You need to sign in or sign up before continuing.
Take a photo of a barcode or cover
alexherder 's review for:
Written by the former head of Google China and one of the pioneers of machine learning in voice recognition, this book is worth reading for anyone who wants to understand AI or US-China relations. I picked this up after working on a video project for an American think tank about China's "Digital Silk Road" initiative and I wanted to know more.
Here are some key takeaways and (below those) some important caveats that he did not write about:
- Task-specific AI and general AI are completely different things. They are probably not all that connected, and the only one that actually exists as of this moment is task-based AI, also known as machine learning (ML).
- ML algorithms are only as good as the data they have, and Chinese companies have access to way more and better quality data than anyone else. There are lots of reasons for this: Communist party centralization, large agglomeration platforms like Alibaba and WeChat, and a lack of concern for laws and regulations on the part of Chinese companies.
- The days of China being the land of cheap knock-offs is over. Chinese companies are simply different from American ones and Chinese users are different. For example, mobile payments are super common there, but still not big over here.
- The lengths that the Chinese state have gone to in order to build up the AI sector in China are amazing, and envy-making. Because of the single party state, Chinese leaders can create the conditions for industry growth that we in the US can only dream about.
- There are four waves to AI: internet AI, business AI, perception AI, and autonomous AI. The US and China have different strengths at each stage but China is better at 1 and 3, whereas the US is better at 3 and 4 is a toss-up because we're not really close to it yet.
- Lee is bearish on the impact of AI on jobs. He thinks we could be looking at job destruction of 50% and that it will hit many sectors quickly. He has a very useful graph showing which kinds of jobs are vulnerable and in what ways. He also thinks that we (humans, not China) need to be figuring out how we're going to deal with this. He's not a big fan of UBI, but thinks that some sort of UBI-for-carers would be humane and lead to better societies.
There are a few glaring blindspots in this book. They're understandable because they would inevitably cast China and the Chinese Communist Party (CCP) in a bad light, but anyone who is really interested in these topics needs to know them.
1) Lee does a good job of describing the techno-utopian POV of the CCP vs. the American "rights-based" framework, but doesn't go into details. The details are disturbing, especially to those of us with rights-based frameworks. For example, the Chinese government have turned the western Chinese state of Xinjiang into a virtual prison. The entire state is under martial law with complete audiovisual and electronic surveillance and current estimates put the number of people in "reeducation camps" (prison camps) over 1.5 million.
2) The Chinese "One Belt, One Road Initiative" is a MASSIVE investment in building a new global infrastructure with Beijing at the center. In addition to ports, railroads, highways, cities, undersea cables, and satelites, they are also exporting governance models that lower the cost of authoritarianism significantly. For example, many countries are accepting Chinese aid in building out their internet infrastructure with the Chinese-style centralization pre-installed. That effectively means dictators can spy on their populations cheaper than before, which creates less of a chance that those dictators will be overthrown.
3) Lee makes a strong case that one of the major reasons that China will "win the AI war" is because they have more and better data. This is partly because their people are less concerned with privacy, but also because their companies and government are less scrupulous with guarding that data. There is lots of evidence that Chinese companies share data with the government and vice versa.
4) Combining points 2 and 3, one of the scary-to-me things to consider happens when Chinese companies are paid (using Chinese loans) to set up surveillance/governance internet infrastructure in other countries, all of that data goes back to Beijing. Zimbabweans recently rejected a large project from Chinese company CloudWalk because this provision alarmed citizens and provoked large protests. Here's why that's concerning: Imagine that Zimbabwe had not rejected the CloudWalk project and, for the next 10 years, used it to manage and monitor their citizens' internet usage and surveil them day-to-day through facial recognition. Perhaps that would deliver better services, but it would also set a high bar for anti-regime success. Then, one day imagine that same autocratic regime wants to move away from Chinese alignment for whatever reason. Now, the Chinese state can back an alternative ruling party and with their superior model of Zimbabwe's own citizens and all of their data, they would probably win any conflict. In the long run, this creates client countries in a more insidious way than even the blunt American corporate empire model ever did.
