Four Battlegrounds Power In The Age Of

Artificial Intelligence Paul Scharre download

https://ebookbell.com/product/four-battlegrounds-power-in-the-
age-of-artificial-intelligence-paul-scharre-51157534

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Four Battlegrounds Power In The Age Of Artificial Intelligence Paul

Scharre

https://ebookbell.com/product/four-battlegrounds-power-in-the-age-of-
artificial-intelligence-paul-scharre-48502186

Four Battlegrounds Power In The Age Of Artificial Intelligence Paul

Scharre

https://ebookbell.com/product/four-battlegrounds-power-in-the-age-of-
artificial-intelligence-paul-scharre-49409992

Four Battlegrounds Power In The Age Of Artificial Intelligence Paul

Scharre

https://ebookbell.com/product/four-battlegrounds-power-in-the-age-of-
artificial-intelligence-paul-scharre-50563844

Four Battlegrounds Paul Scharre

https://ebookbell.com/product/four-battlegrounds-paul-scharre-50563848
Temperament How Music Became A Battleground For The Great Minds Of
Western Civilization Stuart Isacoff

https://ebookbell.com/product/temperament-how-music-became-a-
battleground-for-the-great-minds-of-western-civilization-stuart-
isacoff-4649394

The Effortless Experience Conquering The New Battleground For Customer

Loyalty Dixon

https://ebookbell.com/product/the-effortless-experience-conquering-
the-new-battleground-for-customer-loyalty-dixon-6645318

Battleground Of Desire The Struggle For Self Control In Modern America

Peter N Stearns

https://ebookbell.com/product/battleground-of-desire-the-struggle-for-
self-control-in-modern-america-peter-n-stearns-51757296

Private Business Public Battleground The Case For 21st Century

Stakeholder Companies John Egan

https://ebookbell.com/product/private-business-public-battleground-
the-case-for-21st-century-stakeholder-companies-john-egan-5368586

Beyond The Battleground Classic Strategies From The Yijing And

Baguazhang For Managing Crisis Situations Tom Bisio

https://ebookbell.com/product/beyond-the-battleground-classic-
strategies-from-the-yijing-and-baguazhang-for-managing-crisis-
situations-tom-bisio-5900902
 FOUR
BATTLEGROUNDS

POWER IN
THE AGE OF
ARTIFICIAL
INTELLIGENCE

PAUL SCHARRE
 In memory of Jim Hornfischer

Author, agent, mentor, and friend

Artificial intelligence is the future, not only for Russia, but for all
humankind. It comes with colossal opportunities, but also threats
that are difficult to predict. Whoever becomes the leader in this
sphere will become the ruler of the world.

—RUSSIAN PRESIDENT VLADIMIR PUTIN

Science and technology has become the main battleground of global

power rivalry.

—CHINESE GENERAL SECRETARY XI JINPING

CONTENTS
LIST OF ILLUSTRATIONS
PREFACE

INTRODUCTION

PART I POWER
1. THE NEW OIL
2. DATA
3. COMPUTE
4. TALENT
5. INSTITUTIONS

PART II COMPETITION
6. A WINNING HAND
7. MAVEN
8. REVOLT
9. SPUTNIK MOMENT

PART III REPRESSION

10. TERROR
11. SHARP EYES
12. A BETTER WORLD
13. PANOPTICON
14. DYSTOPIA

PART IV TRUTH
15. DISINFORMATION
16. SYNTHETIC REALITY
17. TRANSFORMATION
18. BOT WARS

PART V RIFT
19. FUSION
20. HARMONY
21. STRANGLEHOLD

PART VI REVOLUTION
22. ROBOTICS ROW
23. PROJECT VOLTRON
24. FOUNDATION
25. THE WRONG KIND OF LETHALITY
26. JEDI
27. DISRUPTION

PART VII ALCHEMY

28. CONTROL
29. POISON
30. TRUST
31. RACE TO THE BOTTOM

PART VIII FIRE

32. ALIEN INTELLIGENCE
33. BATTLEFIELD SINGULARITY
34. RESTRAINT
35. THE FUTURE OF AI

CONCLUSION

ACKNOWLEDGMENTS
ABBREVIATIONS
NOTES
INDEX
 LIST OF ILLUSTRATIONS

Deep Neural Network

U.S. R&D Funding as a Percentage of Gross Domestic Product, 1953–2018
U.S. Share of Global R&D (1960)
U.S. Share of Global R&D (2018)
R&D Spending by Country, 2000–2018
Percentage of World GDP, 1970–2020
Gross National Income Per Capita—China, 1995–2019
Life Expectancy at Birth—China, 1960–2019
Infant Mortality—China, 1969–2019
Adversarial Examples
Adversarial Patch
Adversarial Attack in the Physical World
Camouflaged Adversarial Attack

MAPS
Global Spread of Chinese Public Surveillance Technology, 2008–2019
 PREFACE

There was a singular moment when I realized robots would transform

warfare. It was the spring of 2007 and I had just begun a one-year tour in
Iraq during the height of the U.S. troop surge. Iraq was a chaotic, violent
place. The country was ripping itself apart in civil war. I’d been on three
prior deployments to Afghanistan and my unit had taken casualties every
time—I’d been an Army Ranger and we didn’t shy away from combat—but
I had never seen violence like this. For my Iraq tour I was part of a civil
affairs team deployed to Forward Operating Base Warhorse in Diyala
province. Diyala’s population was split among Sunnis, Shias, and Kurds,
like a microcosm of Iraq. The groups warred with each other and with U.S.
troops for the future of the country. Improvised explosive devices (IEDs), or
roadside bombs, had taken a heavy toll on U.S. troops in the area. The base
had a makeshift monument erected by the 1st Cavalry Division unit
operating out of Warhorse with the names of soldiers killed during their
tour, and when I arrived in country I was taken aback by the long roster of
names stenciled on the concrete barriers. It was a difficult time.
One day on patrol we came across a roadside bomb (as one did at the
time). We saw it first, which is the preferred way of discovering one rather
than running into it, and we pulled the Stryker armored vehicle over to wait
for the bomb disposal techs to arrive. I had occasionally seen U.S. troops
handle roadside bombs by simply shooting at them to detonate the bomb
while leaning out the side of the Humvee, but waiting for the bomb disposal
professionals certainly seemed like the more prudent course of action to me.
It took a while for the explosive ordnance disposal (EOD) team, as they
were formally known, to arrive as they were in high demand at the time, but
eventually they rolled up in their large Buffalo mine-resistant ambush
protected (MRAP) armored vehicle. I had only been in theater maybe a
month or so and had yet to see an EOD team in action, so I popped my head
out of the Stryker hatch to watch with interest. I expected to see the bomb
tech emerge from the MRAP armored vehicle in a giant protective suit.
Instead, out came a little robot. And the light bulb went off in my head:
Send the robot to defuse the bomb!
It seemed obvious, afterwards, to use a robot for such a hazardous
mission. Why put a human at risk if you didn’t need to? My mind went to
all of the missions in war that might someday be done by robots, sparing the
need to put soldiers in harm’s way.
After I left Iraq, I worked at the Pentagon on emerging technologies and
helped push for U.S. investments in military robots. I believed then, and
still do, that these machines would help save American lives and defend the
nation. The military’s plans for increasingly advanced robots led inevitably
to questions about the scope of autonomy in future systems, a thorny issue
with important legal, ethical, and strategic dimensions. I was involved in
drafting the official Defense Department policy on autonomy in weapons,
which gave guidance to weapons developers. In the years following, I had
the opportunity to participate in multiple rounds of diplomatic discussions
at the United Nations on autonomous weapons, which was the topic of my
first book, Army of None. But even as nations debated the role of autonomy
in weapons, the ground was shifting beneath our feet.
The robotics revolution that unfolded in the early 2000s, enabling
Roombas and military robots, has given way to something far deeper and
more profound: a revolution in artificial intelligence (AI). The combination
of exponential growth in data and computer processing power (compute)
has fueled a renaissance in machine learning, an AI technique in which
machines learn from data, rather than follow an explicit set of
preprogrammed rules of behavior. Machine learning is now being applied to
medicine, finance, transportation, and other industries. Many have argued
that AI could be the next industrial revolution. Given the enormous
geopolitical disruptions caused by prior industrial revolutions, it makes
sense that AI is also a fast-growing arena of competition between nations
who are vying for dominance. AI is changing global security and power
dynamics. The consequences of this competition may not be as immediate
as that of the roadside bomb back in Iraq, but the stakes are just as high. AI
is likely to have profound effects on human freedom and global security.
Whoever leads in AI will have a tremendous advantage in shaping the
future, if the technology can be controlled at all. AI will bring about a new
digital order, and this book is the story of those fighting for control over
that future.
 FOUR
BATTLEGROUNDS
 INTRODUCTION

