Artificial Intelligence and the Global Balance of Power: US vs Chinese Perspectives
Introduction
When the Soviets launched Sputnik in October 1957 it sparked fear and imagination across America about the technological prowess of the USSR and their ability to launch nuclear weapons anywhere on Earth. Throughout the Cold War, the superpowers’ development and subsequent proliferation of nuclear weapons determined the global balance of power. Decades after the dissolution of the USSR, America finds itself in a potential new Cold War with a new rival: China. This time the arms race is not focused on nuclear power but rather on the power of Artificial Intelligence.?
China faced its own “Sputnik moment” in March 2016 when the world watched as DeepMind’s AlphaGo defeated Lee Sedol 4-1 in the ancient game of Go (Kanaan 2020; Lee 2018). As described by Kai-Fu Lee, the Chinese-American tech entrepreneur, over 280 million Chinese viewers tuned into the match (more than double the amount of Americans who watched the Superbowl) and the loss to a Western tech company ignited both a “challenge and an inspiration” across Chinese citizens and government officials alike (Lee 2018). Shortly after the loss, in July 2017, the Chinese State Council released a national strategy for the development of AI entitled, “A New Generation Artificial Intelligence Development Plan.” The plan set clear benchmarks starting in the year 2020 and going to 2030 for China to become the world leader in AI technologies (Webster 2017). The objective as laid out by Xi Jinping himself is to ensure that China occupies the “strategic high ground of core AI technologies.” (Zhou 2018)?
In February 2019, President Trump issued his executive order on AI to compete with the Chinese buildup. Trump wrote that “Continued American leadership in AI is of paramount importance to maintaining the economic and national security of the United States and to shaping the global evolution of AI in a manner consistent with our Nation's values, policies, and priorities.”(Trump 2019) In 2021, the US National Security Commission on AI built upon this EO with a detailed national AI Strategy to safeguard America’s military and technological advantage which it explicitly announced was under threat by China (Schmidt 2021). Both the US and China recognize the strategic value of AI to generate new prosperity and reshape the way of war. As noted in the intro of the NSCAI report, “AI technologies are the most powerful tools in generations for expanding knowledge, increasing prosperity, and enriching the human experience.” (Schmidt 2021) Similarly, the Chinese State Council wrote that “China must accelerate the rapid application of AI, cultivating and expanding AI industries to inject new kinetic energy into China’s economic development.” (Webster 2017)
This paper will examine the evolution of the AI arms race between the US and China and its potential impact on the global balance of power. The paper will focus on the key AI strategies between the two superpowers and how differing norms and values shape consequent AI development. The United States government is focused on building AI systems that are “safe, responsible, and equitable.” (Schmidt 2021) The Chinese leadership, meanwhile, is concentrated on building AI which not only enhances overall economic development but also with a special emphasis on “social governance” and “effectively maintaining social stability.” (Webster 2017) The unique open-source nature of AI development defies traditional state control over the direction of the industry. State competition will therefore likely be limited to talent recruitment, data harvesting, hardware manufacturing, and how these algorithms are applied in practice. Nationalist fervor which calls for restricting AI development as a matter of national security harms the overall AI ecosystem which relies on open-source algorithms and free collaboration between researchers to build AI systems which benefit humanity as a whole.?
Brief History of Artificial Intelligence
John McCarthy at Dartmouth coined the term “Artificial Intelligence” and set up a summer research project in 1956 to build a machine that could imitate human intelligence. McCarthy announced that “an attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.” (Ford 2021 p.71) McCarthy genuinely believed that “significant advance [could] be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer.” (Ford 2021 p. 71) Marvin Minsky, one of McCarthy’s carefully selected attendees pronounced in 1970, “In from three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, have a fight. At that point the machine will begin to educate itself with fantastic speed.” (Ford 2021 p. 73) The burgeoning field of AI would enter what is known as an “AI winter” as these overly optimistic predictions failed to materialize and funding dried up. These scientists fundamentally underestimated just how hard it would be to build such a complex AI system. These scientists aimed to build what is known as “Artificial General Intelligence” which is a system that roughly mirrors human intelligence and is flexible enough to learn any human skill. It has taken decades for advancements in data, computing power, and networking just to produce systems which can execute specific tasks with high enough accuracy to be commercially viable.