Here are some key takeaways and (below those) some important caveats that he did not write about:
- Task-specific AI and general AI are completely different things. They are probably not all that connected, and the only one that actually exists as of this moment is task-based AI, also known as machine learning (ML).
- ML algorithms are only as good as the data they have, and Chinese companies have access to way more and better quality data than anyone else. There are lots of reasons for this: Communist party centralization, large agglomeration platforms like Alibaba and WeChat, and a lack of concern for laws and regulations on the part of Chinese companies.
- The days of China being the land of cheap knock-offs is over. Chinese companies are simply different from American ones and Chinese users are different. For example, mobile payments are super common there, but still not big over here.
- The lengths that the Chinese state have gone to in order to build up the AI sector in China are amazing, and envy-making. Because of the single party state, Chinese leaders can create the conditions for industry growth that we in the US can only dream about.
- There are four waves to AI: internet AI, business AI, perception AI, and autonomous AI. The US and China have different strengths at each stage but China is better at 1 and 3, whereas the US is better at 3 and 4 is a toss-up because we're not really close to it yet.
- Lee is bearish on the impact of AI on jobs. He thinks we could be looking at job destruction of 50% and that it will hit many sectors quickly. He has a very useful graph showing which kinds of jobs are vulnerable and in what ways. He also thinks that we (humans, not China) need to be figuring out how we're going to deal with this. He's not a big fan of UBI, but thinks that some sort of UBI-for-carers would be humane and lead to better societies.
There are a few glaring blindspots in this book. They're understandable because they would inevitably cast China and the Chinese Communist Party (CCP) in a bad light, but anyone who is really interested in these topics needs to know them.
1) Lee does a good job of describing the techno-utopian POV of the CCP vs. the American "rights-based" framework, but doesn't go into details. The details are disturbing, especially to those of us with rights-based frameworks. For example, the Chinese government have turned the western Chinese state of Xinjiang into a virtual prison. The entire state is under martial law with complete audiovisual and electronic surveillance and current estimates put the number of people in "reeducation camps" (prison camps) over 1.5 million.
2) The Chinese "One Belt, One Road Initiative" is a MASSIVE investment in building a new global infrastructure with Beijing at the center. In addition to ports, railroads, highways, cities, undersea cables, and satelites, they are also exporting governance models that lower the cost of authoritarianism significantly. For example, many countries are accepting Chinese aid in building out their internet infrastructure with the Chinese-style centralization pre-installed. That effectively means dictators can spy on their populations cheaper than before, which creates less of a chance that those dictators will be overthrown.
3) Lee makes a strong case that one of the major reasons that China will "win the AI war" is because they have more and better data. This is partly because their people are less concerned with privacy, but also because their companies and government are less scrupulous with guarding that data. There is lots of evidence that Chinese companies share data with the government and vice versa.
4) Combining points 2 and 3, one of the scary-to-me things to consider happens when Chinese companies are paid (using Chinese loans) to set up surveillance/governance internet infrastructure in other countries, all of that data goes back to Beijing. Zimbabweans recently rejected a large project from Chinese company CloudWalk because this provision alarmed citizens and provoked large protests. Here's why that's concerning: Imagine that Zimbabwe had not rejected the CloudWalk project and, for the next 10 years, used it to manage and monitor their citizens' internet usage and surveil them day-to-day through facial recognition. Perhaps that would deliver better services, but it would also set a high bar for anti-regime success. Then, one day imagine that same autocratic regime wants to move away from Chinese alignment for whatever reason. Now, the Chinese state can back an alternative ruling party and with their superior model of Zimbabwe's own citizens and all of their data, they would probably win any conflict. In the long run, this creates client countries in a more insidious way than even the blunt American corporate empire model ever did.