“Fight’s on, fight’s on!” The two aircraft twisted in the blue, each
furiously maneuvering to gain advantage. In one simulated cockpit, a
human pilot. In the other, an AI agent.
Less than thirty seconds into the fight, the AI had scored the first kill.
The human pilot’s simulator flashed red as the AI’s guns hit their mark.
Both aircraft dove furiously, but mere seconds later, the AI scored another
hit. Then another. And another.
In under ninety seconds, it was over. The AI had racked up four kills
and the human none.
The fight was the culmination of DARPA’s AlphaDogfight Trials, which
pitted an AI agent against a flesh-and-blood highly experienced F-16 pilot
who goes by the call sign “Banger.” Eight AI teams competed in a
tournament against one another, with the thirty-person Heron Systems
beating out defense giant Lockheed Martin in the finals. Victory over the
other AI agents earned Heron Systems’ algorithm, call sign “Falco,” the
opportunity to compete 1v1 against a human pilot. It performed flawlessly.
In five total rounds, the AI racked up fifteen gunshot kills against Banger.
Banger couldn’t get a single shot on the AI.
 Dogfighting, or air combat maneuvering, is one of the most challenging
skills for human fighter pilots. Even while air combat has evolved to
increasingly rely on air-to-air missiles shot beyond visual range, close-in
dogfighting remains a key crucible in which to test AI performance in aerial
combat. DARPA, the Defense Advanced Research Projects Agency, is the
U.S. military’s “department of mad scientists” that develops technology
breakthroughs, and DARPA’s Air Combat Evolution (ACE) program is
taking air combat to the next level.
Launched in 2019, the ACE program capitalizes on the explosion in AI
under way since 2012 as part of the deep learning revolution. Improvements
in data, computing hardware, and algorithms have enabled rapid
improvements in machine learning, a method for creating intelligent
machines by training algorithms on data. Deep learning, a machine learning
technique that uses deep neural networks, has yielded tremendous
successes, including superhuman performance in games such as the Chinese
strategy game go and complex computer games such as StarCraft II and
Dota 2. With the ACE program, the U.S. military aims to use machine
learning to build more effective AI assistants for human pilots. Not only did
AlphaDogfight achieve superhuman performance in a simulated
environment, it demonstrated some of the superior strengths of AI systems
in combat.
Heron Systems’ algorithm, Falco, displayed superhuman precision in its
flying and fighting, giving it an unbeatable edge over its human opponent.
In an aerial dogfight, planes circle one another in tight spirals seeking to
gain a positional advantage over their opponent so they can maneuver into a
kill shot with short-range missiles or guns. Carefully managing the aircraft’s
energy so they can turn tighter and faster than the enemy is a critical skill in
dogfighting. Pilots must fly the aircraft at the peak of its performance. This
can be challenging for humans, who must not only pay attention to the
velocity and position of their aircraft but also keep track of the enemy
fighter and anticipate next moves. The AI agent flew with superhuman
precision, keeping the aircraft at the peak of its performance and never
wasting energy in suboptimal maneuvers.
The AI agent also demonstrated superhuman precision in employing its
weapons. The AI agent showed a strong favor for what pilots call forward-
quarter gunshots, when the two aircraft are racing toward each other head-
to-head. Human pilots tend to be bad at forward-quarter shots because they
rarely practice them. In training, forward-quarter gunshots are banned
because of the danger of flying directly at another airplane at several
hundred miles per hour. (In combat, there are no rules.) Even with training,
forward-quarter shots would be incredibly difficult. They offer only a split-
second window of opportunity to hit the target, which usually arrives at the
moment when pilots most need to be thinking about avoiding a mid-air
collision. For this reason, human pilots tend to favor rear-quarter shots,
where a fighter pulls in behind another plane at the six o’clock position.
The AI agent had none of these disadvantages. It quickly and precisely
executed forward-quarter shots, all while deftly maneuvering to avoid a
collision. Lieutenant Colonel Justin “Glock” Mock described the AI agent
as executing “a gunshot that is almost impossible,” demonstrating
“superhuman capability . . . to aim that accurately and to be able to do it in
such a dynamic situation.” The AI’s use of a tactic banned in training for
humans might seem unfair, but war isn’t fair. One of the best advantages of
AI systems isn’t just their ability to fight better than humans but to fight
differently.
Lastly, the AI agent could pull maneuvers that a human pilot simply
could not physically withstand, sustaining g-forces that would cause a
human to black out. In their last engagement, the two simulated aircraft
circled each other for nearly two minutes at roughly 9 Gs (nine times the
force of Earth’s gravity), a level of g-forces that even trained fighter pilots
can only withstand for a few seconds. Not only could AI systems exploit
the lack of human physical limitations for superior tactics, future robotic
combat aircraft could be designed to handle far more aggressive maneuvers.
In analyzing AlphaDogfight, Navy fighter pilot Commander Colin “Farva”
Price noted, “The human will always be the limiting factor in the
performance of an aircraft.” By handing over some tasks to AI, militaries
can build more effective combat systems with superhuman performance.
AI systems have seen impressive gains in recent years but still have
many limitations. In the AlphaDogfight Trials, the AI agent was given
perfect situational awareness of the simulated environment, including the
location of the opposing fighter. In the real world, an AI system would need
to rely on sensors and data processing algorithms to find enemy aircraft,
identify them, and correctly distinguish them from friendly or civilian
aircraft. But classification algorithms that can identify objects are
improving as well. DARPA is continuing to mature more capable AI-
enabled autonomy, with the goal of evolving to a live flight demonstration
on a trainer jet. DARPA’s goal isn’t to build a completely autonomous
combat airplane, however. The ACE program manager, Lieutenant Colonel
Dan “Animal” Javorsek, explained that the goal was to build more effective
AI systems that could be used by human pilots. “We envision a future in
which AI handles the split-second maneuvering during within-visual-range
dogfights,” Javorsek said, “keeping pilots safer and more effective as they
orchestrate large numbers of unmanned systems.”
AI is changing war, and it is also changing surveillance, disinformation,
and other aspects of global peace and security. Nations around the globe are
racing to capitalize on AI technology to gain an advantage over others. The
world is just beginning to grapple with the implications of a technology that
could herald another industrial revolution.
AI is not a discrete technology, like railroads or airplanes. It is a
general-purpose enabling technology, like electricity, computers, or the
internal combustion engine, with many applications. Those earlier general-
purpose technologies brought sweeping economic, social, and political
changes. Likewise, the scale of potential change from artificial intelligence
is staggering. By one estimate, nearly half of all tasks currently being done
in the U.S. economy could be outsourced to automation using existing
technology. Wired magazine cofounder Kevin Kelly has argued that “[AI]
will enliven inert objects, much as electricity did more than a century ago.
Everything that we formerly electrified we will now cognitize.” AI has
applications in cybersecurity, surveillance, defense, border security,
disinformation, and economic warfare, yielding major geopolitical
advantages to whoever best harnesses these tools.
The first and second industrial revolutions brought about a societal
process of mechanization that saw the creation of machines that were
stronger than humans for specific tasks. While today’s AI systems are a far
cry from the C-3POs and Terminators of science fiction, AI technology is
real today and a powerful tool. Today’s “narrow,” or task-specific, AI
allows the creation of machines that are smarter than humans for specific
cognitive tasks and will spark a broad process of cognitization across
society. While the process of industrialization transformed society with
physical machines, cognitization will transform society with intelligent
machines. AI has many constructive applications. AI will save lives and
increase efficiency and productivity. It is also being used as a weapon of
repression and to gain military advantage.
This book is about the darker side of AI.
The dangers from AI aren’t the dangers science fiction warned us about.
We needn’t fear robots rising up to throw off their human overlords, or at
least not anytime soon. The dangers from AI today come from people using
the technology maliciously or carelessly, causing harm to others. Militaries
around the globe are investing in AI technology, and authoritarian regimes
are using it to tighten their grip for internal repression. The Pentagon is
applying the same AI techniques used to achieve superhuman intelligence
in poker to much higher-stakes strategic game theory, such as analyzing
what weapons to invest in to deter nuclear war. China is building a techno-
dystopian surveillance state to monitor and repress its citizens through
facial recognition and an Orwellian “social credit” system. Deepfake video
and audio continue to improve, and long-term trends are likely to lead to
fakes that are indistinguishable from reality, undermining truth. And AI
technology could radically alter the nature of war, ushering in a battlefield
“singularity” in which war moves too fast for human control.
Like prior industrial revolutions, the cognitive revolution will reshape
geopolitics in the twenty-first century. AI is likely to lead to shifts in power
on the global stage, empowering some actors and even changing the key
metrics of power. During the industrial revolution, coal- and steel-
producing nations became more powerful and oil became a global strategic
resource. Today, computing hardware, data, and AI talent make up the key
resources in the global struggle for AI dominance.
More than fifty countries have signaled their intent to capitalize on AI
for national advantage, but nowhere is the competition fiercer than in the
accelerating rivalry between the United States and China. China has
launched a national-level AI development plan with the intent to be the
global leader in AI by 2030. China is spending billions on research, training
more AI scientists, and aggressively courting top experts from Silicon
Valley. The White House and Pentagon are undertaking their own
initiatives, reaching out to tech firms and launching billion-dollar cloud
computing projects to build out AI infrastructure.
 The U.S.-China AI competition is complicated by the fact that their AI
ecosystems are deeply intertwined. U.S. military leaders have raised alarm
at China’s model of “military-civil fusion” and the fear that U.S. firms
working in China could, even indirectly, benefit the Chinese military. Some
U.S. tech firms have undoubtedly contributed to the growth of the AI
ecosystem in China including, in at least a few instances, working with
researchers from Chinese military institutions. Partnerships between the
Chinese military and U.S. academia are equally problematic. Since 2007, an
estimated 500 Chinese military scientists have studied in the United States.
From 2006 to 2017, Chinese military scientists coauthored over 1,100
research papers with U.S. scientists. Major U.S. universities including MIT
and Princeton have worked with Chinese companies that have later been
sanctioned for human rights abuses. These ties have led U.S. policymakers
to increasingly push to “decouple” the United States from China to stem the
tide of intellectual capital flowing from U.S. businesses and academia into
the arms of a strategic competitor.
But Chinese tech firms like Baidu, Alibaba, and Tencent are global
leaders in AI in their own right, and China has a vibrant AI start-up
ecosystem. China boasts some of the most highly valued AI start-ups in the
world, including SenseTime at over $16 billion and ByteDance at $350
billion. In some areas, such as facial recognition, Chinese companies such
as SenseTime and Megvii are global leaders and have a major leg up on
U.S. companies such as Amazon, IBM, and Microsoft, which face a much
stricter regulatory environment. Chinese AI firms, meanwhile, benefit from
the Chinese Communist Party’s aggressive domestic surveillance. The
Chinese government is building a burgeoning panopticon with over 500
million cameras deployed nationwide by 2021, more than half of the
world’s surveillance cameras. Even more significant than government cash
buoying China’s AI industry is the data collected, which AI companies can
use to further train and refine their algorithms. With facial recognition
deployment stalled in the United States, Chinese firms are poised to
dominate the global facial recognition industry.
China today is not the China of twenty years ago. While China’s AI
ecosystem has benefited enormously from deep ties to the United States,
both from legitimate research collaboration and intellectual property theft,
China today doesn’t need to steal U.S. research to be a global technology
leader. China already publishes more AI papers than the United States and
is on track to surpass the United States in the top 1 percent most-cited
papers by 2025. More broadly, given current trends China will overtake the
United States in total national research and development (R&D) spending
(across all fields, not just AI) in the next decade.
The tsunami of change that AI is bringing is crashing into the deepening
U.S.-China rivalry, making AI a focus of geopolitical competition. U.S.
policymakers are taking steps to sever what they see as problematic
connections between the two nations, such as banning students from
Chinese universities with military ties, at the same time as making renewed
investments in American competitiveness. In 2022, a bipartisan cadre of
lawmakers passed a massive investment of over $50 billion in the U.S.
semiconductor industry, a foundational hardware underpinning AI and other
digital technologies.
Even as the United States and China race to stay ahead in AI research,
successfully implementing AI for national defense, intelligence,
surveillance, or information operations will require more than basic
research. Governments looking to apply AI for national power will need to
rapidly spin-in and adopt a technology that has largely been invented in the
commercial sector. This book will pull back the curtain on the U.S.
military’s efforts to harness AI and will dive deep into new organizations,
such as the Defense Department’s Joint AI Center and the Pentagon’s
Silicon Valley outpost, Defense Innovation Unit. But despite their best
efforts, these new organizations may not be enough. If the United States
moves too slowly, it could cede military dominance in a critical new
technology to a rising and revisionist China.
Whichever nation leads in AI will have tremendous advantages in
setting the terms of the geopolitical order for the twenty-first century. The
United States and China are vying for military dominance in the Asia-
Pacific region, and AI could help tip the scales to either side. U.S. and
Chinese tech firms are competing for social app dominance, a high-stakes
contest to control the information seen by billions of people. And China is
pioneering a new model of AI-enabled surveillance and repression that is
increasingly being adopted around the world, threatening global freedoms.
On the eve of Russia’s 2022 invasion of Ukraine, Russia and China
announced a “no limits” partnership that encompassed, among other issues,
artificial intelligence. If the United States and other democracies do not
work together to lead in AI and shape the rules for how it is used, they risk
a creeping tide of techno-authoritarianism that undermines democracy and
freedom around the globe.
If China achieves its goal of becoming the world’s AI leader by 2030, it
would empower a regime guilty of gross human rights abuses, threatening
its neighbors, and bullying countries around the world. As China’s power
rises, other nations may seek to adopt its authoritarian tendencies, eroding
global freedoms.
Democracies must work together to lead in AI and present an effective
model for AI governance. Democratic nations have many advantages over
authoritarian regimes in a long-term contest to shape how AI is used.
Collectively, democratic nations have greater talent, military power, and
control over critical technologies. Yet this power is fragmented among
different countries and actors, including governments, corporations,
academics, and tech workers.
Democratic societies will need to manage their internal divisions to
harness their strength. Democratic governments will need to work with tech
firms to govern surveillance technology to ensure that authoritarian regimes
or corporations don’t undermine individual liberties. National security
experts who want to stem the flow of AI technology to competitor nations
will need to work with academics who favor openness. And militaries in
democratic societies will need to demonstrate they will use AI responsibly
to win over skeptical scientists concerned about what their own government
may do with AI.
Caution is warranted, because even as nations race to leverage AI
technology for national advantage, there are risks. AI technology is
powerful but has many vulnerabilities. Machine learning systems can learn
the wrong thing if the data they are trained on is biased or has been
poisoned by an adversary. AI systems can be manipulated by exploiting
vulnerabilities in how the system “thinks,” cognitive hacks that are
analogous to cyberattacks on computer software. These features make AI
technology simultaneously transformative and brittle. Systems may work
brilliantly in one setting, then fail dramatically if the environment slightly
changes. The “black box” nature of many AI methods means that it may be
difficult to accurately predict when they will fail or even understand why
they failed in retrospect. Potentially even more dangerous, the global
competition in AI risks a “race to the bottom” on safety. In a desire to beat
others to the punch, countries may cut corners to deploy AI systems before
they have been fully tested. We are careening toward a world of AI systems
that are powerful but insecure, unreliable, and dangerous.
But technology is not destiny, and there are people around the world
working to ensure that technological progress arcs toward a brighter future.
The contest for who controls AI is global, with many radically divergent
visions for the future. European nations are leaning into regulating AI,
aiming for a “race to the top” on regulatory standards. Grassroots
movements against facial recognition have sprung up across the United
States. Conscientious-objector tech employees have said “no” to
militarizing AI, while defense experts are pushing the military to use AI
responsibly. The fight to control AI includes powerful megacorporations
that control the content for billions of people, human rights activists who’ve
uncovered abuses using AI technology, researchers who are working to
build deepfake detectors, and scientists who are trying to build the next
generation of safer, more robust AI systems. The diversity of voices in
democratic societies debating the future of AI is a strength that can lead to
social benefit in the long run, but only if they build a responsible tech
ecosystem together. Democratic societies need to establish a positive model
of AI governance, or risk a future dominated by authoritarian uses of AI
that undermine truth and personal freedoms.
The future of humanity will be determined in large part by the shape of
AI technology as it unfolds in the world and who determines its destiny. AI
can be used to strengthen democratic societies or authoritarian ones, to
bolster individual freedom or crush it. Russian president Vladimir Putin has
said, “Whoever becomes the leader in [artificial intelligence] will become
the ruler of the world.” The race is under way to lead in AI and write the
rules of the next century to control the future of global power and security.
PART I POWER
 1