Four decades after McCarthy’s summer research project, sufficient advances in computing power allowed IBM’s DeepBlue to defeat Gary Kasparov in the game of chess in 1997. DeepBlue relied on being programmed with every possible move in the game of chess to create what is known as an “expert system” and could simply use brute force to calculate its opponent’s next move. (Greenemeier 2017) IBM again wowed the world in 2011 when it created “Watson” which defeated human champions in the TV game show “Jeopardy!” Watson represented a major advance in the field of AI known as “Natural Language Processing” in which an AI can understand the minutia of human language and return useful answers. (Ford 2021)
The current explosion in AI systems proliferating today is powered by a new approach in AI called “deep neural networks.” The original theory behind neural networks goes back to the work of Cornell psychologist Frank Rosenblatt in the 1950s where he created a “perceptron” - an electronic device that crudely mimicked the firing of neurons in response to information. (Ford 2021) Up until the last decade, AI systems were generally powered by statistical machine learning techniques known as “symbolic reasoning” in which domain knowledge was hand-engineered by subject matter experts, and then computer logic like decision trees was applied to this corpus. The issue was that these systems were very brittle and could not learn anything beyond what was directly fed to them. Deep neural networks by contrast work by allowing an AI to consume vast amounts of data and infer the relationships between the labeled data and its desired outputs. (Domingos 2015; Ford 2021; Kanaan 2020)?
The superiority of deep neural networks was demonstrated in 2012 during the annual ImageNet competition in which a Deep Convolutional Neural Network (CNN) known as “AlexNet” developed by researchers at the University of Toronto soundly defeated its competition. AlexNet was able to correctly label 100,000 test images across 1000 different categories including specific breeds of dog. AlexNet was the first to achieve a sub-25% error rate and beat its nearest competitor by 9.8%. The triumph of AlexNet showed the potential of CNNs to reach human-level performance when powered by the massively parallel computational capacity of Graphic Processing Units (GPUs). (Ford 2021; James Briggs 2022; Korzekwa 2020) The benchmark for human-level performance for image recognition is about 80% while AlexNet was able to achieve 85% accuracy. Just 3 years after Alex Net, Kaiming He and team while working at Microsoft Research Asia in Beijing submitted a 150+ layer CNN which achieved superhuman capacity with an error rate of less than 5%. (Korzekwa 2020) Previous symbolic approaches would have required crafting decision rules for over a thousand different categories, manually telling the computer what a dog’s nose looks like versus a cat, and then repeating such rules for every possible permutation. Manual “feature engineering” collapses when dealing with problems of such a massive scale.?
The Deep Learning revolution has led to a proliferation of tools and algorithms which power everything from Amazon Alexa to Facebook image recognition to language translation from Google. AI powers high-frequency trading algorithms deployed by Wall Street, manages the maintenance of subway systems in Hong Kong, to the autonomous driving features of Tesla. (Ford 2021; KIRILENKO 2017; Hodson 2014) In just over a decade, AI has gone from an academic curiosity to a matter of national strategic advantage. Artificial Intelligence is fundamentally a general-purpose technology and advances in the field do not distinguish between commercial and military applications. The same image recognition algorithms used to power FaceID in iPhones can be applied on a military drone to positively ID targets. Scholars such as Michael Horowitz have described AI as akin to the invention of the internal combustion engine which has gone on to power everything from cars to diesel locomotives to tanks. (Horowitz 2018) The author Martin Ford in his book “Rule of the Robots” has likened it to the “new electricity” which will come to embed itself in every aspect of society and economy. (Ford 2021) While electricity powers our machines, AI has the potential to deliver the intelligence to derive entirely new inventions, designs, and problem-solve in ways that were heretofore impossible for humans. (Ford 2021) As noted by AI pioneer Irvin J Good in 1965, “Thus the first ultraintelligent machine is the last invention that man need ever make.” (Connolly 2015)
AI Arms Race?
