AI Superpowers — Kai-fu Lee

Reading Notes

Finished reading this book during the trip to the Great Smoky Mountains. Overall speaking:

Easy to read: 4.5. This book is easy to read through during flight or fragment time.

Content: 4. The content is good and relatively concise.

Short Cap

This book introduces the concept of machine learning/deep learning and how it surges in popularity in the US and in China at the beginning. It latter spends a large portion of the book talking about the difference of running an Internet company in the US and China, the reason why China is catching up in an unprecedented speed and what are the advantages of each country.

After talking about the roles of the US and China in leading AI for the next 5–10 years. Kai-fu switched gear to talk about non country related and more generic information of AI development, including the trends of AI, the crisis and opportunity of AI.

In the last couple chapters of the book, the author shares his experience fighting cancer and his learning after the cancer attack, and his thoughts on how we, human beings, would live with AI in the future. He gave a couple approaches post AI flourish, like UBI, social investment stipend, etc..

Long cap

1. China’s Sputnik Moment

Kai-fu argues AI development has entered the age of implementation. After years of research into deep neural networks and the explosion of computing power, we have entered an age of implementation for AI algorithms, another breakthrough in AI would not come in next 1–2 decades. In the age of implementation, the importance of elite researchers has dropped and instead, companies or countries who own a huge amount of good researchers and data would succeed faster, and both are China’s advantages. China’s top down approach in pushing AI has been another advantage in speed up AI development and will remain as an advantage for the future.

2. Copycats in the Coliseum

The story starts with introducing Wang Xing, a Chinese entrepreneur who founded Xiaonei (knockoff of facebook), Fanfou (knockoff of Twitter) and Meituan (knockoff of Groupon). The last one, Meituan, even though it started with the idea of mimicking Groupon’s idea, it later expanded to support traveling, food delivery, recommendation and ride hailing. Meituan values at $30 Billion while Groupon’s market cap shriveled to $2.6 Billion.

Kai-fu compared the startups in both the US and China in depth. Culture wise, the U.S. entrepreneurs live in an environment of abundance and encouraging innovation and individualism. Most startups in the U.S are mission driven and detached from financial motivations. In stark contrast, startups in China are market driven — whatever makes them survive and thrive. A cultural acceptance of copying, the scarcity mentality due to the past poverty, and the willingness to dive into any promising new industry form the psychological foundations of China’s internet ecosystem.

The copying process is crude and embarrassing sometimes, but it nurtures a string of forces understanding user interface design, website architecture and backend software development. The market driven entrepreneurs are also forced to put user satisfaction at first and iterate product as soon as they can to fit and compete in the market.

Why silicon valley giants fail in China? Like eBay, Google, Uber, Airbnb, LinkedIn, Amazon, etc. Some blames Chinese government’s protectionism, but that is not true. The juggernaut in the valley approaches China market with marketing mindset instead of a building & tailoring mindset. A resistance to drastic localization greatly slows down the iteration speed. Talent wise, the valley also lose out on top latent as local talents jams into local startups, instead of joining clunky foreign companies in China.

The next story Kai-fu used to incarnate China’s startup culture is Qihoo VS Tencent. Ever since their products step on each other, each started to prompt users to uninstall opponent’s software. It later escalated to a forced choice for the users, you could only install one company’s software. Tencent eventually launched a personal attack on Qihoo and its founder via police.

The methodology Chinese tech entrepreneurs believe in is from The Lean Startup — the founders don’t know what product the market needs, the market knows what product the market needs. Instead of spending years and millions of dollars creating the idea of a perfect product, startups should move quickly to release a minimum viable product that can tease out market demand for different functions. Companies in the valley are sometimes bounded by its mission statement and struggle to expand, Chinese startups have no such dilemma, they follow trend.

3. China’s Alternate Internet Universe

China has been developing its alternate internet universe with government support, a Saudi Arabia of data. China’s massive number of internet users (larger than the US + All of Europe combined) easily gives China a data advantage. Premier Li Keqiang repeated a phrase during 2014 World economic Forum’s Summer Davos “mass entrepreneurship and mass innovation”. This is a top down push and since promotions for local officials in China’s bureaucracy is based on performance evaluations conducted by higher-ups within the Communist party’s internal human resources department. When central government sets a clear metric for lower level officials to demonstrate their competence, local officials would throw them into advancing the goal and proving themselves capable at best.

Compared with the companies in the valley, Chinese companies are leaning towards heavy weighted ones, to build the wall against brutal competitors and reduce cost.

4. A Tale of Two Countries

Once a fundamental breakthrough has been achieved, the center of gravity quickly shifts from a handful of elite researchers to an army of tinkerers — engineers with just enough expertise to apply the technology to different problems. Our present phase of AI implementation fits this model.

The tinkerers have an added advantage nowadays due to the real time access to the work produced by leading pioneers. AI researchers tend to be quite open about publishing their algorithms, data and results.

There are two approaches to distribute AI across the economy: “battery” and “grid”. The grid approach tries to commoditize Ai and turn the power of ML into a standardized service that can be purchased by any company — Google, Amazon, Alibaba. AI startups take the other battery approach to build highly specific battery powered AI products for each use case.

Obama government released a long brewing plan for the US to harness AI power in 2016 but the White House barely registered it in the American news cycle. In 2017, the Chinese State Council’s “development plan for a new generation of AI”. They share a lot in common but Chinese government is pushing for real.

5. The Four Waves of AI

4 waves of AI: Internet AI, Business AI, Perception AI and Autonomous AI.

Internet AI: Leverages the fact that internet users are automatically labeling data as they browser, it give birth to AI driven internet companies. This type of AI remains largely in high tech sector and digital world.

