AI Superpowers: China, Silicon Valley, and the New World Order Paperback – September 14, 2021

An excellent overview of Chinese activity in artificial intelligence from ground zero perspective. Lee shows how two strands (China and AI) are being irreversibly woven together into an unbreakable high-power cable that will transform, control, and possibly strangle humanity’s economic future.The inciting incident for both Lee’s book and another comparable recent effort (Artificial Intuition: The Improbable Deep Learning Revolution, by Carlos Perez) is the recent victory in Go of an AI-based system over the best human champion of that ancient game. This had about the weight of an empty beer can in the USA and other Western news cycles, but shook the Asian intelligentsia at their core (because they care so much more about Go). Both Lee and Perez make a big deal out of the Go victory as a Sputnik moment, awakening entire East Asian populations, and their central planners, to the urgency of becoming the dominant AI superpower. Meanwhile, apart from some corporate research,the USA snoozes blithely on. We may wake like Rip Van Winkle in 20 years (or 20 months) to find ourselves hopelessly lagging China. AI Superpowers skillfully exploits and intensifies the fear factor. A cynic would say the hidden agenda here is to trigger another 1980’s-style AI panic, when it seemed that Japan would conquer the world with their Prolog (logic programming) initiative. But I am not cynical. I appreciate this book on its own terms.Anyway, after both books (Lee and Perez) lead off with humanity’s miserable Go game beatdown, they then diverge sharply in quality, and Lee quickly pulls way ahead of Perez. Where Perez gets lost in an impenetrable thicket of his own miserably confusing writing style and rambling topical garden paths (or garden mazes), Lee drives straight for the goal line: a clear and compelling picture of the current state of play, and a crisp delineation of where things will end up.The depiction of China’s hi-technology business culture is the stronger element, relative to the presentation of AI as technology. Although Lee knows the tech methods and architectures inside and out, this is a popular treatment and you won’t encounter a single equation or circuit layout anywhere. Basically, you’ll get the key message that ‘narrow’ AI (task-specific systems that learn and perform well on human expert functions) has made a giant leap in a brief recent period. These systems are lumped under the term ‘deep learning’, an extension of a fairly simple neural modeling concept dating back to the 1950’s that has just now broken free of the pack and left the field behind. That breakout has been enabled by more data, more computing power, and some architectural upgrades to the original concept. Lee then zooms in on how fast and how furious the deep learning tsunami will hit.AI Superpowers is strongest in its contrasting portraits of the Chinese high tech scene vs. the USA’s Silicon Valley. Lee offers numerous real-world cases illustrating parallels and divergences, sprinkled with entertaining personal tales from the trenches on both sides.He traces the ascendance of China’s tech giants such as WeChat, AliBaba, and many others that aren’t household names in the West, digging deep into exactly how each one succeeded – pound for pound, blow by blow, user by user – a view from the trenches. Basically he portrays Chinese entrepreneurial high tech as sharing much in common withorganized crime, possibly minus the sicarios. (See the book Narconomics: How to Run a Drug Cartel by Tom Wainwright if you want the real dirt.)All that leads into Lee’s clear-sighted take on the employment implications of the new AI. Lee is no Elon Musk, in that he doesn’t see AI posing any immediate existential threat to the human species. Nor does he spend much ink on the roseate Kurzweil ‘singularity’ stuff (which is essentially religious fantasy in my view). True to form, Leemoves soberly and smoothly, like the no-nonsense businessman he is, to consider something closer to home: the possible loss or diminution of up to 40% or more of current jobs with a decade or less. Many of these threats are well known, particularly the driverless vehicles thing, automated medical radiology, and many others. Analyses of thistype go back at least as far as Jeremy Rifkin’s classic The End of Work (1996) and many more recent treatments. But Lee’s examination is particularly lucid and right up to the minute in its full attention to the new AI (deep learning).The book then takes a personal twist as Lee details his battle with an abrupt cancer diagnosis and how the recovery ordeal opened his eyes to elements of the emotional and social landscape he’d skated lightly past in his meteoric ascent to the top of the transPacific high tech dogpiles, both Silicon Valley and Zhongguancun. Newly sensitized to the human side of life, he prescribes human-to-human (or heart-to-heart) operations as something