Kai-Fu Lee's book provides a rich account of the AI state-of-the-art and how much China has caught up with the US in the field. To a Western reader, China's exponential growth in AI-related business and technological development may at times be overwhelming, especially with regard to its scale. For instance, the fact that Didi accounts for more rides in one day throughout China than all of Uber's rides combined across the world. Or that in 2017, Chinese Venture Capitalists made up 40% of AI investors globally, surpassing the US, and that Chinese mobile phone spending outnumbered that in the US by a ratio of fifty-to-one.
Such numbers may be difficult to fathom for someone who has yet to step foot in Beijing or in one of China's many buzzing mega-cities. Yet they are part of a larger and even more impressive story of China’s spectacular development over the past two decades. Most telling of all, the fact that between 2007 and 2017, China went from having zero high-speed train lines to having more miles of high-speed rail operations than the rest of the world combined. Informing and educating the reader about these developments is one thing this book does well and one of its key contributions. The fact that the author knows both sides (the West/Silicon Valley and China's tech ecosystem) undoubtedly helps make some of those developments accessible to a Western audience.
Kai-Fu Lee’s initial historical and theoretical presentation of the field of AI is also well crafted and provides interesting insights to a non-expert audience. The author explains how the field of AI is forked into 2 camps - the "rule-based" approach and the "neural networks" approach. On the one hand, the rule-based approach attempted to teach computers to think by encoding a series of logical rules. This fell apart when the universe of possible choices or moves expanded. On the other hand, the neural networks approach tried to reconstruct the brain itself, layers of artificial neurons that can receive and transmit information akin to our biological neurons. One may thereby feed a lot of examples of a given phenomenon - pictures, sounds - and let the networks themselves identify patterns within the data, the less interference the better. The author points to the fact this “deep-learning” approach can now do a better job than humans at identifying faces, recognizing speech and issuing loans. And this gets a little scary when Kai-Fu Lee tells the story of the day when the World Go champion lost in a tournament against “AlphaGo”, which runs on deep-learning. To make us understand the sheer extent of that technological achievement, the author reminds us that the known possible positions on a "Go" board surpass the number of atoms in the universe – let that sink in.
Throughout the book, Kai-Fu Lee tends to repeat something akin to a mantra about AI, which the reader understands as one of the book’s key take-aways: successful AI systems require 3 things - large amounts of data, computing power, engineers/know-how. Kai-Fu Lee covers each of these aspects in depth in order to support his narrative that competition over AI is at the heart of wider US-China competition and that the skillful application of AI will be China's chance to surpass the US. He describes what he sees as the 4 waves of AI: internet AI, business AI, perception AI, autonomous AI and argues that within 5 years, China will have an advantage of the US in internet AI.
The book nonetheless falls short on several accounts: the reader cannot help but think that the author is a little bit too enthusiastic about AI in general, going over key issues such as the future of jobs, inequality and privacy a bit too casually.
Indeed past technological shocks on society did not happen at the same pace as what is bound to happen with AI: based on current trends, AI will technically be able to replace 50% of jobs in the US within the next 15 years. Even though AI will create new, mostly high-skilled jobs, massive unemployment and poverty among low-skilled workers is a real societal risk, all the while AI tycoons will accumulate astronomical wealth. It is unfortunate that the author does not spend time reflecting upon what this means in terms of up-skilling, training and education policy.
These trends should logically force us to rethink wealth distribution via a revamped taxation system involving taxes on AI while enabling humans to continue creating value for society. The author does attempt to address the idea of a Universal Basic Income, though he does not really believe in the concept. Instead, the author thinks that wealth redistribution through an eventual tax on AI should be used to reinforce salaries and conditions of those existing and new professions that will likely always necessitate a human touch (e.g. health and social work). That is not an uninteresting idea but the author spends too little time on the matter to do it justice.
Finally, the book really fails to acknowledge and provide solutions to AI-related issues of privacy and ethics. The author seems to discard those as mere trade-offs, pointing to the fact that in China people have been far more accepting of having their faces captured and digitized into data, thus trading some privacy for convenience. The author also fails to engage with the related need to address biases in AI systems. Deep learning processes are very impressive but if a machine learns without a built-in ethical system, it will inevitably be biased, for instance towards making profits regardless of human and/or environmental costs. Indeed, the environmental impact of AI is something that is conspicuously absent from the book. As a result, though the book provides some fascinating insights about the field of AI, it remains paradoxically out-of-touch and disconnected from the reality of those risks and challenges that are central to the so-called "New World Order", as it enters the second decade of the 21st century.
Comments