THE NEW OIL

Power is the currency of international relations. It is the difference

between weak states and strong ones, between those who set the terms of
the geopolitical order and those who must accommodate the will of others.
Stronger states don’t always prevail in international disputes, just like my
children don’t always listen to me, despite my power over their screen time.
Strategy, alliances, a willingness to absorb costs, and sheer force of will
also matter in settling disputes, among people and states alike. But being
the bigger dog has its advantages. Germany is a more powerful state than
Luxembourg, for example, giving it more sway over the direction of
European—and global—economics and politics.
Technology is an enabler of both hard (economic and military) and soft
(culture and values) power. Technology can be used to build weapons to
coerce or defend, material goods to trade, bribe, or bolster other states, and
media platforms to spread messages around the world. Technological
revolutions have, in the past, not only transformed the global balance of
power but even the key metrics of power. AI promises to do the same.
One of the starkest illustrations of a technology’s transformative effect
on global power is nuclear technology, whose advent in the mid-twentieth
century quickly divided the world into nuclear haves and have-nots.
Nuclear weapons have awesome destructive power, but translating that
power into political outcomes is hardly straightforward. Threats to use
nuclear weapons are rarely seen as credible except in the most extreme
cases. AI is in many ways the opposite of nuclear technology. AI
technology is widely available, proliferates rapidly, and has a multitude of
uses.
Many experts have suggested that AI, like earlier general-purpose
technologies, could cause changes on the scale of another industrial
revolution. The first and second industrial revolutions saw nations rise and
fall on the world stage based on how rapidly and effectively they adopted
industrial technology. By the time the second industrial revolution ended in
the early twentieth century, the world order and even the key building
blocks of national power had been transformed.
The industrial revolution occurred in two waves: the first industrial
revolution from around 1760 to 1830 and the second industrial revolution
from around 1870 to 1914. Each revolution encompassed a set of mutually
reinforcing technologies. In the first industrial revolution, steam engines,
improvements in iron production, machine tools, and interchangeable parts,
among other inventions, helped to fuel an explosion in economic
productivity, leading to a wave of factory-building and urbanization. The
second industrial revolution was enabled by the internal combustion engine,
electrification, steel, and oil, and led to the construction of railways,
telegraphs, automobiles, and more efficient factory production. These
technological revolutions had an immense effect on global power. As Paul
Kennedy observed in his landmark work, The Rise and Fall of the Great
Powers, “Dramatic changes occurred in the power balances, both inside
Europe and without. Old empires collapsed, and new ones arose.”
The industrial revolution began in Britain and then spread across
continental Europe and beyond, leading to major shifts in the balance of
power. Russia lagged the other European powers in industrialization, with
the result that by 1830 it had a per capita gross national product (GNP) (a
measure of national income-per-person) that was 70 percent that of
Germany and half that of Britain. Yet, for a time, Russia’s superior
population made up for its economic backwardness. Russia was not just the
largest European power by population, it was far and away the largest
European power by GNP, with nearly 30 percent more economic output
than Britain and 45 percent more than Germany. Yet Russia’s size could
only help for so long as other nations surged in industrial productivity.
From 1830 to 1890, Britain and Germany both more than doubled their per
capita GNP while Russia stagnated, increasing by a mere 7 percent total
over a sixty-year period. The result was that while Russia was the largest
European economic power (measured in GNP) in 1830, by 1890 Russia had
been eclipsed by Britain and Germany.
Similar shifts happened worldwide with even more profound
consequences for global power dynamics. In 1790, Europe (collectively),
China, and India (including what is now Pakistan) held roughly the same
shares of global manufacturing output. They all had approximately
equivalent levels of per capita industrialization. But the industrial
revolution skyrocketed European economic productivity and, by extension,
military power. By 1900, Europe collectively controlled 62 percent of
global manufacturing output, while China accounted for 6 percent and India
less than 2 percent.
The resulting European military dominance fueled a wave of colonial
expansion. The effect of technological dominance on military power was
put on stark display in the Battle of Omdurman in 1898. British, Egyptian,
and Sudanese troops were outnumbered two to one by Sudanese Dervishes
yet won the battle by a lopsided sixty-to-one casualty ratio due to superior
firepower. The British troops had better training, but even more significant
was their possession of forty-four Maxim guns, the first machine gun,
which, by virtue of industrial technology, could fire 500 rounds per minute.
By 1914, Europeans had translated their economic and military might into
occupying or controlling over 80 percent of the world’s land surface.
The industrial revolution not only grew some nations’ economic and
military power while leaving others behind; it also changed the key metrics
of power. In a preindustrial era, military power was measured chiefly in
men under arms. As a consequence, national power—and potential military
power—was measured by population. Larger nations could, all things being
equal, field larger armies. And larger armies were more powerful. There
were, of course, many other secondary factors affecting military power,
including geography, alliances, economic and financial power,
technological development, and military organization and training, but
population was a reasonable first-order proxy for potential military power.
The industrial revolution shifted the main metric of military power from
numbers of men under arms to numbers of war machines, such as tanks and
airplanes. Industrialization eclipsed population as the major factor in
potential military power.
By World War II, a nation’s coal and steel production counted far more
toward potential military power than their national population alone.
Scholars of industrial-age national power measure not just national
population and GNP, but also iron/steel production, energy consumption,
manufacturing output, and industrial potential. While economic
productivity does not directly translate to military power, it is a good
approximation of potential military power for industrialized nations.
Factories can be—and were—repurposed for building tanks and airplanes
as nations mobilized for war. In World War II, the Allied powers had three
times the industrial “war potential” of the Axis powers. Once mobilized,
Paul Kennedy explained, “these figures of potential power were being
exchanged into the hard currency of aircraft, guns, tanks, and ships.” At the
height of the war, Allied factories were producing over 3.5 times as many
aircraft and tanks as the Axis powers, burying them beneath an onslaught of
iron.
The information revolution, which began in the 1960s with the
introduction of integrated circuits, or computer chips, brought another shift
in economic and military power. The United States benefited by being a
first mover in the information revolution, which translated into advantages
in both hard power, such as military dominance, and soft power, such as
setting the initial terms of the internet. The military advantages of digitally
enabled systems such as smart bombs, stealth aircraft, GPS satellites, and
computer networks were on full display in the 1991 Gulf War, when the
U.S. military dismantled Saddam Hussein’s Soviet-equipped army with a
thirty-to-one casualty ratio. Iraqi forces were so helpless against American
airpower that the White House eventually ended the war earlier than
planned because media images of the so-called “highway of death” made
American forces seem as if they were “cruelly and unusually punishing our
already whipped foes,” in the words of Gulf War air commander General
Chuck Horner. Yet as information technology proliferated, it leveled the
playing field among nations, allowing other countries to procure their own
precision-guided weapons and build their own balkanized, censored
portions of the internet.
 Even as information technology diffuses around the globe, the
information revolution continues to mature. Big data, Internet of Things
(IoT) devices, high-speed wireless networking, autonomous systems, and
machine learning are just some of the digital systems transforming
industries and society today. The current wave of digital innovation is
highly globalized, with centers of gravity in the United States, China, and
Europe. Each power center is vying to realize their own vision for how
digital technologies should be used in society. This struggle for
geotechnical dominance points to the importance of assessing national
digital power. As the AI revolution unfolds, which countries will benefit the
most, and what digital elements will become key determinants of national
power, like coal, steel, and oil did during the industrial revolution?
Machine learning systems depend on a few key inputs: data, computing
hardware (compute), algorithms, and human capital. Machine learning is
often data- and compute-intensive, using large amounts of computing power
to train an algorithm on enormous datasets. The process isn’t magic, and it
can often be difficult work to get machine learning systems to function
properly, so having talented AI scientists and engineers is crucial to
building effective machine learning systems. When considering AI power,
some of these inputs matter more than others. It’s hard to limit the spread of
algorithms, so the relative availability or scarcity of data, talent, and
compute are major factors (although not the only ones) that affect national
AI capacity. Additionally, social institutions, from the availability of
funding for AI start-ups to public perceptions of AI technologies, also
influence AI adoption. Nations that lead in these four battlegrounds—data,
compute, talent, and institutions—will have a major advantage in AI power.
A number of independent analysts have built global AI indices in an
attempt to comprehensively measure countries’ national AI power. While
the specific metrics vary, many researchers have come to similar
conclusions, which one group summarized as: “the United States currently
leads in AI, with China rapidly catching up, and the European Union behind
both.” These indices consider national AI capabilities across a number of
areas, including funding, data, hardware, research, development, talent, and
adoption. Some indices also include infrastructure, the operating
environment, and government strategy. Two efforts in particular stand out
as noteworthy for their analytic rigor.
 Researchers at the Center for Data Innovation evaluated China, the
United States, and the European Union across thirty quantitative metrics in
2019 and again in 2021. These included the number of top AI researchers,
field-weighted research citations, highly cited patents, R&D spending, AI
start-ups, venture capital and private equity funding, and semiconductor
firms. Their conclusion was that the United States led overall and in the
subcategories of talent, research, development, and hardware, but that
China led in data and adoption.
Researchers at Tortoise released a “Global AI Index” in 2019 and 2021,
assessing sixty-two countries across 143 metrics. They similarly concluded
that the United States led the world overall in AI and in the subcategories of
talent, research, development, and commercial activity, but that China was
the number two country overall and in research, development, and
commercial activity. Additionally, Tortoise researchers ranked China far
ahead of the United States in government strategy and an operating
environment conducive to AI, which includes regulation and public
opinion.
Some caution is warranted in interpreting these results. First, there are
important asymmetries between the American, European, and Chinese AI
ecosystems that can skew results. The biggest among these is size. Like
Russia in the nineteenth century, China has a much larger population, so
metrics that look at raw size, such as internet users, will heavily benefit
China, even if raw size may not be the most relevant indicator in all cases.
Conversely, metrics that are size-adjusted, such as per capita or per worker,
tend to benefit the United States and Europe, which have smaller
populations, even if in other cases sheer size may matter more. Most
importantly, size-adjusted indices discount China’s potential power, which
is enormous if it continues to modernize and harness the potential of its 1.4
billion people.
Another important difference is that, across many metrics, the Chinese
AI ecosystem tends to emphasize quantity over quality compared to the
United States and Europe. For example, the United States has a modest
(1.5×) advantage over China in terms of the overall number of AI
researchers, but it has an overwhelming (5×) advantage when it comes to
top AI researchers. A similar asymmetry exists in publications. Chinese
researchers publish nearly 1.5 times as many papers as U.S. researchers, but
U.S. papers were cited roughly 70 percent more on average, an indication of
their higher average quality. The U.S.-China balance in research and talent
can look very different depending on whether one is looking at quality or
quantity, and both matter in different ways.
There are other ways in which these indices are valuable, but imperfect,
yardsticks for measuring national AI power. While most of the metrics are
quantitative, there is some subjectivity in deciding what to count and, in
particular, how to weight or combine various indices. The indices are also
only a snapshot in time. They capture metrics based on data available when
they were produced, which often is 2017–2020 data, but sometimes are
cumulative metrics dating back several decades. The choices of what to
measure are based on an understanding of AI technology at the time. As AI
technology continues to evolve, its future evolution may change the relative
value of some metrics. For example, the increased use of synthetic data may
diminish the value of real-world data as an input to AI power and increase
the value of computing power, which both analyzes and creates synthetic
data.
The indices generally don’t capture trends, although in some cases
trends can be seen by looking at the underlying data sources. Even if China
is behind in many areas as of 2017 to 2020, the trends heavily favor China.
The Chinese economy has been on a rocket-like trajectory for the last
several decades and has rapidly emerged as a global powerhouse in AI. Just
because China is number two today doesn’t mean that they are destined to
stay there. China’s goal to be the global AI leader by 2030 looks very
achievable based on current trends, absent a course correction by the United
States.
Finally, these indices don’t capture the agency that individuals and
organizations have to chart their future. Governments, corporations,
universities, and civil society groups all have choices about how they
respond to AI technology. These choices can accelerate, shape, or stall AI
adoption. AI indices capture a moment in time; they are not a prediction of
the future.
To better understand which countries have an advantage in military,
economic, and political power in the age of AI, we’ll need to dive deeper
into four key areas: data, compute, talent, and institutions.
 2

DATA

Of all the inputs to machine learning, data has perhaps received the most
attention as an element of national power. In 2017, The Economist
proclaimed, “The world’s most valuable resource is no longer oil, but data.”
The digital revolution has created a globe-spanning network of digital
devices, which in turn are creating unprecedented amounts of data. There
are an estimated 24 billion connected devices in use in 2021, growing 10
percent annually to an estimated nearly 30 billion devices by 2023. As the
internet expands, it is bringing both more people and devices online. By
2023, there will be an estimated 5.3 billion internet users, or two-thirds of
the world’s population. IoT devices, which include smart meters, medical
devices, home appliances, and industrial applications, are growing even
faster than users and by 2023 are expected to account for over half of all
connected devices. These devices create data and share it across a global
network that trafficked an estimated over 250 exabytes of data per month in
2020. Global internet protocol (IP) traffic is growing even faster than
connectivity, at a rate of 26 percent annually, and was projected to increase
to nearly 400 exabytes per month in 2022. Network speeds are increasing to
accommodate this data. Broadband speeds are expected to more than double
and wireless speeds more than triple between 2018 and 2023.
It can be hard to wrap one’s mind around the scale of data that is created
and transmitted over the internet. An exabyte is 10^18, or a million trillion
(1,000,000,000,000,000,000), bytes. To give a sense of scale, if the amount
of data transmitted through global networks every day in 2020 were
represented on paper, it would be a stack of paper that would stretch from
the earth to the sun and back. Every day. By 2022, that stack of paper would
have grown roughly 50 percent taller, doing one and a half round trips from
the earth to the sun every day.
This explosion of data creates enormous opportunities for companies
and governments that can harness it. Data has value both directly as a
resource to be mined and analyzed for information but also for training
machine learning systems. Many of the recent advances in AI are due to
machine learning, an AI method in which algorithms are trained on data.
This method contrasts with rule-based AI systems that follow a set of hand-
crafted rules to govern their behavior, sometimes called “good old-
fashioned AI” (GOFAI).
A subfield of machine learning behind much of the recent AI progress is
deep learning, which uses deep neural networks. Neural networks are
loosely inspired by mammalian brains, in which artificial “neurons” pass
signals to one another through a network. The neural network “learns” by
adjusting the strength of the connection between neurons in the network.
“Deep” neural networks are those with many intermediate layers between
an input layer and an output layer.

Deep Neural Network. Deep learning is a subfield of machine learning in which deep neural
networks are trained on data. Loosely inspired by mammalian brains, neural networks “learn” by
adjusting the weights of connections in the network. Some of the largest neural networks have
billions of connections.

Many of the datasets used to train deep neural networks are massive.
ImageNet, the image database that kicked off the deep learning revolution
in 2012, includes 14 million images. In order for a neural network to learn
what an object looks like, such as a “cat,” “car,” or “chair,” it needs many
examples to develop an internal representation of that object. For any given
object, ImageNet contains roughly 500 to 1,000 images of that object to
allow for a rich set of examples. Deep learning is a more data-intensive
process than writing a set of rules for behavior, but deep learning can also
be vastly more effective at building intelligent systems for some tasks.
Imagine trying to create an AI system to tell the difference between a
picture of an apple and a picture of a tomato. It would be difficult to write a
set of rules to determine the difference. Both are round, shiny, and red (but
sometimes green). Both sometimes have a green stem on top. Yet a child
can tell the difference between them and so can deep neural networks if
they are trained on sufficient numbers of labeled images of apples and
tomatoes. Deep learning image classifiers surpassed human performance on
benchmark tests of image identification back in 2015.
As deep learning has continued to evolve, AI researchers have turned to
ever-larger datasets to train more advanced AI systems. In 2019, OpenAI
announced a language model called GPT-2 trained on 40 gigabytes (GB) of
text. At the time, it was the largest language model that had been trained,
with 1.5 billion parameters. Two and a half years later, Microsoft and
NVIDIA announced Megatron-Turing NLG, a 530 billion parameter
language model that drew its training data from an 825 GB text database.
AI researchers continue to see performance gains from ever-larger models
and datasets, demonstrating the value of large datasets.
The search for historical analogies to understand the geostrategic
significance of digital power, such as data, is understandable, even if such
analogies are bound to be imperfect. “Data is the new oil” became a
common refrain for a brief period, spawning articles and op-eds in outlets
such as Wired, Forbes, and the New York Times. (Not long after came a
counter-wave of data-is-not-the-new-oil articles, including in many of the
same outlets.) Data, of course, is not oil. But it can be a useful metaphor in
certain ways. Not only are data and oil both valuable, they are both critical
inputs into the technological revolutions of their age. Oil is used to fuel
engines that can be used for mechanical work. Data is used to train machine
learning algorithms that can perform cognitive labor. Both data and oil are
resources that can be extracted, pooled, and traded. Both need to be refined
to be used. Amassing data, like oil, confers advantages to those who can
find ways to use it.
 Yet there are important ways in which the comparison breaks down. Oil
is consumed when used, while data is not. Data can be freely shared and
copied, while oil is a scarce resource. But the most consequential difference
—and where claims that China is “the Saudi Arabia of data” fall apart—is
that data is not a fungible resource. Whereas oil can be refined into different
fuels that can be used to power cars, semitrailers, cargo ships, commercial
airliners, tanks, fighter jets, warships, and power plants, data’s uses are very
specific to the type of data that is collected. Data on faces can be used to
train facial recognition systems but won’t be of much use for training AI-
enabled fighter jets in air combat. Imagine, if instead of refining crude oil
into gasoline, propane, and other products, companies had to drill for the oil
to make gasoline, diesel, propane, and jet fuel in different places.
Furthermore, imagine if each make and model of car required its own type
of gasoline. Oil would be less valuable than it is now. (“Data is the new
natural gas” might better draw attention to some of the limitations of data
but is not nearly as catchy a phrase.) Oil remains the new oil.
In fact, machine learning systems are often so narrowly constrained by
the datasets on which they’ve been trained that their performance can often
drop if they are used for tasks that are not well-represented in the training
data. For example, a facial recognition system may perform poorly on
people of races or ethnicities that are not adequately represented in its
training data. A machine learning algorithm used for predictive
maintenance on one aircraft won’t work on another aircraft—it would need
to be retrained on data for the new aircraft. It may not even be effective at
predicting maintenance needs on the same aircraft in a new environment,
since maintenance needs may differ based on environmental conditions,
such as in a desert where sand can clog parts or in a maritime environment
where there is saltwater corrosion.
Nevertheless, data, like oil, has enormous geopolitical significance. Do
some nations have a national advantage over others in how they use data?
The United States, China, and Europe operate under vastly different
regulatory regimes when it comes to personal data. Europe has leaned the
most into regulating data collection, with its General Data Protection
Regulation (GDPR). The U.S. government, by contrast, has taken a more
laissez-faire approach to regulating technology, allowing the growth of
“surveillance capitalism” in which big tech companies collect and store
massive amounts of personal data. (Although political winds in Washington
are starting to shift with a growing “techlash” against big tech firms.) China
represents the starkest difference, with the Chinese Communist Party
building an intrusive and expansive techno-authoritarian surveillance
apparatus, which is imperfect and fragmented for now but will become
increasingly capable over time. Yet China’s data advantage, which has been
espoused both by proponents of China’s AI development and those who
fear it, is overstated. There are some ways in which Chinese companies and
the Chinese government will have access to more data than their American
or European counterparts, but this does not necessarily translate to a major
national advantage, and it certainly doesn’t directly translate to a military
advantage.
In theory, the argument for China’s data advantage stems from the
combination of China’s large population and lax data privacy regulations.
The truth, as always, is more complicated. China has the largest number of
internet users in the world, with an estimated 900 million internet users as
of 2020. India comes in second, with roughly 750 million internet users in
2020, the European Union (collectively) in third with around 400 million
users, and the United States a distant fourth with roughly 290 million
internet users. China and India have an even higher internet user potential
than Europe and the United States simply by virtue of their larger
populations: roughly 1.4 billion for China and India each compared to just
under 450 million in the European Union and 330 million in the United
States. More people means more data, but far more relevant than national
internet user metrics is the size and diversity of the datasets available to
tech companies. Major U.S. tech firms like Meta (formerly Facebook) and
Google have global reach and are hardly constrained by the size of the U.S.
population. Facebook has 2.7 billion users and YouTube over 2 billion,
compared to WeChat’s 1.2 billion. The diversity of data also matters a great
deal for training robust machine learning systems, and, with the exception
of ByteDance’s TikTok, Chinese apps have struggled to gain a foothold
outside of China. Tencent, which owns WeChat and other Chinese social
media apps, will be in an overwhelmingly dominant position to leverage its
user data to train algorithms to predict the behavior of Chinese social media
users, but those predictions may not hold for markets outside China.
 Additionally, Chinese citizens’ alleged “cultural nonchalance about data
privacy,” in the words of Chinese AI pioneer Kai-Fu Lee, is also overstated.
For one, American citizens have acquiesced without much pushback to U.S.
tech companies hoovering up their personal data. Data privacy issues in
China are also more complex than they might first appear. While the
Chinese government exercises enormous surveillance powers over its
citizens, many police databases are decentralized and localized. The
Chinese government has also taken steps to increase privacy protections
from corporate (although not government) surveillance, and the government
may see advantages in reining in the independent power of tech companies.
The Chinese Communist Party has begun cracking down on the power of
Chinese big tech firms, reeling in once-powerful moguls like Alibaba
cofounder Jack Ma.
There are some areas in which Chinese firms will have major data
advantages over U.S. companies, which are likely to translate into technical
advantages in some AI applications. The most notable among these is facial
recognition, which is being widely deployed in China while a grassroots
backlash in the United States has slowed deployment. Several cities and
states across America have banned facial recognition for law enforcement.
In 2020, Amazon and Microsoft placed a moratorium on selling facial
recognition to law enforcement and IBM cancelled its work on facial
recognition. These national differences are likely to give Chinese firms a
major edge in future facial recognition development. This edge in facial
recognition may also help to lift China’s AI ecosystem somewhat overall,
although it isn’t clear how much these advantages will carry over into other
AI applications.
Data is vital to AI, but its advantage is likely to be constrained to
narrow applications. When comparing countries, tech companies’
dominance of global markets is far more important than national
population.
Finally, the value of data itself as an input to AI may be changing, with
AI researchers working on techniques to reduce their reliance on massive
datasets. These include few-shot, one-shot, and zero-shot learning
techniques, which are methods of training models when there are few, one,
or even zero examples of an item in the training data. Another approach is
using computer-generated synthetic data to augment or even completely
replace real-world data.
The evolution of DeepMind’s go-playing AI systems shows the
changing importance of data. DeepMind’s early version of AlphaGo, which
beat eighteen-time world champion Lee Sedol in 2016, was first trained on
a database of 30 million moves by human expert go players. AlphaGo then
refined its performance to superhuman levels through self-play, a form of
training on synthetic data in which the computer plays against itself. An
updated version, AlphaGo Zero, released the following year, reached
superhuman performance without any human training data at all, playing
4.9 million games against itself. AlphaGo Zero was able to entirely replace
human-generated data with synthetic data. (This also had the benefit of
allowing the algorithm to learn to play go without adopting any biases from
human players.) A subsequent version of AlphaGo Zero was trained on 29
million games of self-play. For DeepMind’s next version, AlphaZero, three
different versions of the same algorithm were trained to reach superhuman
performance through self-play in chess (44 million self-play games), go (21
million self-play games), and the Japanese strategy game shogi (24 million
self-play games). For each type of game, 5,000 AI-specialized computer
chips were used to generate the simulated games, allowing compute to
effectively act as a substitute for real-world data.
Strategy games are a special case since they can be perfectly simulated,
while the complexity of the real world oftentimes cannot, but synthetic data
can help augment datasets when real-world data may be limited. The
autonomous car company Waymo stated in 2020 that they had driven over
20 million miles on public roads, building up a large dataset of real-world
driving interactions. To augment this real-world data, Waymo has been
simulating 10 million driving miles every single day in computer
simulations, racking up a total of 10 billion simulated miles as of 2020.
These simulations are another form of synthetic data, which can then be
used to improve autonomous car algorithms. Simulations allow Waymo to
create thousands of variations of situations, ensuring its algorithms are
robust to a range of driving conditions.
Data will continue to be valuable, but if AI evolves toward techniques
using less data or synthetic data, its value may change over time. By using
computing hardware to substitute for data, synthetic data tilts the balance
away from data as a crucial AI input and toward another input: compute.
 3