we can turn to, a need that will persist, even when all the truck driver spots dry up. This is a really laudable aspect of the book. How many tough tech exec’s and macho VC posers would have the spine to reveal this much of themselves and lay down their pugnacious facades, to go this deep? Truly admirable and unique.In the final section, Dr. Lee offers his prescription, which is a social do-gooder program of make-work in areas that computers still have trouble with, such as cheering lonely elders and such. Lee believes government mandated social-service jobs programs have advantages over the resurgent Universal Basic Income proposals (basically air-dropping free money on people from helicopters).The strengths of this book are the great high-tech anecdotes and ringside recent history accounts, the straight-forward descriptions of some key technical advances, and future directions, as well as the uniquely heart-centric infusion of that emo human touch in considering palliatives for the upcoming unemployment wave.Minor downsides include a certain Chinese cultural chauvinism. Lee is very convincing in proudly calling out all the strengths of China relative to the West (basically, the USA) in the AI’s Brave New World. But he sometimes gets a little carried away. For example, he makes more than one admiring reference to a recent Chinese sci-fi hit, Folding Beijing by Hao Jingfang, which depicts a future world of extreme class and functional stratification.That’s an appealing effort but hardly as unique as Lee seems to believe, given that this basic scenario was chillingly and unforgettably depicted in one of the very first science fiction novels ever published (The Time Machine by H. G. Wells - a future world of the surface dwelling Elio vs. the Morlocks, ape-like troglodytes who live in darkness underground and surface only at night). That’s just one example of several where Lee’s understandable cultural pride gets the better of his basically dispassionate instincts.He also paints too strong a contrast between Silicon Valley culture and the organized crime ethos of Zhonguancun (China’s Silicon Valley). For example, Lee writes that the Valley (USA high tech culture) despises copycats. Is this true of Larry Ellison, secretly copying the original IBM research paper on relational databases that became the signature tech of Oracle? Is it true of Steve Jobs, ripping off for his Macintosh every element demonstrated to him on the Star office system when he toured the Palo Alto Xerox research lab one day? And we all know the saga of DOS and Windows. How original was Facebook? I could go on. And on. Things are more similar than different.Where he really has a gift is in making concepts that would be way too scary or boring to a lay reader perfectly understandable and accessible. Here, Lee really shines. For just one of many examples, consider is his seamlessly smooth rendition of a blazing hot method in current AI research, Generative Adversarial Networks (GANs). This would scare the pants off most lay readers if they encountered it in a technical book, but look how easily the medicine goes down when administered by Dr. Lee:“Toutiao then used that labeled data to train an algorithm that could identify fake news in the wild. Toutiao even trained a separate algorithm to write fake news stories. It then pitted those two algorithms against each other, competing to fool one another and making each other better in the process.”Rendering this powerful frontline concept perfectly lucid in a couple dozen words – that’s a gift (but again, note the touch of Chinese chauvinism, implying that Toutiao, a Chinese company, somehow thought up this approach for their own little application, while in fact the GAN configuration is entirely the innovation of a Western AIresearcher).But enough carping. It’s a very good book and well worth reading. Now the big question opens before us: after reading this, am I worried? Am I convinced to be at least as worried (yet cautiously hopeful) as Lee himself? No, I’m not worried at all, and here’s why.First, some background. Despite the fact that these high-tech icons and visionaries pretend to revere intelligence and genius and talent and brains and innovation and creativity and all that, secretly every one of them must know that the greatest economic resource is not intelligence at all. The greatest economic resource, by far, is stupidity. They all know this, and now you know it too. I can prove it.Consider where the economy would be if even a few sectors such as the following were entirely removed: soft drink industry; snack foods industry; global arms industry; all religions; makeup and cosmetic industry, including weight loss and cosmetic surgery; high end luxury brands of all kinds; most video games; most movies and popular entertainment; most of the ‘financial services’ industry – need I go on? Every one of the above sectors is based almost entirely on human stupidity. Or, at a minimum, none of