

123 (53%)

11 (4%)

168 (60%)

279

28 (7%)

260 (63%)

414

23 (5%)

305 (62%)

491

Source: CSET U.S. AI PhD Career Data (see Appendix A).

100 (36%) 126 (30%) 163 (33%)

24 (7%)

204 (57%)

351

123 (35%)

102 (6%)

1,060 (60%)

1,769

607 (34%)

Note: Percentages may not add up to 100 due to rounding.

Total

95 (41%)

234

Center for Security and Emerging Technology14

BOX 3

How does U.S. immigration policy affect career choices?

The choice of career path, like the choice of whether or not to stay in the United

States, can also be affected by immigration rules.

In government, where technical workforce shortages are large, the fact that inter-

national students’ path to citizenship often takes more than a decade means they

cannot ﬁll jobs with security clearance requirements. Since many interesting techni-

cal jobs have such requirements, international students are generally thought to be

much more likely than domestic students to eschew government careers.

In academia, universities are exempt from numerical caps on H-1B visas and are

therefore able to sponsor more international students for employment. Some studies

suggest that this makes international students more likely to “settle” for academia,

even if they might prefer to work in a different sector.

In the private sector, there are two ways in which immigration rules can steer inter-

national students toward large ﬁrms. First, the cost of sponsorship to employers—in

both time and money—means that startups and small businesses are much less

likely to sponsor work visas than large ﬁrms. Because smaller companies often have

more urgent hiring needs, long and uncertain application timelines for visas also

mean that sponsorship is generally off the table even for those willing to incur these

costs.

Second, the lack of a dedicated visa category for entrepreneurs means it is

much riskier for international students to start their own companies. A recent study of

entrepreneurial intentions and outcomes among STEM PhDs ﬁnds that international

students are twice as likely to want to start companies (21 percent, versus 10 per-

cent for domestic students) but are less likely to actually do so (4.6 percent, versus

6.3 percent for domestic students).

In CSET data on the career choices of AI PhD graduates who remain in the United

States, the differences between domestic and international students are consistent

with some of these arguments about immigration rules’ effect on career choice

and inconsistent with others. Most strikingly, far fewer foreign nationals work for or

founded startups even though more foreign nationals go into the private sector. This

suggests the U.S. immigration system is harming AI startups.

However, there is no

evidence that foreign graduates are more likely to “settle” for academia (Figure 8).

Center for Security and Emerging Technology 15

38 percent to Asia (17 percent to China, 7 percent to Singapore, 4 percent to India,

3 percent each to South Korea and Japan, and 4 percent to other Asian countries).

Canada is another popular destination, attracting 7 percent of those who leave.

Past studies of international students who leave the United States generally ﬁnd

that most of them return to their home countries,

and this appears to be true in the

ﬁeld of AI as well. For example, in the CSET dataset, out of those who did their under-

graduate education in China and left the United States after their PhD, 27 out of 38

return to China (with the others mainly going to the United Kingdom and Switzerland).

Western countries attract more international talent than others. Only ﬁve of the 17 who

left for Canada, for instance, got their undergraduate degree in Canada.

Sector of first job among AI PhD graduates who

stay in the United States, across domestic and

international students

FIGURE 8

Foreign Nationals U.S. Nationals

20%

40%

60%

80%

100%

ESTIMATED NATIONALITY

FIRST JOB FOR STUDENTS WORKING IN THE UNITED STATES

Government/Nonprofit Academic Private Sector Private Sector

Large CompanySmall Company

Source: CSET U.S. AI PhD Career Data (see Appendix A).

54%

27%

23%

15%

42%

27%

Center for Security and Emerging Technology16

Departing graduates make somewhat different career choices than those who

stay. Whereas 60 percent of those who stayed in the United States went into industry,

only 43 percent of those who left did. Academic jobs are more common among those

who leave, with 47 percent going to work at a university (compared to 35 percent of

those who stay). Government and nonproﬁt jobs are slightly more common among

those who leave, with 10 percent working in those sectors (compared to 5 percent of

those who stay).

What sector graduates take jobs in varies depending on the country they move

to. Graduates who leave for the India, for instance, tend to work in the private sec-

tor at much higher rates (8 out of 10) than graduates who leave for Germany (1 out

of 15) (Figure 10). More research is needed to understand this variation.

Destination countries among the 230 AI PhD students who leave the

United States at any point after graduating (out of 1,999 total).

FIGURE 9

Source: CSET U.S. AI PhD Career Data (see Appendix A).

United Kingdom

China

Canada

Germany

Singapore

India

Switzerland

Japan

South Korea

Other

15 15

7 7

DESTINATION

STUDENTS

Center for Security and Emerging Technology 17

SUMMARY OF FINDINGS AND TAKEAWAYS

• International students make up a majority of students in AI-related

graduate programs. About two thirds of graduate students in computer

science and electrical engineering are international students (Figure 1 and

Table 1). At the PhD level, roughly 30 percent of international students come

from China, 15 percent from India, 10 percent from OECD countries, and

the remaining 45 percent from other countries (Figure 2).

• International graduates overwhelmingly want to stay in the United

States, primarily for professional reasons. More than 80 percent of stu-

dents in AI-related ﬁelds want to and do stay after graduating (Figures 3-5).

Sector of first job among U.S. AI PhD graduates who left the

United States, by destination country.

FIGURE 10

6 11

India

China

Japan

United Kingdom

South Korea

Canada

Singapore

Germany

Switzerland

Other

20%

40%

60%

80%

100%

DESTINATION COUNTRY

GRADUATES

Government/Nonprofit Academic Private Sector

Source: CSET U.S. AI PhD Career Data (see Appendix A).

Center for Security and Emerging Technology18

This reﬂects the historical dominance of the United States in science and

technology; survey evidence shows that job opportunities and other profes-

sional factors are the main reasons graduates want to stay, and retention

rates are generally higher for graduates from developing countries such as

India and China (Figures 6-7).

• Immigration difﬁculties are an important reason for graduates to

leave. Students who want to stay face uncertainty and long waits in their

immigration process, and surveys indicate that this makes many graduates

less likely to stay. Immigration rules also affect the type of jobs open to stu-

dents; data on graduates’ career choices suggest that U.S. immigration rules

prevent international students with entrepreneurial ambitions from working

for startups or starting their own companies (Figure 8).

• When graduates leave, they primarily go to U.S. allies and partners.

China is the second most common destination country among those who

leave (17 percent), but a large majority go to countries whose relations with

the United States are much more friendly, such as the United Kingdom, Can-

ada, or South Korea (Figure 9). In some countries, such as the India, those

who leave primarily go on to work in the private sector, whereas in others,

such as Germany, they primarily work in academia (Figure 10).

Of the factors affecting graduates’ choice of whether to stay or leave, immigra-

tion policy is the factor most directly under the control of U.S. policymakers, which is

why Chapters 2 and 3 focus on immigration policy.

Other factors, even if they are not as easy for U.S. policymakers to control,

serve as an important backdrop for policymaking—and this backdrop is largely

bad news for the United States. As Chapter 2 discusses, the United States is losing

its status as the world’s sole science and technology superpower, with other coun-

tries making both large science and technology (S&T) investments and liberalizing

their high-skill immigration systems to attract S&T talent. Given that professional

considerations have been the main reason for most international graduates to stay,

these reforms could draw talent away from the United States. This makes action in

immigration and other areas U.S. policymakers can control all the more important.

Center for Security and Emerging Technology 19

hapter 1 showed that the vast majority of graduate students in

AI want to—and do—stay in the United States, but that some are

denied this opportunity due to problems with the U.S. immigration

system. To set the stage for policy recommendations in Chapter 3, this

chapter ﬁrst outlines the immigration process graduate students have to

go through to stay and provides numerical estimates for the size of the

international student population in different parts of the U.S. immigration

system. Second, it discusses recent policy reforms and trends in both

allied and competitor countries that have been laying the foundations for

increased AI talent competition with the United States.

The Policy Context

Immigration pathways available to international students.

FIGURE 11

Arrows represent possible transitions between different immigration steps.

OPTIONAL

PRACTICAL

TRAINING

(OPT)

PERMANENT

RESIDENT

CITIZENSHIP

TEMPORARY

RESIDENT

(E.G. H1-B)

STUDENT

Center for Security and Emerging Technology20

DOMESTIC POLICY CONTEXT: INTERNATIONAL GRADUATE

STUDENTS’ IMMIGRATION PROCESS

Students who stay in the United States will typically proceed through some or

all of the following steps in their immigration process: post-graduation Optional

Practical Training (OPT), temporary residency (e.g., an H-1B visa), permanent

residency, and ﬁnally citizenship.

Students and OPT

Students in AI-relevant ﬁelds generally come into the U.S. on F-1 visas.* The

number of F-1 visas that can be issued each year is unlimited, and F-1 students

can generally stay in the U.S. for the duration of their degrees.

Graduates who were on F-1s are also entitled to up to three additional years

(one year for those studying non-STEM subjects) of Optional Practical Training

(OPT), during which they are authorized to work full-time while retaining their F-1

status. The program is widely used but also controversial. A court challenge to OPT’s

legality has been pending for years, and the Trump administration in fall 2017

declared its intention—though it took no action—to roll back the program, which is

regulatory rather than legislative in nature and can therefore be changed or elim-

inated through executive action alone.

Changes to OPT were re-added to the

administration’s fall 2019 regulatory agenda.

After studying or OPT, graduates can try to get either temporary residency on a

non-immigrant work visa or jump directly into permanent residency (“green card”

status), depending on an employer’s willingness to sponsor and the availability of

green card slots (discussed below).

Temporary residency

Graduates can get temporary work authorization if an employer sponsors them

for a “non-immigrant” employment visa (so-called because ”immigrant” techni-

cally means someone who intends to reside in a new country permanently, not

temporarily).

The most commonly used temporary work visa is the H-1B visa. H-1B visas

are typically valid for three years and can be renewed for another three years

once, with additional indeﬁnite one-year extensions for individuals who are in

the green card queue. The annual number of new H-1B issuances is capped at

85,000, though universities and many nonproﬁts are exempt from this cap. Because

*This paper uses the term “visa” colloquially to describe a legal right to be physically present in the

United States or a document conferring that right. Legally speaking, a visa is a document allowing

a noncitizen to travel to a port of entry to seek admission to the United States. The separate right

to be present in the United States is often referred to as “legal status” or just “status.” See “Student

Visa vs. Student Status: What is the Difference?,” Department of Homeland Security, https://

studyinthestates.dhs.gov/2016/01/student-visa-vs-student-status-what-is-the-difference.

Center for Security and Emerging Technology 21

the number of applications usually far exceeds the number of available slots, U.S.

Citizenship and Immigration Services (USCIS) runs a once-yearly lottery to select

awardees. Less than half of eligible applicants have been able to get an H-1B visa

in recent years.

Many other non-immigran

t employment visas exist. They are not widely used by

international graduates speciﬁcally and are thus not discussed here, even though some

of them, like L-1 and J-1 visas, are important to the AI workforce more generally.

Permanent residency

Employers can also sponsor graduates for permanent residency (also known as

“green card” status), either while they are still students (on an F-1 visa) or while

they have a dedicated work visa (most commonly on an H-1B visa).

Relevant employment-based permanent residency categories include EB-1 (for

those with “extraordinary ability” or “outstanding professors and researchers”),

EB-2 (for those with “exceptional ability,” most commonly used by those with grad-

uate degrees), and EB-3 (college graduates). Since this paper is mainly focused

on graduate students, the EB-2 category is the most relevant of the three (though

graduate students are generally also able to apply for EB-3 visas).

About 80,000 slots are available for EB-2 and EB-3 applicants each year, allo-

cated on a ﬁrst-come-ﬁrst-served basis. Spouses and children of green card hold-

ers count toward this cap, so the annual number of slots for workers is signiﬁcantly

lower (typically less than half of the total number of available slots).

There are also

caps on what proportion of green cards can go to people born in a given country in

a single year.

People from India and China face signiﬁcant backlogs and delays because

there are so many employment-based applicants from those countries. For example,

one study projects the time spent in the “green card queue” for new EB-2 applicants

from China is six years; for Indians, that number is a staggering 54 years.

It is

difﬁcult for employees to get promoted or to switch companies while in this queue,

which means these large backlogs can carry signiﬁcant professional costs for

prospective immigrants from these countries.

(After getting a green card, changing

jobs or employers no longer requires approval from immigration agencies.)

Citizenship

rmanent residents can generally apply for U.S. citizenship after ﬁve years on an

employment-based green card. This step is, for practical purposes, optional; some

permanent residents do not apply for citizenship and instead stay in the United

States by renewing their green card every ten years. From an employment per-

spective, the main difference between permanent residency and citizenship is that

citizenship is required for many government- or defense-related jobs.

Center for Security and Emerging Technology22

The public picture of the immigration pipeline is incomplete due to lack of accessible

data. However, by piecing together information from different data sources, it is still pos-

sible to assess the approximate number of AI-relevant international graduates entering

the U.S. immigration system each year and which status categories are most important to

them (Table 3). When combined with estimates of how much AI talent there is in the U.S.

immigration system as a whole, as outlined in the CSET report Immigration Policy and

the U.S. AI Sector, these ﬁndings suggest that former international students make up well

over half of all foreign-born workers entering the U.S. AI labor market each year.

OPT. The most notable recent visa-related trend among international students has been

the very rapid rise in the use of OPT. For example, between 2014 and 2017 the number

of students granted OPT per year increased from 133,000 to 276,500.

Data obtained

by Pew Research Center through a Freedom of Information Act request indicates that

among graduate students, who account for over two thirds of all OPT grantees, roughly

half of those on OPT hold AI-relevant degrees (CS or engineering).

H-1B. Little is known about how many and what kinds of international graduates are on

H-1B visas due to a lack of publicly available data. However, it is clear that internation-

al graduates are an increasingly important source of H-1B entrants. Roughly 50,000

students transitioned from F-1 to H-1B status in 2018, and between 2012 and 2018

international students went from accounting for less than a quarter of H-1B entrants to

accounting for more than half.

It seems likely that a majority of students granted H-1B

status hold AI-relevant degrees given that nearly three quarters of H-1B holders work

Graduates’ U.S. immigration pathways, by the numbers.

BOX 4

Annual number of AI-relevant international

graduates from U.S. universities entering

into OPT, H-1B status, or permanent residency.

TABLE 3

OPT

125,000

H-1B

EB(-2/-3)

270,000

80,000

54%

70%

90%

55%

100%

42%

>35%

50%

49%

TOTAL NUMBER OF

ENTRANTS IN FY2017

PERCENT WHO WERE

INTERNATIONAL STUDENTS

PERCENT WHO HELD

GRADUATE DEGREES

PERCENT WHO HELD

AI-RELATED DEGREE

Center for Security and Emerging Technology 23

The policies and data reviewed point to several immigration challenges for

international graduates. First, large and growing bottlenecks in the immigration

pipeline harm international AI graduates’ prospects. Bottlenecks have grown

because the number of international students—and the number of other immigrants

competing for the same spots—has steadily risen while the numerical caps on the

number of available H-1Bs and green cards have not changed for decades. The

result is that AI graduates face signiﬁcant uncertainty about whether short-term or

long-term immigration is possible at all, and even those who do manage to get

through the system face long and costly wait periods. Large queues, processing

backlogs, and uncertainty have been a problem for a while, but all have notably

increased in recent years.