COMPUTE

Algorithms aren’t trained on data by magic. They are trained using

computing hardware, also called “compute” for shorthand. If data is over-
hyped as an input of national AI power, the geopolitical significance of
compute is often underappreciated. The deep learning revolution has led to
an explosion in compute requirements for cutting-edge machine learning
research. Control over AI hardware has emerged as a key battleground
between the United States and China and could be a critical determining
factor in the future of AI power. A global semiconductor shortage in 2021
highlighted the vulnerabilities of disrupted supply chains and the
importance of a few key actors who control them.
Machine learning uses compute in two ways. The first is to train
algorithms, a very compute-intensive process. The second is for inference,
in which trained AI models are used to process data. For neural networks,
the training process consists of adjusting the weights of the connections in
the network so that, as training progresses, the neural network adapts to
make more accurate predictions. Training an AI model on large datasets can
take an enormous amount of computing power. The compute requirements
for trained AI models doing inference are much less, sometimes orders of
magnitude less than what was needed for training. While AI models are
often trained at large data centers, the lower compute requirements mean
that inference can increasingly be done on edge devices, such as
smartphones, IoT devices, intelligent video cameras, or autonomous cars.
 Both training and inference are done on computer chips, and advances
in computing hardware has been fundamental to the deep learning
revolution. Graphics processing units (GPUs) have emerged as a key
enabler for deep learning because of their ability to do parallel computation
(which is valuable for neural networks) better than traditional central
processing units (CPUs). A McKinsey study estimated that 97 percent of
deep learning training in data centers in 2017 used GPUs. As machine
learning researchers have turned to training bigger models on ever-larger
datasets, they have also needed increasingly massive amounts of compute.
From 2010 to 2022, the amount of compute used in cutting-edge
machine learning research projects increased ten billionfold. Compute
usage is growing exponentially, doubling every six months. For
comparison, Moore’s law, which guided the development of ever-smaller
computer chips for decades, had a doubling period of every two years. For
some current AI applications, researchers are now using tremendous
amounts of compute. In training an algorithm to achieve superhuman
performance at the computer game Dota 2, researchers at OpenAI used
“thousands of GPUs over multiple months.” Because the computer could
play games at an accelerated speed, the training was equivalent to a human
playing for 45,000 years. In another project, an OpenAI team trained a
robotic hand to manipulate a Rubik’s cube in 13,000 years of simulated
computer time.
The massive amounts of computing power used for machine learning
research doesn’t come free. Leading AI research teams at organizations
such as OpenAI, DeepMind, and Google Brain are spending millions on
compute pursuing the latest advances in AI. These exorbitant sums are only
possible because the labs are backed by some of the world’s largest
corporations with deep pockets. DeepMind, which is owned by Google’s
parent company Alphabet, lost nearly $650 million in 2019 and had a $1.5
billion debt waived by Alphabet. In 2019, Microsoft announced plans to
invest $1 billion in OpenAI. (Microsoft and Alphabet were both in the top
five largest publicly traded companies in the world by market capitalization
as of 2022.)
The recent explosion of interest (and money) in AI has been closely
followed by the development of specialized hardware better suited for deep
learning. For example, by 2022 an estimated three-fourths of all
smartphones shipped—1.25 billion devices—will have an AI-specialized
processor on board. These chips will improve the devices’ ability to
perform facial recognition, image identification, natural language
processing, and other AI tasks onboard the device.
AI algorithms and software tools are widely available, with
programming frameworks like TensorFlow, PyTorch, and Keras free online,
yet the hardware ecosystem is concentrated among a small number of
actors. A limited number of companies—and an even smaller number of
countries—wield outsize influence over key chokepoints in global chip
supply chains. Additionally, the compute-intensive nature of deep learning
means that those who have access to specialized AI chips and the deep
pockets needed to pay for compute-heavy training have major advantages
over those who do not. Control over compute may turn out to be the crux of
the global competition for AI power. Among the key countries are some
surprising players, such as Taiwan and the Netherlands, who play an outsize
role in hardware supply chains and, by extension, AI geopolitical power.
The semiconductor, or computer chip, industry is highly globalized,
with supply chains for a single electronics product running through several
countries. Yet these supply chains are also highly concentrated in a few key
points, such as chip fabrication. Computer chips undergo three stages
during production: design, manufacturing, and a post-manufacturing phase
that includes assembly, testing, and packaging. Design and manufacturing
account for roughly 90 percent of the value of the chip, split evenly
between the two, with assembly, testing, and packaging accounting for the
remaining 10 percent of a chip’s value. Some semiconductor companies,
such as Samsung (a South Korean firm) and Intel (a U.S. company) are
integrated device manufacturers that do all three phases of semiconductor
production. Other companies specialize in certain production phases.
“Fabless” companies specialize in chip design and outsource fabrication
(chip manufacturing) to other firms. Conversely, pure-play foundries
specialize in chip fabrication, manufacturing chips on contract for other
firms.
The United States, South Korea, Taiwan, Japan, Europe (collectively),
and China are the biggest players in the global semiconductor market. Yet
these actors are not equally distributed across the semiconductor value
chain. For example, while U.S.-headquartered firms account for 47 percent
of global semiconductor sales, the United States has only 12 percent of
global chip manufacturing. U.S. and South Korean companies
overwhelmingly dominate in integrated device manufacturers, with Intel
and Samsung the two biggest players. U.S. chip-design firms like
Broadcom, Qualcomm, and NVIDIA are some of the leading fabless
companies, and U.S. firms account for 65 percent of the global fabless
market. Yet nowhere in the semiconductor value chain is the market more
concentrated than in pure-play foundries specializing in chip fabrication.
One single firm overwhelmingly dominates the contract foundry
market: Taiwan Semiconductor Manufacturing Company (TSMC). TSMC
is the third-largest semiconductor company globally behind Intel and
Samsung and alone accounts for over half of the global pure-play foundry
market. Combined with other smaller foundries, Taiwanese companies
make up 65 percent of the foundry market. (The United States, by contrast,
has 10 percent of the foundry market.) Raw sales figures understate
Taiwan’s significance, however. Not all semiconductors are created equal,
and TSMC is not only the largest contract foundry but also a technology
leader.
The overarching trend in semiconductors, since the 1960s, has been
toward smaller transistors, which have allowed for increased density (more
transistors per square inch) on chips. This trend, sometimes characterized as
Moore’s law, has resulted in an exponential growth in computing power as
chips have packed increasingly more computing power per square inch.
This, in turn, has enabled not only more powerful but also increasingly
miniaturized electronics, enabling everything from smartphones and
watches to IoT devices and 5G wireless networks. While the pace of
advancement has been slowing for the past decade or more as smaller
transistors approach fundamental physical limits, chip performance
continues to improve, albeit at a slower pace. Yet the counterpart to
Moore’s law is Rock’s law, which predicts exponential growth in the costs
for new semiconductor fabrication facilities. Over the last two decades, fab
costs have increased 11 percent annually and chip design costs 24 percent
annually. A new leading-edge foundry can cost $10 billion to $20 billion.
The massive capital expenditures required make it hard for other companies
to compete with industry leaders. TSMC planned to spend between $40–44
billion in capital expenditures in 2022 alone. The skyrocketing cost of chip
fabrication has consolidated the semiconductor market, driving out many
companies.
At each evolution of chip production, the number of companies at the
leading edge has shrunk. In 2001, when the 130 nanometer (nm) node (a
term used to denote the manufacturing process) went into mass production,
there were over two dozen companies at the leading edge of chip
fabrication. As chips advanced over the next two decades, with each
introduction of a new manufacturing process fewer companies were able to
fabricate leading-edge chips. (More advanced chip fabrication processes are
characterized by smaller sizes, such as 90 nm, 65 nm, 45 nm, 22 nm, 10 nm,
and 7 nm. In practice, node naming has long been divorced from actual
transistor size.) Only two companies, Samsung and TSMC, are producing
chips at the current leading-edge 5 nm process node.
Taiwanese foundries churn out one-fifth of global chip fabrication (from
both contract foundries and integrated device manufacturers), but over 90
percent of leading-edge chips at the 7 nm node and below. For AI chips,
TSMC manufactured eight of the ten AI-specialized chips available at the
16 nm process or below as of 2020, including both leading GPUs (designed
by AMD and NVIDIA). (Intel and Samsung manufacture the other two
leading AI chips.) Taiwan’s outsize role in leading-edge fabrication places it
at the geopolitical fulcrum of compute. Statements like “China is the Saudi
Arabia of data” don’t match reality, but it isn’t a stretch to claim that
Taiwan is the Saudi Arabia of compute.
China, meanwhile, is a relative backwater in chip production but is the
kingpin of chip consumption. Over the last several decades China has
emerged as a global hub of microelectronics manufacturing. Chips that are
designed in the United States and manufactured in Taiwan are often
destined for China, where they will be assembled into final finished
products such as smartphones, computers, or other devices that are then
shipped around the world. China accounts for 60 percent of the global
demand for semiconductors. Because Chinese domestic chip production has
struggled, China imports most of its chips. China runs a massive chip trade
deficit, importing over $400 billion worth of chips annually.
Nowhere in the domain of AI competition is China further behind than
in hardware. Every AI expert I spoke with in China was acutely aware of
China’s hardware gap. China’s dependency on chip imports is not just a
massive economic cost but also a national security vulnerability. As the
U.S.-China competition has heated up, the U.S. government has used
China’s chip dependency to kneecap one of its tech “national champions,”
Huawei, a global leader in 5G wireless technology. The U.S. government’s
willingness to exploit China’s vulnerability in semiconductors has thrust
compute to the fore of geopolitical competition in AI. As the United States
and China compete in AI, control over compute will be a critical factor in
who controls the future of AI.
 4