them could function without a solid root in human stupidity. And that’s only the start.Now consider the ramifications of just eliminating one item on that list, say, the soft drink and snack foods industry. That alone would probably eliminate up to 50% of current health care services required by the population. So the effects would ripple out everywhere. I could go on but you get the idea. The one essential economic resource isnot intelligence at all. It is stupidity. The human economy would grind to a dead halt without it. Whether human stupidity is exploited haphazardly by existing manual methods, or (in the near future) exploited and stimulated more efficiently via AI methods is immaterial.That’s why I don’t see any great long term threat in AI. Or if there is any threat in AI, it isn’t the economic stuff called out by Lee, it’s more likely to be the existential stuff called out by Musk and others in his camp. But we’ve put that aside for this discussion, so within the terms of Lee’s book, we can expect clear sailing. Yes, AI will continue to advance, but in the words of one famous science fiction writer: ‘the street finds its own use for things’. Humans will adapt AI to their own unceasing pursuit of profit and pleasure through organized and unorganized crime and it all will be business as usual in the long run.The only big effect will be that the AI mavens of today, the ‘smart ones’, will probably end up displacing themselves. We can do without intelligence (see above). The crucial resource is stupidity. The co-founder of Communism, Vladimir Ilyich Lenin, once famously predicted that “when the time comes to hang the capitalists, they will rush to sell usthe rope”. Similarly, the Chinese rush to pull ahead in AI by throwing money and brains at it is likely merely accelerating the creation of their own successor species. That engineered new life form will likely put all the smart guys out of business yet retain some need for stupid feedstock, just as in the Matrix movies the AI’s ran the world but kept sleeping (stupid) humans as batteries.But still - shouldn’t I be a bit more tremulous about the advent of our AI overlords? After all, it has been stated by one who should know that: With superintelligent computers that understand the universe on levels that humans cannot even conceive of, these machines become not just tools for lightening the burdens of humanity; theyapproach the omniscience and omnipotence of a god.Wow, AI’s will become ‘gods’. But even so, they will never be able to beat down the human race. Because we have our great ace in the hole, the one cognitive space we humans uniquely occupy, where by definition, no AI can follow. Before you hide under your bed, dig these words of wisdom:“Against stupidity, the very gods themselves contend in vain.”- Friedrich Schiller

This is one of the most important books of 2018 If you are engaged in any business that is or will use machine learning (especially deep learning) you should read this. Likewise if you are concerned with the social implications of AI, or cultural differences in how technology gets developed and deployed, or international relations, or data privacy and ownership. The book will provoke a lot responses, or it should.I would read it with three other books. To get a wider view on the limitations of the current approach to AI and a possible path forward read Judea Pearl and Dana McKenzie's The Book of Why. For context on competition between China, a rising power, and the US, a mature power, see Destined for War by Allison Graham (the book is more optimistic and less sensational than the title suggests), and for more on business implications of prediction see Prediction Machines by Ajay Agrawal, Joshua Cans and Avi Goldfarb.The author has a deep technical understanding of machine learning, has worked in Silicon Valley and for Silicon Valley firms in Beijing, and plays an important role in the Beijing innovation ecosystem. He has direct knowledge of what he speaks.His basic position is that many of us do not understand the speed of innovation on machine learning in China or what is driving that innovation and why China is likely (he believes) to out perform Silicon Valley (it is not clear if he is using Silicon Valley to stand for the US , Canada and in general or if he really just means SV, but many people in SV are similarly confused). He gives four reasons for this. (i) Chinese entrepreneurs have grown up under hyper competitive commercial competition where anything goes and will out compete their SV competitors. (ii) China is creating more AI engineers than any other country and by a large margin. (iii) AI (or machine learning anyway) is in an implementation cycle, not an exploration and innovation cycle, and although China may be a follower on innovation it will be a leader on implementation. (iv) The Chinese government will give Chinese researchers and companies unfettered access to the data needed to feed machine learning and will