Second, rollbacks to OPT would be catastrophic for international AI

graduates. While H-1B visas often dominate many immigration conversations,

OPT—largely due to bottlenecks further down the immigration pipeline—has be-

BOX 4 CONTINUED

in computing-related jobs,

though we have not been able to ﬁnd data on degree ﬁelds

among H-1B applicants.

Permanent residency. Labor certiﬁcation (PERM) data released by the Department of

Labor provides a somewhat detailed picture of permanent residency applicants.

Based

on this data, from 2015 to 2018, roughly 42 percent of workers sponsored for perma-

nent residency had studied at U.S. universities, among whom 90 percent held graduate

degrees and 49 percent had studied in AI-relevant ﬁelds (CS or EE). Because PERM

data contains professional information, it is possible to zoom in on green card applicants

whose job title and employer indicate that they could do speciﬁcally AI-related work (see

Appendix A). From 2015 to 2018, there were about 26,000 green card applicants who

met these criteria, of whom a large majority (71 percent) had previously studied at U.S.

universities. Their most common countries of origin were India (47 percent), China (24

percent), and Canada (6 percent).

We do not have data on post-permanent-residency naturalization rates among AI-rel-

evant graduates. One study ﬁnds that roughly 30 percent of all international doctoral

graduates in the United States naturalize within 12 years of graduating.

However,

many PhD ﬁelds have stay rates considerably lower than those in computer science and

engineering, so we expect AI-relevant naturalization rates to be higher than this overall

naturalization rate.

Center for Security and Emerging Technology24

come perhaps even more essential for (initial) graduate retention, with tens of thou-

sands of AI graduates utilizing the program each year. If current legal and policy

challenges to OPT were to succeed and no compensating reforms enacted, these

graduates would likely have to leave the United States.

Third, the employer-driven and inﬂexible nature of the U.S. immigra-

tion system places serious constraints on international AI graduates. For

example, none of the immigration programs open to graduates are designed for

entrepreneurs, and time- and funding-constrained startups often cannot bear the

costs of visa sponsorship. Moreover, because graduates are generally bound to the

employers that sponsor them, they can face signiﬁcant difﬁculties switching jobs or

getting promoted. These features of the system make the United States signiﬁcantly

less attractive as a place for ambitious AI graduates.

INTERNATIONAL POLICY CONTEXT: INCREASED

COMPETITION FOR TALENT

When international U.S. graduates decide where to work after graduation, they

think not only about whether they are able to stay in the United States but also

about how attractive their alternatives are. It is therefore useful to brieﬂy examine

other countries’ policies and reforms aimed at attracting AI talent.

First, “receiving countries” that take in a large number of international students

are important talent competitors for the United States. These countries can also

provide lessons learned and models for policy change, for example in immigration

policy. Second, a small number of “sending countries” produce most internation-

al students. These countries’ policies, such as whether and how they incentivize

post-graduation return, affect the United States’ ability to recruit and retain students.

Receiving countries. After the United States, which had 1,094,792 inter-

national students in 2018, the top receiving countries are the United Kingdom

(506,480), China (489,200), Australia (371,885), and Canada (370,710).

Most

of these countries have a dedicated pathway for top students to become permanent

residents that they are actively strengthening and promoting. Many countries have

also launched programs to attract tech talent more generally, programs for which

U.S. graduates are eligible and a prominent recruitment target. These receiving

country efforts are discussed in more detail in Appendix B.

Sending countries. The top sending countries for international students, both in

general and in AI-relevant ﬁelds, are China and India. Two trends contribute to an

increase in the attractiveness of returning home after studying abroad. First, many

sending countries are becoming more professionally and personally attractive due

to their economic development (e.g., more robust domestic tech ecosystems, high-

Center for Security and Emerging Technology 25

er quality of life). Second, sending-country governments are increasingly offering

incentives to attract returnees (e.g., start-up subsidies, tax breaks, and scientiﬁc

funding),

although experts have questioned the effectiveness of these programs,

for example China’s Thousand Talents program. Appendix C discusses these points

and the relevant evidence in more detail.

The broader lens through which to look at policy developments in these other

countries is the globalization of science, R&D, and innovation. As noted in Chapter 1,

career considerations tend to be the most important factor shaping the migration choices

of high-skilled STEM talent. This means that welcoming immigration policies will not

work without attractive professional ecosystems, and that attractive professional eco-

systems can to some extent compensate for bad immigration policies.

For top technical talent, the United States’ status as the sole global S&T super-

power was historically attractive enough to compensate for the ﬂaws of the U.S.

immigration system. Today, however, the United States is losing its S&T superpower

status and facing competitors with increasingly robust private sector and academic

ecosystems of their own.

For example, between 2013 and 2018, the United States

went from accounting for more than 70 percent of funding deals for AI startups to

40 percent.

A recent analysis of 18 national AI strategies found that “AI talent

policies” were included in every strategy.

The view from China: “Talent is an important factor for the future

development of AI. Currently, the US remains the world’s gathering

place for research talent, but the strength of Chinese people in the

ﬁelds of AI research and applications, together with the Trump ad-

ministration’s immigration policies, have provided China opportuni-

ties to bolster its ranks of high-end talent.”

—CCID, state-run Chinese consulting ﬁrm, CSET translation of “White Paper

on Chinese Cities’ Development of Artiﬁcial Intelligence” (2018)

Center for Security and Emerging Technology26

In summary, as surveys of prospective and current STEM students conﬁrm,

“the U.S. is no longer an automatic choice for obtaining the best PhD education in

science and engineering,”

nor can one assume it will remain the automatic choice

for top careers. The effects of these trends, if not yet apparent in retention statistics,

are seen clearly in enrollment ﬁgures.

For example, in 2016, 14 percent of col-

leges had students decline admission offers because they decided to study at home,

and 19 percent because they went to a third country. Two years later, in 2018, these

proportions had risen to 39 percent and 59 percent.

The view from Canada: “Talent is a key factor of success in the era

of the Fourth Industrial Revolution. Canada’s world-class research

universities already attract international STEM talent and organiza-

tions. … As the U.S. continues to build a wall to exclude researchers

from countries that it deems hostile, Canada should not only keep

its doors open, but also actively attract and retain international

talent seeking opportunities outside the U.S.”

— Asia Paciﬁc Foundation of Canada (2019)

Center for Security and Emerging Technology 27

reserving the United States’ leadership position in science and

technology generally and AI speciﬁcally is essential for the

country’s economic and national security. Talent has been and

will continue to be a crucial factor in that effort. From an economic

perspective, more AI talent means more growth and innovation—labor

market indicators point to a large talent shortage in AI, and experts are

concerned that this shortage will persist and “slow the rate of diffusion

of [AI] and any productivity gains that accompany it.”

From a security

perspective, more AI talent means more people who can work toward

ensuring that AI systems are effective, safe, and secure.

As explained in Chapter 1, international graduate students are a large

source of AI talent for the United States, accounting for two thirds of grad-

uates in AI-related ﬁelds. And more than 80 percent of these international

graduates have historically stayed in the United States. However, Chapter

2 highlighted two trends that could erode this U.S. strengths in AI graduate

student attraction and retention: increasing immigration obstacles for grad-

uates who want to stay in the United States and increasing international

competition for AI talent.

This chapter builds on these ﬁndings to offer concrete actions for pol-

icymakers to work toward two overarching priorities: ﬁrst, attracting and

retaining international graduate students, and second, addressing security

concerns about foreign talent.

ATTRACTING AND RETAINING INTERNATIONAL

GRADUATE STUDENTS

Attracting and retaining top AI graduate students can be broken down

into three steps. First, students should want to come to the United States

Priorities and

Options for U.S.

Policymakers

Center for Security and Emerging Technology28

to study. Second, they should want to remain and work in the country after grad-

uating. Third, they should be able to obtain permanent residency so that they can

stay indeﬁnitely and have a pathway to citizenship. Policies targeting these steps

will be mutually reinforcing. If long-term job prospects are good, more students

will want to come study in the United States. Similarly, if students feel welcomed

when they ﬁrst arrive and are helped with their initial transition into the labor mar-

ket, they are also more likely to want to stay long-term.

Today, U.S. immigration policies hamper both the attraction and retention of top

AI talent. In a 2018 survey, more than 80 percent of universities said visa delays

and denials contributed to declines in the number of international students accept-

ing offers, up from 36 percent in 2016.

Moreover, students who do come face

several constraints and limitations that can reduce their desire to stay and—if they

decide to stay—their ability to contribute to innovation and growth.

The following policy options address these and other problems with the current

U.S. system. Together, they would improve both attraction and retention and sub-

stantially increase U.S. competitiveness in the international battle for AI talent.

Reform student visa regulations and procedures

Codify the Optional Practical Training program. Available data suggest tens

of thousands of graduate students with AI-relevant degrees use the F-1 visa’s OPT

program every year and depend on it for initial entry into the U.S. labor force

(Table 3). However, OPT was created via regulation and faces serious legal and

policy challenges.

To safeguard the status of this important part of the AI talent

pipeline, Congress should codify the existence of OPT in legislation. (The codi-

ﬁcation of OPT would be unnecessary if, as discussed below, Congress were to

create a statutory student-to-work pathway separate from the F-1 student visa.)

Address backlogs in F-1 and OPT processing. Signiﬁcant increases in

processing times of F-1 and OPT applications have forced many students to delay

or entirely forego education and employment.

For example, in the summer of

2019, processing times for OPT employment authorization regularly exceeded the

90-day window in which students were allowed to apply for OPT, meaning they

were not allowed to show up to work on their jobs’ purported start date. To address

these issues, USCIS could reinstate a recently rolled back internal rule mandating

processing of employment authorization requests within 90 days,

and similar time

constraints could be introduced, where feasible, for the interagency security review

(SAO) process that some F-1 applicants must go through. Congress, for its part,

could conduct further oversight over backlogs and allocate additional resources to

agencies where necessary.

Retain and improve ﬂexibility in student visa conditions. Experts predict

that pre-existing administrative backlogs—and the costs they impose on students—

Center for Security and Emerging Technology 29

would grow even larger if certain pending F-1 regulatory proposals were imple-

mented. Several policies, such as changing the duration of status from a ﬂexible to

a ﬁxed term, could also directly and negatively impact students. To avoid this, the

administration should amend or withdraw the relevant F-1 regulations and guid-

ance.

Given the desire expressed by policymakers across the political spectrum to

have students stay in the United States after graduating, Congress could also amend

the Immigration Nationality Act (INA) to allow students to express intent to stay in

the United States long-term without putting their visa at risk.

Streamline post-graduation transition into U.S. labor market

The policy options outlined in this section mostly concern temporary (“non-immi-

grant”) employment programs. However, if Congress adopts legislation allowing

graduate students to receive permanent residency (“immigrant” status) immedi-

ately after graduation, reform to non-immigrant programs could be partially or

entirely redundant. Policy options for permanent residency reform are discussed in

the next section.

Create a statutory student-to-work pathway. In contrast to Canada and

some other countries, the United States has no dedicated post-graduation employ-

ment visa for international students, and there are many more graduates than there

are available visa slots in current non-student-speciﬁc programs. To alleviate this

bottleneck and to help U.S. universities compete for top international talent, Con-

gress could create a student-speciﬁc temporary employment visa program akin to

successful programs in other countries with many international students (see Appen-

dix B). Such a program could include especially favorable conditions (e.g., duration

of visa, increased processing speed, more ﬂexible certiﬁcation requirement) for

students with job offers in labor-constrained ﬁelds like AI.

Allow entrepreneurial graduate students to start companies. There is

currently no visa program tailored to entrepreneurs in the U.S. immigration system,

as most visas require a formal employer-employee relationship. To ﬁll this gap,

Congress could create a visa category that allows international graduate—or

foreign-born workers more generally—to obtain either temporary or (condition-

al) permanent residency status if they start their own company. Barring legislative

changes, the White House could reverse its process of rolling back the International

Entrepreneur Rule, an Obama-era regulatory program intended to facilitate entre-

preneurship among immigrants.

Improve the ﬂexibility of employment visas. Graduates and employers in

fast-changing emerging technology ﬁelds such as AI face additional barriers due

to certain inﬂexibilities in the immigration system. For example, there are reports of

Center for Security and Emerging Technology30

people with physics degrees having trouble getting H-1Bs for data science roles

on the grounds that their degrees are insufﬁciently related to their jobs.

Other

ﬂexibility issues such as difﬁculties transferring H-1Bs between jobs or employers

also affect AI talent, as does spousal employment authorization. Because statuto-

ry language on such matters is typically broad, many of these problems could be

addressed by USCIS through guidance or regulations.

Reduce the burdens of visa allocation processes. To facilitate more efﬁcient

and timely hiring, the relevant agencies should work to increase the frequency of

visa allocations and decrease processing times. For example, USCIS could hold an

H-1B lottery quarterly as opposed to just once a year,

and the Department of La-

bor could reduce labor certiﬁcation requirements in labor-constrained ﬁelds such as

AI.

These measures would be especially beneﬁcial to smaller employers, like start-

ups, who work on tighter timelines and face more resource constraints than large

ﬁrms. (AI employers have likely already beneﬁted from certain recent USCIS-led

changes that favor higher-paying and higher-degree H-1B applicants.

)

Shorten the path to permanent residency and citizenship

Remove country-based caps on the number of available green cards.

Country-based caps have led to prohibitively long green card backlogs among

Chinese and especially Indian nationals. As noted in Chapter 2, an Indian PhD

entering the green card queue today is projected to face a wait time of around

50 years. This harms U.S. competitiveness in AI because approximately half of

international graduate students in AI-relevant ﬁelds are from India and China

(Figure 1). To reduce backlogs for AI talent source countries such as India and

China, Congress could eliminate country-based caps from the INA (or raise caps

more generally).*

Automatically grant green cards to postgraduate degree holders. An-

other approach to the green card backlog problem for AI students is to guarantee

international graduate (or only PhD) students conditional permanent residency upon

graduation from U.S. universities, either in general or for a smaller set of labor-con-

strained and strategically relevant ﬁelds such as AI.

†

To alleviate critics’ concerns

about automatic green cards incentivizing “diploma mills,” labor market pressures,

*A bill to this effect, the “Fairness for High-Skilled Immigrants Act,” passed the House and is

currently stalled in the Senate. The consequences of removing country-based caps are complex,

and before taking this step Congress should carefully assess its potential negative side-effects.

Increasing the overall number of available green cards could achieve many of the same beneﬁts to

the AI sector as removing country-based caps without the latter’s potential negative side-effects, but

such a measure is likely politically infeasible today.

