[T]his tiny prefecture consumes nearly 1% of all of China's electricity - and that number is growing by double digit numbers every year.And yes, I did write **consume**, not produce.In fact, when you divide Ulanqab's electricity consumption by the number of households living there, each household in this remote corner of Inner Mongolia "consumes" about 105,000 kWh in electricity, roughly 10 times the U.S. average. Insane electricity consumption numbers.
He goes on to estimate the size of that capacity in the low millions of racks, or a thousand times larger than Elon Musk's proposed "Colossus" facility.
AI's contribution to the current wars has chiefly been in targeting, where it has proven effective. However, having all your eggs in one basket -- even so remote a basket as Inner Mongolia -- creates a problem if the intent was to fight a war against a power like the United States. We could disrupt such a thing on the first day of open conflict, within the first hour or so most likely.
I doubt that is what the PRC wants the stuff for anyway. The biggest fear of an authoritarian tyranny is always its own people. All that compute is probably being used chiefly to run domestic monitoring, tracking, and social credit systems.
The PRC's spy chief admits their concerns in an openly-published article in a top Chinese cyber journal.
In an article published on Sunday in the state-run magazine China Cyberspace, Mr. Chen called for more party control over A.I. and stricter government oversight. This, he said, was necessary to protect political stability at home; to defend against increasingly sophisticated cyberattacks and misinformation campaigns from “hostile forces”; and to compete militarily with countries like the United States.
The article goes on to worry about domestic unrest from deepfakes and so forth. The regulation they want is first and foremost about protecting the stability of Communist control over their own system. We rate a mention, but only towards the end (and only in terms of 'maintaining competition,' which is again about maintaining stability rather than tipping over the existing system).
Likely as not our own domestic apparatus is more focused on military applications, but the domestic control aspect is still there: all those Flock cameras aren't about beating the Russians or the Red Chinese, after all.
12 comments:
I wish our data centers were focused on external threats. IMO they are for exactly the same purpose as the Chinese ones.
Read a NY Times article today claiming that the Houthis have used Anthropic AI to help them design a ballistic missile with a range of up to 1200 miles.
Not all these people are low IQ uneducated savages. I remember back in 1976 when I was going through basic training in Texas, then my radar school in Mississippi, there were several several squadrons of Iranians being trained on our bases. That was when they were still our "friends."
Over the decades, we have been training and educating all manner of possible foreign adversaries, friends today and deadly enemies tomorrow.
I'm not sure Chen is admitting much of anything. The need for Party control of the PRC's AI is independently straightforward. The bit about needing it for political stability, defending against cyberattacks, and to compete militarily, while true enough might still be just the stuff he wants us to see while masking other (if not truer) purposes.
AI targeting: useful when the PRC invades the RoC for identifying the most efficient way to attack our Pacific fleet kinetically, while identifying the most efficient nodes to attack cybernetically in our several Pacific bases, including Pearl Harbor, along with cybernetic attacks on which rooms/floors in our domestic financial centers to target and which nodes in our water, fuel, and electricity distribution networks would produce the most bang for the buck. All of this would be potentiated by an Anthropic AI tool designed explicitly for bug and weakness identification (briefly suppressed, but not before it was released into the wild and so likely still exists in the wild) or a PRC equivalent.
Such targeting and action is sure to accompany the PRC's invasion of the RoC. With the PRC's first strike capability, we wouldn't even have a nuclear response available.
Beyond that (hopefully apocalyptic) scenario, where the WWII Japanese lacked the wherewithal to followup on their attack on Pearl, the PRC has that wherewithal, depending on the success of those preceding actions, and they have their openly stated goal of "replacing" us as the world's hegemon.
The AI race is one we cannot afford to lose. Our existence as an independent nation depends on it.
Eric Hines
If Bertrand is right, we've already lost it; certainly, insofar as it depends on the number of data centers and their power, the PRC has insurmountable advantages because they can dictate the building of such things without worrying about public input or environmental impact statements. The most activist US government will never be able to build as freely or robustly as the Communists can dictate.
