病菌与钢铁 redux

by liberal japonicus

A decent discussion of the approaches to AI in China versus the US is found here, but to me, because it starts from a misguided assumption, is China and the US are playing the same game with regards to AI, they can’t see the forest for the trees.

The US and China really have two different goals. China wants AI to support its society, US AI labs (because the government is hands-off) aim to break society and salvage the stuff that is valuable and then sell it back to us. You may not like the society that Chinese AI is supporting, but Chinese AI IS subordinated to Chinese society. US AI doesn’t give a shit about the society it exists in and is happy to tear it down if it means more money.

I don’t think the Chinese approach emerged because of benevolence on their part, hence the title, which is the Chinese translation for Jared Diamond’s Guns, Germs and Steel. I’d argue we have a similar situation with AI here. A quick summary of Diamond’s thesis, the West became ascendent not because of a better morality, but because the three mentioned factors gave them an advantage that could not be made up by other cultures.

At the beginning of the talk, they discuss how China does not have the same amount of compute as US AI firms, so China is behind the US. It seems to me that this lack of compute has been a blessing, allowing them to work within those limits rather than making increasingly powerful models which require more compute, which push firms to create even more powerful models, etc etc. If you only have a hammer, every problem looks like a nail, and if you think that continually adding compute is going to address your problems, you are basically treating every AI problem as a compute problem.

On the other hand, China, constrained by the lack of compute, also has to deal with the US trying to keep the best chips out of China. So you don’t have Chinese labs plagiarizing research paths of mathematicians to solve Millenium prizes, because they aren’t obsessed with frontier models. China has set their research on ways of getting around these limitations, which is what I mean by China and the US not playing the same game.

A similar bottleneck is the lead the West (specifically ASML) had in photolithography machines, which they need if they want to have indigenous chip manufacturing. The pdf suggests that the 7nm machines could be coming by 2032 at the earliest, but the pdf notes the ASML developed the processes and machines in 2010, before AI came on to the scene, and being able to use AI in the development could streamline the process further.

I’d also note how we see China developing AI. SNS always has funny clips of Chinese robots running into walls or freaking out and self destructing. But this belies the significant amount of research is going into robotics coupled with AI rather than pure AI.

The talk goes on to point out the open vs closed weight approach and how you can take open weight models and tweak them to fit your own needs, which is why Airbnb, DoorDash or Uber (US companies using Chinese AI models) are turning to them. If you think it is vital that they know more about Tianamen, well, knock yerselves out, but I’d bet that it doesn’t concern them too much. In fact, if I were a company outside of the US, I sure as well wouldn’t want to be locked into Microsoft, Google and Amazon. Just ask Tomoko Akane or the other 11 judges on the ICC.

The discussion takes up RSI (recursive self-improvement), but China, as opposed to the West, has governmentbodies (plural) overseeing AI implementation and use. Reading this wikipedia page, I’m struck how many more checkpoints and choke points exist ii the Chinese AI industry. Much of this grows out of China’s attempts at content control and moderation. Sheehan suggests that Chinese labs are latecomers to the discussion, but because they are not dealing with AI qua AI, but concentrating on applied AI, RSI does not suggest the existential challenge that Western researchers have suggested for US AI companies.

Taking another viewpoint, looking at most Western science fiction, machine sentience is generally adversarial and often along the lines of Frankenstein. In earlier comments, people are discussing Klein and Taylors End Times Fascism, which draws on that same motif. On the other hand, the Chinese science fiction I’ve seen tends to focus on how AI and sentience needs to be integrated into society. I’ve found this collection here on the Internet Archive of 10 translated Chinese science fiction stories, along with a preface and an afterword for each story entitled AI 2041: Ten Visions of our Future. It dates from 2021, so perhaps someone with a wider reading in science fiction can check me, but there is a different quality to these stories than to Western fiction on AI, which often revolve around questions of individuality and individual rights rather than what I take as a Chinese view of it. It seems to me that for Chinese, AI progress is a process that is persistently inching forward rather than everyone looking for singularity.

I also think that the whole tech-bro culture, obsessed with eternal life, has them always grasping for the next big thing. While there is a thread of Chinese history and myth of searching for eternal life, it seems to function much more as a cautionary tale than as a possible goal. There was Putin and Xi’s hot mic exchange about extended lifespans, but while I imagine there are a number of people in the West who want to follow in Peter Thiel’s footsteps, the number of Chinese who want to be the next Jack Ma is probably fewer.

The talk did leave out how the Chinese courts are stopping AI from replacing humans. From this article

In a landmark labor dispute case published by the Beijing Municipal Bureau of Human Resources and Social Security, a map data collector named Liu was fired after his company deployed AI to handle data collection. Arbitrators and courts ruled the termination illegal. They stated that adopting AI is a voluntary operational decision to boost efficiency, not an uncontrollable event, and companies cannot shift business transformation risks onto employees.

Wait a minute, that’s a possibility? From the wikipedia page on the Chinese AI industry.

Writing from a Chinese Marxist view, academics including Gao Qiqi and Pan Enrong contend that capitalist application of AI will lead to greater oppression of workers and more serious social problems.  Gao cites how the development of AI has increased the power of platform companies like Meta, Twitter, and Alphabet, leading to greater capital accumulation and political power in fewer economic actors.  According to Gao, the state should be the primary responsible actor in the area of generative AI (creating new content like music or video).  Gao writes that military use of AI risks escalating military competition between countries and that the impact of AI in military matters will not be limited to one country but will have spillover effects.

Ya think?

Given all this, it is unsurprising that Chinese attitudes towards AI are much more favorable that Western attitudes. The pdf notes that 78% of respondents in China agree that AI-powered products and services have more benefits than drawbacks, compared to 35% in the United States, 37% in Australia, and 31% in France.

Anyway, hope that is of interest.