†

Guaranteed green cards can also take the form of cap-exemptions, meaning that graduate

students from U.S. universities who meet certain requirements could obtain permanent residency

without entering lengthy green card queues. This was the approach taken in the Border Security,

Economic Opportunity, and Immigration Modernization Act, S. 744, 114th Cong. (2013).

Center for Security and Emerging Technology 31

and dual residency,

100

the program could be made speciﬁc to students graduating

from highly-ranked universities and include strict domestic residency requirements.

Create and utilize accelerated paths to citizenship in exchange for gov-

ernment service. Congress and agencies should work together to create programs

whereby in-demand international talent can receive green cards and citizenship

on accelerated timelines in exchange for a number of years of government service.

To inform program design, studies should be commissioned to inventory and derive

lessons learned from related past programs such as the Military Accessions Vital

to National Security (MAVNI) program.

101

(While a secure statutory program is

preferable, MAVNI was created by executive order, suggesting similar AI-speciﬁc

programs could be as well.)

Policy priorities besides immigration

It is not enough for a country to be welcoming. Graduates, in order to come and

stay, also need to see a country as professionally attractive. The United States

is already strong on this front, but it is facing increased competition from China,

Canada, the U.K., and other countries. Policies that bolster the U.S. academic and

commercial AI ecosystems—for example by addressing AI faculty shortages at

universities—would therefore also aid graduate retention. Future CSET reports will

provide more speciﬁc policy options on this front.

ADDRESSING SECURITY CONCERNS ABOUT FOREIGN TALENT

Policymakers should also act to address security concerns around the training and

presence of foreign talent in dual-use ﬁelds such as AI.

Other countries, most notably China, are actively engaged in trying to ex-

tract and absorb AI-relevant technology and knowledge from the United States to

strengthen their economies and militaries.

102

Students are one vector through which

these countries hope to achieve such technology transfer. U.S. law enforcement

agencies are now focusing on “non-traditional collectors” and calling out gradu-

ate students as a population of concern,

103

and commentators outside of govern-

ment have echoed their concerns.

104

Recent FBI enforcement actions and policy

changes at federal funding agencies have, in turn, sparked pushback from affect-

ed communities.

105

Given the dual-use nature and strategic value of AI, there are legitimate reasons

for U.S. policymakers to worry about technology transfer. Some speciﬁc concerns,

however, seem to be based at least partially on misperceptions. A prominent 2018

report by the Defense Innovation Unit notes that 25 percent of graduate students in

STEM ﬁelds are Chinese and that “nearly all [of them] will take their knowledge and

skills back to China” because they “do not have visas to remain in the U.S.,” the im-

plication being that U.S. universities are educating the country’s competitors without

Center for Security and Emerging Technology32

much beneﬁt to the United States.

106

As this report shows, that is not the case—with

the vast majority of Chinese graduate students in fact staying in the United States—

despite longstanding efforts by the Chinese government to draw them back.

Other disagreements stem from real uncertainty about the nature and extent

of risk. For example, CSET data collection found that out of the small number of

students who return to China after completing their PhD in the United States, most

go to work for the private sector and none directly for the Chinese government or

military (Figure 10). To what extent this assuages concerns about “educating our

competitors” depends on whether one believes China’s military-civil fusion plans,

whereby the government intends to convert civilian technological successes into a

long-term military advantage, will actually bear fruit. This is a question that China

experts continue to debate.

107

Other areas of potential disagreement include how

easy it is to militarize AI, or how difﬁcult it will be to distinguish between civilian and

military lines of AI R&D as the technology advances.

108

A lack of systematic data

and analysis on these and related questions makes it hard for governments and

other stakeholders to conduct risk assessments and develop countermeasures.

109

However, even if technology transfer risks are potentially substantial, policies

must also account for the serious security risks from not attracting and retaining

foreign students in the same way the United States does today. Given that there are

large talent shortages in AI, any decrease in international talent inﬂow and retention

would hurt U.S. industry and, by extension, the defense industrial base. Talent short-

ages also increase the risk of technology transfer by incentivizing U.S. companies

to set up labs in countries where protections against transfer are worse than in the

United States, as has already started happening in the AI sector.

11 0

Conversely, competitors would beneﬁt from decreased U.S. openness. One

study of top Chinese-trained AI researchers ﬁnds that 75 percent currently live

outside of China, nearly all of them in the United States.

111

China’s aggressive talent

recruitment efforts (discussed in more detail in Appendix C) show that its leadership

is unhappy with this situation and thinks its competitiveness in AI would increase

if more of this talent returned from abroad. Lastly, because the United States does

not have a monopoly on AI, it is likely that any unilateral counter-transfer measures

would simply displace, not decrease, transfer activities.

For these reasons, we do not recommend adopting policies targeting

broad student populations, such as the recent shortening of visa durations for

Chinese graduate students in certain technology ﬁelds.

112

Such measures put a core

U.S. advantage in technology competition at risk. Many are also likely to be inef-

fective. For example, Nicholas Eftimiades, a former senior U.S. counterintelligence

ofﬁcial and author of a book on Chinese intelligence practices, said of proposals

Center for Security and Emerging Technology 33

to broadly enhance screening of Chinese students and scholars, “Not a good idea.

The process will almost certainly fail at determining an individual’s future course of

action.”

113

Instead, the United States should adopt a more targeted approach to

countering the risks of technology transfer via talent ﬂow. First, it is important

to acknowledge that there is a clear need to encourage more domestic students to

pursue graduate degrees in AI-related ﬁelds,* and that most large-scale transfers of

data and intellectual property involve not individual researchers but cyber breach-

es or private investment and acquisitions. Much can still be done on those fronts

without running the risk of harming U.S. talent competitiveness.

114

Even when it comes

to addressing transfer occurring through individual students and scientists, however,

there are relatively low-risk measures U.S. policymakers can take today that would

help lay the groundwork for better and more targeted policy decisions down the

line. The remainder of this section outlines some of these measures.

Enhance domestic and international policy coordination

Create an interagency task force charged with improving both screening

and retention. A wide range of organizations, including the Departments of

State, Defense, Homeland Security, Commerce, and Education; the intelligence

community (IC); and various science funding agencies, are involved in creat-

ing and implementing U.S. policies relevant to foreign S&T talent screening and

retention. There is no U.S. government entity with a policy focus and a mission that

includes both the screening and retention of foreign S&T talent, two goals essen-

tial to U.S. security and affected by many of the same policy decisions.

†

This can

lead to a disproportionate focus on some policy goals at the expense of others.

In designing a new entity to ﬁll this gap, policymakers could take inspiration from

the membership structure and certain other aspects of the Committee on Foreign

Investment in the United States (CFIUS). The tradeoffs CFIUS navigates—between

the national security beneﬁts and risks of foreign investment—are similar to those

that crop up with foreign S&T talent and similarly require cross-agency input and

*How to encourage more domestic students to do graduate studies—and understanding domestic

students’ educational and career decisions more broadly—will be the focus of future CSET research.

†

There are entities that bring together different agencies to execute speciﬁc functions, such as making

individual student and scholar visa vetting decisions, or adjudicating deemed export control licenses.

However, these groups are typically not focused on policy-level activities and have narrow remits.

Recently, a subcommittee within the Ofﬁce of Science and Technology Policy, JCORE, took on a

coordinating function for certain research security policies. This is a step in the right direction, but

JCORE’s mission does not explicitly include talent retention, and it is unclear whether the committee

will have sufﬁcient resources or institutional clout. See “Letter to the Research Community,” Ofﬁce of

Science and Technology Policy, September 16, 2019, https://www.whitehouse.gov/wp-content/

uploads/2019/09/OSTP-letter-to-the-US-research-community-september-2019.pdf

Center for Security and Emerging Technology34

expertise. At minimum, a new interagency task force or other entity should get an

advisory and coordinating role.

Engage international allies about knowledge transfer concerns. To

avoid having to act unilaterally, the White House should engage allies with robust

emerging technology ecosystems to discuss how harmful knowledge transfer can

be avoided in an internationally coordinated fashion. Absent coordination, U.S.

countermeasures are likely to steer international talent toward other AI hubs such

as the U.K. and Canada without achieving any of the desired outcomes. Recent

multilateral initiatives organized by the State Department are a good ﬁrst step and

should be expanded.

Raise awareness and improve screening through open-source collec-

tion and dissemination

Allocate more resources to open-source intelligence collection. Many indi-

cators of transfer activity and evidence of potentially illegal behavior are avail-

able in the public domain, as recent indictments and agency enforcement actions

demonstrate.

115

For example, one recently indicted Chinese researcher accused

of hiding his employment at a Chinese university had his dual afﬁliation listed on

multiple public scientiﬁc papers. Unfortunately, open-source intelligence (OSINT)

collection and analysis currently get short shrift in the intelligence community,

leading one former deputy CIA director to argue “open-source intelligence

deserves its own agency.”

116

One advantage of such OSINT reforms would be to

improve vetting and monitoring capability when it comes to foreign researchers

in sensitive ﬁelds. Another is that open-source analysis can—to some extent—be

shared with professors and university administrators, some of whom are currently

mistrustful of transfer-related warnings from the intelligence community because

they are not able to scrutinize the evidence for themselves.

Consider expanding the scope of the Foreign Agents Registration Act

(FARA) to cover foreign talent recruitment efforts. FARA is a law aimed at

increasing public disclosure of foreign propaganda activity, with the intent not of

banning such activity but of educating the American public and relevant stakehold-

ers.

117

Analysts have suggested this same framework might be appropriate for tech-

nology transfer activity.

118

Congress should study whether and how it should adapt

or expand FARA to increase the visibility of foreign recruitment efforts of U.S.-based

talent and other transfer activities.

Collect more and better data about student retention trends

Expand agency surveys of U.S. student populations to include master’s

students and more retention-related questions. The National Science Foun-

dation and other agencies administer many useful surveys of student populations

in the United States, some of which this report draws on. However, these surveys

Center for Security and Emerging Technology 35

can be expanded in both who and what they cover. For example, there are cur-

rently no regular government agency surveys of master’s students—even though

there are now many times more master’s students than there are PhD students in

important ﬁelds like computer science (see Table 1). Existing surveys also often

do not ask how long students intend to stay after graduating or what factors drive

those decisions. To ﬁll these informational gaps, Congress should ask—and allo-

cate the necessary resources to—the National Science Foundation to expand its

existing portfolio of student surveys. Congress could also require periodic reports

on retention trends and their causes from the relevant agencies.

Center for Security and Emerging Technology36

Center for Security and Emerging Technology 37

alent is a crucial competitive advantage for the United States in

AI, in large part because of the country’s ability to attract and

retain the best technical minds in the world. Many of these minds

ﬁrst come to the United States as students and subsequently stay to be-

come some of the country’s best AI scientists, engineers, and leaders. In

doing so, they help ﬁll critical talent gaps in the U.S. AI sector.

Despite its historical strength in student attraction and retention, the Unit-

ed States needs to act if it wants to maintain the country’s AI talent ad-

vantages. These advantages are being challenged by two trends. First,

the U.S. immigration system is becoming increasingly difﬁcult to navigate

even for the most highly skilled. And whereas U.S. primacy in science

and technology used to be such that the United States was virtually the

only place where AI talent could do cutting-edge work, today other

countries are rapidly building up their domestic tech and AI ecosystems

and becoming attractive professional destinations.

While there is little that U.S. policymakers can do to halt the rise of com-

petition, they can control whether the United States remains an attractive

and welcoming destination for international talent. The primary tool they

have for this is immigration policy. Immigration reforms must go hand in

hand with policies aimed at expanding the domestic talent pipeline. But

domestic talent policies will take years to pay dividends, and domestic

talent will never be able to fully substitute for international talent.

Conclusion

Center for Security and Emerging Technology38

Overview of immigration policy options

BOX 5

Potential legislative actions Potential executive actions

REFORM STUDENT VISA REGULATIONS AND PROCEDURES

• Codify the Optional Practical Training

program in statute.

• Increase oversight of processing backlogs

in F-1 applications and OPT employment

authorizations.

• Allow students to express a desire to re-

main in the U.S. upon graduation (“dual

intent”).

• Avoid restrictions on the OPT program

that hurt AI employers.

• Adopt and implement rules that limit F-1

and OPT processing times to reasonable

timeframe

STREAMLINE POSTGRADUATION TRANSITION INTO U.S. LABOR MARKET

Potential legislative actions

Potential executive actions

• Create a new student-to-work pathway

with a dedicated employment visa for

former international students.

• Create a visa category for entrepreneurs

that allows international students to start

their own companies upon graduation.

• Hold H-1B lotteries quarterly in order to

increase time-constrained AI startups’

ability to utilize the H-1B visa.

• Reduce labor certiﬁcation requirements for

labor-constrained ﬁelds such as AI.

• Reverse the rollback process for the Inter-

national Entrepreneur Rule.

SHORTEN THE PATH TO PERMANENT RESIDENCY AND CITIZENSHIP

Potential legislative actions Potential executive actions

• Eliminate country-based caps on the

annual number of green cards that can be

issued.

• Automatically grant green cards to (or

exempt from green card caps) international

graduate students from select universities.

• Create accelerated paths to citizenship for

international graduates in exchange for

government service.

• Create accelerated paths to citizenship for

international graduates in exchange for

government service.

Several of these actions could be implemented in ways that either apply to international

graduate students broadly or to AI graduates speciﬁcally, as discussed above.

Center for Security and Emerging Technology 39

At the same time, the United States must balance the security and economic bene-

ﬁts it derives from being the global hub for AI talent with security concerns around

training foreign talent that might later work for U.S. competitors. This balancing

act is a delicate one. The easiest way for the United States to hurt itself in compet-

ing with China is to help the Chinese government recruit U.S.-based AI talent by

becoming less welcoming. To avoid such outcomes, U.S. policymakers should ﬁrst

focus on implementing necessary counter-transfer reforms that pose fewer risks to

talent competitiveness, such as strengthening investment and cyber protections,

improving domestic and international policy coordination, and shoring up the

country’s open-source intelligence apparatus.

Center for Security and Emerging Technology52

Center for Security and Emerging Technology 41

In the future, CSET aims to build on the research presented in this paper in several ways:

• Conduct career choice surveys. In its future surveys of AI students and faculty, CSET will

ﬁll informational gaps in existing surveys and learn more about career choices, the factors

that inﬂuence moving decisions, and the destination countries students consider attractive.

• Further analyze administrative immigration data. Available data from immigration

agencies is helpful but incomplete. Pending responses to agency requests, CSET will

conduct further analyses of administrative immigration data to form a better picture of the

pathways students travel through the immigration system and where they most frequently

encounter problems.

• Continuously track retention rates. Existing survey-based approaches to measuring

stay rates often provide incomplete coverage and suffer from delays in data availability.

In the future, CSET aims to continuously track retention rates via publicly available re-

sume data in order to provide timely information about changing patterns and trends in

post-graduation choices.