The idea of an AI arms race is rooted in the fear that a nation could develop a sufficiently advanced AI which would provide it a decisive strategic advantage both militarily and in terms of overall economic competitiveness. Pedro Domingos, a fellow with the Society of the Advancement of Artificial Intelligence and winner of the highest award in the field of data science, echoed the sentiment of IJ Good in his 2015 book the Master Algorithm. Domingos wrote, “In fact, the Master algorithm is the last thing we’ll ever have to invent because, once we let it loose, it will go on to invent everything else that can be invented.”(Domingos 2015 pg. 25) Such a Master Algorithm, or artificial general intelligence, could invent hypersonic planes, coordinate drone swarms, or devise mutating cyber weapons. It is this potential, and the fear it inspires, which likely led Russian President Vladimir Putin to remark, “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.” (Kanaan 2020) Xi Jinping was filmed with a copy of the “Master Algorithm” on his bookshelf during his 2018 Chinese New Year address suggesting that this book has informed his bullish views on the prospect of AI. (Shead 2018) Xi has stated that “China must develop, control, and use artificial intelligence (AI) to secure the country’s future in the next technological and industrial revolution.” (Zhou 2018)
The real competition in AI between the superpowers is unlikely to revolve around a single algorithm to “rule them all” but rather a plethora of discrete advancements in AI frameworks and applications which enable a wide range of “narrow” AI systems which excel at specific tasks. In 2017, the Pentagon attempted to leverage advances in computer vision to develop AI systems that could automatically identify and categorize targets from drone footage in a project called Maven. The goal was to free up human personnel from the monotonous job of reviewing thousands of hours of sensor footage and allow them to eventually command a fleet of semi-autonomous drones. (Chin 2018; Pellerin 2017) In 2020, DARPA tested “AlphaDogFight” built by Heron AI which defeated the Air Force’s top human pilot in a simulated dog fight 5-0. A key advantage of AlphaDogfight was that it was not limited by engrained training and regulations like a human pilot and could react to its opponent in nanoseconds. (Everstine 2020) At an even higher cognitive level, in 2019, CYBERCOM deployed Project IKE which is designed to be an operational planning aide for cyber commanders. The AI can recommend the best exploits, network paths, and even which cyber mission team is best suited for a particular operation based on their previous missions. (Buchanan and Imbrie 2022; Pomerleau 2021)
Open-source information pertaining to Chinese military AI developments is particularly more limited compared to the stream of press releases from the Pentagon. The 2017 AI development plan called for, “strengthen[ing] the use of AI in military applications that include command decisionmaking, military deductions, and defense equipment.”(Webster 2017; Kania 2020) Research by the Brookings Institute documented that the PLA has invested in unmanned ground vehicles to support logistics, unmanned surface vessels, autonomous submarines, and increasingly advanced Unmanned Combat Aerial Vehicles (UCAVs). PLA Rocket Forces are incorporating AI for remote sensing to automate the targeting process and their ballistic missiles are being upgraded to become more “intelligent” such as determining their own flight path to avoid enemy interceptors. (Kania 2020) PLA Strategic Support Force which handles space operations has deployed a semi-autonomous “killer satellite” that can seek out and knock enemy satellites out of orbit or hurl them into space. (DIA 2022; Harrison 2022)?
? In the first stage of development, military AI systems will be utilized to automate dull dirty tasks such as sifting through drone imagery or transporting supplies to the front lines. However, as these systems advance, they have the potential to rapidly accelerate the sensor to shooter operational tempo beyond what humans can process and react too. (Horowitz 2018) As AI-powered weapons increasingly proliferate on the battlefield from armed drones to intelligent cyber weapons, it may become an outright necessity to deploy countervailing AI defenses in order to remain operationally relevant.?
Measuring an Arms Race
The intangible nature of AI makes it far more difficult to measure progress in any perceived arms race as one cannot simply count warhead yields or dreadnought tonnage as in the arms races of the 20th century. Stanford University’s Human-centered AI program has attempted to develop a systematic ranking of countries' AI capacity through proxy measures such as academic journal output, citations, code repository contributions, venture capital funding, skill listings on LinkedIn, and open job postings. These metrics are focused on academic and economic measures versus those of military application. (Clark and Perrault 2022)?
The United States remains the world leader in AI as measured in terms of total private investment, newly funded AI companies, the total number of AI patents granted, AI code shared in public repositories like GitHub, and overall citations at AI conferences. (Clark and Perrault 2022) Private investment in American AI companies has reached over $50B compared to China’s $15 Billion in 2021. For every AI company which received funding in China, two are funded in the United States. American AI companies were granted nearly 10,000 patents in 2021 compared to only 1,000 in China. (Clark and Perrault 2022)
China is leading the world in terms of the absolute number of AI patent applications, conference publications, and total journal citations. (Clark and Perrault 2022) Data compiled by the Allen Institute for AI shows that China has rapidly advanced in publishing AI articles that rank in the top 10% and 1% of impact as measured by the number of times these papers were subsequently cited by other researchers. The research predicts that China will overtake the United States in the most cited AI academic publications in 2021 for most papers in the top 10% and in 2023 for most papers in the top 1%. (Yang 2021) Chinese advances in AI research have grown exponentially since 2013 with the publication of several national-level policy documents. (Roberts et al. 2021)
The data from Stanford shows that neither the US nor China is leading in terms of aggregate AI talent and skills. India is actually the world leader in the number of open AI jobs, AI talent concentration, and relative skills penetration. McKinsey survey data also shows that as of 2021 Indian companies were the most likely to adopt AI technologies in their business processes followed closely by those in South Korea and Japan. (McKinsey 2021; Clark and Perrault 2022) Stanford talent data was attained by culling data from LinkedIn showing how many people living and working in these countries showcased using AI skills at their current job or had acquired AI-related skills certifications. The United States ranked 3rd for skills penetration, 6th for open AI jobs, and 13th for talent concentration out of the top 22 countries measured. China by contrast ranked 9th for skills penetration, 15th for open AI jobs, and 7th for talent concentration. (Clark and Perrault 2022) The Stanford and McKinsey data shows that the race for AI is not limited to a superpower duopoly but rather is broadly distributed across the world’s most technologically advanced states including Israel, South Korea, India, Singapore, and others.?
