6 February 2019

WHY CHINA CAN DO AI MORE QUICKLY AND EFFECTIVELY THAN THE US


WHEN ENRICO FERMI decided to leave Benito Mussolini’s Italy and emigrate to the United States, he changed the global balance of power. After arriving in the US, Fermi led the world’s first self-sustaining nuclear reaction at the University of Chicago and played an indispensable role in the Manhattan Project, which led to the end of World War II in the Pacific and laid the groundwork for a new world order and America’s prominent role.

So it is not surprising that some Americans think the same should be true with AI. Emigrant AI researchers like Geoff Hinton, Yann LeCun, Yoshua Bengio, Andrew Ng, and Fei-Fei Li are the Enrico Fermis of AI and should secure an American (and Canadian) hegemony in AI. Indeed, the US and Canada have 100 percent of the top 10 AI researchers, and 68 percent of the world’s best 1,000 or so researchers.


But technological revolutions are not only driven by big discoveries. Often, once a fundamental breakthrough has been published, the center of gravity quickly shifts from a handful of elite researchers to an army of tinkerers—engineers with enough expertise to apply the technology to different real-world problems and customer needs. This is particularly true when the payoff of a breakthrough is diffused throughout society rather than just concentrated in a few labs.

Mass electrification exemplified this process. Following Thomas Edison’s harnessing of electricity, the field rapidly shifted from invention to implementation. Thousands of engineers everywhere began tinkering with electricity, using it to power new devices and reorganize industrial processes. Those tinkerers didn’t have to break fundamentally new ground like Edison. They just had to know enough about how electricity worked to turn its power into useful and profitable machines.

Our present-day phase of AI implementation fits this latter model. But you might say: A constant stream of headlines about AI breakthroughs (AlphaGo, Stanford beating doctors in cancer diagnosis, Microsoft beating human in speech recognition, etc.) shows that we are still in an age of discovery. In reality, we are witnessing the application of the same fundamental breakthrough—deep learning and related techniques—to many different problems. That’s a process that requires well-trained AI engineers, the tinkerers of this age, to apply and tweak deep learning for each domain. Today, those tinkerers are applying AI’s superhuman powers of pattern recognition to making loans, driving cars, translating text, and powering our Amazon Go and Amazon Alexa.

Deep-learning pioneers like Geoffrey Hinton, Yann LeCun, and Yoshua Bengio—the Enrico Fermis of AI—continue to push the boundaries of artificial intelligence. And they may yet produce another game-changing breakthrough, one that scrambles the global technological pecking order. But in the meantime, the real action today is with the tinkerers, those who implement AI and make it solve real-world problems.

And this is where China comes in—while the US is the world’s leader in AI discoveries, China is actually the leader in AI implementation. What matters in AI implementation is speed, execution, product quality, data, and government support. Chinese companies are equal to or ahead of their American counterparts in each of these areas.

First, let’s look at speed. The Chinese environment is a combination of a huge market, ample capital, and daring entrepreneurs who are the embodiment of “lean startup”: fail fast, fail early, and fail often. Chinese entrepreneurs are fast to find market opportunities, build products, and pivot when needed.

For example, after companies like Uber and Didi proved the viability of ride-sharing, China’s startup world caught sharing fever, trying out every possible iteration of it: shared bicycles, shared mopeds, shared concrete mixers, and shared mobile phone chargers. The vast majority of these died off quickly (known as fail fast), but a few rose to become unicorns, such as Mobike, which reached 20 million rides per day and was sold for $2.7 billion three years from founding.

This rapid prototyping and tweaking approach have also enabled Chinese AI companies to find winners. For example, Megvii (Face++) is a computer vision company that originally tried face morphing games and face unlock of phones before striking gold on finance fraud avoidance as the killer app (think of credit card companies using face scan instead of customer service “quizzes” on suspicious credit card usage). As mobile payments ballooned to $18.8 trillion in China, this turned out to be the winner. In another example, AInnovation started on AI sales forecast but added hardware products like computer-vision driven vending machines and whole-basket check-out kiosks, all within six months. There are many other examples like this in my book, AI Superpowers.

In execution, Chinese entrepreneurs are unafraid of the tedious, messy, and risky tasks, if they help achieve the ultimate result. Chinese CEOs usually have absolute power within the company, which makes execution much more effective.

As an example, the “Chinese Groupon” Meituan was tenacious in focusing on users’ needs and found a way to nail that problem. They found that people wanted to eat take-out food, but it had to be delivered with 30 minutes (including cooking time), and the cost must be brought down to about 70 cents per delivery in order to profitably offer “free delivery.” Meituan then maniacally worked on this problem for years, eventually hiring 600,000 people on mopeds, adopting an Uber model, finding riders willing to work during meal hours, tweaking the lowest-cost delivery vehicle (battery-operated mopeds), solving the battery life problem for mopeds, inventing AI matching and routing algorithms. After billions of dollars of losses and many years of iterations, Meituan managed deliver food to any destination within 30 minutes and under 70 cents. That totally changed the way Chinese people eat. And that differentiated it from the US Groupon, Yelp, and OpenTable, which collectively are worth less than one-tenth of the $60 billion valuation.

