Kai-fu Lee:
"AI Superpowers" (2018)

(Copyright © 2018 Piero Scaruffi | Terms of use )
Yet another book that tells us how soon AI (Artificial Intelligence) will rule the world. Yet another book that makes the ridiculous prediction that by 2040 or so AI eliminate 40-50% of jobs in the USA. Why would anyone develop AI at all if this is the price to pay for it? Lee seems proud of being an AI scientist: why be proud of having contributed to the rise of such a demonic force?

Lee may have a PhD in AI from the 1980s, but this book shows that he knows very little about the status of today's AI. AI is far from fulfilling the expectations generated by Hollywood movies. It has certainly improved face recognition and speech recognition, but no you cannot have a conversation with machines and no machines don't understand much of this text that you are reading and no they cannot drive a car and no they cannot replace jobs other than repetitive ones. And it will be this way for a long time. And it is hard to pinpoint a job that can be replaced by AI while it is very easy to list jobs that are being created by AI. Lee argues that we need to "reconstruct our economies and rewrite our social contracts" but that's as scary as the communist manifesto: do not let these people "reconstruct our economies". Economies reconstruct themselves all the time in response to technology and so far progress has taken us from the caves to Silicon Valley, and the wealthiest countries with the lowest unemployment are the ones with the most advanced technologies, including the technologies of automation.

Luckily that's not the core of Lee's book. Lee's book is very informative when it deals with the Chinese system.

The key argument that gives the title to the book is that Lee believes that there is a race in AI between China and the USA and that China will win. I didn't see the race to start with: the US government has shown little or no interest in boosting domestic AI. But let's assume there is such a race.

Lee's argument is that China is about to surpass the USA because research is no longer the most important factor: implementation is. He makes this grand statement because he thinks that AI is fundamentally different from all previous technologies. But if, for a second, you accept that there is nothing special about AI (other than its awful name), that it is just computational mathematics (and mostly very boring one), then he is saying something trivial, that has been true for at least three decades: the West does all the research and all the inventing, and China simply applies Western inventions within China while at the same time protecting its firms from Western firms such as Google and Facebook (which are banned in China just like dozens of other Silicon Valley firms). As i have written and discussed at length (including in China itself), China has used and is using the West as its research laboratory (free of charge): the West theorizes, invents, experiments, and China simply waits to see what succeeds and what fails. Lee thinks that this strategy is a winning strategy in AI. I think that China has confined itself to being a follower in pretty much every field and AI, which is neither the most important one nor the most lucrative one, is just another case of China following instead of leading. Lee thinks that research is not essential anymore to be the leader in AI. It is a silly argument: the application of research is always what makes money, and the place that knows it best is Silicon Valley, which has invented precious little but applied very well the inventions made elsewhere (the computer was invented in Britain, the transistor in New Jersey, AI in Boston, the smartphone in Finland, the Web in Switzerland, etc). But the country that has the research (that "invents") has the obvious advantage of setting the agenda for the rest of the world: why is China investing in AI instead of XYZ? Because the West invented AI.

The story of AI will simply mimic the story of all previous technologies. China will massively apply the (wildly overrated) AI technology coming from the West, protecting its firms from the competition of Western firms, while the West will invent the next big thing. Today's big thing in AI is neural networks, and specifically deep learning, invented in Canada's universities about ten years ago. China is frantically applying this ten-year-old tecnology, but investing almost zero on developing the next big thing. Western universities, instead, are investing massively in the next big thing, that could well be unsupervised learning or some breakthrough that today doesn't have a name. When the new big thing makes the headlines, China will frantically start copying it and applying it to its vast market. And the loop goes on. That's been the history of business for the last 30 years, and, if Lee is right, China is happy with this loop: let the West invent, and let's just copy what succeeds in the West.

By Lee's own account, China's advantage in today's AI is all based on data, and deep learning (today's AI) works only if you have data to train the neural network that it uses. But what if the next big thing is an AI that doesn't need data? China will soon look like the country that still makes old-fashioned deep learning.

I suspect that this is no longer true. I suspect that the Chinese leaders themselves disagree with Lee. China's "Sputnik moment" came early in 2018 when China's telecom multinational ZTE almost went bankrupt: the US Congress decided to punish it for violating US laws and China discovered that a (supposedly) high-tech firm like ZTE cannot compete without the (real) high-tech chips made in Silicon Valley. I suspect that the top ranks of the Chinese leadership are rethinking the 30-year-old strategy of simply copying and applying what the West discovers/invents. Someone up there is realizing what Lee fails to realize: that China will be stuck with neural networks and deep learning until some scientist in a Western university invents a new AI method and dozens of Western firms popularize that method, and China will, yet again, be ten years late in adopting it.

