I have been a hater, I mean... who knows... but for the sake of argument let's accept it as a possible outcome.
Given that assumption, is there anything more scary than a space-based, distributed data center in the sky, with a non-aligned agent? How does one unplug that?
Think about this for a moment. Starlink is the Internet 2.0. It is a very difficult to destroy network with millions of clients/base stations.
The proposed always-in-sun-orbit inferance net is our most likely "great filter?"
If there is even a .01% chance of non-aligned ASI, then should we not ban distributed inference in space?
How does an artificial intelligence improve without upgrades or access to resources? We're past the steep acceleration curve of innovation - linear improvements now require exponential increases in resources.
Also, it's remarkable how much of the supply chain for components still rely on human labor. Maybe not the assembly and fab of chips - but some parts of the supply chain still rely on tooling equipment made in the 1960s! None of that is moving to space anytime this century.
Starlink is perhaps one of the most vulnerable infrastructures available. Every node de-orbits after 5 years and has to be replaced. And they can be destroyed by any number of terrestrial or low orbit solutions.
> Starlink [...] is a very difficult to destroy network with millions of clients/base stations.
I don't know a single professional who agrees with this statement. Starlink is currently being jammed very effectively, and it's apogee is not nearly far enough away to hide from ASAT weapons a-la Molniya satellites.