Meituan chose this approach rather than resting on its laurels like Groupon and Yelp, because Chinese entrepreneurs are tenacious and if you have a profitable “light-weight” business, you will find yourself surrounded by entrepreneurs who want a share of your profits.
Fierce competition pushes entrepreneurs to improve the product at lightning speed, with incredible work ethic, always pressured to develop impregnable business models. As a result, Chinese products often evolved into better products than their American counterparts (e.g., Wechat vs. Whatsapp, Weibo vs. Twitter, Taobao vs. eBay).

The Chinese market rapidly embraces new products and new paradigms. Just within the last 3 years, mobile payments have emerged as the dominant transaction tool, replacing cash and credit cards. Total transaction in 2017 was $18.8 trillion, even larger than China’s GDP. China’s mobile payments are built on the world’s best infrastructure: nearly zero-transaction-fee, micropayment-capable, and peer-to-peer. Over 700 million Chinese users can pay each other, whether for online, offline, loan, or gift, whether to your child, a farmer in a village, or even a beggar.

All of this is amplified by China’s enormous market size, which generates the treasure trove of data that fuels AI (AI is usually more improved by more data than better AI engineers). China’s data edge is three times the US based on mobile user ratio, 10 times the US in food delivery, 50 times in mobile payment, and 300 times in shared bicycle rides. All this rich data is used to make Chinese companies’ AI work better.

Finally, the Chinese government’s support of AI development will prove important in the age of AI, though not in the way most Western analysts believe. Western narratives trivialize the government’s role as subsidizing winners and protecting them from foreign competition. But actually, sophisticated government support comes in three forms, which are neither blind nor anti-competitive: (1) Central government sets the tone, which can legitimize a burgeoning industry like AI and influence companies and consumers to adopt AI (and smart young people to go into AI); (2) techno-utilitarian policy, which allows unproven technology to be launched early and quickly and adds regulation only if necessary later (enabling a nation-wide displacement of cash and credit cards with mobile payment); (3) infrastructure-building, such as rebuilding cities and highways with special roads with sensors and special lanes or “levels” for autonomous vehicles (contrasted with the US decision to slow down autonomous truck testing due to job displacement concerns).

As a result of all of the above, China has the world’s most valuable companies in computer vision, drones, speech recognition, speech synthesis, and machine translation. Sinovation Ventures, my VC company, now has five AI unicorns, with a combined valued of $23B. These companies were founded only two to four years ago.

The speed, execution, product focus, access to data, and government support are significantly higher than their American counterparts. So can we conclude that China is the de facto winner of the AI race? No, because researchers worldwide are still plunging ahead. Not long ago, Geoff Hinton made a call for top researchers to abandon deep learning and develop brand-new machine learning algorithms that can come closer to human intelligence.

So what can this balance of strengths and weaknesses tell us about international leadership in AI? Here, it helps to zoom in on perhaps the most coveted of all AI-powered products: fully autonomous vehicles. Companies in both countries are feverishly chasing the dream of mass deployment of cars that drive themselves. They’ve made great strides toward this goal, but it still remains an open question which company or country will get there first.

That question gets right to the crux of the US-China dichotomy between visionary research and practical implementation. Waymo (an Alphabet company) jumped out to a major lead in autonomous vehicles because it was willing to think big and take risks.

Once Waymo demonstrated that this was possible (in large part due to breakthroughs like deep learning), companies in both the United States and China started playing catch-up. At the same time, local Chinese officials began competing with each other to attract autonomous-vehicle startups, pledging to build entirely new public infrastructure projects to facilitate deployment: highways lined with sensors that communicate with vehicles and elevated roads reserved for training autonomous vehicles.

The winner in this race will likely depend on whether the final bottleneck is about core technology or implementation details. If the bottleneck is technical—major improvements for core algorithms—then advantage US. If the bottleneck is about implementation—smart infrastructure or policy adaptation—then advantage China.

What should the US do in light of the rising Chinese AI capabilities? I have three suggestions. First, move away from trivializing the Chinese approach (e.g., copycat, government protectionism), recognize that the Chinese approach has merit, and be open to learning from it. Second, double down on fundamental research, where the US has a huge lead over the rest of the world. American universities attract top students around the world, who learn AI in the US, and many will choose to stay. Third and most important, recognize that AI development is not the new cold war. AI is more like electricity than nuclear weapons. US and China have much to learn from each other. And the opportunities and challenges from AI are much larger than any threat or competition from any one country to another.

No comments: