Kai-Fu Lee, “AI Superpowers: China, Silicon Valley, and the New World Order
A book written by an AI researcher-turned venture capitalist. Here is a summary of the book.
AI is GPT
AI is very likely to be the third general purpose technology (“GPT”), a technology that shapes the society to the next level. The first was industrial machines, and the second was ICT. Andrew Ng, the world’s top AI researcher, compared AI to electricity.
A brief history of AI
AI’s history dates back to 1950s. Back then, there were two schools — Expert System and Neural Network (“NN”). NN experienced 3 booms — the first was from 1950s to 1960s, then 1990s and mid 2000s. Geoffrey Hinton and the other researchers came up with abetter way to train the deep neural networks and ‘rebranded’ it as “Deep Learning.” The NN-based algorithm won a vision detection contest in 2012 and then the frenzy for AI began. 2016 marked the huge event for many Asians — Alpha Go completely beat the best human professional.
AI hegemony
Best AIs necessitates 4 things: large data, a large number of data scientists, quality entrepreneurs and government support. The author concludes that China and the US will be the two superpowers fighting for the AI hegemony. The author argues that China has enough reason to surpass the US AIs given the following reasons:
(1) the country produces the largest digital data in the world
(2) China has a good number of data scientists who eagerly learn things
(3) quality of the Chinese entrepreneurs are world-class thanks to the cut-throat competition environment — the “gladiatorial start-up ecosystem” makes the quality entrepreneurs
(4) the government is backing the AI entrepreneurs to win the AI hegemony in the world
There are now 7 AI juggernauts — Google, Amazon, Facebook, Microsoft, Tencent, Alibaba and Baidu. Google is ahead of the others now, but not sure in the future.
4 waves of AI
According to Kai-Fu Lee, there are 4 waves of AI.
(1) Internet AI
The AI companies leverage the browsing data in the Internet. Recommendation engine is what Internet AI is all about, i.e., shopping, search results, news feed, etc. Google and Amazon are good examples.
(2) Business AI
The companies use the existing well-structured business data. Financial data (at financial institutions), health data (at hospitals) and legal data (courts or law firms) are good examples. By applying deep learning, the companies can find out hidden correlations among data, and the results are used to improve loan assessment, diagnosis, judgment, etc.
(3) Perception AI
The companies leverage data which are originally not digital. The information about voice (e.g., Amazon Echo), vision (e.g., Face ID), etc. is transformed into digital data. The moves significantly widen the access point between the Internet and the real world (originally it was usually PC & phones only), and in other words, they enable machines to recognize the real world. Initially, it was called O2O (online to offline) but now what is happening is OMO (online merge offline).
The advancement blurs the line between the physical and the digital world and revolutionizes how we experience things. For instance, the author gives an example of ordering a meal using a smart speaker. Here, we’re using an Internet service, but the experience is like a physical experience with a servant.
(4) Autonomous AI
There is a striking difference between automation and autonomy. The latter means the machines do things by its own. Car and drones are just a beginning. Later, farming, food making, and many other labors will be replaced by machines which autonomously do what they have to do. A good example is Amazon’s warehouse, where a tiny number of people exist.
The author argues that the Chinese AI giants will catch up the counterparts in the US. Below is the table by the author.
AGI — will it come?
The author seems to be not confident about the early advent of the artificial general intelligence (“AGI”) which will mark the singularity.
Deep Learning is just one of the breakthroughs to make AGI possible. To develop the AGI, the current AI needs to do the following additionally:
- Multidomain learning
- Domain independent learning
- Natural language understanding
- Common sense reasoning
- Planning
- Learning from small number of examples
- Learn other human factors such as self-awareness, humor, love, empathy, appreciation for beauty, etc.
Real short-term AI threat — inequality
Rather than the “Terminator” story, the real AI crisis may be its impact on exacerbating inequality. The advent of GPT had always resulted in the rise of inequality.
The two negative things will occur due to AI advancement.
The first is obvious. We need less labor forces. The author says that 40–50% of the jobs in the US will be replaced by machines in the next 15 years. Some jobs may be created newly, but it is likely that those jobs require high intelligence and education. The below are the “endangered jobs” map, both physical and cognitive labors.
The second is what I realized for the first time — the gap between developing and developed nations will not shrink rapidly. Less wealthy countries had leveraged their cheaper labor forces to compete in the global market for their economic development. If AIs significantly reduce the low-skilled labor, then the labor cost composition of products will be smaller, resulting in less competitive power of the goods from the pooer nations.
The author explained 3R — retrain, reduce (the working hours) and redistribute, but they are not beyond what many politicians have discussed so far.
Some remarks:
- I personally am feeling that India will join the league in the next 10 years (large data, great researchers and entrepreneurs and government backing are all there)
- We may need to think about how the unit economics of my business would look like when the AI application becomes full-fledged.