Why China's Race For AI Dominance Depends On Math

by Michael Auslin

Source Link

THE WORLD first took notice of Beijing’s prowess in artificial intelligence (AI) in late 2017, when BBC reporter John Sudworth, hiding in a remote southwestern city, was located by China’s CCTV system in just seven minutes. At the time, it was a shocking demonstration of power. Today, companies like YITU Technology and Megvii, leaders in facial recognition technology, have compressed those seven minutes into mere seconds. What makes those companies so advanced, and what powers not only China’s surveillance state but also its broader economic development, is not simply its AI capability, but rather the math power underlying it. 

The race for AI supremacy has become perhaps the most visible aspect of the great power competition between America and China. The world’s dominant AI power will have the ability to shape global finance, commerce, telecommunications, warfighting, and computing. President Donald Trump recognized this last February by signing an executive order, the “American AI Initiative,” designed to protect U.S. leadership in key AI technologies. In just a few years, American corporations, universities, think tanks, and the government have devoted hundreds of policy papers and projects to addressing this challenge.

Yet forget about “AI” itself. It’s all about the math, and America is failing to train enough citizens in the right kinds of mathematics to remain dominant.

AI IS not simply a black box that will grow if unlimited funds are poured into it. Dozens of think tank projects and government reports won’t mean anything if Americans can’t maintain mastery over the fundamental mathematics that underpin AI. Calls for billions of dollars in related investments won’t add up without the abstract math ability needed to transform the economy or military.

What we call “AI” is in fact a suite of various algorithms and distinctive developments that draw heavily from advanced mathematics and statistics. Take deep neural networks, which have understandably become a CIO/CTO buzzword, as an example. These are not artificial brains. They are stacks of information-transforming modules that “learn” by repeatedly computing a chain of what are known as gradients (something rarely taught in high school calculus), which are the backbone of a family of algorithms known as backpropagation.

Similar dissections can be made for all of machine learning, which is a study of how to program computers to learn a task rather than execute a rigid pre-coded one. The ability to rapidly classify massive amounts of data, identify patterns, “predict” outcomes, and “self-learn,” all comes down to ever more sophisticated algorithms paired with increasingly powerful computational power and a commensurate amount of data.

From iPhones to Summit—the world’s most powerful supercomputer, located at the Oak Ridge National Lab—and from Google to Facebook, these computing platforms and programs use incredibly complex mathematical calculations to do everything from model nuclear detonations to provide web search results.

And contrary to what some prominent AI advocates—like Kai-fu Lee, author of the AI Superpowers—argue, it’s not simply all about data. Lee is famous for saying that, today, data is the oil of the early twentieth century, and that China, which has the most data, is the new Saudi Arabia. Yet without the right type of math, and those who can creatively develop it, all the data in the world will only take you so far—and certainly not far enough into the future AI advocates boldly envision.

That is why cutting-edge mathematics focuses, among other things, on being able to work with partial information loss and sparse data, or to discard useless information that is collected along with the core data. No matter how you slice it, the world runs on ones and zeros—and on the white boards where the algorithms that manipulate them are thrashed out. Yet one can’t simply jump into creating more powerful and elegant algorithms; it takes years of patient training in ever more complex math.

Unfortunately, American secondary school and university students are not mastering the fundamental math that prepares them to move into the type of advanced fields, such as statistical theory and differential geometry, that makes AI possible. American fifteen-year-olds scored thirty-fifth in math on the OECD’s 2018 Program for International Student Assessment tests—well below the OECD average. Even at the college level, not having mastered the basics needed for rigorous training in abstract problem solving, American students are often mostly taught to memorize algorithms and insert them when needed.

This failure to train students capable of advanced mathematics means fewer and fewer U.S. citizens are moving on to advanced degrees in math and science. In 2017, over 64 percent of PhD candidates and nearly 70 percent of Master’s students in U.S. computer science programs were foreign nationals, and fully half of doctoral degrees in mathematics that year were awarded to non-U.S. citizens, according to the National Science Foundation. Chinese and Indian students account for the bulk of these, in large part because the most advanced training in American universities still outstrips that in their home countries, though the gap is closing with respect to China. Yet that also means that the majority of those being prepared by U.S. universities to open new frontiers in computer science and abstract math are not Americans. Some of these non-citizens will stay here. But many will return home to help grow their countries’ burgeoning tech industries.

There are good reasons to argue that U.S. visa restrictions on skilled workers should be eased, tempting more of those foreign nationals to stay in the United States after their studies have been completed. But the bottom line is that not enough American citizens are choosing to major in advanced math, which has corresponding implications for everything—from foreign competition to Silicon Valley’s startup culture, from national security concerns to whether or not U.S. corporations consider themselves American.

AMERICA’S SELF-INFLICTED math wounds matter because the Chinese Communist Party has made global AI dominance a national goal by 2030, and is leveraging its resources to make it so. Indeed, the world now sees the battle over AI as a battle between China and the United States. Under General Secretary Xi Jinping, China has invested heavily in AI-related technologies, making it a core focus for the modernization of Chinese industry. This effort underpins Beijing’s “Made in China 2025” initiative, which seeks to make the country dominant in most high-tech processes.

China’s AI market is now estimated to be worth around $3.5 billion, and Beijing has set a goal by 2030 of a one trillion yuan AI market ($142 billion). The government has pledged the equivalent of $2.1 billion to build an AI industrial park outside Beijing, among other major investments. Leading the effort is Huawei, which has established AI research laboratories in London and Singapore, unveiled a new generation of “AI processor” chips, and laid out an “all scenario” AI strategy.

Much of China’s spending is directed towards facial and voice recognition technologies like those of Megvii and SenseTime, along with natural language processing. The focus on these particular technologies is purpose-driven: Beijing is using its country’s facility in applied mathematics and AI, whether honed in America or at home, to create a digital surveillance state that is unrivaled in history. For example, a new law requires all individuals registering new mobile phone numbers to have a facial scan. The world’s most advanced algorithms are being used to aid in monitoring and controlling Chinese society and bolster the country’s security services.

Some of this is already plainly visible. Beijing is notoriously creating a “social credit” system based on facial recognition and other technologies that rewards or penalizes certain behavior—jaywalking, credit unworthiness, insufficient patriotism, and the like—so as to shape individuals’ private and public behavior. The two far western provinces of Xinjiang and Tibet have become virtual police states within China, as their Uighur Muslim and Buddhist Tibetan populations are ceaselessly monitored and controlled through the application of facial recognition and forced DNA collection.

China’s AI focus has global security implications as well, given Beijing’s “military-civil fusion” policy which mandates that all high-tech advancements be made available to the Chinese armed forces for incorporation in weapons systems. Just as insidiously, Beijing is reportedly recruiting the country’s smartest high school students to train them as AI weapons scientists. A recent National Science Foundation report noted that Chinese government policies do not share “U.S. values of science ethics,” raising concerns over U.S.-trained Chinese scientists employing advanced research that benefits the CCP ’s surveillance state and military.

Even as China’s AI industry works to catch up to its American counterpart in terms of talent, the country is investing in its mathematical ability. Chinese students ranked number one in the world in math (as well as science and reading) in the latest pisa tests. While there is good reason to doubt the veracity of at least some of the Chinese scores, there is no question that China is focusing heavily on STEM education, outstripping America and European nations. The recently announced “Strong Base Plan” will recruit the country’s top students to study mathematics, as well as physics, chemistry, and biology, among other fields.