Business AI: business AI leverages the fact that traditional companies have also been automatically labeling huge quantities of data for decades. It mines these databases for hidden correlations that escaped human eyes. Business AI draws on all the historic decisions and outcome within organizations and uses labeled data to train an algorithm that outperforms most experienced human practitioners.

Perception AI: perception AI can group pixels from photos or videos into meaningful clusters and recognize objects in much the same way our brain does — natural language processing, image recognition, machine translation etc.

Autonomous AI: the integration and culmination of previous waves, fusing machines’ ability to optimize from extremely complex data sets with their newfound sensory powers. There are also two approaches to develop this: Google’s way by building the perfect product and make the jump straight to full autonomy once the system is safer than humans. Tesla’s way of incremental development to make up ground as soon as they are available. China is following Tesla’s approach in developing its autonomous AI.

Kai-fu’s estimate of AI development in China & the US in next 5 years

| Today | 5 years from Now| China vs the US

Internet AI 5:5 -> 6:4

Business AI 1:9 -> 3:7

Perception AI 6:4 -> 8:2

Autonomous AI: 1:9 -> 5:5

6. Utopia, Dystopia, and the Real AI Crisis

The utopians see the dawn of AI as the final frontier in human flourishing and an opportunity to expand our own consciousness and conquer mortality. The other side fears AI as the biggest risk we face as a civilization. The dystopian camp worries the Super Intelligence would seek to achieve goals in the most efficient way possible without the same instincts as humans — they might wipe out humans from the planet if they see humans as blocker to their goals.

The real AI crisis lies in the possible disruption of our society due to economic division and challenges to our sense of human dignity and purpose.

The Luddite fallacy is the simple observation that new technology does not lead to higher overall unemployment in the economy. New technology doesn’t destroy jobs — it only changes the composition of jobs in the economy.Jan 15, 2017

Optimists think the same way as Luddite fallacy, shifts in technology might lead to some short term displacement, but as millions of farmers became factory workers, those laid-off factory workers can become yoga teachers and software programmers.

Blind optimism, however, has some risks. There has been only 3 GPTs (general purpose technology) that extend their reach into many corners of the economy and radically altered how we live and work — the steam engine, electricity and information and communication technology. 3 is rare enough to warrant evaluation on their own. The Second Machine Age argues over the past 30 years, the US has seen steady growth in worker productivity but stagnant growth in median income and employment. The economy gains of ICT increasingly accruing to the top 1% of the population.

What AI Can and Can’t Do? The risk of replacement graphs:

Different studies have very different predictions for job disruptions. A pair of researchers from Oxford University said in 2013 that 47% of U.S jobs could be automated within the next decade or two. In 2016, OECD (Org of Economic Cooperation and Development) used an alternate model and predicted 9%. The different is that Oxford’s approach is occupation based and OECD is task based. Later in 2017, PwC used the task based approach and produce their estimate of 38% with a slightly different algorithm in the calculations. Kai-fu argues there is a missing aspect that there are two kinds of job loss: 1–1 replacement and ground up disruptions. Lower costs and superior services from AI based companies would force other companies to adapt from the ground up and restructure their workflows to reduce employees.

Moravec’s paradox is the discovery by artificial intelligence and robotics researchers that, contrary to traditional assumptions, high-level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources.

It is easier to build AI algorithms than to build intelligent robots. Even though China will face wrenching labor-market transition due to automation, large segments of that transition may arrive later or move slower than the job losses wracking the American economy. Core to this logic is Moravec’s paradox.

PwC estimates the US and China are set to capture a full 70% of the $15.7 trillion that AI will add to the global economy by 2031 and the rest of the world will be left to pick up the scraps. This process will exacerbate and grow the divide between the AI haves and have-nots.

7. The Wisdom of Cancer

Kai-fu retrospected on his past life on work life balance. His goal in life was “To maximize my impact and change the world”. He gave an example of himself debating if he should stay by his wife’s side for the delivery of his daughter or rush off to an important meeting. After the strike of cancer, he met Master Hsing Yun and Hsing Yun told him:

When people speak in the way of “maximize impact”, it’s often nothing but a thin disguise for ego, for vanity. The quantification of everything eats away at what’s really inside of us and what exists between us. It suffocates the one thing that gives us true life: love.

After recovery from cancer, Kai-fu has come to cherish time with those closest to him and spent much more time with them. Of all AI’s astounding capabilities, the one thing that only humans can provide turns out to be what is most needed in our lives: love.

8. A Blueprint for Human Coexistence with AI

In this chapter, Kai-fu laid out a couple ways to coexist with AI for the future.

Reduce, Retrain and Redistribute: retrain the workers, reduce work hours and redistribute the workforce.

Universal Basic Income: every citizen in a country receives a regular income stipend from the government — no strings attached. UBI camp argues job retraining and clever scheduling are hopeless in the face of widespread automation and only guaranteed income will let us avert disaster during the jobs crisis that looms ahead. How exactly a UBI would be implemented remains to be seen.

Social Investment Stipend: Kai-fu suggests a stipend to those who invest their time and energy in those activities that promote a kind, compassionate, and creative society. These would include three broad categories: care work, community service and education.

9. Our Global AI Story

Steve Jobs cautioned against trying to chart one’s life and career in advance:

You can’t connect the dots looking forward, you can only connect them looking backwards. So you have to trust that the dots will somehow connect in your future.

Kai-fu reflected he has spent much his life obsessively working to optimize his impact and turn his brain into a finely tuned algorithm for maximizing his own influence. It took him a cancer diagnosis and the unselfish love of his family for him to finally connect all these dots into a clearer picture of what separates us from the machines we built.

We are not passive spectators in the story of AI, we are the authors of it.

hacker, lifetime learner