• Mapping graduate ﬂows internationally. This report focused on graduate students

in the United States, but the career and destination choices of graduate students in other

countries also provide valuable information about the attractiveness of different AI ecosys-

tems and global talent ﬂows. CSET aims to expand data collection efforts to other coun-

tries as well, including Canada, the United Kingdom, and China.

• Track enrollment trends. From a workforce perspective, it matters not only where stu-

dents go after graduating but also where they decide to enroll in the ﬁrst place. In future

work, CSET hopes to get a better picture of enrollment trends both in the United States and

other countries across different degree levels and AI-relevant ﬁelds.

• Better understand AI source ﬁelds. Since AI is a rapidly growing and changing ﬁeld,

the picture of the educational and employment routes into the AI workforce is imperfect.

Examining the career histories of those currently employed in AI jobs can offer a better

sense of the backgrounds and skills AI employers look for. This information could inform,

for example, immigration ofﬁcers' decisions about whether an applicant's skills and edu-

cation meet the legal criteria for employment-based visas, or Department of Labor assess-

ments of labor supply and shortages.

Future Work

Center for Security and Emerging Technology42

• Further assess security concerns. Student retention rates are only one part of the

discussion about the potential security risks of foreign talent in dual-use ﬁelds such as AI.

CSET will also study other questions that are part of this discussion, including further data

collection and assessments of potential countermeasures.

We welcome questions, feedback, and collaboration proposals on these topics; please feel

free to contact remco.zwetsloot@georgetown.edu.

Center for Security and Emerging Technology 43

A. DATA SOURCES

We drew on several data sources in Chapters 1 and 2 of this report. This Appendix brieﬂy discusses each of them,

describing the information available in the sources and the analytical choices we made in working with the data.

Original CSET data collection: top U.S. AI PhD graduates

For data on the education and career histories of AI PhD graduates from U.S. universities, CSET launched a large-

scale collection effort in early 2019. We started by creating a population list of recent PhD graduates on the basis

of a ProQuest dissertation search using AI-ML keywords. For this initial effort, we looked at students who graduated

between 2014-2019 from one of the 20 universities with the highest-ranked AI departments in the U.S.

119

A team of

research assistants took this list and manually collected data on their pre- and post-PhD educational history, their

professional activities, and their scientiﬁc publications from sources such as LinkedIn and Google Scholar. This paper

reports results on 1,999 graduates for whom we have collected complete histories so far. This data forms the basis for

Figures 4, 5, 7, 8, 9, and 10, and Table 2.

Data collection for the United States is ongoing, and collection is also expanding into other countries (initially

Canada, the United Kingdom, and China). In future publications we will expand on the data collection methodology

and highlight further ﬁndings. Readers interested in discussing the data in more detail in the meantime are welcome to

Administrative immigration data

Several U.S. departments charged with administering and enforcing immigration regulations regularly release

data on the number and characteristics of immigrants. The most detailed of these is the Department of Labor’s

“PERM” data, which includes information on foreign workers for whom employers initiated a labor certiﬁcation

process since 2010. This process is a prerequisite for green card sponsorship. Data on these workers includes their

country of citizenship, their educational and professional backgrounds, and characteristics of their prospective jobs.

Approximately 888,000 individuals are included in the 2010-2019 data.* We used the PERM dataset to study the

main feeder ﬁelds into AI jobs (Chapter 1) and the backgrounds of foreign workers sponsored for green cards (Box

4). To do this analysis, we needed working deﬁnitions of “AI employers” and “AI jobs.” We outline these deﬁnitions

brieﬂy and will elaborate on them in future work.

There is no widely agreed-upon deﬁnition or list of “AI employers.” For the purposes of this paper and the PERM

dataset, where we were primarily interested in identifying individuals with AI skills, our goal was to ﬁnd employers

who have hired signiﬁcant amounts of AI talent, even if AI represents a minority of what the employer does. We

ultimately compiled a list of several hundred AI companies from several sources: (1) companies that had been

identiﬁed in Crunchbase as specializing in AI and that (a) are publicly listed, (b) have 50 or more employees, or (c)

have raised at least $10 million in funding;

120

(2) leading AI startups, as identiﬁed by market research ﬁrm CB Insights;

and (3) companies that are especially active in hiring personnel with AI skills, as identiﬁed by market research ﬁrm

Paysa.

121

There were 50,391 entries (sponsored individuals) at these AI employers in our database.

It is similarly difﬁcult to deﬁne what counts as an “AI job,” especially with access to only a job title. A “software

engineer” at Google, for example, could work on an AI team or could be doing something else entirely. For this

paper we took a broad approach, reporting statistics for people who worked in broadly “technical” roles.

122

This

approach prioritizes minimizing the number of “false negatives” over minimizing the number of “false positives,”

Appendix

*Data for 2019 were downloaded in spring and are thus not comprehensive for that year.

Center for Security and Emerging Technology44

leaving in some non-AI-related technical jobs but at least excluding management, sales, accounting, and similar

“non-technical” jobs. Roughly 15 percent of jobs at AI employers in the PERM database were non-technical or

difﬁcult to classify and thus dropped, leaving us with a total of 43,070 potentially AI-related jobs at AI employers.

Other publicly available immigration datasets and statistics (e.g. on OPT or H-1B entrants) contained too little

information to do similarly detailed analysis, but they were used in Table 2 for aggregate statistics. To address

these data gaps and allow us to conduct further analysis, we have several FOIA requests pending with the relevant

agencies.

Existing surveys

We also analyzed data from existing surveys of AI-relevant student populations and university departments. Our

primary sources of survey data were:

The Survey of Earned Doctorates (used in Figures 2, 3a, and 6), an annual census of PhD graduates conducted

by the National Science Foundation (NSF).

123

The survey collects individual-level data on graduates’ academic

ﬁeld, country of origin, and career plans, generally achieving response rates exceeding 90 percent. Researchers

must request a license in order to access the data.

The Survey of Graduate Students and Postdoctorates in Science and Engineering (used in Figure 1 and

note 15), an annual census of research university science and engineering departments conducted by the NSF.

124

It asks departments for the number of enrolled students and postdocs and their demographic proﬁles, among other

information. It does not contain questions on students’ speciﬁc countries of origin, only whether they are domestic or

international. Its 2017 survey, the latest for which data is publicly available, was the ﬁrst that asked departments to

report statistics for master’s and PhD students separately (see note 15).

The Taulbee Survey (used in Figure 3b and note 14), an annual survey of PhD-granting computer science,

computer engineering, and informatics departments in the United States and Canada run by the Computing

Research Association (CRA).

125

It asks departments to provide information about PhD, master’s, and bachelor’s

students’ demographic background, and about PhD graduates’ ﬁelds of specialization and post-graduation plans.

CRA surveys nearly 300 departments, with response rates of 60-70 percent. Notes 14 and 52 also draw on data

from CRA’s separate Data Buddies survey.

126

We thank these organizations for sharing their data and insights with us, which we will continue to draw on in future

publications. CSET will also ﬁeld its own surveys; readers interested in providing input on such surveys are welcome

to reach out.

B. “RECEIVING” COUNTRIES: COMPARING STUDENT RETENTION

POLICIES

The United States leads the world in international student enrollment, followed by the United Kingdom, Canada,

and Australia. Comparing student retention policies across these countries, however, the United States does

substantially worse.

Student-to-work pathways. Most countries with signiﬁcant international student populations have created

speciﬁc “student-to-work” pathway. These are generally similar to Optional Practical Training (OPT) in the United

States. Unlike in the United States, however, these countries’ student-to-work routes are generally more securely

codiﬁed in law and more widely supported among policymakers. Indeed, many countries have recently reformed

and expanded the immigration options available to international students.

• In Australia, students can obtain a post-study work visa that is valid for two to four years (depending on

education level) and participate in a year-long professional program that prepares students for a career

in Australia.

127

These two professional activities yield graduates extra points in Australia’s points-based

immigration system, thereby increasing a graduate’s chances of obtaining permanent residency.

128

Center for Security and Emerging Technology 45

• In Canada, the Post-Graduation Work Permit (PGWP) allows students to work up to three years without

restrictions after getting their degree. As in Australia, a major beneﬁt of the PGWP is that graduates can use

Canadian work experience to boost their rank in the recently launched points-based Express Entry system

to secure permanent residency.

129

In early 2019, Canada liberalized PGWP regulations by extending the

deadline to apply from 90 days after graduation to six months.

130

• The United Kingdom terminated its two-year post-study work visa in 2012, and since then students have

had to compete with other immigrants for a general work visa capped at 20,700 spots per year. In a recent

effort to carve out a new student-to-work pathway, the government has proposed extending the time gradu-

ates can seek a job, reinstating the post-study work visa, and eliminating the cap on the general work visa by

20 21.

131

In October 2019, the government already exempted PhD level occupations from the general work

visa cap.

132

Other countries with straightforward student-to-work pathways include France, Germany, and New Zealand.

133

Besides having codiﬁed student-to-work pathways, other countries’ immigration systems also differ from the U.S.

system in other ways relevant for the AI workforce.

Entrepreneurial visas. Nearly 20 other countries—including Canada, France, the United Kingdom, and Israel—

have recently introduced speciﬁc visa programs for entrepreneurs (especially in tech), which make it possible for

domestically- and internationally-trained students to start businesses after graduating.

134

As previously noted, the

United States currently has no operational visa category for entrepreneurs.

Permanent residency. Many other countries provide easier paths to permanent residence than the United States,

which—since most student-to-worker pathways are initially temporary in nature—is important for graduates’ ability to

stay in the long term. In the United States,

135

as noted in Chapter 2, the number of eligible graduates and workers far

exceeds the number of available green card slots, especially among Indian and Chinese nationals. The United States

is not unique in this—prior to its recent reforms, caps on the number of skilled workers that can get permanent residency

in the United Kingdom led to similar bottlenecks in their system.

13 6

C. “SENDING” COUNTRIES: COMPARING CHINESE AND INDIAN

STUDENTS AND RETURNEES

The majority of foreign-born AI-relevant graduate students come from just two countries: China and India. While

some generalizations can be made about why students decide to stay in or leave the United States, in many cases

the relative importance and salience of different factors depends on country-speciﬁc details. Because of the outsize

importance of China and India, and to see stay-departure decision-making play out at a more granular level, this

section describes returnee dynamics among these two countries’ students speciﬁcally.

To summarize, there are both similarities and differences between the two communities. The main similarity is that

professional opportunities resulting from economic development are the prime reason that students and workers return

home. A salient difference is that Indian students face a much tougher immigration environment due to green card

queues. Many more Indian returnees report going home for immigration-related reasons. On the policy front, the

Indian government has been much less aggressive than the Chinese government in trying to recruit full-time returnees,

instead focusing on attracting diaspora ﬁnancial investment and remittances.

Indian students and policy

• Professional opportunities. Around half of Indian returnees cite professional opportunities as their main

reason for returning across a range of surveys.

137

Many private sector returnees cite budding startup ecosys-

tems in places like Bangalore and Hyderabad as particularly attractive.

13 8

• Immigration rules. As noted in Chapter 2, green card queues for Indian in the U.S. are very long; current

PhD graduates can expect to wait around 50 years for permanent residency.

139

Some high-tech immigrants

stuck on temporary status or in the queue now consider their decision to move to the United States the “worst

Center for Security and Emerging Technology46

decision of my life.”

140

In a survey of academics who returned to India, 26 percent of respondents said that

immigration problems were an important factor in their decision. For many academics, though, the main

problem was their inability to get work authorization for their spouse, as opposed to their own status.

141

another survey of mostly private sector returnees, 30 percent said visa issues were signiﬁcant.

142

• Personal and cultural considerations. In one survey of returnees, over 30 percent state they had fami-

ly-related reasons for returning.

143

In another, about 25 percent report either family reuniﬁcation or “Indian

cultural identity” as their primary reason.

144

A study of Indian medical doctors based abroad found that

40 percent were “ready to return,” “mainly for personal reasons.”

145

Interestingly, there have not been many Indian policy efforts to attract returnees, in contrast to China. One study

concludes that “the government in New Delhi has done little or nothing” to encourage high-skilled in-migration

because it is more concerned about increasing investment and remittances from abroad.

146

Another similarly

concludes that “there has not been any concerted effort to induce return migration in India, except in some

indirect and limited ways,” hypothesizing that an excess of domestic labor reduces the need for returnees and that

bureaucratic problems hampered the few nascent programs that did get launched.

147

Chinese students and policy

• Professional opportunities. One survey of Chinese returnees found that professional opportunities were

a “very” or “extremely” important factor for 70 percent of respondents.

148

Another study that focused

speciﬁcally on returned STEM faculty in China found that “job opportunities” was the most common rea-

son for return (cited by 46 percent of respondents).

149

Experimental research with U.S.-educated Chinese

STEM students has found that salary is an important determinant of returnee preferences;

150

the fact that

economic growth and investment in high-tech ﬁelds have driven up salaries and created more interesting

jobs in China is a big factor for many returnees.

151

Many returnees also feel they have better professional

networks in China than in the United States.

152

• Immigration rules. Visa problems do not seem to be a very important reason for return among academ-

ics

153

or prospective entrepreneurs (20 percent of whom rated expiring visas as an extremely to moderate-

ly important reason for return),

154

though in some surveys Chinese graduate students do report difﬁculties

ﬁnding employment as non-citizens,

155

which is likely at least partially visa-related.

• Personal and cultural considerations. In the same study of STEM faculty who returned to China pre-

viously cited, “family” was the second most important reason for returning, cited by 45 percent of re-

spondents.

156

Other commonly cited reasons were “wanting children to receive a Chinese education” (18

percent) and “did not adjust well to foreign culture” (10 percent). Another survey ﬁnds that 84 percent of

Chinese graduate students in the United States who intended to return home after their studies considered

“missing family/friends” an important reason for returning, with the second most important factor being

“cultural comfort” in China (79 percent).

157

There are many factors that disincentivize U.S.-educated students from returning to China. Salary gaps between the

United States and China are still large, even though Chinese salaries have been rising,

158

and roughly 80 percent

of returnees report lower-than-expected salaries.

159

Returnees also often experience challenges integrating into the

Chinese labor market, with 70 percent saying that their position did not match their experience and skills.

160

Many in

academia consider the research environment in China to be less merit-based, leading to frequent complaints about

plagiarism and the political nature of promotions.

161

Governments and university administrators also try to dictate

much of the content and output of research. One survey of academics ﬁnds that “the goal of research in China is

no longer seen as about the pursuit of knowledge; rather, it has become a pursuit designed to meet quantitative

indicators for one’s evaluation.”

162

Potential returnees who are thinking about going into the private sector have felt

deterred by the political environment as well.

163

The Chinese government has been very active in trying to encourage return, for example by offering ﬁnancial

incentives to both returnees and Chinese institutions, using a combination of embassy and other ofﬁcial networks to

Center for Security and Emerging Technology 47

create linkages, and providing job search and other administrative support. The most well-known returnee program is

the “Thousand Talents Program,” but there are many more.