Journal citations and new venture capital funding are a poor proxy measure of AI impact because it is fundamentally unknown which of these bets will truly have an outsize impact on the world. A single research paper like “Attention is all you need” which was produced by researchers at Google Brain in 2017 gave rise to an entirely new field of neural networks called transformers which form the basis of state-of-the-art machine translation and other natural language programs. (Vaswani et al. 2017) Ian Goodfellow’s 2014 paper “Generative Adversarial Networks” opened the potential for AI-generated images, video, and art which can fool human observers with their realism; giving rise to the scourge of “DeepFakes.” (Goodfellow et al. 2014) Such groundbreaking research is published openly for the world to use and benefit from and is not owned by any single country to claim the advantage. The open-source nature of the AI industry has not deterred both the United States and China from putting forth national strategies to attain and/or maintain hegemony in this new domain.
The United States and AI
The US has enjoyed a privileged position for decades as home to the most cutting-edge developments in IT and especially AI due to the unique open collaborative R&D ecosystem between private companies, academia, and federal government labs. The National Academy of Sciences wrote for Congress that, “The quality and openness of our research enterprise have been the basis of our global leadership in technological innovation, which has brought enormous advantages to our national interests.”(National Academies of Sciences, Engineering, and Medicine 2022) Major private investments by American tech giants have been instrumental in building the underlying technologies and recruiting top AI talent to build the global AI infrastructure. Google sponsored the work of Dutch researcher, Guido van Rossum, from 2005 to 2012 where he spent half his time developing the Python programming language - the “lingua franca” of AI. (“Guido van Rossum” 2022; Kaye 2022)? Professor Andrew Ng of Stanford co-founded Google’s AI research lab in 2011 and helped invent the AI framework TensorFlow in 2015. Facebook released its own AI framework “PyTorch” in 2016 and together these two frameworks built with Python form a duopoly in the global AI industry. (Kaye 2022)?
These AI frameworks serve as the foundation for subsequent AI application development for the rest of the world - akin to the Android vs iOS duopoly which exists with smartphones today. Unlike the “walled garden” of mobile operating systems, these AI frameworks are open source and available to be used and modified by anyone - even competitors. PyTorch and TensorFlow are openly available across the major cloud competitors: Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP), and can be downloaded locally onto any sufficiently advanced computer. AWS has integrated both TensorFlow and PyTorch as backends to their AI SageMaker service. While Google maintains control of the TensorFlow project, Facebook has transferred control of PyTorch to the Linux open-source consortium. (Kaye 2022) An aspiring AI developer sitting anywhere in the world can use an Apple Macbook Pro running Microsoft’s Visual Studio Code to test AI models in JupyterNotebook using TensorFlow which are then containerized using Docker and run using AWS Elastic Kubernetes Service and committed to GitHub repositories. This interoperability built by researchers across American tech companies and research institutions forms the bedrock of modern software development and makes AI development possible globally. As noted by Kevin Goldsmith, CTO of Anaconda which packages various AI tools for developers, “I don’t think we’d have the kind of machine learning boom we are having without open source. I just don’t think it would have been possible. If this was all proprietary solutions being sold, it never would have happened.” (Kaye 2022)
US AI policies such as EO 13859 “Maintaining American Leadership in Artificial Intelligence,” the National AI Initiative Act of 2020, and the 2021 National Security Commission Report of AI (NSCAI) have sought to strike a balance between maintaining the vibrancy of the American research ecosystem while also safeguarding critical AI technologies from being usurped by foreign competitors like China. (Trump 2019; NAIIO 2020; Schmidt 2021) These laws, policies, and strategies focus on increasing federal investment into AI research to include dedicated AI research centers, supporting the development of the American AI workforce, prioritizing a domestic manufacturing base to produce high-end microchips, and establishing AI regulations that safeguard “civil liberties, privacy, and American values.” (Trump 2019)
National security imperatives to safeguard critical technological advantages have clashed with the open-source ethos of the AI industry. The US Department of Commerce attempted to put open-source software like TensorFlow and Kubernetes used in AI development on the export control list when such technology is already freely available worldwide. Google protested the restrictions in 2019 noting, “These examples point to the fact that the information-sharing ecosystem for AI development is inherently international, with joint development occurring simultaneously across borders, and with a significant open-source culture. U.S. persons in the United States, working for companies with U.S. offices only, do not have a monopoly on such technology.” (Kaye 2022) Techno-nationalist policies which attempt to balkanize AI technology as a key national advantage undermine the advance of the larger industry.?
