As you may have heard, the Chinese have unveiled a new AI, Manus, that, like DeepSeek before it, is the source of consternation among the Silicon Valley digerati. Tyler Cowen's been posting about it, Manus assorted links, a more recently, The political economy of Manus AI. That last post opens this way:
Early reports are pretty consistent, and they indicate that Manus agentic AI is for real, and ahead of its American counterparts. I also hear it is still glitchy Still, it is easy to imagine Chinese agentic AI “getting there” before the American product does. If so, what does that world look like?
The cruder way of putting the question is: “are we going to let Chinese agentic bots crawl all over American computers?”
The next step question is: “do we in fact have a plausible way to stop this from happening?”
The post is about the political implication of Manus, and its successors, both for China and the Chinese Communist Party, and for America. Thus, almost at the end:
But are the AI systems truly aligned in terms of having the same limited, selective set of information weights that the CCP does? I doubt it. If they did, probably they would not be the leading product. [...]
Does the CCP see this erosion of its authority and essence coming? If so, will they do anything to try to preempt it? Or maybe a few of them, in Straussian fashion, allow it or even accelerate it?
Let’s say China can indeed “beat” America at AI, but at the cost of giving up control over China, at least as that notion is currently understood. How does that change the world?
Solve for the equilibrium!
Who exactly should be most afraid of Manus and related advances to come?
Here's the comment I made in response:
I've not been following the Manus news, nor, for that matter, have I thought all that much about the CCP in this context. I note, however, the Chinese engineers are subject to the same constraints and potentials and occidental engineers. They don't inhabit a different physical universe.
It seems to me that in the past two or three years Silicon Valley has been engaged in a hype competition which is, at the same time, something of a mutually congratulatory circle-jerk, to be rather crude about it. With DeepSeek and now Manus, the Chinese seem to have revealed that to be a bit silly. That's a good thing, no?
Even if your primary analytic lens is that of great power competition, if the players on one side are trapped in cycles of self-congratulatory mimicry, they'll loose out to competitors who are more realistic. Are the Chinese in fact more realistic? I don't know.
But Dread Pirate Marcus is by no means the only critic of [the] supersize mafia. There's Yann LeCun, there's David "Watson" Ferrucci, who stays out of these debates, but has his hands full developing his own hybrid tech through his company, Elemental Cognition. There are others. And there's some interesting academic work as well. Consider this video, by an MIT graduate student, Pratyusha Sharma, Discovering & Engineering the Computation Underlying Large Intelligent Agents. She presents work on LLMs and on sperm whale communication. One of the things she's discovered about transformers is that, within (at least some) very large models, you can find a smaller model that's just as effective. Whoops! That suggests there's lots of room for actual deliberate engineering.
I suspect that's more stuff like that out there. The Chinese can access it as easily as we can. And it looks like India are just decided to ramp up its AI work. Are we headed for a three-way competition, USA, China, and India? Where's Putin's Russia? Can we afford to write Europe off?