TALENT

The machine age is being built by people. Human talent is needed to

implement AI technology today and unlock the next generation of
advancements. AI talent is a major source of competition and a potential
differentiator among countries seeking to enhance their national AI power.
The United States, Europe, and China all have robust AI research
communities. Europe (collectively) leads in total size, the United States
leads in quality, and China is the fastest-growing region in both quantity
and quality of AI research. As in other areas, data that captures only a
snapshot in time can be misleading if it doesn’t account for trends. Most of
those trends favor China.
China overtook the United States in total AI publications in 2006 and
Europe in 2017. These figures don’t account for quality, though. U.S.
papers are cited, on average, roughly 30 percent more than European papers
and around 70 percent more than papers from China. China has been
rapidly improving its research quality over the last two decades, however.
The number of Chinese authors contributing to top AI journals increased
twelvefold from 2009 to 2019 and is now roughly 2.5 times the number of
U.S. authors contributing to those journals. China surpassed the United
States in the total number of AI journal citations in 2020. According to a
2019 analysis from the Allen Institute, China was expected to overtake the
United States in the top 10 percent of most-cited papers in 2020 and the top
1 percent of most-cited papers by 2025. Yet as fast as China is growing
talent, it is bleeding it abroad.
A 2020 study by MacroPolo tracked the flow of international AI talent
based on a sample of papers accepted to one of the top deep learning
conferences, NeurIPS. Using authors of papers accepted to the conference
as a proxy for top AI researchers, the study found that more of the top
researchers did their undergraduate education in China than any other
country. Yet the vast majority of top Chinese undergraduates studying AI
left China to pursue their graduate work abroad. Less than a third stayed in
China and more than half came to the United States for their graduate work.
Of those Chinese undergraduates who moved to the United States for their
PhD studies, over 90 percent chose to stay and work in the United States
after graduation. China is the biggest source of AI talent, producing more
students who go on to be top researchers than any other country, but the
United States is the biggest magnet for AI talent, drawing in the best and
brightest from around the world—including China—and keeping them in
America.
The United States is overwhelmingly the top destination for AI
scientists from around the world. Two-thirds of the top AI researchers in the
world work in the United States, the vast majority of whom are originally
from outside the United States. Less than one-third of the top AI researchers
currently working in the United States did their undergraduate studies at
U.S. universities. The remaining did their undergraduate work abroad and
came to the United States, with the top sending regions China, India, and
Europe.
American institutions—both universities and companies—are a major
draw for AI researchers from around the world. Google, Stanford, Carnegie
Mellon, MIT, and Microsoft Research top the charts for affiliations for
researchers published at top deep learning conferences. Of the top fifteen
institutions publishing deep learning research, thirteen are American
universities or corporate research labs. Only one, Tsinghua University, is
Chinese.
There is no denying that quality matters far more than quantity when it
comes to advancing basic research. The competition for AI talent is fierce,
with fresh PhDs reportedly making between $300,000 to $500,000 or more
a year in salary and stock options. Top AI researchers are highly sought
after and can make millions. Jack Clark, then policy director for OpenAI,
told the New York Times, “For much of basic AI research, the key ingredient
in progress is people rather than algorithms.” Algorithms are not a scarce
resource. People are.
Yet when it comes to translating AI into national power, leading in basic
research may matter less than implementation. Because the AI community
is so open, advances made in top AI labs may be rapidly copied and put into
use by others. Building practical AI applications depends upon not only
having a ready pool of AI talent, but also empowering AI researchers and
entrepreneurs with financing, resources, and the freedom to build. In terms
of implementing AI, the global playing field is much more level. Even if it
lags in cutting-edge research, China is equal to or ahead of others in AI
deployment.
The U.S. government can take steps to double down on America’s
advantages in talent. In 2020, more than twenty American universities and
companies, including Carnegie Mellon, Stanford, Amazon, Google, IBM,
and NVIDIA, called on the federal government to create a national AI
research cloud to empower academic researchers with greater compute and
data resources. One of the unfortunate consequences of the exponential
growth of compute in machine learning research has been to divide the AI
research community into haves and have-nots. Compute-intensive research
has become increasingly concentrated in the hands of corporate-backed
labs, such as DeepMind, OpenAI, and Google Brain. Anima Anandkumar,
director of machine learning at the chip company NVIDIA and a professor
at Caltech, noted that academics “do not have the luxury of a large amount
of compute and engineering resources” even to replicate, let alone compete
with, research from corporate-backed labs. Increased funding can help
academics stay engaged in compute-intensive research. Giving universities
the necessary resources will help them continue to attract top faculty and
students to American universities and make the most out of the talent that
resides there.
The United States has natural advantages in human capital, and
concerted U.S. government action can help double down on those
advantages. The bipartisan National Artificial Intelligence Initiative Act of
2020 established the creation of a National Artificial Intelligence Research
Resource to provide students and faculty with necessary data and compute
resources. Similarly, investments in new semiconductor fabs in the United
States will not only help secure chip supply chains; they will also grow the
talent base of U.S. microelectronics engineers working on leading-edge
fabrication technology, helping to ensure American leadership in the next
generation of semiconductors.
The Chinese government has long prioritized scientific talent as part of
its national competitiveness, and no aspect of international talent flows has
drawn more controversy than China’s Thousand Talents Plan. The program
offers participants high salaries, bonuses, research funding, lab space, and
housing in exchange for bringing their scientific knowledge back to China.
Thousand Talents is only the most prominent of China’s estimated over 200
talent recruitment plans. The Chinese government uses programs such as
Thousand Talents to tap into the massive pool of Chinese scientists
overseas, which one study estimated might be as large as 400,000
individuals. China’s talent programs are not limited only to Chinese
nationals, however. In 2021, a jury convicted the former chair of Harvard’s
Chemistry and Chemical Biology Department of lying to federal authorities
and tax fraud for failing to disclose a $50,000 a month salary he was paid
for participating in Thousand Talents. (Participating in a talent program is
not illegal; failing to disclose foreign income and conflicts of interest on a
U.S. government grant application can be.)
The U.S. government has increased its attention on Chinese talent
recruitment programs in recent years as part of a broader effort to crack
down on Chinese espionage and intellectual property theft. In response to
increased scrutiny, the Chinese government deleted mention of Thousand
Talents online, yet recruitment efforts continue. In 2019, the Chinese
government consolidated Thousand Talents and other recruitment programs
into the new High-End Foreign Expert Recruitment Program. Emily
Weinstein, a research analyst at the Center for Security and Emerging
Technology who built a comprehensive Chinese Talent Program Tracker,
described Chinese talent programs as “a moving target.”
Through increased investigations and improved visa screening, U.S.
policymakers aim to blunt China’s technology transfer campaign without
cutting off the flow of Chinese talent to the United States—a difficult
balancing act. High-skilled immigration from outside China presents less of
a challenge, and the United States has an opportunity to double down on its
advantages in AI talent by increasing the inflow of top researchers from
around the world.
The congressionally appointed U.S. National Security Commission on
AI warned in its 2021 final report, “The United States is in a global
competition for scarce AI talent. . . . For the first time in our lifetime, the
United States risks losing the competition for talent on the scientific
frontiers.” The commission called for a new National Defense Education
Act, mirroring the one the United States passed in the wake of the Soviet
Union’s launch of Sputnik in 1957, to reinvigorate math and science
education. Yet a nation of 330 million people will always be at a
disadvantage competing against a nation of 1.4 billion if the United States
restricts itself to home-grown talent. America’s unique advantage in the
competition for AI talent is its ability to draw on the best and brightest out
of the world’s 8 billion people. The biggest barrier today is America’s own
immigration policies. In a survey of top machine learning researchers
residing in the United States, nearly 70 percent said visa and immigration
problems were an obstacle to recruiting foreign talent. The National
Security Commission on AI called for expanding high-skilled immigration,
arguing “immigration reform is a national security imperative.”
In early 2022, the Biden administration announced a series of measures
to ease obstacles to international science, technology, engineering, and
mathematics (STEM) talent coming to the United States. These included
increasing foreign exchange programs, expanding the availability of visas,
and updating immigration policies for those with advanced degrees. These
are valuable steps, yet additional actions are needed to make it easier for
U.S. universities and companies to recruit talented scientists and engineers
from abroad.
Human capital is at the core of the global competition for AI power.
Innovation comes from people. China is improving its domestic education
and recruiting foreign scientific knowledge through programs like
Thousand Talents, but it cannot compete with the United States as a global
draw for students and entrepreneurs. The competition for talent is, unique
among core AI inputs, the United States’ to lose. The United States remains
a global leader in AI talent despite inadequate attention to domestic
education and neglected opportunities for increasing high-skilled
immigration. With reforms, the United States could secure another
generation of global science and technology leadership.
 5

INSTITUTIONS

Institutions are the final critical input for national AI power, transforming
the raw elements of data, compute, and talent into practical applications.
Having a strong foundation of the latent elements of national AI power is
valuable but doesn’t automatically translate to a military, political, or
economic advantage. To use AI to boost military power, military
institutions must be able to access AI technology, successfully import and
adopt it, and find ways of using it for military advantage. To use AI to
shape the information environment, nations must be able to build tech
platforms that dominate the information ecosystem, such as Facebook or
WeChat, along with platform governance rules that advance their national
interests. Because AI is such a diffuse technology, many countries will have
access to the necessary inputs. How nations apply these raw inputs will
have a tremendous influence on how the AI revolution unfolds.
The AI revolution poses a particular challenge for militaries, since the
bulk of AI innovation is occurring outside of traditional defense industries.
In the 1960s, the federal government dominated the U.S. research and
development ecosystem, funding nearly two-thirds of all U.S. R&D. But
public sector R&D funding declined over time and is now less than one-
quarter of all U.S. R&D, with private companies picking up the slack.
Much of this commercial innovation has military applications. Eleven of the
Defense Department’s fourteen “critical technology areas” are commercial
technologies. To access many cutting-edge, transformative technologies,
Random documents with unrelated
content Scribd suggests to you:
—Je t'ai moins vu que lui. Je ne t'ai pas vu souvent, je ne t'ai pas vu
longtemps. Il y a une fatalité, une prédestination qui nous ont
poussés l'un vers l'autre: il n'y eut pas de fatalité entre lui et moi,
tout fut humain, presque petit, tout se tissa de pitié: ce fut un étroit
et gris couloir d'émoi.
—Ah! chérie, comme tu es cruelle. Je veux échapper à cet homme
qui est en face de moi et tu me le renvoies et tu le jettes sur moi—
en beauté, il me cerne de toutes ses vertus et de toutes les larmes
que tu vas verser sur lui—car comme tu vas pleurer, chérie!
—Je pleure, mon ami, je pleure mais ce sont des pleurs sans
méchanceté et je pleure sur toi, sur lui, sans savoir pourquoi.
—Ah! pleure sur moi, chérie, pleure beaucoup. Tu m'admires: tu as
tort. Je suis un pauvre petit garçon et j'ai vieilli sans le vouloir et j'ai
conservé tous mes défauts, toutes mes impatiences, toutes mes
débilités et toutes mes susceptibilités et toutes mes timidités. Pleure:
j'ai de très vieux parents quelque part, qui pensent à moi et qui
pensent à la mort et qui sont seuls dans de pauvres murs, dans de
pauvres meubles, qui ont reçu les années, à bout portant et à
l'ancienneté, sur leurs têtes, sur leurs jambes, sur leurs bras—et à
qui il n'a pas été fait grâce d'une infortune, d'une maladie et qui les
ont eues l'une après l'autre, en cadence, à la suite... Pleure: j'ai un
passé terne qui se double de cauchemars et quand je me le rappelle,
je ne me le rappelle pas bien et je ne sais pas si je passe des
calamités, des monotonies—ou si j'en ajoute. Pleure: j'ai des doutes.
Pleure: j'ai un avenir qui hésite, qui se sauve, qui se fait tirer à moi,
qui résiste—et je n'ai pas le courage de le tirer.
—N'insiste pas: ne me demande pas de trop pleurer sur toi, je ne
puis pas. Tu m'as, moi, tu m'as toute.
—Toute?
—Oui, toute.
—Et ton mari, tes regrets, tes remembrances?
—Ah! ne me demande pas d'explications. Ce sont des sensations,
des nuances.
—Tu m'as parlé de nuances, tout à l'heure—pour lui.
—Ça ne fait rien. Je t'aime, je l'aime. Je l'aime—et je n'aime que toi:
voilà. Tu ne crois pas?
—Ah! chérie, chérie, si je crois! je ne suis pas sûr parce que la
certitude est encore du raisonnement, de la ratiocination, de la
machinerie, de la marchandise à logique, mais je suis plein de toi,
plein de foi et je suis irradié de ta divinité. Et je dis des bêtises.
—Dis toujours.
—Non! j'ai besoin de silence, d'un silence pour enfant, pour enfant
qui a peur la nuit et qui implore, jusqu'à ce qu'il les entende, de
souples ailes de fée sur son sommeil. Et l'enfant est inquiet tout de
même, parce qu'il n'est pas seul, parce qu'il a peur du cortège de la
fée, de l'omnipotence de la fée, de la bonté de la fée, parce qu'il
s'avoue que tout cela est trop grand, trop surnaturel pour lui—et j'ai
besoin du silence d'une chambre de petite fille où un grand frère de
dix ans veille sur sa petite sœur et j'ai besoin du silence des
évocations, du silence des magies, du silence de création et du
silence de néant. Parle, toi, car tu parles bien, car tu dis des mots
nécessaires, que je ne puis prévoir en leur simplicité et qui me
surprennent comme le génie.
—Je ne te parlerais que de lui.
—Eh bien! veux-tu que je lui dise ce que tu dis de lui? que je lui
rapporte tes louanges et tes glorifications?
—Tu ne le pourrais pas. Tu ne te rappellerais pas. Ce sont des mots
qui s'évaporent comme la rosée, qui s'évanouissent comme des
nymphes élégiaques, qui ne bruissent que dans le mystère et qui se
perdent comme les petits vagabonds, dans les forêts de légende. Et
si tu veux essayer...
—Je ne sais par où commencer et c'est un discours difficile,
d'homme à homme.
—Ah! ah!
—Et puis je n'ai pas le temps: il se lève, il déclare: «Je dois rentrer:
ma femme m'attend»; il me serre la main et il s'en va. Il te rejoint,
toi, toi! Ah! parle-moi, parle-moi de n'importe quoi, de lui, pour que
j'entende—en moi—ta voix, pour que je ne sois pas seul, assis sur
mon bonheur comme sur la pierre d'un tombeau.
Ah! ton mari! il a eu plus de compassion que toi, il est parti, par
modestie, pour ne plus m'infliger son éloge.
Mais non.
Il a coupé, traîtreusement, notre conversation de sa fuite et il a fui
vers toi, vers ta caresse, vers les litanies d'adoration que tu viens
d'improviser et que tu perpétues.
Ah! n'est-ce pas? tu t'arrêtes? tu arrêtes net ton affection qui se
précipite et qui se cabre, tu achèves en un murmure ton oraison
ardente, claire et haute.
Je ne t'entends plus. Je n'entends plus rien. Il t'entendra encore, lui:
il t'entendra discuter, conter, babiller, imiter, te moquer, que sais-je?
Il aura la fanfare diverse et journalière de tes opinions, de tes
manies et il aura, en des paroles, en des gestes menus, ta nature et
ton humanité.
Des heures... des heures... Et les mêmes heures se dresseront pour
moi, vides, rèches, sèches, obscures, qui me tortureront de ton
fantôme épars, qui me jetteront ton absence dans les jambes et
dans le cœur.