The Biden administration in October 2022 severely restricted the sale of advanced GPUs necessary to run highly complex machine-learning algorithms in an effort to curtail China’s progress in developing AI. (Cherney 2022) GPUs made by American companies like NVIDIA and AMD are critical hardware components for AI making it possible to train on very large datasets in parallel and saving potentially weeks of compute time as compared to traditional CPUs. (Buchanan and Imbrie 2022) NVIDIA and AMD have a global duopoly in the GPU market with NVIDIA alone representing 80% of the global market share and AMD making up the other 20%. (Mujtaba 2022) Thus the US can use export restrictions to tremendous effect to choke off Chinese access to these components. The issue is that NVIDIA and AMD do not have their own chip fabrication plants and are reliant on Taiwan Semiconductor (TSMC) to physically produce the chips in fabs in Taiwan. In a potential US-China conflict over Taiwan, the US supply of chips for major companies like Apple, AMD, NVIDIA, Qualcomm, and others would be cut off. (Statista 2021) This national vulnerability, highlighted by the supply chain chaos caused by the pandemic, has led to calls to redouble investment in domestic chip production. The Biden administration signed the CHIPS act into law in August 2022 which dedicated $50 billion towards supporting US semiconductor manufacturing in a direct counter to China. (T. W. White House 2022) TSMC and Intel have been in a race to secure subsidies to produce new fabs in Arizona. (Pham 2020) The current restrictions on GPU sales to China are a wasting asset as it only further encourages Chinese competitors like Semiconductor Manufacturing International Corporation (SMIC) to accelerate their own chip fab capacity to compete with their American rivals. China plans to move beyond the need for GPUs entirely by investing in brain-inspired “neuromorphic” processors in which the artificial neurons are embedded directly into microchips rather than simply being simulated with software. (Webster 2017; Seeker 2018)
The United States has a distinct advantage in that its top research universities and tech companies are able to source the best talent from around the world and bring them to America. Andrew Ng, was born in the UK and raised in Singapore, was sponsored by Carnegie Mellon, later went on to achieve a Master's at MIT and his Ph.D. at UC Berkeley and now teaches at Stanford. Yann LeCunn, the inventor of the convolutional neural network, was born in France, but worked at the US Bell Labs, NYU, and most recently as the Chief AI Scientist at Meta (formerly known as Facebook). Geoffry Hinton, a British-Canadian, inventor of backpropagation vital to enabling modern deep learning systems, and mentor to the team which won ImageNet, now works for Google. (Buchanan and Imbrie 2022)?
The American AI industry is heavily dependent on the continual influx of foreign talent to maintain its advantage, especially those from China. A study done by the MarcPolo found that Chinese educated researchers comprised a full third of AI conference publications but the vast majority of them lived in America working for American tech firms and universities. The report found that the US had a wide lead in AI talent, with 60% of the world’s top-tier researchers living and working in America. China is the largest source of this top-tier talent with more than half leaving China to study, live, and work in America. (MarcoPolo 2022) The NSCAI report emphasizes the need to invest in both American students but also to expand the immigrant talent pool. The report calls for expanding the flexibility of H-1B and O-1 extraordinary talent visas while simultaneously investing in greater STEM education for the American K-12 system. The NSCAI advocates for establishing dedicated AI talent pipelines with a new Digital Service Academy and a civilian National Reserve to groom students for future national security-focused AI roles. (Schmidt 2021) The report notes that “The human deficit is the government’s most conspicuous AI deficit and the single greatest inhibitor to buying, building, and fielding AI-enabled technologies for national security purposes.” (Schmidt 2021) The issue for the US government is that while the talent exists it simply cannot match the salaries of the major tech companies and military applications of AI have seriously irked the AI community. Google had to cancel its Maven project with the DOD after its own engineers went on strike because they did not wish to conduct any work which supported the US drone program. (Statt 2018)?
