AI Superpowers - critical summary review

For the past few decades, the world of artificial intelligence (AI) – and, as a result, our modern world, as well – has been shaped by a few trailblazing American companies. In his provocative debut, “AI Superpowers,” Taiwan-born AI pioneer and China expert Kai-Fu Lee argues that China could become the center of global innovation in AI very soon thanks to its demographics and unique sociopolitical environment. Get ready for the details.

A brief history of deep learning: rule-based AI vs. neural networks

After World War II, a few scientists –  now described as the pioneers of AI – set themselves “an impossibly lofty but well-defined mission: to recreate human intelligence in a machine.” However, they couldn’t agree on the most efficient way to set about this mission, so, in the mid-1950s, the field of AI had divided researchers into two camps: supporters of the “rule-based” approach and advocates of “neural networks.”

In rule-based AI (sometimes called “symbolic systems” or “expert systems”), computers are “taught” to think through scripts that encode a series of logical rules – e.g., if X, then Y. In the case of simple and well-defined games (“toy problems”), rule-based AI works well. For example, to teach a computer what a cat is, you just need to write a script with a few simple rules, such as – “If there are two triangular shapes on top of a circular shape, then there is probably a cat in the picture.” However, the more complicated the situation – that is, the more choices or moves it requires – the more difficult it is to clearly outline the behavior of the computer. As a rule of thumb, the more human programmers, the better these rule-based AI systems are.

Inspired by the architecture of the human brain, and aiming to reconstruct it, neural networks researchers took an altogether different approach. Namely, instead of programming an existing computer to think through rules, they tried to build systems (neural networks) that could perform tasks by considering previous models. For example, to teach a neural network what a cat is, they would feed the system millions of sample photos manually labeled “cats” and “no cats” and let the program itself figure out the necessary pattern of traits. As opposed to rule-based AI, in neural networks, the less human interference there is, the better the final result.

The first neural networks showed some promising results during the 1960s, but, in 1969, an article by AI pioneers Marvin Minsky and Seymour Papert demonstrated their shortcomings and convinced the world that they were “unreliable and limited in their use.” However, in the mid-2000s, British Canadian computer scientist Geoffrey Hinton – “The Godfather of Deep Learning” – discovered a way to efficiently train more layers of neural networks, effectively multiplying their power capacity several times over. Almost overnight, in the form of deep learning, neural networks hit the mainstream, and the world began to dream of robots who could talk, decipher human speech, predict consumer behavior, and even drive cars or beat humans in complex games.

A view from Beijing: AlphaGo and China’s Sputnik moment

By the 1990s, rule-based researchers were already able to devise AI programs capable of beating grandmasters at chess. In 1997, a rule-based chess-playing computer named Deep Blue did the previously unthinkable and became the first AI system to defeat a reigning world champion, Garry Kasparov. However, in the opinion of many, this was as far as rule-based AI could go. 

Games more complex than chess – such as the Chinese abstract strategy board game called Go (also known as Weiqi and Baduk) – seemed prohibitively difficult for this more conservative AI method. Go seemed out of reach for deep-learning neural networks as well – but not anymore after Hinton’s breakthroughs. 

In 2015, DeepMind – a British AI company that had been just acquired by Google – developed a computer Go program called AlphaGo. At the end of that year, AlphaGo beat the European Go champion, Fan Hui. In 2016, it beat Lee Sedol, one of the highest-ranked players in the world. Then, in May 2017, AlphaGo won a three-game marathon match with Ke Jie, an exceptional Chinese Go player who at the time held the world No. 1 ranking continuously for two years by then. 

Since Go is not such a popular game in the West, most of these breakthroughs remained unnoticed by American and European media. When mentioned, it was simultaneously with things like “American dominance” and “Western technology.” AlphaGo’s achievements were not only the latest proof that the United States had won the technological war against the rest of the world, but also one of the first poised to continue this dominance into the age of AI. Routinely dismissed because of its copycatting practices, China had a different idea. 

When in 1957, the Soviet Union launched Sputnik – the first man-made satellite into orbit – almost overnight, the event profoundly altered the American psyche and the U.S. government policy. AlphaGo’s victories had the same effect on Beijing. Ignored by the West, the program’s victory over Sedol was watched by 280 million Chinese viewers, triggering an AI fever. Less than two months after Jie’s losses, “the Chinese central government issued an ambitious plan to build artificial intelligence capabilities,” and become the center of global innovation in AI by 2030.

A tale of two countries, five differences and seven giants

Just a few years ago, China’s attempt to become the undisputed leader of the AI world in the following decade could be scoffed with a laugh and a lecture. After all, Chinese companies were mere copycats, their finest creations being reverse-engineered Silicon Valley products. 