There's another huge disadvantage to China's strategy of "copying and applying" instead of "inventing": this strategy sounds dull and boring, i.e. it is unlikely to attract foreign talents the way Silicon Valley and North American universities do. If you are young and brilliant, there is virtually no amount of money that would convince you to waste your brain analyzing Chinese data. What you want to do is to invent something that will change the world. As long as research is done in the West, the West will continue to enjoy a colossal advantage over China: talents from all over the world, including and especially from China, will want to move to the West.

The West is also beginning to lose its patience with a country that is so openly protective of its domestic firms. China knows that it is just a matter of time before the USA retaliates against China's ban of Google, Facebook, Twitter and countless other US firms. Why is Wechat available all over the West while the equivalent Western platforms are banned in China?

Lee assumes that Chinese companies have been the best in the world at applying today's AI but the truth is that they don't compete with Western ones. Nobody outside China uses Baidu: how do you know that Baidu is a good application of AI to search? (In fact, i don't think it is: i think that Google is far superior). Lee also wildly exaggerates the merits of Chinese AI firms. The most popular Chinese platform of speech recognition is awful, the most popular Chinese platform of face recognition can easily be fooled, and Chinese robots are low quality. Lee misses an important point: that the Chinese government has been protecting its AI companies from Western competition, and the result, as it is often the case, is that they are not as competitive as it looks. The fact that there is a lot of application within China of China's face recognition technology does not tell us much about the quality of that face recognition technolology.

As of 2018, the most successful application of A.I. yet is Google's translation app, that is orders of magnitude better than anything China has produced in machine translation. Google is banned in China but Google Maps is more accurate than China's own mapping apps. Allow me to suspect that this is true of all the software that is banned in China: the original Western application is much better than the Chinese copy that exists only because the Western original is banned by the government. As of 2018, the most successful application of A.I. to a store is the Amazon Go store: no shop in China even remotely does what Amazon Go's pilot stores do. But i am sure that there will soon be copycat shops all over China... once Amazon Go shows China how to do it.

Lee divides technology into research and application, but in reality that division is arbitrary: is AlphaGo research or application? Is the Amazon Go store a research project or an application? The real division is between original and copy, early adopter and late adopter, innovator and immitator. China is simply one of the many prudent countries that prefers to copy, to imitate, to wait and see. Nothing wrong with that, but it is misleading to characterize it as "application vs research". China, in fact, tends to copy applications, not research: Baidu is the Google of China, Renren is the Facebook of China, Weibo is the Twitter of China, Alibaba is the ebay of China, Didi is the Uber of China, and so on: these are not original applications of an advanced Western research but copies of a Western application of relatively old research. And all of them have been carefully protected from the competition of the Western originals.

Coming to A.I., i am not sure that A.I. applications are as important as Lee thinks they are. I personally don't think they amount or will amount to much in the grand scheme of things.

Ultimately, this book is a great marketing idea because "scare" stories always sell, and Lee's book reads like a convincing scare story. Western universities and research labs love this kind of books because the scare story helps them get from funding from government agencies. Everybody is happy.

Well, at least the publishing industry seems to be done with the silly "singularity" thing, and now we're discussing business and politics, which are way more reasonable topics.

The one that doesn't get any help from this kind of books is China, a country that probably needs to be told exactly the opposite: you are paying a price because you don't invest enough in research, and the day of reckoning is not far. China has created a mindset (just copy Western applications) that won't be easy to change.

Lee mourns the fact that the poor nations of the world "will be left to pick up the scraps while these AI superpowers will boost productivity at home and harvest profits from around the globe." My advice to the poor nations of the world is NOT to follow the lead of China, which would make them followers of a follower, but to invest in research, so that they can leapfrog China directly to the next big thing.

For the record, Lee leads Sinovation Ventures, an AI-focused incubator based in China. He has a vested interest that the USA invests more in AI. He was also influential in convincing the Chinese government to make AI a strategic national goal, a fact that has lent credibility and money to AI startups in China. This book and his interviews are clearly trying to influence the US government to do the same. In an interview with the MIT Technology Review, he warned that the USA is falling behind because it is failing to invest in A.I. the way China did, and painted A.I. as a race that China is winning. The problem is that A.I. is not a race that China is winning but a race that China is the only one running: no other government has been convinced so far that A.I. is such a strategic technology. The US government has not acted. The big investment in A.I. has come from companies such as Google, Nvidia and Facebook, whose profits are actually coming from totally different businesses. So there is also a chance that the opposite is true: China may be squandering crucial resources after a technology that won't matter that much while other countries are investing in the technologies that will matter a lot.

My take on AI: "Intelligence is not Artificial".

TM, ®, Copyright © 2018 Piero Scaruffi