A key aspect of American AI strategic development under both the Trump and Biden administration is ensuring that the systems which are built uphold American values like privacy, civil liberties, and equality. The Trump EO emphasized that “The United States must foster public trust and confidence in AI technologies and protect civil liberties, privacy, and American values in their application in order to fully realize the potential of AI technologies for the American people.” (Trump 2019) Eric Schmidt, former CEO of Google and Chair of the NSCAI Commission noted in his personal letter which accompanied the report, “The AI competition is also values competition.” The strategy which his team produced was a direct repudiation of Chinese AI development which he saw as a tool of repression and domestic control. Schmidt urged that “We must work with fellow democracies and the private sector to build privacy-protecting standards into AI technologies and advance democratic norms to guide AI uses so that democracies can responsibly use AI tools for national security purposes.”(Schmidt 2021) To this end, President Biden unveiled an “AI Bill of Rights” for the American people in October 2022. The Bill of Rights focused on ensuring that Americans would be protected from ineffective systems, algorithmic discrimination, and would safeguard data privacy. The Bill of Rights also encourages that Americans should be informed when AI systems are in use and making decisions and have the right to opt-out for a human alternative if they so choose. (O. White House 2022) The current “Bill of Rights” is non-binding and non-regulatory and cannot be enforced as law. The “Bill of Rights” serves as a blueprint rather than a formal codification of American AI principles which should be used to inform the development of subsequent AI systems both in the government and in private companies. (Butterfield 2022) The United States lags behind the EU in crafting explicit AI regulations like the forthcoming 2024 EU AI act which provides specific legal protections for citizens regarding what type of AI systems can and cannot be deployed within the bloc. (EU 2021)?
AI with Chinese Characteristics
The State Council, the chief administrative authority of the People's Republic of China, announced in July 2017 after seeing the loss of Lee Sedol to a Western AI the formulation of a new national strategy for AI entitled the “New Generation Artificial Intelligence Development Plan” (AIDP). (Webster 2017) The strategy commanded that China, “take the initiative in planning, firmly seize the strategic initiative in the new stage of international competition in AI development, to create new competitive advantage, opening up the development of new space, and effectively protecting national security.” (Webster 2017) This strategy is part of a broader series of national technical policies in which China is working to transform itself to not only compete with Western dominance but to outpace them and create a new world tech order with Chinese characteristics. (Mozur and Myers 2021) The explosive growth of the AI field presents a unique opportunity for China to rapidly catch up and then potentially dominate the field. The Chinese are able to accelerate their AI industry rapidly because the key technologies and theories are open source. While the strategy calls for developing a “first mover” advantage, China has been extremely successful in harnessing (and outright stealing) American and other Western firms' key inventions and then innovating on top of this to produce state-of-the-art results.
Kai-Fu Lee describes the entrepreneurial culture of China’s tech ecosystem as one of “gladiatorial combat” in his book “AI Supowers.” (Lee 2018) Where in America’s Silicon Valley patents and intellectual property are held sacred, there are no such barriers in China. In the Chinese tech world, if it can be built, it can be cloned and thus it produces a ruthless cycle of innovation and competition to stay ahead the copycats and remain relevant in the market. As described by Kai-Fu Lee, “But it was precisely this widespread cloning—the onslaught of thousands of mimicking competitors—that forced companies to innovate. Survival in the internet coliseum required relentlessly iterating products, controlling costs, executing flawlessly, generating positive PR, raising money at exaggerated valuations, and seeking ways to build a robust business “moat” to keep the copycats out.” (Lee 2018) While American firms like Google’s DeepMind or OpenAI may produce the most state-of-the-art systems today, Chinese entrepreneurship based on these technologies will likely proliferate them from the lab to mass market adoption, at least domestically within China.
It is through this domestic entrepreneurship base that China seeks to become a peer competitor in AI with the United States and other countries at the globally advanced level by 2020. According to the AIDP, the first “echelon” will be reached when Chinese AI companies produce $150 billion RMB in value (~$20B). The Minister for Industry and Information Technology has announced that China as of 2022 has already achieved its 2025 benchmark of a 400 Billion RMB (~$50B) valuation for core AI companies. (Times 2022) In context, these market valuations are small relative to Silicon Valley where Uber alone has a market cap of $50B and major tech companies like Google, Amazon, Apple, and Microsoft exceed $1 trillion market caps. The global market for AI is estimated to be approximately $400B+ as of 2022 according to Bloomberg, in which China only makes up about 12% market share. (Bloomberg 2022b)?