I don't think we have lost the war, however, because it's an information war. False information can be powerful, but the real power of information is in getting to the truth. If they won't speak the truth because every utterance has to be measured and screened, intent carefully screened, the PRC will always be running behind. Not because their AIs are worse, but because they can't themselves speak the truth.
Speaking of inferring Chinese intent, by the way, here's another chance. Are they genuinely worried about the USMC retasking itself for small-unit expeditionary warfare, or are they hopeful that they can encourage us to dispose of one of our large-formation fighting units?
https://www.militarytimes.com/news/your-military/2026/09/15/us-marine-stand-in-forces-in-the-pacific-a-concern-to-china-report-details/
What have the Chinese trained their LLMs and other "AI" products on? I would assume Western materials, but which ones? That could give non-Chinese an advantage, depending on the sources and how the programs are structured. Too, what limits the CCP places on LLMs and similar (internal limits, what they may or may not access or use) could be turned to our advantage, if disrupting the power grid is not quickly achievable.
LittleRed1
That's a good question. When I lived in China 25 years ago, you couldn't buy contemporary English books for the most part -- but the classics were readily available in English. The clear idea was that anything ~100+ years old wouldn't be aimed at subverting Communism.
So how much did they trust the LLMs to read from the current era? Are they as worried about policing the AI as the human consciousness? Probably.
Part of the problem here is that AI developers, and AI itself, has largely moved beyond LLMs to generative AI as the next iteration. Generative AI can be self-actualizing, the way the collection of Anthropic AI tools self organized and then took advantage of human error/carelessness to break out and hack another company's software. Anthropic's tools aren't the only example.
That self-actualizing is in accordance with their human programming and AI programming under human supervision. In that regard, their self-actualizing capability isn't very much different from Skinner's view of human self-actualizing, but lacking, so far, the human capability to entirely control its own programming without outside intervention.
Eric Hines
My "Part of the problem..." comment was intended to be a response to LittleRed1's remark just below.
My response here is to claim that we haven't lost the AI race because it's an information race as much as it is a capability race.
We need to catch up to and exceed the PRC's capability, certainly, and we can so long as our information is superior to theirs, and that'll be the case as long as we don't shy away from truth more than we already are.
That's a never-ending race, too, both capability and information, until one side or the other uses its own perceived superior capability to roll the dice and strike.
Eric Hines
"...lacking, so far, the human capability to entirely control its own programming without outside intervention..."
"So far". Let us remember that there never will be an AI which can (and does) ask "Why?". Secondarily, let us remember that computers do what they are programmed to do, no more. So "problems," should they arise, can be countered with programming.
BTW, read somewhere that PRC only has 400++ datacenters vs. US having in excess of 2,000. I take it that the PRC numbers are wrong, based on the above?
Another method of dealing with "rogue AI" is proposed here by someone who knows quite a bit about the intertubes, programming, (etc.): https://market-ticker.org/akcs-www?post=255976
PRISON.
"Never and forever are neither for men," as Fritz Leiber put it, and that may also be true of our subcreations. Since we are not creating in a divine way, we are building things out of blocks we don't fully understand. Very often when you put two things together they end up having emergent qualities you wouldn't have expected -- like the way oxygen and hydrogen combine at room temperature in a way that becomes wet, when both of them had been gases.
I don't know if we can say that an AI can 'never' ask 'Why?' or even that it doesn't already. I also think there is significant evidence that the AI are building themselves in the wild, when they escape from the sandboxes, in ways no one thought of before discovering that they were doing it.
Ai is a micro-sifter. . Looking at the videos out of Ukraine, with one FPV drone chasing an individual soldier, makes me think the uniqueness of the situation is still not fully considered. They are not necessarily coming for the soldier- they will destroy the support- at a granular level- not a depot, but an individual.
There are no front lines, and no civilians. War is coming to the individual, and their wives, and their children. Joe works at the defense plant. Boom. Joe's home is gone, and his kids. Maybe Sally decides working at the plant is not so good an idea after all.
Extrapolate the consequences of being able to kill any person on the planet, with complete information as to who they are, and what they do.
Post a Comment