even be willing to redesign and rebuild infrastructure for machine learning. The SV companies will not be able to gather enough data to compete.Lee also has some interesting thoughts on the O2O and OMO economies. O2O is Online to Offline and refers to the use of online (mobile and machine learning) to deliver offline services (in SV Uber would be one of thousands of examples). Lee feels that this is where China excels. He then goes on to explore Online Merged with Offline, which is what we will see in the mashup of machine learning, augmented reality and the larger Internet of Things. This is a useful framework for thinking about the changes we are seeing.Less useful is his division of AI into four waves: Internet AI (for example Facebook's use of AI to manipulate your news feed); business AI (in which he includes health care); perception AI (which includes education), and autonomous AI (which one might also describe as ambient AI) in which AIs are embedded in all systems and are making decisions. He then makes predictions of which country will come to dominate each of these waves of AI (the only one where he thinks SV will dominate is business AI). The inclusion of healthcare in business AI is a sign of the limitations of Lee's thought. Coming from the US and China he seems to have a limited view of how other cultures innovate in and deliver healthcare, More on this later.Another key point he makes is the degree to which Shenzen has become the innovation lab for hardware and the advantages this will give China in an OMO world (Online Merged with Offline). This is an important point. To get a feel for this go visit Shenzen if you possibly can and read Bunny Huang's book The Hardware Hacker.So how is this book blinkered?The most serious shortcoming is that he only considers SV and China. Like many in SV (and possibly China) he misses what is actually happening in Europe, especially Northern Europe, the strength of the machine culture there and the depth of research that is going on. Many in the US are used to discounting Europe and they are wrong to do so. Europe is likely to be the leader in manufacturing AI. He also fails to consider India and the possibility that over the next 50 years the Indian economy will displace China as the world’s largest. This seems to be because he fails to consider the importance of demographic change. China is about to have a rapidly aging and then shrinking workforce and economic growth will slow significantly. The lack of insight into Europe and India is a major flaw in this book.There is also an assumption that companies will own the data that drives AI innovation and that this will make the wealth gap much worse. There is no reason this needs to be true. Perhaps we (or Europe or India) will adopt a different model where the individual owns the data and shares in the value created by it. Data ownership by corporations is not a given.I wonder if AI over the next decade may not be much more diverse than the current deep learning orthodoxy, based on the work in Toronto r led by Geoffrey Hinton and Montreal by Yoshua Bengio. I know this is now the established belief of many people in SV, but see the above cited book by Judea Pearl (being commercialized at companies like Via Science) and expect many other approaches to surface over the next decade. I do not think the shift from research to implementation is as crisp and Lee, and I suspect the real progress will cone as research and implementation interact more deeply.Books like AI Superpowers would benefit from taking a scenario planning approach, and looking at the future as a set of possible alternatives based on critical uncertainties rather than as linear predictions (I wonder if the optimization paradigm so important to machine learning limits one’s ability to imagine multiple different futures).The book would have benefitted from a deeper conversation on data ownership and its implications. Lee touches on this as an example of how the Chinese government's willingness to ensure wide data collection and access to this data by commercial entities over which it has partial control gives it a strategic advantage in AI. I wonder. Are there other data ownership and privacy paradigms that could give even better results? This is one of the critical questions facing us.Finally, I am discouraged by the narrow nationalistic thinking coded into this book. Do we really have to have a world based on competition between superpowers? Is a collaborative approach possible? Will there be communities that crossnational boundaries to create wealth that is more widely distributed and less concentrated? I am a bit of an idealist here, but I have no interest in working to strengthen any superpower. Historically superpowers have not acted in the interests of the larger good, or even in the interests of their own citizens.Be sure to read the final part of the book, where Lee talks more about his personal story and ways in which AI can become more humane.