The 2025 phase of the plan calls for China to produce “world-leading level” fundamental AI theories and technologies, especially in voice and facial recognition systems. (Webster 2017; Kanaan 2020) China is working to supplant the current American hegemony in AI by producing its own AI frameworks: Baidu’s PaddlePaddle and Huawei’s Mindspore. In a Whitepaper released by the China Academy of Information and Communication Technology in 2022, the authors lament the current American-led duopoly held by TensorFlow and PyTorch. The authors discuss that “the AI framework is the operating system of the smart economy era.” (Ding 2022) As discussed previously, these AI frameworks are the foundation for future AI applications which are built using them. They present a unique soft power advantage for whatever framework becomes globally dominant similar to monopolistic power enjoyed by Microsoft Windows or the Google search engine except that the CCP would have strict oversight and control over how these frameworks are utilized. In the final phase of the plan, China plans to establish absolute supremacy in AI by 2030, which will serve as “an important foundation for becoming a leading innovation style nation and an economic power.” (Webster 2017) Such bullish aspirations are part of the larger “Made in China 2025” strategy in which the key future technologies of the world are designed and made in China from AI to quantum computing to 5G to electric vehicles. (McBride 2019) This is all part of a broader plan for China not only to free itself from dependence on Western technologies but to be in a position to dictate tech standards for the rest of the world and establish a “new tech world order.” (Mozur and Myers 2021)
A new Chinese world tech order is of particular concern to those in liberal democracy given the authoritarian nature of the CCP. Eric Schmidt writes that “China’s domestic use of AI is a chilling precedent for anyone around the world who cherishes individual liberty. Its employment of AI as a tool of repression and surveillance—at home and, increasingly, abroad—is a powerful counterpoint to how we believe AI should be used.” (Schmidt 2021) The AIDP states that one of the major principles of developing AI in China is to augment “social governance” and that it should “play an irreplaceable role in effectively maintaining social stability.” Since 2015, the CCP has been developing a “social credit” system in order to monitor the “social trustworthiness” of its 1.4 billion people. (Kanaan 2020) The system uses AI embedded into mobile apps like WeChat to monitor social media posts, private messages, phone calls, and financial transactions in order to construct a “social credit score” which is effectively a quantized measure of how loyal a citizen you are to the party. A Chinese citizen with a low social credit score determined by these algorithms can be barred from public transportation, loans, housing, and job applications. (Kanaan 2020; Udemans 2018) The official aim of the program is to, “allow the trustworthy to roam everywhere under heaven while making it hard for the discredited to take a single step.” (Kanaan 2020 p. 183) Such mass surveillance was heavily used during China’s COVID-19 lockdowns where simply purchasing fever medication would trigger mandatory COVID testing and could land you in quarantine. (Bloomberg 2022a)
The use of AI-powered surveillance for domestic control has been taken to an extreme in the Chinese province of Xinjiang, home to the ethnic Uighur minority who are predominantly Muslim. The CCP has engaged in a campaign of ethnic cleansing against the Uyghur people in the name of “counter-terrorism” operations. The CCP is using a national network of AI-powered surveillance cameras to identify and track the unique ethnic features of Uighurs across China in an explicit campaign of racial profiling. (Kanaan 2020) If such cameras catch an Uyghur engaging in a proscribed activity such as reading the Quran or attending worship at a mosque, it could automatically trigger their detention and expulsion to “reeducation camps” (read: concentration camps). (Ford 2021; Kanaan 2020)? Research from the Wilson Center has shown that the Chinese government has built a global surveillance network to track and intimidate Uyghurs. (Jardine 2022) Through the use of its “Smart City” AI-powered surveillance systems exported abroad, Chinese intelligence services could extend their reach globally to track Uyghurs and other dissidents. (Ford 2021; Kanaan 2020) The Chinese have exported such surveillance systems to Ecuador, Zimbabwe, Pakistan, Kenya, Venezuela, Angola, and the UAE. (Kanaan 2020) China is not just exporting AI-powered surveillance systems but it is exporting an authoritarian model of AI explicitly designed to empower oppressive regimes to maintain control of their populace.?