Many analysts, however, doubt that government policies have been central in shaping returnee patterns to China,

and they may even unintentionally increase emigration. As a literature review on this topic states, “Most studies argue

that governments have limited impact on the return tide. Preferential policies for returnees can, in fact, increase the

numbers going abroad, since preferred beneﬁts are available only to returnees.”

164

Tensions between returnees and

“domestic” workers and scientists are currently commonplace, with those who have never gone abroad alleging

discrimination and those who have ﬁnding it hard to break into longer-standing domestic networks.

165

Moreover,

returnees are often thought to be of lower quality on average than those who remain abroad.

166

Center for Security and Emerging Technology48

Center for Security and Emerging Technology 49

1. “Artiﬁcial Intelligence for the American People,” White House, https://www.whitehouse.gov/ai/.

2. CCTV, “Why Is Xi Jinping’s ‘First Resource’ So Important? (“习近平眼里的“第一资源”为何如此重

要”), July 18, 2018, http://politics.people.com.cn/n1/2018/0718/c1001-30155931.html.

3. Elsa Kania, “China’s AI Talent ‘Arms Race,’” The Strategist, April 23, 2018, https://www.aspistrategist.

org.au/chinas-ai-talent-arms-race/. People in industry, such as David Wipf, a Principal Researcher at

Microsoft Research, similarly talk about a “battle for talent” and its importance to AI leadership; see e.g.

David Cyranoski, “China Enters the Battle for AI Talent,” Nature, January 15, 2018, https://www.nature.

com/articles/d41586-018-00604-6.

4. “Interim Report to Congress,” National Security Commission for Artiﬁcial Intelligence (November 2019),

16, https://drive.google.com/ﬁle/d/153OrxnuGEjsUvlxWsFYauslwNeCEkvUb/view.

5. Daniel Culbertson, “Demand for AI Talent on the Rise,” Indeed Hiring Lab (blog), March 1, 2018,

https://www.hiringlab.org/2018/03/01/demand-ai-talent-rise/.

6. For a more extensive discussion of talent shortages in AI and the proportion of foreign-born workers in

the U.S. AI workforce, see Remco Zwetsloot, Roxanne Heston, and Zachary Arnold, Strengthening the U.S.

AI Workforce: A Policy and Research Agenda (Center for Security and Emerging Technology: September

2019), https://cset.georgetown.edu/wp-content/uploads/CSET_U.S._AI_Workforce.pdf.

7. “CSRankings: Computer Science Rankings,” CSRankings, accessed August 12, 2019, http://csrankings.

org.

8. Lesleyanne Hawthorne, “Attracting and Retaining International Students as Skilled Migrants,” in

High-Skilled Migration: Drivers and Policies, Mathias Czaika (Oxford University Press, 2018), http://

www.oxfordscholarship.com/view/10.1093/oso/9780198815273.001.0001/oso-9780198815273-

chapter-10. This is also better for the employees, who tend to ﬁnd better employment matches more easily

and receive higher starting wages; see John Bound, Murat Demirci, Gaurav Khanna, and Sarah Turner,

“Finishing Degrees and Finding Jobs: US Higher Education and the Flow of Foreign IT Workers,” National

Bureau of Economic Research (University of Chicago Press, 2015): 64-65, https://www.journals.

uchicago.edu/doi/pdfplus/10.1086/680059.

9. Bound et al., “Finishing Degrees and Finding Jobs”; Christiane Kuptsch, “Students and Talent Flow - the

Case of Europe: From Castle to Harbour?,” in Competing for Global Talent, ed. Christiane Kuptsch and

Pang Eng Fong (International Institute for Labour Studies, 2006): 41, http://www.ilo.org/wcmsp5/

groups/public/@dgreports/@dcomm/@publ/documents/publication/wcms_publ_9290147768_

en.pdf#page=44.

10. Stuart Anderson, “Immigrants and Billion Dollar Startups,” National Foundation for American Policy,

March 2016, https://nfap.com/wp-content/uploads/2016/03/Immigrants-and-Billion-Dollar-Startups.

NFAP-Policy-Brief.March-2016.pdf. Between 1995 and 2005, roughly 50 percent of U.S. technology

and engineering companies founded by immigrants were founded by former international students; see

Vivek Wadhwa, Ben A. Rissing, AnnaLee Saxenian, and G. Gerefﬁ, “Education, Entrepreneurship and

Endnotes

Center for Security and Emerging Technology50

Immigration: America’s New Immigrant Entrepreneurs, Part II,” SSRN, June 2007, https://papers.ssrn.

com/sol3/papers.cfm?abstract_id=991327.

11. CSET analysis of Department of Labor PERM data, 2015-2019 (see Appendix A). Only 60 percent

of applications describe workers’ education history, so this is an estimate with some uncertainty. Among

applicants who reported their highest degree, 60 percent reported a master’s degree and 9 percent

reported a PhD degree. This compares to 41 percent and 5 percent at non-AI employers. No comparably

detailed data is available for domestic workers or for foreign-born workers on other visa categories.

12. Jean-Francois Gagne, “Global AI Talent Report 2019,” https://jfgagne.ai/talent-2019/.

13. CSET analysis of Department of Labor PERM data, 2015-2019 (see Appendix A). Other ﬁelds all

accounted for less than two percent of workers in the dataset. Note that computer engineering (CE) is a

subﬁeld of electrical engineering (EE), and often not reported by students or universities as a separate

ﬁeld. Universities also often do their CS and EE training within a single department; Berkeley, for example,

teaches AI in its “Department of Electrical Engineering and Computer Science.” For these reasons, we

report statistics on CS/EE student populations below when CE-speciﬁc data is unavailable.

14. CSET analysis of Computing Research Association data (see Appendix A). In the Fall 2017 CRA Data

Buddies survey of graduate students in computing programs, 1,437 out of 4,550 respondents (32 percent)

report “Artiﬁcial Intelligence” as one of their specialties. In the 2018 CRA Taulbee survey of PhD students

in computing programs, roughly 20 percent selected “Artiﬁcial Intelligence / Machine Learning” as their

specialty, up from 14 percent in 2016 and 2017 and 10 percent in prior years.

15. The 2017 version of this survey was the ﬁrst to publish separate statistics for master’s and PhD students.

At that time there were, at the master’s level, 75,618 CS students (of which 58 percent were temporary

visa holders) and 29,816 EE students (of which 68 percent were temporary visa holders). At the PhD level,

there were 14,291 CS students (of which 63 percent were temporary visa holders) and 17,936 EE students

(of which 71 percent were temporary visa holders). Master’s students thus make up roughly 75 percent of

all graduate students. See “Table 4-4b. Citizenship, Ethnicity, and Race of Master’s Students, By Detailed

Field: 2017,” National Science Foundation, last accessed August 12, 2019, https://ncsesdata.nsf.gov/

gradpostdoc/2017/html/gss17-dt-tab004-4b.html and “Table 4-4c. Citizenship, Ethnicity, and Race of

Doctoral Students, By Detailed Field: 2017,” National Science Foundation, last accessed August 12, 2019,

https://ncsesdata.nsf.gov/gradpostdoc/2017/html/gss17-dt-tab004-4c.html.

16. Students from Russia, a geopolitically relevant country with signiﬁcant ambitions in AI, accounted for

only 0.4 percent of U.S. CS PhD graduates in 2016.

17. CSET calculations based on data from NSF’s Survey of Graduate Students and Postdoctorates in

Science and Engineering.

18. Michael G. Finn and Leigh Ann Pennington, Stay Rates of Foreign Doctorate Recipients from U.S.

Universities, 2013 (Oak Ridge Institute for Science and Engineering Education: January 2018), Figure 2,

https://orise.orau.gov/stem/reports/stay-rates-foreign-doctorate-recipients-2013.pdf.

19. Patrick Gaule, “Who Comes Back and When? Return Migration Decisions of Academic Scientists,”

Economcis Letters 124, no. 3 (September 2014), https://www.sciencedirect.com/science/article/abs/

pii/S0165176514002699.

20. Richard van Noorden, “Global Mobility: Science on the Move,” Nature 490 (October 2012),

https://www.nature.com/articles/490326a.

21. For example, Yolanda Gil, writing on behalf of the Association for the Advancement of Artiﬁcial

Intelligence (AAAI), states that “International PhD graduates are leaving the US in larger numbers than

Center for Security and Emerging Technology 51

before, in part due to immigration constraints but also due to the availability of attractive opportunities

for AI overseas.” See “AAAI Response to NITRD RFI: National Artiﬁcial Intelligence Research and

Development Strategic Plan,” Association for the Advancement of Artiﬁcial Intelligence, October 26, 2018:

8, https://www.nitrd.gov/rﬁ/ai/2018/AI-RFI-Response-2018-Yolanda-Gil-AAAI.pdf. Some, such as

Vivek Wadhwa and co-authors, already stated that “it appears likely that the trend of skilled workers to

return has accelerated [since 2000]” roughly a decade ago; see Vivek Wadhwa, AnnaLee Saxenian,

Richard B. Freeman, and G. Gerefﬁ, “America’s Loss is the World’s Gain: America’s New Immigrant

Entrepreneurs, Part 4,” SSRN, March 2, 2009, 6 https://papers.ssrn.com/sol3/papers.cfm?abstract_

id=1348616. For China-speciﬁc examples of claims about increased returnee rates, see Meng Jing, “Why

Trump’s Clampdown on Academia is Forcing Many Chinese Researchers into a Difﬁcult Corner,” South

China Morning Post, June 13, 2019, https://www.scmp.com/tech/policy/article/3014185/why-trumps-

clampdown-academia-forcing-many-chinese-researchers, Bloomberg Businessweek, “Chinese Workers

Abandon Silicon Valley for Riches Back Home,” January 10, 2018, https://www.bloomberg.com/news/

articles/2018-01-10/chinese-workers-abandon-silicon-valley-for-riches-back-home, and Richard Engel

and Kennett Werner, “China’s Rising Tech Scene Threatens U.S. Brain Drain as ‘Sea Turtles’ Return Home,”

NBC News, July 14, 2019, https://www.nbcnews.com/tech/tech-news/china-s-rising-tech-scene-

threatens-u-s-brain-drain-n1029256.

22. Emily Feng, “Chinese Students are Under Suspicion in China and the U.S.,” NPR, June 4, 2019,

https://www.npr.org/2019/06/04/729510902/a-foot-in-both-worlds-students-under-suspicion-in-

china-and-the-u-s; Elizabeth Redden, “Science vs. Security,” Inside Higher Ed, April 16, 2019, https://

www.insidehighered.com/news/2019/04/16/federal-granting-agencies-and-lawmakers-step-scrutiny-

foreign-research.

23. Some research indicates that there have been increases in the number of Chinese students who say

they eventually want to return home, even if they initially stay; see Ryan P. Kellogg, “China’s Brain Gain?

Attitudes and Future Plans of Overseas Chinese Students in the US,” Journal of Chinese Overseas 8, no.

1 (September 2010), https://www.researchgate.net/publication/228120877_China's_Brain_Gain_

Attitudes_and_Future_Plans_of_Overseas_Chinese_Students_in_the_US. In a recent survey, 54 percent

of Chinese STEM PhD students in the U.S. said they intended to work in the U.S. ﬁrst before eventually

returning home; see Michael Roach, Henry Sauermann, and John Skrentny, “Are Foreign STEM PhDs More

Entrepreneurial? Entrepreneurial Characteristics, Preferences and Employment Outcomes of Native and

Foreign Science & Engineering PhD Students,” NBER Working Paper, September 2019, https://www.nber.

org/papers/w26225. However, in the past many such surveys have seen higher intention to depart rates

than were later found in actual departure rate statistics, so it’s hard to know how much faith to put in such

numbers.

24. Country of undergraduate education is commonly used as a proxy for nationality in academic studies

when researchers do not have access to underlying citizenship data. The proxy has clear shortcomings,

however, the main one of which is that many non-Western international students do their undergraduate

education in the United States or Europe and that this proxy would misclassify their nationality. The

undergraduate proxy in our data would imply that about 47 percent of AI PhDs in our data are American

and 15 percent from OECD countries. From NSF survey data it is known that these ﬁgures are lower for

CS PhDs more broadly (35 and 5 percent, respectively; see Figure 1). This suggests our proxy could

misclassify the nationality of about 20 percent of our sample; a large number, but at the same time not so

large that it could qualitatively change the results of the paper.

25. Grant C. Black and Paula E. Stephan, "The Importance of Foreign PhD Students to U.S. Science," in

Science and the University, ed. Paula E. Stephan and Ronald G. Ehrenberg (Madison: The University of

Wisconsin Press, 2007): 113-130.

Center for Security and Emerging Technology52

26. Jeffrey Grogger and Gordon H. Hanson, “Attracting Talent: Location Choices of Foreign-Born PhDs

in the United States,” Journal of Labor Economics 33, no. 3 (2015); Linda Van Bouwel and Reinhilde

Veugelers, “An ‘Elite Brain Drain’: Are Foreign Top PhDs More Likely to Stay in the U.S.?,” SSRN,

January 15, 2012, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2109278. Other studies ﬁnd

no correlation between measures of students’ or scientists’ ability (e.g. publications, degree-granting

institution, awards) and their decision to return to their home country. See for example Gaule, “Who

Comes Back and When?”

27. CSET plans to collect this data in future research.

28. Other factors that have been found to matter that are not in this list include domestic political stability,

the level of democracy, and general quality of life (e.g. environmental) factors. See for example Dongbin

Kim, Charles A. S. Bankart, and Laura Isdell, “International Doctorates: Trends Analysis on Their Decision

to Stay in US,” Higher Education 62, no. 2 (August 2011): 145-146 and Grogger and Hanson, “Attracting

Talent.”

29. Drawing on survey evidence, Xueying Han and Richard P. Appelbaum, “Will They Stay or Will They

Go? International STEM Students Are Up for Grabs,” Ewing Marion Kauffman Foundation, July 2016,

found 74 percent of international graduate students in STEM ﬁelds reported future career opportunities as

a reason to come, though only 22 percent came speciﬁcally because they wanted to stay in the United

States. See also Paula Stephan, Giuseppe Scellato, and Chiara Franzoni, “International Competition for

PhDs and Postdoctoral Scholars: What Does (and Does Not) Matter,” Innovation Policy and the Economy

15, ed. William R. Kerr, Josh Lerner, and Scott Stern (Chicago: The University of Chicago Press, 2015):

73-133. Researchers have also found evidence of international students’ educational choices changing

depending on post-graduation job prospects; see Takao Kato and Chad Sparber, “Quotas and Quality:

The Effect of H-1B Visa Restrictions on the Pool of Prospective Undergraduate Students from Abroad,” The

Review of Economics and Statistics 95, no. 1 (March 2013), https://doi.org/10.1162/REST_a_00245;

Bound et al., “Finishing Degrees,” and Catalina Amuedo-Dorantes, Delia Furtado, and Huanan Xu, “Did

OPT Policy Chances Help Steer and Retain Foreign Talent into STEM?,” IZA Discussion Papers No. 11548,

Institute of Labor Economics (IZA), 2018. https://www.econstor.eu/bitstream/10419/180566/1/

dp11548.pdf.

