The slide’s authenticity was confirmed by a Navy spokesperson, who cautioned that it was not meant to be an in-depth analysis.
The slide shows that Chinese shipyards have a capacity of about 23.2 million tons compared to less than 100,000 tons in the U.S., making Chinese shipbuilding capacity more than 232 times greater than that of the U.S.
The slide also shows the “battle force composition” of the countries’ two navies side-by-side, which includes “combatant ships, submarines, mine warfare ships, major amphibious ships, and large combat support auxiliary ships.” The ONI estimated that China had 355 such naval vessels in 2020 while the U.S. had 296. The disparity is expected to continue to grow every five years until 2035, when China will have an estimated 475 naval ships compared to 305-317 U.S. ships.
Another section of the slide provides an estimate on the percentage each country allocates to naval production in its shipyards, with China garnering roughly 70% of its shipbuilding revenue from naval production, compared to about 95% of American shipbuilding revenue.
Because of China’s centrally planned economy, the country is able to control labor costs and provide subsidies to its shipbuilding infrastructure, allowing the Chinese to outbid most competitors around the world and dominate the commercial shipping industry, Sadler said.
Alternative title - “Central planning is more efficient than markets” confirms US Navy
I read a while back that the PLAN was using AI specifically to quickly generate many iterations of more efficient and compact electrical grid layouts for their ships, and I think that’s a really neat and levelheaded way to use AI. No marketing bullshit, no insane and dangerous cost-cutting measures, just “generate a path between these points while staying at least x centimeters away from y and z subsystems at all times, give me 50 variants.”
That’s just called metaheuristics and they’ve been a thing since the 70s. Super super interesting, but calling it an AI is a stretch.
Even back then there was a bunch of grifters peddling the stuff, I remember reading a book about 1900s rocket engineering. It included an anecdote where the army paid a company a bunch of money to make a computer spit out the most energetic compounds for use as rocket fuel. Naturally none of them were physically possible to synthesize or exist for more than a nanosecond at .1 K.
One of the lead designers on Spore described this design process as [paraphrasing] automating the thought processes of subject matter experts. In the case of Spore, that subject matter was creature design, but every single model had to be fine tuned to exclude large swaths of the design space which the model made possible, but were undesirable in some way. Automation of expertise followed by expert review is the way to go. It makes expertise more efficient and also tends to lower the bar to access to that expertise.
Edit: also, this has been a standard technique for decades and “AI” is often used as a marketing term to rebundle old tech that’s gotten faster.