By definition, this meant they were always lagging one step behind the United States. It changed significantly, however, during the past few years, and, in Lee’s opinion, the reason why few are slow to wake up to this fact is the West’s “outdated assumptions about the Chinese technology environment.” In reality, China already has the advantage – because of at least five differences between the functioning of the booming AI sector at Silicon Valley and China.

Difference No. 1. Private vs. government-backed innovation

The reason why the United States has been miles ahead of other countries in so many different spheres of life is laissez-faire capitalism. No other economic system produces a more effective cycle of personal freedom and healthy competition – the two prerequisites for innovative entrepreneurship. When it comes to collecting data, however, and implementing AI, the freedom-based economic mechanism and fierce combative nature also are its biggest problems.

Lee claims that capitalism aggressively punishes missteps or waste in funding technological upgrades. China’s centralized communist economy has the luxury of implementing a different techno-utilitarian approach. Being the country’s most important economic actor, the Chinese government is free to support innovation as it thinks fit – rewarding proactive investment and adoption, and even working closely with private companies. Setting aside questions of moral superiority, China’s system is far more suited for the implementation of AI.

Difference No. 2. Innovation vs. implementation

Even though newspapers would have you believe that we are still living in the age of discovery, the truth is that almost all of the milestones in AI of the past few years are the result of applying past decades’ breakthroughs to new problems. Soon after Hinton’s revolutionary improvements in neural networks, humanity left the age of innovation and entered a new age – implementation. Implementation is what China has always done better than the rest of the world.

Difference No. 3. Expertise vs. data

Innovation may require geniuses and freedom – but implementation requires large amounts of data, centralized data-gathering systems, and brute force; the United States lacks all three. China seems as built to thrive in an age of data. Because of a few government interventions and a market where reverse-engineering is allowed, grunt work is what differentiates successful companies from failed startups, and China has become the Saudi Arabia of data. “By immersing themselves in the messy details of food delivery, car repairs, shared bikes, and purchases at the corner store,” in just a few years, Chinese companies have turned the country into the world’s largest producer of digital data. The gap between China and the United States is widening by the day. 

Difference No. 4. Mobile-first internet users and the O2O revolution

Unlike all the American would-be Bezoses and Jobses who dream of becoming billionaires without leaving their San Francisco loft, the unique Chinese business environment forces companies to “get their hands dirty in the real world.” As a consequence, the data they collect is not only quantitatively superior to the data collected by Silicon Valley companies, but it is also qualitatively more meaningful. 

Google, Facebook, and Amazon may follow what you do online, but Chinese AI startups know what you do offline as well. Once mobile-first internet users and mobile payments became a thing, the Chinese government encouraged this behavior among the population, and an online-to-offline (O2O) startup revolution began. Originally a Facebook Messenger copycat, Tencent WeChat has grown to become a national superapp, “a kind of digital Swiss Army knife for modern life.” There’s almost nothing a user can’t do with WeChat, from messaging and localization through internet searching and scanning QR codes to booking taxis and paying. 

Difference No. 5. Light touch vs. heavyweights

As strange as it may sound, WeChat is not an exception: many Chinese AI companies are essentially heavyweights, and most startups strive to become supercompanies as early as their mission statements. It’s different in Silicon Valley. American internet companies tend to take a “light approach” and stick to their core strength: information platforms. They come up with elegant solutions to computer problems and leave “the brick-and-mortar businesses handle the on-the-ground logistics.” 

On the other hand, Chinese companies go heavy: “they don’t want to just build the platform – they want to recruit each seller, handle the goods, run the delivery team, supply the scooters, repair those scooters, and control the payment.” Two comparisons clearly illustrate the point. 

First, there’s Uber, a peer-to-peer ridesharing company - its far more successful Chinese rival DiDi owns a large fleet of vehicles, gas stations, and auto repair shops. Similarly, while Airbnb is largely an internet platform where you can list your home, Tujia manages a large chunk of rental properties and takes care of all the grunt work necessary to prepare an apartment for the next visitor.

The new kids on the block

Until recently, four American companies dominated the AI market: Google, Facebook, Amazon, and Microsoft. However, Baidu, Alibaba, and Tencent have virtually driven out all of them from China and have earned a head start in several developing markets. More importantly, they operate closely with the government in the most populous country in the world – and among people quite used to sacrifice privacy for convenience. Will this make them household names outside China soon? Only time will tell.

Final Notes

Provocative and immensely readable, “AI Superpowers” is both a riveting narrative and an eye-opening analysis of the present and future of AI. 

A must-read – and not only for tech enthusiasts.

12min Tip

A provocative takeaway instead of a tip: the United States may be the global AI leader today, but China is perfectly positioned to dominate the AI era of tomorrow.