30. Hawthorne, “Attracting and Retaining,” 202.

31. Shulamit Kahn and Megan MacGarvie, “The Impact of Permanent Residency Delays for STEM PhDs:

Who Leaves and Why,” National Bureau of Economic Research, October 2018, http://www.nber.org/

papers/w25175. See also Gaule, “Who Comes Back and When,” and Grogger and Hanson, “Attracting

Talent.” Except in Kahn and McGarvie, “Impact of Permanent Residency,” who use measures of a country’s

domestic S&T environment, most studies use GDP per capita in home country as proxies for professional

opportunities. It is hard to parse out how much of GDP’s effect comes from job opportunities speciﬁcally

versus general quality of life improvements.

32. Philip G. Altbach, “Impact and Adjustment: Foreign Students in Comparative Perspective,” Higher

Education 21, no. 3 (April 1991), https://www.jstor.org/stable/3447137.

33. Richard Freeman, Paula Stephan, and John Trumbour, “Career Patterns of Foreign Born Scientists and

Engineers Trained and or Working in the U.S.,” National Bureau of Economic Research Workshop Report

(January 2008): 3, https://users.nber.org/~sewp/Workshop.Report.November.2007.pdf.

34. In Han and Appelbaum, “Will They Stay,” 80 percent of those who wish to remain in the United States

cite “job opportunities” as their main reason, and over 50 percent also mention professional networks and

salary as important reasons. See also Mark Musumba, Yanhong H. Jin, and James W. Mjelde, “Factors

Inﬂuencing Career Location Preferences of International Graduate Students in the United States,” Education

Center for Security and Emerging Technology 53

Economics 19, no. 5 (September 2009): 501-517, https://doi.org/10.1080/09645290903102902.

35. Chiara Franzoni, Giuseppe Scellato, Paula Stephan, “Foreign Born Scientists: Mobility Patterns for

16 Countries,” Nature Biotechnology 30, no. 12 (December 2012), Figure 2, https://www.nature.com/

articles/nbt.2449.pdf; Reinhilde Veugelers and Linda Van Bouwel, “Destinations of Mobile European

Researchers: Europe versus the United States,” in Global Mobility of Research Scientists: The Economics of

Who Goes Where and Why, ed. Aldo Geuna (London: Academic Press, 2015), 215-237, also ﬁnd that,

among European-born researchers who got their PhDs in Europe, “career motivations are more strongly

related to mobility to the United States.” See also Susan Guthrie, Catherine Lichten, Jennie Corbett, and

Steven Wooding, International Mobility of Researchers: A Review of the Literature (Santa Monica, CA:

RAND Corporation, 2017), https://royalsociety.org/~/media/policy/projects/international-mobility/

researcher-mobility-report-review-literature.pdf.

36. While immigration difﬁculties are universally agreed to be important, some consider them secondary

to professional considerations. For example, William Kerr argues: “Restrictive US immigration policy plays

a role [in encouraging return migration], but this role is likely secondary to the attractive opportunities for

many in returning home.” William R. Kerr, “US High-Skilled Immigration, Innovation, and Entrepreneurship:

Empirical Approaches and Evidence,” in The International Mobility of Talent and Innovation: New

Evidence and Policy Implications, ed. Carsten Fink and Ernest Miguelez (Cambridge: Cambridge

University Press, 2017).

37. Han and Appelbaum, “Will They Stay,” 15.

38. Hawthorne, “Attracting and Retaining,” 201-202.

39. Kahn and McGarvie, “Impact of Permanent Residency.” Some of the effects found in this paper are

complex and non-linear (i.e. the effects of an extra year of delay likely depend on how much delay

students already faced).

40. Pooja B. Vijayakumar and Christopher J. L. Cunningham, “An Indentured Servant: The Impact of

Green Card Waiting Time on the Life of Highly Skilled Indian Immigrants in the United States,” Industrial

and Organizational Psychology Translational Research and Working Papers, https://scholar.utc.edu/

iopsy/2/.

41. Vivek Wadhwa, The Immigrant Exodus: Why America is Losing the Global Race to Capture

Entrepreneurial Talent (Philadelphia: Wharton Digital Press, 2012).

42. Sonia Paul, “In Limbo, Some Families in the US with H-1B Visas Begin to Make Other Plans,” PRI,

September 18, 2018, https://www.pri.org/stories/2018-09-18/limbo-some-families-us-h-1b-visas-

begin-make-other-plans.

43. Han and Appelbaum, “Will They Stay,” ﬁnds that 80 percent of international STEM graduate students

who intend to leave the U.S. cited family as an inﬂuential factor in their decision. The next most common

factors are “cultural reasons” (37 percent), “friends” (35 percent), and “social reasons” (32 percent).

44. Heike C. Alberts and Helen D. Hazen, ““There Are Always Two Voices…”: International Students’

Intentions to Stay in the United States or Return to Their Home Countries,” International Migration 43, no. 3

(November 2005); see also Guthrie et al., International Mobility of Researchers.

45. Han and Appelbaum, “Will They Stay.” These challenges are typically greater for students from

countries whose ﬁrst language is not English, such as China.

46. Jing, “Why Trump’s Clampdown.”

47. Wadhwa et al., “America’s Loss,” Figure 6.

48. There are not many instances of dual afﬁliations in our dataset, i.e. of people who hold simultaneous

positions in two sectors. Out of a sample of 1,000 individuals for whom we looked into this question, 48

have simultaneous afﬁliations with academia and the private sector, 12 with academia and government or

nonproﬁt institutions, and 12 with the private sector and government or nonproﬁt institutions.

49. Catalina Amuedo-Dorantes and Delia Furtado, “Settling for Academia? H-1B Visas and the Career

Choices of International Students in the United States,” The Journal of Human Resources 54, no. 2 (Spring

2019): 401-429. Different studies come to different conclusions about how much of this “settling for

academia” effect can be explained by immigration restrictions versus students’ preferences (see Xiaohuan

Lan, “Permanent Visas and Temporary Jobs: Evidence from Postdoctoral Participation of Foreign PhDs in

the United States,” Journal of Policy Analysis and Management 31, no. 3 (2012): 623-640; Ina Ganguli

and Patrick Gaulé, “Will the U.S. Keep the Best and the Brightest (As Post-Docs)? Career and Location

Preferences of Foreign STEM PhDs,” NBER Working Paper Series 24838, National Bureau of Economic

Research, Cambridge, MA, July 2018.

50. Michael Roach and John Skrentny, “Why Foreign STEM PhDs are Unlikely to Work for US Technology

Startups,” Proceedings of the National Academy of Sciences 116, no. 34 (August 2019), https://doi.

org/10.1073/pnas.1820079116.

51. Roach et al., “Are Foreign STEM PhDs More Entrepreneurial?”, Figure 1.

52. An alternative explanation for this ﬁnding is that foreign nationals could simply have different career

preferences. However, research has consistently shown greater, not lesser, entrepreneurial ambitions

among international than among domestic students, which favors the immigration regulation-focused

explanation of our ﬁndings; see Roach et al., “Are Foreign STEM PhDs More Entrepreneurial?”. CSET

analysis of CRA Data Buddies data, available upon request, conﬁrms that this ﬁnding also holds for

graduate students in AI speciﬁcally.

53. Kahn and McGarvie, “Impact of Permanent Residency”; Roach et al., “Are Foreign STEM PhDs More

Entrepreneurial?”

54. Possible explanatory factors include differences in the relative attractiveness of countries’ universities

and private sectors in AI, differences in general career preferences among nationals from these countries,

differences in the availability of domestically trained talent, and differences in the openness of countries’

universities to those trained abroad. Some of the variation may also be due to random chance, since many

countries have relatively few data points.

55. Ofﬁce of Information and Regulatory Affairs, “Practical Training Reform,” Fall 2017, available at

https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=201710&RIN=1653-AA76. On

subsequent inaction, see Laura D. Francis, “Foreign Student Training Program Gets Little Notice From Trump

(2),” Bloomberg Law, March 26, 2019, https://news.bloomberglaw.com/daily-labor-report/foreign-

student-training-program-gets-little-notice-from-trump-2.

56. Ofﬁce of Information and Regulatory Affairs, “Practical Training Reform,” Fall 2019, available at

https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=201910&RIN=1653-AA76. The Fall

2019 rule description is more generic than the Fall 2017: “ICE will amend existing regulations and revise

the practical training options available to nonimmigrant students on F and M visas” (2019), compared to

“ICE will propose this rule to improve protections of U.S. workers who may be negatively impacted by

employment of nonimmigrant students on F and M visas. The rule is a comprehensive reform of practical

training options intended to reduce fraud and abuse” (2017).

57. Laura D. Francis, “H-1B Program ‘Alive and Well,’ Latest Visa Lottery Results Show,” Bloomberg Law,

April 17, 2019, https://news.bloomberglaw.com/daily-labor-report/h-1b-program-alive-and-well-latest-

visa-lottery-results-show.

Center for Security and Emerging Technology54

Center for Security and Emerging Technology 55

58. The L-1 visa is used by multinational U.S. AI companies to transfer employees from foreign ofﬁces to

U.S. ofﬁces; since students are already based in the country this visa category is at best indirectly relevant

(e.g. if students ﬁrst go abroad before coming back to the United States). The J-1 visa is used by interns,

post-docs, and scholars. The O-1 visa is for individuals with “extraordinary ability,” a qualiﬁcation that

most recent PhD graduates do not meet (according to the current legal interpretation of the term).

59. U.S. Citizenship and Immigration Services, “Estimated Dependent Multiplier for Employment Based

Visa Categories, FY2016,” 2018, https://www.uscis.gov/sites/default/ﬁles/ﬁles/nativedocuments/

Count_of_Approved_I-140_I-360_and_I-526_Petitions_as_of_April_20_2018_with_a_Priority_Date_

On_or_After_May_2018.PDF.

60. David Bier, “Immigration Wait Times from Quotas Have Doubled: Green Card Backlogs Are Long,

Growing, and Inequitable,” CATO Institute, June 18, 2019, Table 4, https://www.cato.org/publications/

policy-analysis/immigration-wait-times-quotas-have-doubled-green-card-backlogs-are-long.

61. Jennifer Hunt and Bin Xie, “How Restricted is the Job Mobility of Skilled Temporary Work Visa

Holders?,” Journal of Policy Analysis and Management 38, no. 1 (Winter 2019), https://onlinelibrary.

wiley.com/doi/full/10.1002/pam.22110.

62. Other beneﬁts of citizenship (compared to permanent residency) include the eligibility to vote and

the ability to apply for a green card for relatives. One common explanation for why permanent residents

choose not to naturalize is that such beneﬁts are not worth the cost and effort of the required application

process. Another is that certain countries do not allow dual citizenship, meaning naturalization in the

United States would require someone to give up their citizenship in their country of birth.

63. Foreign-born workers who had not previously studied in the United States would most commonly enter

the U.S. labor force on H-1B, L-1, J-1, or O-1 visas. Out of those, the H-1B is by far the most numerous,

but as Table 3 notes, more than half of new H-1B entrants are now former international students. We have

been unable to ﬁnd good data on the background of J-1, L-1, or O-1 holders, but even if none of them

were previously international students—which we consider highly unlikely—they would be outnumbered

by OPT and H-1B holders. For more details, see Zachary Arnold, Roxanne Heston, Remco Zwetsloot, and

Tina Huang, Immigration Policy and the U.S. AI Sector: A Preliminary Assessment (Center for Security and

Emerging Technology: September 2019), Table 1, https://cset.georgetown.edu/wp-content/uploads/

CSET_Immigration_Policy_and_AI.pdf#page=14.

64. Neil G. Ruiz and Abby Budiman, “Number of Foreign College Students Staying and Working in

U.S. After Graduation Surges,” Pew Research Center, May 10, 2018, https://www.pewresearch.org/

global/2018/05/10/number-of-foreign-college-students-staying-and-working-in-u-s-after-graduation-

surges/.

65. “F-1 Students Obtaining Another Non-Immigrant Classiﬁcation: Fiscal Year 2008-2018 Approvals,”

U.S. Citizenship and Immigration Services, https://www.uscis.gov/sites/default/ﬁles/USCIS/

Resources/Reports/Report_-_F-1_Students_Obtaining_Another_Nonimmigrant_Classiﬁcation.pdf, Figure

3. Whereas most other ﬁgures in the table are for FY2017, this percentage is available for FY2018.

66. “Characteristics of H-1B Specialty Occupation Workers: Fiscal Year 2017 Annual Report to

Congress,” U.S. Citizenship and Immigration Services, https://www.uscis.gov/sites/default/ﬁles/ﬁles/

nativedocuments/Characteristics_of_H-1B_Specialty_Occupation_Workers_FY17.pdf, Table 6.

67. USCIS has, to our knowledge, not released information about the educational ﬁelds that H-1B holders

received their training in. We estimate that more than 35 percent of H-1B holders got their training in

AI-relevant ﬁelds because roughly 70 percent of H-1B holders work in “computer-related occupations,”

and, because H-1B workers are required to have degrees relevant to the job they work in, it seems

very likely that at least half of these workers trained in computer science or computer engineering. See

Center for Security and Emerging Technology56

“Characteristics of H-1B Specialty Occupation Workers,” Tables 8A and 8B.

68. CSET calculations from Department of Labor PERM data (see Appendix A). Percentages indicate

proportions among principal applicants, excluding dependents.

69. Neil G. Ruiz and Abby Budiman, “Number of Foreign College Graduates Staying in U.S. to Work

Climbed Again in 2017, but Growth has Slowed,” Pew Research Center, July 25, 2018, https://www.

pewresearch.org/fact-tank/2018/07/25/number-of-foreign-college-graduates-staying-in-u-s-to-work-

climbed-again-in-2017-but-growth-has-slowed/.

70. Speciﬁcally, master’s students account for 57 percent of all OPT grantees between 2004-2016 (for

a total of 841,000), with doctoral students accounting for approximately another 15 percent. Among

master’s students, 27 percent of OPT grantees had degrees in engineering and 22 percent in computer and

information sciences. Among doctoral students, 34 percent of grantees had degrees in engineering and 6

percent in computer and information sciences. The data reported by Pew is not disaggregated within the

ﬁeld of engineering. See Ruiz and Budiman, “Number of Foreign College Students.”

71. “F-1 Students Obtaining Another Non-Immigrant Classiﬁcation: Fiscal Year 2008-2018 Approvals,”

U.S. Citizenship and Immigration Services, https://www.uscis.gov/sites/default/ﬁles/USCIS/

Resources/Reports/Report_-_F-1_Students_Obtaining_Another_Nonimmigrant_Classiﬁcation.pdf.