Dormir... dormir...
Quand j'étais petit et quand j'avais mal c'était le mot qui matait ma
douleur, dont j'essayais de me couvrir, de m'enlinceuler. Dormir...
dormir... Le sommeil est si vaste, si libre et si vague que je pourrai te
héler et t'appeler en barque, que tu pourras me tendre les bras du
haut d'une montagne, que tu pourras surgir pour moi d'une étoile ou
d'un ciel.
Mais il faut mériter le sommeil et achever d'abord sa journée: on ne
s'endort pas, comme ça, parce qu'on a envie de rêver, il faut qu'il
soit l'heure, car il est l'heure de dormir—comme l'heure de mourir.
Et je reste l'otage des amis de ton époux qui commentent les
événements, gravement, et qui en ont négligé, en route.
Ah! messieurs, il s'est accompli aujourd'hui un prodige plus
remarquable: une ère s'est ouverte, aujourd'hui, qui est la seule ère.
Et la volupté est née aujourd'hui.
Ce n'est pas une chose à dire mais mes lèvres ont frémi,
apparemment, car ces hommes se sont tournés vers moi et
m'interrogent. Je leur dois une réponse, je leur dois ma quote-part
de propos car j'ai été bien sage jusqu'ici et bien discret.
Et je suis si prisonnier de ton souvenir, si esclave de cet homme qui
vient de s'en aller, si esclave de tout ce que tu as chanté, de loin, sur
lui, que je me décide.
—Tortoze, avant de partir, ne vous a pas tuyautés sur son invention?
Et je l'invente, cette invention, au hasard, je la bourre
d'invraisemblance, je la complique de perfection, je l'élargis de
sublime et je vais, je vais: l'invention prend corps, éclate, se
consolide, s'attable en face de moi et les amis écoutent, s'étonnent,
admirent, se courbent devant l'ombre de celui qui te rejoint, là-bas,
et constatent: «Ça c'est tout à fait, tout à fait épatant!»
IV
LE CŒUR, LE CERVEAU ET LES YEUX
Le lit où je me suis couché est un lit que tu ne connais pas: il est
situé au bout du monde, comme il convient, à l'autre bout du
monde.
Un corridor y conduit, bossué, bosselé, écartelé, très long, très étroit
et jaloux.
Ma chambre déborde de livres, de livres inutiles, car je n'y lis jamais:
c'est une chambre d'attente et une chambre de rêves.
C'est une chambre d'alchimiste où j'ai forgé des avenirs, où j'ai pétri
des ambitions, où j'ai façonné l'univers à mon caprice, à ma
convoitise, à ma fantaisie et à ma raison.
Mais voici longtemps que, en un envoûtement passionné et en un
agenouillement sans fin, je n'y ai plus songé qu'à toi, où je n'ai pétri
—d'une main si tremblante et si malhabile—que l'avenir où tu
souriais, où je n'ai forgé que l'ambition où tu te dressais, où je n'ai
façonné l'univers qu'à ton caprice, à ton caprice où tu m'admettais.
Ton image, comme un clown d'au-delà, a dansé, a sauté ici à travers
toutes les auréoles—et cette chambre est restée—de toi—boiteuse,
borgne, folle.
C'est la chambre où, comme au haut des tours pour fillettes frêles,
on monte pour voir venir, pour interroger les astres et pour
s'interroger mieux, en liberté. C'est une chambre où j'ai eu faim, où
j'ai douté, où j'ai pleuré, où j'ai été plus seul que partout et que
nulle part, où je me suis senti—des soirs—vraiment dieu et, d'autres
soirs vraiment néant, où j'ai eu des regrets, des espérances et des
remords et ces remords, ces regrets, ces espérances, cette
humanité, cette divinité, cette humilité, ces larmes, ces doutes, ces
faims, cette misère éparse et ces désirs demeurent, s'obstinent,
s'éternisent dans un pli de livre, dans un tournant de mur, dans un
retroussis du tapis sordide, et dans les papiers et les hardes qui
s'amoncellent, et se confondent.
Rien n'est plus résolument triste, rien n'est plus parlant et plus
silencieux qu'une chambre d'hôtel, rien n'est plus accommodant à
votre âme—quand vous avez une âme.
Ma chambre est une cellule de couvent, altière et nue, et c'est
depuis quatre ans le désert même.
J'y ai reçu des lettres et de mauvaises nouvelles sans un mot, sans
une plainte et je n'ai pas bronché, pas rougi, pas rugi. Elle a gardé
sa majesté et son énigme; elle a été le nid et le refuge, le reposoir
et la caverne.
Elle m'a envoyé et renvoyé ton portrait de ses parois sans miroir, et
cette nuit, plus discrète encore que les autres nuits, épaississant son
silence, épurant son mystère, elle s'est endormie sur ton souvenir,
sur ta présence, sur ton obsession, sur ton immensité.
Et elle m'a endormi, moi aussi: j'avais peur de ne pas dormir et de te
chercher, de mes mains de fièvre: j'ai dormi.
J'ai bien dormi, en une extase.
Mais le réveil me rapporte le bourdonnement de mon bonheur et de
mon anxiété, le réveil me rapporte mon veuvage.

Et cette chambre est trop vide, trop pleine aussi de toi. Elle est trop
accoutumée à mon infortune, à ma faim: c'est une chambre de
patience, de résignation, c'est une chambre d'où l'on prend son élan
—et il me faut rentrer—de plain-pied—dans la joie.
Je m'y rue. Les rues se filent, se coupent, les rues s'enfantent l'une
l'autre, sans fin, qui mènent à ma pauvre chambre du bout du
monde d'en face, du bout du monde opposé et c'est un entrelac de
boulevards et de carrefours, ce sont des arrêts de voitures, des
lenteurs et d'autres lenteurs: tout se met en travers de mon rêve et
je monte, je monte—car mon temple est situé en haut d'une
montagne, pour que je puisse avoir Paris à mes genoux, quand je
serai à genoux.

Et me voici à ma petite chambre, à notre chambre: j'ouvre la porte

d'un coup sec, d'un coup brusque.
Je ne veux pas que la chambre continue à dormir, je veux qu'elle
s'éveille en sursaut, qu'elle me saute à la gorge, qu'elle crie et
chante vers moi, qu'elle soit tyrannique, agressive et câline, qu'elle
m'étouffe de tendresse, de grâce, d'amour, que tous ses souvenirs,
que la masse de son émoi m'écrasent, me piquent, me crucifient, de
leur âpre et chaude volupté.
Mais la chambre est suppliante: elle a mal dormi, sans nous.
Et elle ne se rappelle rien que notre histoire, l'histoire que je lui ai
tissée hier des cheveux fins de ma chérie et de tous les fils de la
Vierge qui traînèrent en nos après-midis et en nos crépuscules.
J'ai tant de temps à tuer et à tuer sans méchanceté jusqu'au
moment où elle viendra, où elle sautera de mon cœur dans ma vie.
Je suis très las, vieux désespérément.
Je m'étends sur le lit et je songe.
Je songe pour la chambre et pour moi.

Et voici les pâles et fiévreuses évocations qui, lentement, une à une,

des antipodes et d'à côté arrivent et me reprennent. Car tu partis
seule pour l'Italie, seule avec ton mari.
Et je dus quitter la sainte caverne bleue qui s'était dressée pour
nous sur la mer. Petit Poucet mélancolique, je m'éloignai plus vite et
le cœur plus gros que le Petit Poucet: pour retrouver mon chemin
vers la tacite caverne bleue, je semais ton souvenir sur le chemin, je
semais et ton souvenir grandissait au fur et à mesure, de temps en
temps j'arrêtais ma fuite, je descendais en une ville pour être, en ma
fuite, plus près de toi.
L'horreur grasse de Marseille, ses fenêtres étroites, sa mer mangée
de vaisseaux et de barques, ses voiles rouges, ses rumeurs
piémontaises, tout me cria ta grâce et ton azur, ta fraîcheur, ton
élégance, ton charme net. Cette ville facile, trop amène, se prêtant
trop, cette ville prostituée et racoleuse me jeta à la face, de son
impudence et de son impudeur la pudeur de notre rencontre et de
notre destin, et les arbres—où il y en a—me furent, comme partout,
consolants et prometteurs. Du haut de sa montagne, Notre-Dame de
la Garde se dressa pour nos fiançailles et, comme pour les mariages
des reines, à rebours, je t'y épousai par procuration.
C'était mon cœur qui te figurait, qui te représentait, mon pauvre
cœur qui m'avait quitté pour te suivre et qui quittait un moment
cette Italie confuse où Florence, Venise, Rome et Naples se ruaient
l'une dans l'autre et s'aggloméraient pour enfermer toute beauté,
toute fatalité, toute divinité et tout souvenir, quittait les âpres routes
aussi et la pitié éparse et morne dans les lagunes et dans les golfes,
dans les montagnes et les volcans—afin de se prêter à cette
cérémonie et de mettre le Dieu des marins, des aventuriers et des
pirates, le Dieu des forçats, des veuves et des fiancées, le Dieu des
misérables, et des simples en nos espérances, en notre fièvre, en
notre histoire.
Mon cœur et moi nous n'entrâmes pas dans l'église. De très loin, de
très bas, au ras du port, le dos en des mâtures et des voilures, en
des grelins légers et des cordages fins, et mon cœur à côté contenu,
soutenu, arrêté par les treillis bruns et blonds, par l'harmonieux
enchevêtrement des gréements, du chanvre et du lin, en une prison
de soie et de fer, nous fîmes descendre lentement, doucement,
l'église sur nous.
Derrière nous l'univers se pressait dans le gréement, les cordages et
les voiles des vaisseaux, l'univers était là, tassé, immobile et les
siècles aussi (car il y avait de très vieux bateaux, des bateaux qui ne
naviguent plus—et si lourds de leurs coques et de leurs carènes, de
leurs attributs désuets et de l'univers attardé, des siècles endormis
qu'ils gardent à leur bord, parmi leur équipage fantôme, des bateaux
si muets et si tristes qu'on les laisse dans le port mourir quand ils
voudront—et les pays de songe, les pays de bataille, les pays de
glace et les pays de soleil demeuraient attachés aux câbles et
gonflaient les voiles, gonflaient les cheminées aussi et l'univers, les
siècles, toutes les mers, les âmes des marins et ce qu'il subsiste de
fatalité dans les soutes des bâtiments de commerce, d'héroïsme sur
les tillacs des frégates désaffectées, ce furent nos témoins et les
invités de nos noces).
Il n'y eut point de chants trop graves et trop nourris pour effarer les
antiques mâtures: le silence, un silence lyrique et liturgique, deux
souffles d'âme, l'éternité de deux «Oui» et Notre-Dame de la Garde
remonta sur sa montagne et je quittai sans un «Au revoir» mes
témoins les vaisseaux, les univers et les siècles.

Les cailloux pointus d'Avignon me parlèrent de toi, mon aimée et,

tout droit, d'un seul jet, d'un seul effort tranquille, le château des
Papes ne bougea pas et nous bénit de haut, imperceptiblement, raidi
en son austère magnificence, en sa hautaine nudité, en sa sobre
fierté, et les siècles encore entre les pavés pointus, entre les portes
sculptées et les balcons, entre les jardins et les places, les siècles me
prirent, marchèrent à moi et voulurent me conter des choses des
croisades, des guerres et de foi. Je leur dis: «Je ne suis pas seul» et
ils dansèrent autour de nous des rondes connues, des rondes
d'enfantelets au bord du Rhône et sur le pont, des rondes bien
conservées et ronronnantes de bonhomie et des rondes plus
secrètes, plus anciennes et des rondes qui n'étaient pas des rondes
et qui étaient des danses de nonnes, des danses sarrasines, des
danses de moines, des danses de cardinaux, de papes et
d'hérésiarques. Ces danses nous entraînèrent sur le rocher gris, vert
et blanc qui se penche sur le fleuve et qui s'en moque un peu, elles
nous jetèrent dans l'île qui flotte sur le fleuve,—et l'île, le rocher, la
ville, les jardins et le ciel chantaient des chants de troubadours, des
cantilènes et des sirventes, des chants de guerre contre les ennemis
qui pourraient menacer notre amour.

Et je m'enfuis loin de cette ville de rondes: Lyon se précipita au-

devant de moi, énorme, grise, toute en montées. Je crus que je
montais vers toi et je montai, je montai. Les escaliers s'espaçaient,
se succédaient, semblaient se cacher pour surgir tout près et ce fut
une ascension pénible, une montée à vide, un vain pèlerinage.

Je ne reçus pas de tes nouvelles. Pourquoi m'aurais-tu écrit?

Et j'avais peur de recevoir une lettre de toi.
Je nous prêtais un si agréable, un si tragique dialogue, je sentais—
sans les entendre—des paroles si impossibles, si caressantes, si
enveloppantes et si aiguës, je te prêtais une telle éloquence et une
telle poésie que jamais tu n'y eusses atteint. A vrai dire, ces paroles
étaient si belles que je ne pouvais même pas les imaginer.
Murmure des sources, murmure des étoiles, murmure des feuilles
dorées au-dessus des étangs, plaintes des oiseaux et sourires
psalmodiés des cieux, c'étaient toutes les idées et tous les langages
de la nature et de l'au-delà, tout, excepté des paroles. Murmure qui
me faisait murmurer: «Que c'est joli!» et qui me faisait fermer les
yeux, fermer ma mémoire pour entendre encore, pour être tout à ce
murmure, pour être tout murmure.
Et Paris, où j'étais revenu, que j'avais lancé sur moi comme un
écrasant manteau de maisons et de soucis, Paris me permit ce
murmure à tous ses carrefours, à tous ses coins. Je t'y retrouvai—si
peu!
Et des instants se rencontrèrent où je te parlai.
J'étais condamné à un jargon de convention, à un jargon travesti, à
cause des gens et à cause que, par une honte pieuse et par
impuissance, je ne retombais pas, les lèvres en avant, sur ce
murmure unique et suave qui creva pour nous le firmament.

Je cause avec toi, la bouche tremblante et tordue de contrainte, des

mille événements qui rident notre indifférence, de ce monsieur, de
cette dame et de ce livre. Des gens, les gens plongent en notre
conversation et s'y perdent, et s'y oublient.
Parfois pourtant je puis te dire: «Vous savez que je vous aime».
Ah! ton sourire, chérie, le drame de ton sourire! J'y perçois tout
l'azur, tous les azurs de notre entrevue et de notre destin! Mais ce
sont des azurs, c'est un sourire que je dois garder pour moi tout seul
et je ne puis y faire aucune allusion, je ne puis les tremper en ce
marécage, en ce vaudeville de la vie.
Je dois me contenter de te dire: «Vous savez que je vous aime» ou
«à propos, vous savez que je vous aime» et me contenter—me
contenter!—de ta réponse: «Vous ne serez donc jamais sérieux?»
Tu te débats contre le lyrisme de ton existence et contre ta fatalité:
je n'y puis rien. Je ne puis te plonger dans ta beauté comme on
plongea Achille dans le Styx, je ne puis que rester à côté, sottement,
à attendre que tu te souviennes et boire autour de toi, happer en
ton sourire, comme un chien avide, ton azur, ton immatérialité, ton
immensité!
Et dès que je t'ai quittée, en tramway, dans les rues, les mots me
viennent qu'il m'aurait fallu dire, puis c'est le retour de ce murmure
divin, où je cause avec toi et où tu me réponds, dans du sublime. Et
je ne t'aurai pas vue longtemps, car tu t'en vas avec ton mari.