Balance of Power in AI
A balance of power in AI is achieved through transnational research collaboration and the open-source nature of the field which creates a level playing field for anyone to contribute. This open-source ethos seriously erodes the concept of a "strategic high ground" when anyone can have access to massive cloud computing resources and state-of-the-art algorithms. Where there may be room for competition is the acquisition of global AI talent and how they utilize the available data - which is the fuel of deep learning systems. China has a unique advantage here in terms of the widespread adoption of smartphones by its populace and laissez-faire data privacy regulations. This allows Chinese AI companies to train with massive datasets and thus achieve very high levels of accuracy in their models. For example, Didi, China’s Uber competitor, processes 9 billion routes a day and 1,000 ride requests every second. (Li, Tong, and Xiao 2021) This creates a positive feedback loop whereby the company that has the best algorithm is able to attract the most users, more users mean more data, which further refines and augments the AI algorithm, leading to a better user experience and even more users. Despite this seeming advantage, China has yet to produce world-dominating AI solutions to compete with its Western counterparts - with the distinct exception of TikTok produced by ByteDance. TikTok does not represent any groundbreaking AI invention. Its algorithms are extremely similar to those employed by Facebook, Twitter, and Instagram to predict what content users will want to interact with. TikTok's explosive growth, and its threat to the dominant position of Facebook, are based on the simplicity of its user experience. China’s path to potential AI dominance will most likely lie in Chinese entrepreneurs finding ways to embed AI into every conceivable service and product and then outcompeting entrenched Western competitors after establishing themselves domestically.
America’s distinct advantage is being able to win over the top talent from around the world - and poaching it from China. As aforementioned, 60% of the world’s top-tier AI researchers live and work in America. (MarcoPolo 2022) Data from Georgetown CSET shows that 90% of Chinese AI students who completed their doctoral studies in the US decided to stay for at least 5 years after graduation - often working for US tech companies and research institutes. (Mozur and Metz 2020) These top researchers go on to produce the leading breakthroughs which advance the entire field from convolutional neural networks to the GANs to reinforcement learning systems like AlphaGo. The key determinant in a potential AI cold war with China may come down to the intellectual firepower that each country is able to amass. As Jack Clark, policy director of OpenAI noted, “For much of basic A.I. research, the key ingredient in progress is people rather than algorithms. There’s a lot of open-source technology lying around for researchers to use, but relatively few researchers with the sorts of long-term idiosyncratic agendas that yield field-changing advances.” (Mozur and Metz 2020) The Trump administration’s crackdown on Chinese researchers as potential spies seriously harms this otherwise robust talent pipeline. Lisa Li, a Chinese engineer who graduated from Johns Hopkins remarked, “Sacrificing international students is killing the goose that lays the golden egg. It will eventually destroy the future competitiveness of America.” (Mozur and Metz 2020) If the US administration continues to put further restrictions on the immigration of Chinese researchers out of national security concerns, it can be quite certain that Beijing will welcome them back with open arms.?
This “arms race” is less about who will build the first super-empowered AI in a 21st-century redux of the Manhattan project but rather a broader long-term ideological competition. This new AI “Cold War” between the US and China is about whether we will see AI which safeguards privacy and civil liberties or will we see a world where authoritarian regimes are augmented by the power of Chinese-built AI. The historian Yuval Noah Harari is pessimistic about the prospect of AI and argues that technology favors tyranny as it allows a singular elite to control the masses like never before with AI-generated propaganda and unrelenting surveillance. (Harari 2018) What is likely to occur is a broader decoupling between the US and China in technology standards and values creating distinct technological spheres of influence - determined by which brand of AI and 5G networks countries choose to align themselves with. (Inkster 2020) This competition will play out most aggressively in the developing world in places like Latin America, Africa, and Southeast Asia as countries in these regions make strategic choices in how to modernize their economies.?
Conclusions
The true value of Artificial Intelligence is that it is a general-purpose technology - quite possibly the “electricity” of the 21st century. Human mastery of electricity spawned not only modern illumination and power for machines; but also gave rise to the telephone, semiconductors, computers, and the Internet. Where electricity powers machines, AI will power a whole new series of innovations and inventions in every field. We as humans can use AI to solve monumental challenges - from protein folding to designing better spacecraft to writing computer programs automatically. The capacity to think, plan, design, and invent at machine speed - and to do so in ways that no human ever would- presents a significant potential strategic advantage and a distinct challenge to the current balance of power. The US and China are locked in an arms race to monopolize the new commanding heights of the global knowledge economy; building up their tech giants and poaching talent.? But AI is an extremely broad field of technologies - it is not owned by any one nation or company. The algorithms, notebooks, and data are freely available to the world to learn from, modify, and contribute too. Invention in this field can come from anyone with sufficient training and a broad enough imagination. This new age of AI will give rise to inventions which are both the stuff of dreams - and nightmares. The great challenge here is not that some rival power will master it first, but rather, that there are too few among us that understand and can tame this new fire; leaving its power and potential concentrated amongst only an elite few who utilize it to optimize for profit rather than optimizing for humanity.?
领英推荐
Works Cited