72. Neil G. Ruiz, “Key Facts About the U.S. H-1B Visa Program,” Pew Research Center, April 27, 2017,

https://www.pewresearch.org/fact-tank/2017/04/27/key-facts-about-the-u-s-h-1b-visa-program/.

73. Note that this data is on applicants for permanent residency, not those actually granted permanent

residency. However, because the PERM process “involves a set of technical, expensive, and highly time-

consuming steps [and] complex strategies that extend over a period of typically six to eight months,”

employers are likely to complete the process only for highly desirable employees with a strong chance of

obtaining green cards and staying employed in the United States; see Anna Angel, “5 Key Considerations

When Initiating a PERM Labor Certiﬁcation for Your Employee,” Ogletree Deakins, published June 9,

2017, https://ogletree.com/insights/2017-06-09/5-key-considerations-when-initiating-a-perm-labor-

certiﬁcation-for-your-employee/. For more on the permanent residency application process, see Maggio

Kattar, “The Three Stages of Employer Sponsored Permanent Residence via PERM,” https://maggio-

kattar.com/three-stages-employer-sponsored-permanent-residence-perm/.

74. Michael G. Finn and Leigh Ann Pennington, Stay Rates of Foreign Doctorate Recipients from U.S.

Universities, 2013 (Oak Ridge Institute for Science and Foundation: January 2018), https://orise.orau.

gov/stem/reports/stay-rates-foreign-doctorate-recipients-2013.pdf#page=13. CSET plans to calculate

AI-speciﬁc naturalization rates in future research pending the necessary data access.

75. For data on long-term trends, see David Bier, “Immigration Wait Times from Quotas Have Doubled:

Green Card Backlogs Are Long, Growing, and Inequitable,” CATO Institute, June 18, 2019, https://

www.cato.org/publications/policy-analysis/immigration-wait-times-quotas-have-doubled-green-card-

backlogs-are-long. For data on recent changes, see National Foundation for American Policy, “H-1B

Denial Rates: Analysis of H-1B Data for First Three Quarters of FY2019,” NFAP Policy Brief, October 2019,

https://nfap.com/wp-content/uploads/2019/10/H-1B-Denial-Rates-Analysis-of-FY-2019-Numbers.

NFAP-Policy-Brief.October-2019.pdf.

76. “Infographics and Data,” Institute for International Education, accessed August 13, 2019, https://

www.iie.org/Research-and-Insights/Project-Atlas/Explore-Data.

77. Original CSET translation of “中国人工智能城市发展白皮书” by 赛迪顾问股份有限公司, available

in Chinese at http://xqdoc.imedao.com/164f00e92bb573fd4b9e6ce1.pdf. Full translation of the

document available from CSET upon request.

Center for Security and Emerging Technology 57

78. Xueying Han, Galen Stocking, Matthew A. Gebbie, and Richard P. Appelbaum, “Will They Stay

or Will They Go? International Graduate Students and Their Decisions to Stay or Leave the U.S. upon

Graduation,” PLoS ONE (March 2015), Table 1.

79. For evidence of this on a range of S&T metrics, see Task Force on American Innovation, Second Place

America? Increasing Challenges to U.S. Scientiﬁc Leadership 2019 Benchmarks, May 2019, http://

www.innovationtaskforce.org/benchmarks2019/.

80. Dion Rabouin, “The World is Catching up on AI,” Axios, July 29, 2019, https://www.axios.com/us-

share-artiﬁcial-intelligence-investment-cdf9bf0f-d036-4b21-b96b-7ab2e2d8332a.html.

81. Tim Dutton, Brent Barron and Gaga Boskovic, Building an AI World: Report on National and

Regional AI Strategies, CIFAR, December 2018, 10, https://www.cifar.ca/docs/default-source/ai-

society/buildinganaiworld_eng.pdf.

82. “Technology and Geopolitics: Navigating a Future of Tech Uncertainty,” Asia Paciﬁc Foundation

of Canada (conference report), October, 2019, https://www.asiapaciﬁc.ca/sites/default/ﬁles/apf_

canada_technology_and_geopolitics_conference_report.pdf.

83. Han et al., “Will They Stay,” 15.

84. National Foundation for American Policy, “Declining International Student Enrollment at U.S.

Universities,” NFAP Policy Brief, February 2018, https://nfap.com/wp-content/uploads/2018/02/

Decline-in-International-Student-Enrollment.NFAP-Policy-Brief.February-2018-2.pdf.

85. NAFSA, Losing Talent: An Economic and Foreign Policy Risk America Can’t Ignore, May 2019,

https://www.nafsa.org/_/File/_/nafsa-losing-talent.pdf#page=8.

86. Ann Saphir, “As Companies Embrace AI, It’s a Job-Seeker’s Market,” Reuters, October 15, 2018,

https://www.reuters.com/article/us-usa-economy-artiﬁcialintelligence/as-companies-embrace-ai-its-

a-job-seekers-market-idUSKCN1MP10D.

87. Alex Usher, “Has President Trump Scared Away All the Foreign Students? The Facts Behind Fears of a

Higher-Education Revenue Recession,” Education Next, https://www.educationnext.org/has-president-

trump-scared-away-foreign-students-facts-behind-fears-higher-education-revenue-recession/.

88. Speciﬁcally, the three challenges to OPT are:

• An active lawsuit by the Washington Alliance of Technology Workers arguing that the Department of

Homeland Security overstepped its regulatory authority in creating OPT.

• A proposed rule on the DHS Fall 2019 regulatory agenda making changes to the OPT program (discussed

and referenced in Chapter 2 of this paper).

• A proposed bill that would eliminate OPT, the Fairness for High-Skill Americans Act of 2019, H.R. 3564,

116th Cong. (2019).

For more background, see Elizabeth Redden, “Is OPT in Peril?”, Inside Higher Ed, November 26,

2019, https://www.insidehighered.com/news/2019/11/26/lawsuit-challenges-program-allows-

international-students-work-us-after-graduating.

89. Erica L. Green, “Visa Delays at Backlogged Immigration Service Strand International Students,”

The New York Times, June 16, 2019, https://www.nytimes.com/2019/06/16/us/politics/visas-

international-students.html.

90. Elizabeth Redden, “Waiting for Work Authorization,” Inside Higher Ed, June 25, 2019, https://www.

insidehighered.com/news/2019/06/25/international-students-applying-work-authorization-face-

longer-wait-times. Attorneys report that even when the rule was in place, USCIS had stopped adhering

to it in recent years. Reinstatement of the formal rule would thus have to be paired with supportive

guidance and, where necessary, additional resources.

Center for Security and Emerging Technology58

91. Congress recently conducted its ﬁrst backlog oversight hearing: “Policy Changes and Processing

Delays at U.S. Citizenship and Immigration Services,” U.S. House Committee on the Judiciary, July 16,

2019, https://judiciary.house.gov/legislation/hearings/policy-changes-and-processing-delays-us-

citizenship-and-immigration-services.

92. For more details on the various relevant F-1 policies, see Marnette Federis, “Visa Rules Are Restricting

the Future of International Students in the US,” PRI’s The World, June 20, 2019, https://www.pri.org/

stories/2019-06-20/visa-rules-are-restricting-future-international-students-us; and Stuart Anderson,

“USCIS Policy Change Could Bar Many International Students,” Forbes, June 1, 2018, https://www.

forbes.com/sites/stuartanderson/2018/06/01/uscis-policy-change-could-bar-many-international-

students/#54ed03b81600.

93. Currently, students are not allowed to express an intention to stay in the United States past the

expiration of their nonimmigrant (temporary) student visa. This rule has been waived for certain other

nonimmigrant visa holders, including those on H-1Bs, meaning they can express what is generally referred

to as “dual intent.”

94. International students in fact do sometimes start companies today, either during their OPT time or

by structuring their company in such a way that they can technically be employed by it so that they

become eligible for an employment visa such as an H-1B. Both of these options involve large downsides,

however. The former means that students are only guaranteed residency in the United States for three

years (the maximum duration of OPT), which is often too short a timeline for an ambitious entrepreneur.

The latter often involves the immigrant founders having to forfeit legal ownership, which makes the

prospect of starting companies less attractive. A third option is to use an O visa, but only a small number

of prospective entrepreneurs will be eligible. See Pavithra Mohan, “With the U.S. Startup Visa on Hold,

Immigrant Entrepreneurs Get Creative,” Fast Company, September 17, 2019, https://www.fastcompany.

com/90403404/what-immigrant-entrepreneurs-can-do-without-a-startup-visa.

95. CSET interview with U.S. immigration lawyer (anonymous), August 2019.

96. For a more detailed discussion, see Arnold et al., Immigration Policy and the U.S. AI Sector, 17-22.

97. William R. Kerr, The Gift of Global Talent (Stanford: Stanford University Press, 2019), 174.

98. To obtain a labor certiﬁcation from the Department of Labor, as is currently required for EB-2 and

EB-3 green card applicants, sponsoring companies are required to attempt recruitment of alternative

candidates for a certain period with proscribed methods, request a “prevailing wage determination” for

the job they are hiring for, and submit complex paperwork. This can take months or even over a year,

thereby making hiring prospective green card applicants burdensome (or, for certain short-term needs,

impossible). If the Department of Labor added AI professionals (appropriately deﬁned) to its “Schedule

A” list of occupations with labor shortages, employers sponsoring AI professionals for EB-3 green cards

would be exempted from the recruitment requirements. Currently, only physical therapists and professional

nurses are counted at Schedule A professions; see “Permanent Labor Certiﬁcation Details,” Department of

Labor, https://www.foreignlaborcert.doleta.gov/perm_detail.cfm#schedule.

99. For more discussion of how changes to the H-1B allocation process beneﬁt (or not) U.S. AI employers,

see Arnold et al., Immigration Policy and the U.S. AI Sector, 17-18.

100. For concerns about diploma mills and labor market pressures, see Ron Hira, “Congressional

Testimony: The Impact of High-Skilled Immigration on U.S. Workers,” Economic Policy Institute, March 1,

2016, https://www.epi.org/publication/congressional-testimony-the-impact-of-high-skilled-immigration-

on-u-s-workers-4/. On the possibility that automatic green cards could incentivize dual residency, analyst

Matt Sheehan explains: “I’ve spoken to multiple Chinese PhDs in Silicon Valley who told me they stayed in

Center for Security and Emerging Technology 59

the United States until they got a green card. Once they had the security of knowing they could return to

the United States any time for the next ten years, they headed back home for better opportunities.” Matt

Sheehan, “Who Loses from Restricting Chinese Student Visas?” Macro Polo, May 31, 2018, https://

macropolo.org/who-loses-from-restricting-chinese-student-visas/.

101. For one in-depth evaluation of MAVNI, see Beth J. Asch, Jennie W. Wenger, Troy D. Smith, The

Military Accessions Vital to the National Interest (MAVNI) Program, RAND Corporation, July 21, 2017,

https://www.documentcloud.org/documents/4578092-MAVNI-RAND-Report.html. For a legislative

model of an accelerated citizenship path in exchange for national security service, see the Border

Security, Economic Opportunity, and Immigration Modernization Act, S. 744, 114th Cong. (2013), Sec.

2307(c)(2).

102. William Hannas and Huey Chang, China’s Access to Foreign AI Technology: An Assessment

(Center for Security and Emerging Technology: September 2019), https://cset.georgetown.edu/wp-

content/uploads/CSET_China_Access_To_Foreign_AI_Technology.pdf. For an illustrative case, see,

e.g., Zachary Cohen and Alex Marquardt, “US Intelligence Warns China Is Using Student Spies to

Steal Secrets,” CNN, February 1, 2019, https://www.cnn.com/2019/02/01/politics/us-intelligence-

chinese-student-espionage/index.html.

103. See, e.g., FBI Director Christopher Wray’s speech at the Council on Foreign Relations in April

2019 (available at https://www.cfr.org/event/conversation-christopher-wray-0) or December 2018

Congressional testimony by the head of the FBI’s China Initiative, John Demers (available at https://

www.justice.gov/sites/default/ﬁles/testimonies/witnesses/attachments/2018/12/18/12-05-2018_

john_c._demers_testimony_re_china_non-traditional_espionage_against_the_united_states_the_threat_

and_potential_policy_responses.pdf).

104. See, e.g., Dan Cadman, How U.S. Foreign Student and Exchange Visitor Policies Undercut

National Security, (Center for Immigration Studies: August 2019), https://cis.org/sites/default/

ﬁles/2019-08/cadman-foreign-students-19.pdf.

105. For overviews of these debates and pushback, see, e.g., Elizabeth Redden, “Science vs. Security,”

Inside Higher Ed, April 16, 2019, https://www.insidehighered.com/news/2019/04/16/federal-

granting-agencies-and-lawmakers-step-scrutiny-foreign-research, and SupChina, “The U.S. Sinophobia

Tracker: How America Is Becoming Unfriendly to Chinese Students, Scientists, and Scholars,” https://

signal.supchina.com/the-u-s-sinophobia-tracker-how-america-is-becoming-unfriendly-to-chinese-

students-scientists-and-scholars/.

106. Michael Brown and Pavneet Singh, China’s Technology Transfer Strategy: How Chinese

Investments in Emerging Technology Enable A Strategic Competitor to Access the Crown Jewels of

U.S. Innovation (Defense Innovation Unit Experimental: January 2018), 18, https://admin.govexec.

com/media/diux_chinatechnologytransferstudy_jan_2018_(1).pdf. Another recent Department of

Defense report similarly states that “American universities are major enablers of China’s economic

and military rise.” See Ofﬁce of Industrial Policy, “Assessing and Strengthening the Manufacturing

and Defense Industrial Base and Supply Chain Resiliency of the United States,” 44, https://media.

defense.gov/2018/Oct/05/2002048904/-1/-1/1/ASSESSING-AND-STRENGTHENING-THE-

MANUFACTURING-AND-DEFENSE-INDUSTRIAL-BASE-AND-SUPPLY-CHAIN-RESILIENCY.PDF.

107. For background on China’s military-civil fusion strategy and its implementation, see e.g. Lorand

Laskai, “Civil-Military Fusion and the PLA’s Pursuit of Dominance in Emerging Technologies,” China

Brief 18, no. 6 (2018), https://jamestown.org/program/civil-military-fusion-and-the-plas-pursuit-of-

dominance-in-emerging-technologies/, and Elsa Kania, “In Military-Civil Fusion, China Is Learning

Lessons from the United States and Starting to Innovate,” RealClearDefense, August 27, 2019, https://

Center for Security and Emerging Technology60

www.realcleardefense.com/articles/2019/08/27/in_military-civil_fusion_china_is_learning_lessons_

from_the_united_states_and_starting_to_innovate_114699.html.

108. See, e.g., Michael Horowitz, “Artiﬁcial Intelligence, International Competition, and the Balance

of Power,” Texas National Security Review 1, no. 3 (2018), https://tnsr.org/2018/05/artiﬁcial-

intelligence-international-competition-and-the-balance-of-power/ and Maaike Verbruggen, “The Role

of Civilian Innovation in the Development of Lethal Autonomous Weapon Systems,” Global Policy 10,

no. 3 (September 2019), https://onlinelibrary.wiley.com/doi/epdf/10.1111/1758-5899.12663.

