AI Superpowers, by Lee

Saturday September 26, 2020

There's a lot Kai-fu Lee isn't wrong about in AI Superpowers. The major break-through of deep learning has occurred, and applications are now more engineering than invention (cf. Perez). China has data, entrepreneurs, engineers, and government support for AI. Compared to the US, China has fewer ethical qualms and democratic controls, which Lee propagandizes as advantages for China. I think he's right that we should worry more about human society than about AGI, but I think he exaggerates the exceptionalism of AI.

The boundary between AI and non-AI computing isn't clear. There have been industrial robots for years. Does it matter whether Google search is powered by PageRank or BERT? It's hard to pin down exactly what's AI and what isn't. But this isn't my main objection.

There has already been a trend toward concentration of corporate power, economic inequality, and unemployment/underemployment. The main problem with Harari's "useless class" idea is his use of the future tense. The useless class is here already.

It might be better to discuss issues of the future of society in connections with AI rather than not at all, but I worry that the focus on AI can distract from trends and approaches that aren't tied to AI. It isn't useful to make new AI a scapegoat for existing phenomena.

I think Lee overestimates the magnitude of AI's direct impact on jobs, but maybe not the magnitude of the overall trend in jobs. Regardless, he then spends the back of the book opining on public policy after his come-to-Jesus (not literally) experience with cancer. He praises volunteers and then proposes that we must pay people for things they would volunteer to do if they didn't have to worry about money. I don't oppose paying teachers (etc.) more, but I think he's fundamentally wrong about how humans work, and I think he's way outside his area of expertise.

Selected quotes and (rarely) commentary

"In my view, that [Chinese/US] lead in AI development will translate into productivity gains on a scale not seen since the Industrial Revolution." (page 18)

"Based on the current trends in technology advancement and adoption, I predict that within fifteen years, artificial intelligence will technically be able to replace around 40 to 50 percent of jobs in the United States. Actual job losses may end up lagging those technical capabilities by an additional decade, but I forecast that the disruption to job markets will be very real, very large, and coming soon." (page 19)

Published in 2018, so that's capability by 2033 and job losses by 2043.

"The AI world order will combine winner-take-all economics with an unprecedented concentration of wealth in the hands of a few companies in China and the United States. This, I believe, is the real underlying threat posed by artificial intelligence: tremendous social disorder and political collapse stemming from widespread unemployment and gaping inequality." (page 21)

"Entry into the country's imperial bureaucracy depended on word-for-word memorization of ancient texts and the ability to construct a perfect "eight-legged essay" following rigid stylistic guidelines." (page 27)

"But what these critics miss is that this process [of Chinese central government support] can be both highly inefficient and extraordinarily effective." (page 65)

"Lucky for them, these [Chinese] cities are also home to large pools of migrant laborers who would gladly bring that service to their door for a small fee." (page 69)

The Migrant Workers Behind China’s Economic Miracle Are Miserable

AI is "the defining technology of the twenty-first century." (page 80)

"As I laid out earlier, creating an AI superpower for the twenty-first century requires four main building blocks: abundant data, tenacious entrepreneurs, well-trained AI scientists, and a supportive policy environment." (page 82)

"The complete AI revolution will take a little time and will ultimately wash over us in a series of four waves: internet AI, business AI, perception AI, and autonomous AI." (page 105)

"And judicial biases can be far less malicious than racism: a study of Israeli judges found them far more severe in their decision before lunch and more lenient in granting parole after having a good meal." (page 116)

Even if this well-known judge study were replicated and broadly believed (which it is not), Lee's reference is still rather appalling excuse-making for racist algorithms.

"Chinese and American internet companies have taken different approaches to winning local markets, [China supporting local companies, America expanding into markets directly] and as these AI services filter out to every corner of the world, they may engage in proxy competition in countries like India, Indonesia, and parts of the Middle East and Africa." (page 139)

"We looked at technical fixes that seek to smooth the transition to an AI economy: retraining workers, reducing work hours, and redistributing income through a UBI. While all of these technical fixes have a role to play, I believe something more is needed. I envision the private sector creatively fostering human-machine symbiosis, a new wave of impact investing funding human-centric service jobs, and the government filling the gaps with a social investment stipend that rewards care, service, and education." (page 224-225)

"In revamping our education systems, we can learn much from South Korea's embrace of gifted and talented education. These programs seek to identify and realize the potential of the country's top technical minds, an approach suited to creating the material prosperity that can then be broadly shared across society." (page 229)

Describes Smart Finance as a technological leap-frog AI success story (page 112-113) but many would describe it as a privacy and fairness nightmare.