Dans les déplacements et villégiatures que publient les journaux

mondains et les journaux graves:
à Royan... M. et Mme Godefroy Tortoze. C'est tout: déplacements et
villégiatures de mon cœur, déplacements et villégiatures de ma vie!
Nous ne nous sommes dit ni: Adieu ni: Au revoir et tu es partie sans
un baiser. Chérie, chérie, les mers sont méchantes, les mers
mondaines et les chemins de fer sont méchants. Et tout est
méchant, les montagnes et les casinos, les voitures, les bateaux, les
chiens...
Et il faut que je patiente, que j'invoque les éternités, que je me
réfugie en mon rêve. Il faut que je me donne à tous les leurres et
Paris vide de toi, est si grand, si long, si chaud...
Chérie, chérie, j'ai peur de te perdre en cette chaleur, en cette
poussière, en cette atmosphère, en ce malaise.
Tu pèses si peu et ton souvenir est si léger, si inconsistant! il tremble
tant au fond de mon âme, au bord de mon âme—et j'ai si mal. Je ne
sais plus si je t'aime, je ne sais si je te désire, je sais seulement que
je suis ici—où tu n'es pas. Et je sais que je suis devenu si timide
devant mon amour, si pauvre, si peureux! et que j'ose à peine
m'aventurer sur les routes, craignant de perdre sur les routes cette
misérable tendresse—et ma chère, ma chère douleur.
C'est la saison exquise où les forêts s'entr'ouvrant à peine et
s'entr'ouvrant pourtant, se laissent violer doucement, à demi, et se
font intimes, odorantes et charmantes pour ceux et celles qui
veulent se risquer parmi elles en pèlerins, en flâneurs, en amants;
c'est la saison où la mer berce en elle le soleil, la lune et les cieux,
se joue avec les bras et les bouches des femmes; c'est la saison où
tout est idylle, où c'est une idylle entre la nature et les hommes, où
les montagnes et les arbres, les fleurs, les océans, les fruits, les
petites sources et les fleuves se permettent les plus subtiles
coquetteries, à cette fin de rendre la santé, la gaîté, le repos aux
touristes qui les apportent avec eux, dans leur bagages, pour être
plus sûrs de les retrouver.
Parisiens épars, insoucieux de Paris, Parisiens venus sans dieux lares
et avec les fétiches locaux qui sont nécessaires à tel casino ou à tel
autre, Parisiens par la force des choses qu'un Dieu malin essaima
vers les Auvergnes récupératrices et les provinces vengeresses, les
gens se fatiguent et peinent pour oublier leur lassitude.
Ma lassitude est autre et je ne suis qu'élégie et espoir. Les gens sont
au bord de cette panacée moutonnante et liquide en quoi ils ont
déguisé la mer. Ils y découvrent leur tub un peu moins personnel, un
peu plus inconfortable, une piste pour courses sans automobiles, où
le sable ne manque pas, mais est trop bas et trop sale,—et une
arène pour concours d'anatomies.
On leur a dit qu'il fallait rêver: ils y tâchent, mais ce n'est pas facile.
On leur a dit qu'il fallait s'abandonner aux caresses fécondes de la
lune; à toutes les chansons que vient importer la marée haute et au
soupir mélancolique et profond de la marée qui s'en va lentement et
qui revient pour s'en aller, et qui revient et qui s'en va, cependant
que les heures tombent, traînent avec l'eau pour aller se blanchir là-
bas, là-bas où sont les vagues blanches et les nuages blancs, et pour
reparaître (jour nouveau) argentées et lentes et hâtives. C'est
difficile de s'abandonner. Et c'est un plaisir rare qu'indiquent tous les
tarifs d'hôtel.
Jamais il n'a fait plus chaud.
Jamais il n'a fait plus triste.
Jamais il ne fit plus envie de partir, de fuir des souvenirs et des
désirs, de fuir une ombre fraternelle, une ombre ennemie qui se
glisse entre les arbres, entre les rues, pour sourire de son horrible
sourire chaste, de son sourire câlin, de son horrible sourire fidèle, de
cet horrible sourire derrière lequel il n'y a rien, que le vide,
l'impossible, de cet horrible sourire qui est tout sourire, tout charme,
toute vie, tout au-delà.
Jamais il ne fit tant besoin de posséder à la fois tous les arbres, tous
les ciels, toutes les solitudes, les palais historiques et les plus
secrètes chaumières, toutes les sources et toutes les mers.
Jamais une telle soif ne me brûla d'immensité et d'intimité, de larges
espaces à parcourir et d'une couchette étroite—où rêver de toi. Rien
n'est trop loin, rien n'est trop haut; il n'est pas de mer assez trouble,
de montagne assez âpre. Et je n'ose m'arracher à Paris.
Je prends les rues au hasard, comme elles se suivent et c'est une
ville si imprévue, élégiaque, nostalgique qui, nonchalamment,
paresseusement, lève ses voiles et se révèle ville de douceur et de
larmes—pour toi.
Pourquoi aller chercher les canaux dorés de Hollande lorsque les
quais de la Seine, les quais où l'on ne passe jamais, les quais
d'après-Bastille et de la Cité offrent leur lèpre blonde au baiser du
soleil mourant, lorsqu'ils s'entr'ouvrent, se fendillent, se découpent
et s'éternisent sans autre monotonie que celle de la misère et de la
mélancolie?
Autour, ce sont des boutiques de rêve, des devantures de marchands
de vin (oui, de marchands de vins!) où s'étalent des pièces d'or et
d'argent déjà démonétisées au temps du déluge, des officines de
tailleurs où l'on martèle—pour quels cyclopes?—des salopettes de
zinc et des tabliers d'acier, ce sont des maisons croulantes qui ne
croulent pas, des maisons qui s'avancent, qui s'élargissent de bas en
haut, vers le fleuve, pour que les pauvres qui les habitent, puissent
s'y précipiter plus facilement et ce sont, à côté, des maisons Henri IV
où les fenêtres longues, hautes et profondes comme le jugement
dernier se font opaques de leur mystère tricentenaire.
Et voici des jardins publics ignorés, sortis d'on ne sait quels contes
de fées—contes de fées où les fées ne sont pas riches—où les
enfants errent sans s'amuser et où l'on trouve dans le sable rare des
larmes et l'apprentissage en culottes courtes—de l'horreur.
Et voici des églises à la mode de Caen, des casernes archaïques, des
idylles en camisole rose et des amas de bric-à-brac où l'on n'ose
feuiller de peur d'y rester, comme en certaines fontaines pétrifiantes.
C'est le décor désert et peuplé qui convient à ma songerie, la misère
me fait tant te revoir, vague comme tu l'es à mes yeux, femme que
je n'embrassai jamais, femme qui te dressas devant moi, un jour où
j'étais beau de pensée et où tout était beau autour de moi, femme
qui, de la lenteur rythmique et rituelle d'un paysage, de la souple
immobilité des montagnes et de la mer, de la magnificence
hiératique d'un crépuscule, de la jeunesse, de la naïveté, de la
perfection d'un soir, te précipitas en mon cœur, de très haut et du
fond des mers et qui te révélas à moi en même temps que la grâce,
la beauté et Dieu.
Le paysage est triste et les êtres sont misérables: ces quais, ces
squares, tout, jusqu'au crépuscule, est médiocre et désolé.
Et ton souvenir se colle à moi, contre la médiocrité du dehors et tu
m'es un bouclier et tu es cette chose de buée, ce nuage qui
enveloppa des héros contre les dangers.
Ne me protège pas trop et aimons les pauvres. Je suis pauvre, de ne
t'avoir pas, je suis pauvre d'avoir de si pauvres rêves, de si pauvres
évocations et de ne pouvoir fixer ton image devant moi, brutale et
nette.
Et je suis pauvre de tout, et de moi. Je ne puis m'établir sur ces
rives: il faut encore trop d'argent pour vivre avec les pauvres, et j'ai
des amis qui viendraient me tirer à eux, me forcer à rire avec eux.
Et, tout de même, d'avoir trouvé en Paris un nostalgique et exotique
Paris, je veux de la vraie nostalgie et de l'exotisme d'ailleurs.

Il y a dans une petite ville où il est né, un homme qui m'a invité et
qui m'attend. C'est un humoriste. C'est le plus célèbre des
fantaisistes; il a sécularisé le bizarre et rendu l'étrangeté
quotidienne. De sa table de travail, de sa table de café, du milieu du
boulevard il a saisi le cauchemar à bras-le-corps, si j'ose dire, l'a
coiffé d'un chapeau comique, l'a déshabillé, l'a dénudé, l'a scruté et
examiné, puis l'a vêtu sans hâte d'une casaque mi-partie, de la
casaque qu'il voulait, en a fait sa chose et l'a offert ensuite au public
sans hauteur, sans roideur, gentiment, comme un apéritif ou un
cigare. Il ne s'est pas mis à l'affût des mouflons à cinq pattes ou des
sangliers du Thibet. Il a erré, musé parmi les boulevards,
s'intéressant à tous les passants et à tous les néants et, tout à coup,
de deux doigts, il a saisi, conquis, retenu quelque chose dans l'air—
et c'était le rire, et c'était le burlesque, le grotesque, la rapide et
immense féerie. Il a derrière lui, comme une escorte, comme un
état-major, comme une armée, le rire de tout une ville et de tout un
peuple. Il a été l'imagination de la foule, il a été le paradoxe de tous,
la folie quotidienne, cette dose de folie, de furie, de mépris des
choses, d'indifférence, de stoïcisme, d'héroïsme aussi, d'épopée
changeante, de farce multiple qu'il faut chaque jour à un chacun,
pour lui permettre d'être ensuite aussi vide, aussi morne, aussi sage,
aussi pauvre que la veille.
Et, un jour, il est sorti de ses phantasmes pour me tendre la main et
pour me dire des phrases sans magie, des phrases de simplicité où il
me promettait le succès, le triomphe et où il m'annonçait qu'un jour
je mangerais à ma faim. C'était une rue large où je me sentais plus
petit; des voitures roulaient autour de moi pour que je me sentisse
plus à pied, c'étaient des librairies pour que je sentisse que je ne
pouvais pas acheter de livres et des brasseries pour me sentir plus à
jeun.
Il m'offrit deux bocks, des rires sur ma copie—inédite—et du
courage et il s'en fut, sa tâche faite. Je ne le retrouvai que bien plus
tard et il me fut un compagnon aisé, un aîné très paternel.

Il me demande de travailler avec lui, là-bas.

Je sais que nous ne travaillerons pas.
Ce n'est pas le moment. L'été, la mer, sa fonction d'humoriste, ma
peine d'amour, tout nous fera rêver, tout nous fera taire. Nous
resterons de longues heures sans parler, devant la mer et nous
serons tristes, lourdement.
Je m'achemine vers ma tristesse.
«Bonjour, Cahier.
—Bonjour, Maheustre!»
Nous nous serrons les mains, nous sourions, par habitude, nous
souriant moins l'un à l'autre que souriant de la vie, des gens, des
choses, de je ne sais quoi, souriant pour sourire et nous allons tout
de suite voir la mer.
Elle est grise, elle est partout.
Elle vient furieuse jusqu'aux falaises, elle monte, descend, tourne,
s'emprisonne en des quais, en des apparences de canaux,
s'appauvrit, s'amaigrit, s'étrangle.
Nous allons sur une langue de bois, considérer la mer, du bout de la
jetée, du milieu de la mer.
Ce n'est pas la mer qui m'a fiancé, ce n'est pas la mer bleue aux
coulis et aux coulées bleues, gonflée de bleu, qui s'apaisa devant
moi à Monte-Carlo. C'est une mer pâlie, verdie, passée, grondante,
aigre, une mer d'écume et de rage, une mer qui gémit, qui se
balance, qui s'irrite, qui s'excite.
Le regard de mon ami Cahier plonge en elle, s'y perd, je baigne, moi
aussi, ma ferveur dans la mer. Ma ferveur est trouble: je sens en moi
un moutonnement semblable à celui de la mer, une hésitation
sifflante devant la vie, un gémissement, un élan, un désespoir, une
fureur qui écume et qui pleure, qui jaillit, qui recule et qui s'alanguit.
Mon pauvre humoriste, tu t'épouvantes devant la sévérité molle de
la mer et devant sa roideur et je me trouve une âme aussi écrasée,
aussi grouillante, aussi pauvre, aussi hésitante qu'elle, une âme de
désir et d'impuissance, avide et craintive, une âme grise et verdâtre,
excitée, irritée, lente et dormante. J'ai envie de tout parce que je
n'ai envie que d'une femme—et en ai-je envie?
Je l'aime sans plus et je ne sais si je l'aime, je suis lointain, sans
force, sans prise, «sous l'influence», comme on dit en médecine,
captif—et si misérable et si gratuit captif!
Prisonnier de chimères, je t'ai suivi, Cahier, en des cafés aux
plafonds bas, en des cafés où nous avons joué aux cartes avec de
vieilles gens, officiers en retraite ou marins à l'arrière. Mon amour
est venu me bercer entre les cartes et m'étouffer sous les plafonds
bas.
Nous sommes retournés tous les jours sur la mer et tu m'as parlé de
la mort, de tes camarades qui avaient passé le porte-plume ou le
crayon à gauche et qui s'en étaient allés vivre ailleurs. Et la mer,
sans relâche, t'apportait de la tristesse et te la jetait au visage, t'en
souffletait doucement, à petits coups, comme pour te punir de la
gaieté que tu avais infligée aux autres.
Et elle te punissait gentiment parce que tu n'avais pas été gai toi-
même et que ta gaieté était sans grossièreté, nerveuse, hâtive, âpre
et pas convaincue. Et tu t'humiliais délicieusement.
La mer ne m'apportait pas de tristesse: où en eût-elle trouvé pour
moi?
J'étais si triste et si plein d'espoir que j'étais sans pensée, sans
envie, sans espoir, presque pas triste, abruti—en un couloir de
préparation, en une antichambre de fatalité.
Et me voici—je me rappelle et je rêve vite, fuyant la mer et cette
bourgade, tombant à Paris, chancelant en plein amour—tout de
suite.

Car je la rencontrai à la gare, mon aimée—comment? je ne sais pas,

—revenant de je ne sais où, pas de Royan, prenant en mes yeux
désolés, en ma torpeur, en mon ardeur torve tout moi, toutes les
heures que je lui avais consacrées, tous les baisers que je lui avais
gardés, prenant mon cœur, mes lèvres, ma peine et me disant d'un
seul regard qu'elle me comprenait, qu'elle plaignait mon martyre,
qu'elle allait tâcher à me payer, à me récompenser, à me consoler.
Et c'est notre premier baiser, mon baiser timide et son baiser à elle,
en retour, si vite, si gentil qu'il me parut presque traître, qu'il me
surprit, qu'il me fit fermer les yeux, que je n'y crus point. C'est son
abandon en mes bras, c'est sa voix changée, sa voix d'amante et
c'est—ah! mon Dieu! me pardonneras-tu mon bonheur!—le
tutoiement soudain où elle m'enveloppa, dont elle me garrotta, dont
elle m'attacha à soi.
Ah! le tutoiement!
Le mystère du tutoiement! toutes les barrières franchies, brisées,
rayées, tous les voiles arrachés et la facilité de l'existence! Aux
temps où j'étais très solitaire et où je m'accoutumais à Paris et à
l'infortune, je faisais des lieues—à pied naturellement—pour voir une
cousine et une tante et pour avoir quelque chose à tutoyer.
Quelquefois je ne les rencontrais pas et je rentrais avec mes «tu»,
avec ma soif de confidence, ma familiarité et ma fraternité.
Et voici que tu me tutoies, comme dans les idéologies, comme dans
les traités de Platon, les épopées et les drames antiques, voici que
nous nous rajeunissons de ce tutoiement et que nous sommes
devenus pareils aux petits enfants qui s'interrogent sur leurs
nourrices et leurs poupées. Et voici des entrelacs de baisers, voici
une tendresse légère et voici des mélancolies à deux, chaudes,
ambrées, des mélancolies de flamme, tissées d'humanité et de
divinité.
Comment pouvons-nous nous embrasser? Je ne sais pas. Comment
pouvons-nous nous engager nos vies? je ne sais pas.
Personne ne passe par là que notre étoile et Dieu nous sourit de
haut et ne sourit même pas, car il nous respecte en notre amour. Et
voici que mon cœur crève, que mes larmes éclatent et coulent et
qu'elles purifient, qu'elles sanctifient, qu'elles baptisent notre amour!
Ç'a été l'étreinte pour l'étreinte, étroite, dure, haletante, expirante, le
baiser dont on se contente amèrement et qui mord jusqu'au sang,
ç'a été l'éploi de nos virginités, de la mienne, de la tienne qui
revenait pour vibrer et pour s'inquiéter et nous avons été heureux
jusqu'à la souffrance, inclusivement, nous avons été
douloureusement, fièrement amoureux jusqu'à ne pas nous satisfaire
pour rester plus amoureux, pour avoir plus—et autant—à désirer.
Nous avons entretenu le mal de nos corps et de nos âmes, de
baisers naïfs, de baisers à vide, de baisers de promesse et de
tristesse, nous nous sommes usé les yeux à nous regarder dans les
yeux et à chercher en nous des délices prochaines, à considérer en
face notre éternité; nous nous sommes attendris si longtemps, si
pieusement, entre deux portes et nous avons été, dans de l'émotion,
les chers malades qui restent malades précieusement,
incurablement, pieusement—l'un pour l'autre.
Nous avons ouvert une ère, languissamment et ç'a été un
apprentissage de la joie, sans fin.