109. For systematic work beginning to be done in this space (though not on AI speciﬁcally), see, e.g.,

Alex Joske, “The China Defence Universities Tracker” (Australian Strategic Policy Institute, November

25, 2019), https://www.aspi.org.au/report/china-defence-universities-tracker; Alex Joske, “Picking

Flowers, Making Honey” (Australian Strategic Policy Institute, October 30, 2019), https://www.

aspi.org.au/report/picking-ﬂowers-making-honey; and Marcel Angliviel de la Beaumelle, Benjamin

Spevack, and Devin Thorne, “Open Arms: Evaluating Global Exposure to China’s Defense-Industrial

Base” (C4ADS, October 17, 2019), https://www.c4reports.org/open-arms.

110. See Zwetsloot, Heston, and Arnold, Strengthening the U.S. AI Workforce, 5.

111. Joy Dantong Ma, “China’s AI Talent Base is Growing and then Leaving,” MacroPolo (blog), July 30,

2019, https://macropolo.org/chinas-ai-talent-base-is-growing-and-then-leaving/.

112. Josh Lederman and Ted Bidris, “AP Sources: US to Impose Limits on Some Chinese Visas,”

Associated Press, May 29, 2018, https://apnews.com/82a98fecee074bfb83731760bfbce515.

113. Nicholas Eftimiades, “China’s Theft & Espionage: What Must be Done,” Breaking Defense, April

19, 2019, https://breakingdefense.com/2019/04/chinas-theft-espionage-what-must-be-done/.

Similarly, a decade ago, a high-level advisory committee on knowledge transfer concerns advised

against using visas for screening because “the personnel supporting the visa processing system are, in

most instances, not equipped to make judgments as to the commercial and security implications of fast-

changing leading-edge scientiﬁc and technologic advancements.” Deemed Export Advisory Committee,

The Deemed Export Rule in the Era of Globalization, December 20, 2007, https://fas.org/sgp/

library/deemedexports.pdf. A former CIA Director recently argued that “placing restrictions on foreign

graduate students … conﬂicts with the open structure of admission, research, and publication that keeps

the US innovative ecosystem fresh, exciting, and agile.” John Deutch, “Is Innovation China’s Next Great

Leap Forward?,” Issues in Science and Technology 34, no. 4, (Summer 2018), https://issues.org/is-

innovation-chinas-next-great-leap-forward/.

114. See, e.g., James Andrew Lewis, Emerging Technologies and Managing the Risk of Tech Transfer to

China (Center for Strategic and International Studies: September 2019), 7-10 and 21-22, https://csis-

prod.s3.amazonaws.com/s3fs-public/publication/190904_Lewis_ChinaTechTransfer_WEB_v2_1.pdf.

115. Peter Waldman and Robert Burnson, “U.S. Says Scientist Hid Job in China. Web Search Tells

Otherwise,” Bloomberg, August 21, 2019, https://www.bloomberg.com/news/articles/2019-08-

21/u-s-industrial-researcher-charged-with-hiding-his-job-in-china. For an example of how Chinese tech

transfer activities can be analyzed in-depth exclusive using open-source data, see William C. Hannas,

James Mulvenon, and Anna B. Puglisi, Chinese Industrial Espionage: Technology Acquisition and

Military Modernization (Routledge: 2013).

116. Michael Morrell and Amy Zegart, “Spies, Lies, and Algorithms: Why U.S. Intelligence

Agencies Must Adapt or Fail,” Foreign Affairs, May/June 2019, https://www.foreignaffairs.com/

articles/2019-04-16/spies-lies-and-algorithms.

117. Cynthia Brown, “The Foreign Agents Registration Act (FARA): A Legal Overview,” Congressional

Research Service, December 4, 2017, https://fas.org/sgp/crs/misc/R45037.pdf.

Center for Security and Emerging Technology 61

118. See, e.g., Elsa Kania, “Chinese Military Innovation in Artiﬁcial Intelligence,” testimony before the

U.S.-China Economic and Security Review Commission, June 7, 2019, 35, https://www.uscc.gov/

sites/default/ﬁles/June%207%20Hearing_Panel%201_Elsa%20Kania_Chinese%20Military%20

Innovation%20in%20Artiﬁcial%20Intelligence.pdf.

119. To deﬁne our list of universities we relied on the U.S. News & World Report, which put the following

universities in the top 20 in 2018: Carnegie Mellon University, Massachusetts Institute of Technology,

Stanford University, University of California Berkeley, University of Washington, Cornell University,

Georgia Institute of Technology, University of Illinois-Urbana Champaign, University of Texas-Austin,

University of Michigan, University of Massachusetts-Amherst, Columbia University, University of

Pennsylvania, University of California Los Angeles, University of Southern California, University of

Maryland-College Park, Princeton University, Harvard University, California Institute of Technology, and

University of Wisconsin-Madison. See “Best Artiﬁcial Intelligence Programs,” U.S. News & World Report,

https://www.usnews.com/best-graduate-schools/top-science-schools/artiﬁcial-intelligence-rankings.

120. “Crunchbase Database,” Crunchbase, https://data.crunchbase.com/docs.

121. “AI 100: The Artiﬁcial Intelligence Startups Redeﬁning Industries,” CB Insights, February 6, 2019,

https://www.cbinsights.com/research/artiﬁcial-intelligence-top-startups/; “Learning and the Machine,”

Paysa, April 2017, https://www.paysa.com/blog/wp-content/uploads/2017/04/Paysa-AI-Tech-

Investment.pdf.

122. As examples of the kinds of roles included under this label, the most common “technical” job titles in

the database—not tagged as potentially “AI-related” unless the job was at a company categorized as an

AI employer—were Software Engineer, Computer Systems Analyst, Senior Software Engineer, Software

Development Engineer, and Software Developer.

123. “Survey of Earned Doctorates,” NSF National Center for Science and Engineering Statistics,

https://www.nsf.gov/statistics/srvydoctorates/.

124. “Survey of Graduate Students and Postdoctorates in Science and Engineering,” NSF National

Center for Science and Engineering Statistics, https://www.nsf.gov/statistics/srvygradpostdoc/.

125. “The CRA Taulbee Survey,” Computing Research Association, https://cra.org/resources/taulbee-

survey/.

126. “The Data Buddies Project,” CRA Center for Evaluating the Research Pipeline, https://cra.org/

cerp/data-buddies/.

127. “Professional Year Program,” ACS, https://www.acs.org.au/cpd-education/professional-year-

program.html.

128. “Points Table for Skilled Independent Visa (Subclass 189),” Australian Government Department of

Home Affairs Immigration and Citizenship, https://immi.homeaffairs.gov.au/visas/getting-a-visa/visa-

listing/skilled-independent-189/points-table.

129. “Post-Graduation Work Permit Program (PGWPP),” Government of Canada, 2019, https://www.

canada.ca/en/immigration-refugees-citizenship/corporate/publications-manuals/operational-

bulletins-manuals/temporary-residents/study-permits/post-graduation-work-permit-program.html.

130. Green and Spiegel, Important Changes to the Post Graduate Work Permit Program, March 2019,

https://www.gands.com/knowledge-centre/blog-post/insights/2019/03/25/important-changes-to-

the-post-graduate-work-permit-program.

131. Home Ofﬁce in the Media, Fact Sheet: Graduate Immigration Route, October 2019, https://

homeofﬁcemedia.blog.gov.uk/2019/10/14/fact-sheet-graduate-immigration-route/; HM Government,

Center for Security and Emerging Technology62

The UK’s Future Skills-Based Immigration System, December 2018, https://www.gov.uk/government/

publications/the-uks-future-skills-based-immigration-system.

132. Parliament of the United Kingdom, Statement of Changes in Immigration Rules (London: The U.K.

Parliament, September 9, 2019), https://assets.publishing.service.gov.uk/government/uploads/

system/uploads/attachment_data/ﬁle/830153/CCS001_CCS0919963034-001_HC_2631_text.

pdf.

133. Migration Advisory Committee, Impact of International Students in the UK, September 2018, 43,

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/

ﬁle/739089/Impact_intl_students_report_published_v1.1.pdf. In the European Union, international

students from other EU countries have an automatic and indeﬁnite right to stay and work.

134. Dominic Lau, “The International Startup Visa Guide,” Medium, February 25, 2018, https://medium.

com/@dominiclau/the-international-startup-visa-guide-eb01bd1a46e8.

135. Luke Klimaviciute, “To Stay or Not to Stay: The Calculus for International STEM Students in the

United States,” Migration Policy Institute, January 4, 2017, https://www.migrationpolicy.org/article/

stay-or-not-stay-calculus-international-stem-students-united-states.

136. “Visa Refusals: The Figures, the Problem, and the Solution,” Campaign for Science and Engineering,

May 16, 2018, http://www.sciencecampaign.org.uk/news-media/case-comment/tier-2-refusals.html.

137. Meghna Sabharwal and Roli Varma, “Grass is Greener on the Other Side: Return Migration of

Indian Engineers and Scientists in Academia,” Bulletin of Science, Technology & Society 37, no. 1,

https://journals.sagepub.com/doi/pdf/10.1177/0270467617738463, 34-44; Wadhwa et al.,

“America’s Loss.”

138. Elizabeth Chacko, “From Brain Drain to Brain Gain: Reverse Migration to Bangalore and

Hyderabad, India’s Globalizing High Tech Cities,” GeoJournal 68, no. 2 (June 2007): 131-140.

https://www.researchgate.net/publication/225654622_From_Brain_Drain_to_Brain_Gain_Reverse_

Migration_to_Bangalore_and_Hyderabad_India's_Globalizing_High_Tech_Cities/download.

139. Bier, “Immigration Wait Times.”

140. Vijayakumar and Cunningham, “An Indentured Servant,” 29.

141. Sabharwal and Varma, “Grass is Greener,” Table 1, 40.

142. Wadhwa et al., “America’s Loss,” Figure 28.

143. S Irudaya Rajan, V Kurusu, and Saramma Panicker, C.K, Return of Diasporas: India’s Growth Story

vs. Global Crisis, Ministry of Overseas Indian Affairs Research Unit on International Migration, 2013,

Figure 4, http://cds.edu/wp-content/uploads/2013/10/ReturnMigrantsReport.pdf.

144. Sabharwal and Varma, “Grass is Greener,” Table 1.

145. Rupa Chanda and Niranjana Sreenivasan, “India’s Experience with Skilled Migration,” in

Competing for Global Talent, ed. Christiane Kuptsch and Pang Eng Fong (International Institute for

Labour Studies, 2006): 233, http://www.ilo.org/wcmsp5/groups/public/@dgreports/@dcomm/@

publ/documents/publication/wcms_publ_9290147768_en.pdf#page=226.

146. Alfonso Giordano and Giuseppe Terranova, “The Indian Policy of Skilled Migration: Brain Return

Versus Diaspora Beneﬁts,” Journal of Global Policy and Governance 1, no. 1 (December 2012): 20, 25,

https://link.springer.com/content/pdf/10.1007%2Fs40320-012-0002-3.pdf.

147. Chanda and Sreenivasan, “India’s Experience,” 240-242.

148. Wadhwa, Saxenian, Freeman, and Gerefﬁ, “America’s Loss is the World’s Gain,” Figure 16.

Center for Security and Emerging Technology 63

149. Xueying Han and Richard P. Appelbaum, “China’s Science, Technology, Engineering, and

Mathematics (STEM) Research Environment: A Snapshot,” PLoS ONE 13, no. 4 (April 2018), https://

journals.plos.org/plosone/article?id=10.1371/journal.pone.0195347#sec008.

150. Robert Zeithammer and Ryan P. Kellogg, “The Hesitant Hai Gui: Return-Migration Preferences of

U.S.-Educated Chinese Scientists and Engineers,” Journal of Marketing Research 50, no. 5 (October

2013), https://www.jstor.org/stable/42002792.

151. Xue Hao, Kun Yan, Shibao Guo, and Meiling Wang, “Chinese Returnees’ Motivation, Post-Return

Status and Impact of Return: A Systematic Review,” Asian and Paciﬁc Migration Journal 26, no. 1

(March 2017): 147, https://journals.sagepub.com/doi/10.1177/0117196817690294; David Zweig,

“Learning to Compete: China’s Efforts to Encourage a “Reverse Brain Drain,”” Competing for Global

Talent, ed. Christiane Kuptsch and Pang Eng Fong, (International Institute for Labour Studies, 2006): 208

http://www.ilo.org/wcmsp5/groups/public/@dgreports/@dcomm/@publ/documents/publication/

wcms_publ_9290147768_en.pdf#page=198.

152. Han and Appelbaum, “China’s STEM Research Environment.”

153. Han and Appelbaum, “China’s STEM Research Environment,” 8.

154. Vivek Wadhwa, Sonali Jain, AnnaLee Saxenian, G. Gerefﬁ, Huiyao Wang, “The Grass is Indeed

Greener in India and China for Returnee Entrepreneurs: America’s New Immigrant Entrepreneurs - Part

VI,” SSRN, April 8, 2011, Figure 14, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1824670.

155. Kellogg, “China’s Brain Gain?,” Table 10.

156. Han and Appelbaum, “China’s STEM Research Environment,” 7.

157. Hao et al., “Chinese Returnees’ Motivations.”

158. Zeithammer and Kellogg, “Hesitant Hai Gui.”

159. Mandy Zuo, “Why China’s Overseas Students Find Things Aren’t Always Better Back Home,”

South China Morning Post, September 1, 2018, https://www.scmp.com/news/china/society/

article/2162229/why-chinas-overseas-students-ﬁnd-things-arent-always-better-back.

160. Zuo, “Why China’s Overseas Students.” Job application callback rates from Chinese employers

are actually higher for Chinese-educated than for U.S.-educated students, primarily because

employers consider U.S.-educated students to be more in demand and harder to attract and retain;

see Mingyu Chen, “The Value of U.S. College Education in Global Labor Markets: Experimental

Evidence from China,” (PhD Diss., Princeton University, 2019), https://static1.squarespace.com/

static/5b511a7112b13fce46a57901/t/5c99544df4e1fc96efbda239/1553552462159/Chen_

Mingyu_JMP.pdf.

161. Cong Cao, “China’s Brain Drain at the High End,” Asian Population Studies 4, no. 3 (November

2008), http://doi.org/10.1080/17441730802496532, Hao, “Chinese Returnees’ Motivations,” 147-

148.

162. Han and Appelbaum, “China’s STEM Research Environment,” 17.

163. Zweig, “Learning to Compete,” 210.

164. Zweig, “Learning to Compete,” 188.

165. Hao et al., “Chinese Returnees’ Motivations.”

166. Zweig, “Learning to Compete,” 203-205.

Center for Security and Emerging Technology 77

CSET.GEORGETOWN.EDU | CSET@GEORGETOWN.EDU