Sans fin? Non, car il t'a fallu repartir.

Tu n'avais pas épuisé les vacances, les vacances qui vous arrachent
à votre âme pour vous jeter en pâture à des pays, à du vert, à du
ciel, à des wagons, les vacances qui nous font payer cher l'apparente
santé qu'elles octroient et qui t'emmenèrent en Hollande, en Frise,
au cap nord, que sais-je?
Tu n'avais fait à Paris qu'une escale.
Et je voulus, moi aussi, n'avoir fait qu'une escale à Paris, m'y être
arrêté un instant, le temps de m'initier aux pires, aux plus doux
mystères, d'y avoir engagé ma vie, d'y avoir perdu—ou gagné—mon
cœur. Tout dans cette ville—et notre secret n'y avait tenu que si peu
de place—me parlait de toi, de moi, de nous deux, brutalement, de
tout près, et je voulais songer à toi, ne songer qu'à toi, mais
délicatement, timidement, fiévreusement. Je voulais que la
mélancolie dorée de notre extase s'encadrât de l'or de l'automne et
je voulais des bruissements légers autour de mes soupirs—et un ciel
vague et distrait.
Je voulais un exil où rêver, où revivre notre hâtive vie.

Le fatidique Cahier me rappela auprès de lui: ses idées de travail, de

collaboration le reprenaient. J'obéis. Le train matinal qui m'emporta
mal éveillé, cahoté de notre idylle, me berça, me perça de notre
tutoiement: les paysages qui se succédèrent, cette orgie de verdure
ample, pareillement large, touffue, ordonnée et pittoresque me
jetèrent au cœur tes cheveux et tes tu.
Et il me semblait que je me rapprochais de toi.
C'est que je me rapprochais en effet et que cette mer au bord de
laquelle j'allais rêver était la mer au bord de laquelle tu rêvais et
que, plus loin, au ras des flots plus gris peut-être et plus pâles sous
un soleil plus blanc tu me jetais parmi les remous plaintifs en une
bouteille intangible et sacrée tes pensées, tes espoirs et l'armure
blanche de tes caresses, que, fiancée secrète, tu imaginais des
voyages sur cette mer, où je t'aurais rejointe d'avance et où tous
deux mollement, indissolublement enlacés, blêmes d'ardeur et de
fidélité, nous allions chercher le pays des aventures, les palmiers de
repos et ces mystiques forêts vierges où les serpents et les fauves
sont aimants à ceux qui savent aimer. C'est que tu me lançais tes
réflexions, tes remarques, les petits riens de ta conversation et que
tu me lançais toutes tes heures, toutes tes minutes, tous tes loisirs,
tous tes ennuis, que tu me faisais un collier de tes solitudes et que
tu regardais fuir vers moi les barques marchandes aux voiles ternes
qui ne me reconnaîtraient pas, qui ne me diraient rien et qui
viendraient simplement s'amarrer lourdes et béantes et où je lirais
sans maître et sans truchement ton clair regard parmi l'embrun, ton
humide baiser parmi les paquets de mer.
Car tu m'écrivais, mais tu ne m'écrivais que des choses d'amour, tu
ne m'envoyais par la poste que des lyrismes et quels lyrismes sûrs,
parfaits, discrets et sauvages!
Tu ne m'envoyais par la poste que ton idéal, ta passion et ton rêve;
ce n'était pas ta vie, ta pauvre vie et tu voulais tant me l'offrir telle
quelle, mal occupée, hachée, vide, pour m'offrir tout toi, pour ne pas
m'offrir seulement ce que tu n'avais pas, le ciel, les fleurs, ce qu'on
s'offre en amour.
Pour répondre à tes lettres, pour te renvoyer un peu de ton ciel, de
ton univers, de ton au-delà, pour enclore un peu d'infini en une
enveloppe, j'étais obligé de descendre par des rues pointues et
glissantes jusqu'à la poste, j'étais contraint de traverser un marché
aux poissons et je pensais qu'en ta petite ville, là-bas, tu avais une
poste aussi difficile, que ton idéal, avant de se mettre en route vers
moi, devait traverser d'identiques relents, et je te plaignais et je
t'admirais et je découvrais en ces petites épreuves un charme de
plus, un peu humble et câlin comme une tache d'huile.
Lorsque tes lettres me parvenaient, je remontais pour les lire en ma
chambre et j'enfermais à double tour mon exaltation, mon
amertume et mon délice.
Cahier y serait tombé avec un haussement d'épaules. Cet homme
avait noyé l'amour dans les mille tracas de la vie: tendre certes, et
tristement tendre, il avait une tendresse tiède, lourde, irritée, courte,
une tendresse timide, sinueuse, sans férocité: il aurait souri de ma
ferveur furieuse, de ma jeunesse en amour. Vingt fois par jour j'avais
la tentation de lui avouer ma fièvre et vingt fois je me taisais devant
ses yeux gris. Il m'avait fait venir pour que nous fussions deux à être
tristes: il avait besoin du vide de mon âme, besoin que mon âme fût
vide. Hébété d'amour ou hébété par la vie, je lui plaisais.
Et il laissait sa pensée et ma pensée faire des ronds fraternels dans
l'eau et des voyages parallèles à la rencontre des bateaux: il se
reposait de ses longs repos de Paris et demandait de la simplicité à
l'horizon et à l'immensité.

Cependant les ciels continuaient à me parvenir et à se ruer hors de

leur prison cachetée et les éblouissements, les éblouissements pour
moi tout seul s'évadaient, caracolaient, incendiaient. Mon aimée,
sans effort, variait sa sublimité et sa subtilité et c'étaient des
chansons où elle se permettait de dire: je t'aime, en un insouci des
répétitions, chaque fois qu'elle avait à dire: je t'aime,—toujours.
Elle cueillait des anémones de mer et des anémones de ciel, des
algues roses et des algues mauves, des étoiles indociles, du soleil,
de la fraîcheur et un je ne sais quoi des nuits de là-bas, de ses nuits
veuves où le regret se mariait au désir et qu'elle m'envoyait à mon
réveil pour m'endormir dessus, voluptueusement, et pour que je
contraignisse le matin à être encore la nuit, une heure, deux heures,
à faire semblant d'être la nuit, pour étouffer, pour apaiser mon
amour, et pour que mon amour ne fût plus qu'une nuance, profonde,
éternelle, sur la mer et sur le jour.
Les grandes formes qui se lèvent sur la mer et qui peuplent sa
netteté et parfois son brouillard, les grandes formes pures qui
traversent les siècles comme les paquebots traversent le monde, les
grandes formes qui s'endorment et qui veillent sur la mer, entre des
ailes d'albatros et des ailes de mouettes, ces grandes formes qui
sont la poésie et le mal de vivre, qui sont les phares de rêverie, les
phares de l'imprécis et de l'irréel, qui sont des déesses et des
mourantes et qui sont la mer elle-même, d'hier et de demain, ces
grandes formes me paraissaient se relayer jusqu'à toi, se succéder
jusqu'à toi et te porter intact mon songe, intacte la grandeur et la
pureté de mon être en t'évoquant.
Et je t'évoquai sur la mer, souple, penchée, ondulante, je t'évoquai
souriant comme je ne t'avais vu sourire et j'évoquai tes larmes aussi
que je ne t'avais jamais vue verser.
Et tout cela était simple, naturel, si mystérieux que personne ne s'en
doutait, pas même Cahier et que je t'écrivais en ayant l'air de
n'écrire à personne, d'écrire pour le public.

Et me voici quittant cette petite ville qui me fut hospitalière et où je

m'attendris à ma soif. Me voici à Paris, te précédant, en quête d'un
appartement où abriter notre secret. Me voici, solitaire, en des rues
inconnues, longeant des squares, traversant des avenues, trouant
des déserts. Voici des concierges et voici des amis heurtés sans
plaisir, aimables, prévenants, conseilleurs.
Et te voici enfin, te voici délivrée de tous tes séjours, de tes stations,
de tes paysages, te voici chancelante et amoureuse qui t'abat, sur
ma poitrine et qui t'évanouis en moi, si lasse, si lasse de ne m'avoir
pas eu—depuis si longtemps.
Te voici...

Mais voici que tu n'es pas là. Voici que des heures et des heures, les
yeux mi-clos, j'ai commandé au temps, aux souvenirs, que j'ai
groupé autour de moi l'escadron volant du passé. Je n'ai pas mangé.
Je t'ai attendue à jeun et j'ai laissé glisser ce jour sur les jours
d'antan, et je me suis souvenu lentement, comme on prie.
Tu m'as laissé me souvenir et alentir mes souvenirs et me souvenir
péniblement et tu n'es pas entrée au beau milieu. Je me suis
souvenu jusqu'au bout—hélas!
Viendras-tu maintenant?
Il est tard, très tard. La chambre est noire depuis des temps,
pitoyable, un peu dédaigneuse. La lampe qui ne s'est pas allumée et
qui s'épaissit inutile, le fauteuil où tu n'as pas jeté tes vêtements, la
glace qui n'a pas happé ton reflet, la clef que tu n'as pas touchée,
tout est âpre, vindicatif, geignard, tout est famélique et pauvre,
pauvre! Je n'ai pas besoin de savoir l'heure aujourd'hui.
Il est l'heure de fuir et ce n'est pas, après tout, une heure méchante,
puisqu'elle me chasse de ma géhenne.
Je n'ai pas beaucoup souffert.
Je n'ai pas subi cette journée. Puisqu'elle n'a pas voulu être bonne,
elle n'a pas été.
J'ai été le nostalgique prisonnier de mes autres journées, des
journées de genèse, des journées qui s'éclairaient du reflet
grandissant de l'avenir.
Et je m'en vais dans du noir. Je m'en vais sans hâte parce que je n'ai
aujourd'hui aucune hâte, et parce que tu peux arriver encore. Je
m'en vais comme je suis venu. C'est du noir.
Je ne veux pas heurter les meubles. Je suis discret comme un voleur.
J'ai volé cette chambre.
Et je n'ai pas à l'endormir puisque je ne l'ai pas éveillée.
J'ai la tête lourde comme si le passé y était rentré et pesait deux
fois.
Je cours pour me réfugier plus vite en ma vie ancienne, en ma vie
sans splendeur et sans feu, en ma vie du temps où je ne vivais pas.
Je me jette en un omnibus déjà parti, où il y a des gens, n'importe
quoi, n'importe qui.
Je m'écroule sur la banquette, je m'anéantis. Ma tête roule, mon
corps s'effondre, j'étouffe. Je me suis traîné vers de l'air, sur la
plateforme, j'ai ouvert ma bouche agonisante pour respirer un peu
de vie et je sors—oh! en des secondes—de mon engourdissement
chaud de sang, la vie me reprend en me débarrassant des
bandelettes de l'évanouissement et c'est la ténèbre autour de moi, la
ténèbre opaque, qui subsiste, qui s'éternise.
De mon doigt je me suis assuré que mes yeux étaient grands
ouverts—et ils ne voient pas.
Ces mers, ces champs, ces paysages, ces lunes et ces couchers de
soleil, ces soleils et ces longs jours se sont précipités sur mes yeux
et en s'enfuyant, ont emporté mes yeux larme à larme. Mes pleurs
anciens—et j'ai tant pleuré—sont revenus, sont repartis avec mes
yeux. Ou plutôt—pourquoi chercher en mon malheur—c'est ta vision,
ma bien-aimée, c'est ta fugitive et lente vision qui m'a aveuglé—et
c'est de ne t'avoir pas vue que je ne vois plus.
Misérable trompeur de moi-même! Je me cachais mon émotion, je
me contais des contes—mon conte—en sérénité, en confiance: je
trouvais ça très touchant et très amusant.
Et, sous mon épiderme raidi en sa volonté d'indifférence, tout mon
être—secrètement, doucement, pour que je ne m'en aperçusse pas
—tout mon être en sanglots, en révoltes, en désespoirs, se gonflait
et s'en allait à la dérive du fleuve d'amour, s'en allait comme il était
venu—sans baisers.
Et je me croyais calme, résigné!
Je me mourais—sous moi.
Mes yeux ne verront plus: la voiture descend, c'est une rue avec des
lumières et des gens me frôlent et me touchent pour passer, pour
monter: du noir, du noir, du noir que je ne puis même plus trouer de
ta chère silhouette, de tes cheveux,—du noir, un noir total...
Je me rappelle maintenant: c'est le jour des morts; hier ce n'était
que le jour de la mort, aujourd'hui ce sont les morts, un par un,
ceux qui ont un nom, ceux qui n'en ont pas et je suis leur
compagnon, leur prisonnier, un mort qui a des souvenirs, sans
images, un souvenir muet, un souvenir à vide, un souvenir si lointain
qu'on ne peut le saisir. Et que m'importe de voir puisque je ne t'ai
pas vue!
Si, si, il me faut mes yeux pour plus tard, pour te retrouver, pour te
revoir!...

Les yeux me sont revenus, en deux fois. La nuit m'avais repris et m'a
lâché et maintenant timidement, je regarde—pour voir quoi?
Des gens qui s'apitoient, des gens que je n'aurais jamais dû voir—
mes yeux se sont fermés pour les avoir vus, pour les avoir trop vus!
Ah!
Je n'aurais jamais dû voir que toi, chérie, et j'aurais dû garder mes
yeux pour toi, mes pauvres yeux qui voient trop, qui se fatiguent sur
ces gens, en ce soir des morts où je ne t'apercevrai pas, en ce soir
de mort qui agonise si lentement et qui s'épand, qui s'allonge à
l'infini de notre amour et qui l'enferme d'un tombeau mourant et
glissant, d'un tombeau qui grandit, qui grandit devant mes pauvres
yeux, devant mon envie de pleurer, mon désespoir et mon désir.
Comme je t'aime!
V
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com