[TheOtherHobbes]

Or maybe Von Neumann machines can't be built because they're impractical.

Consider that you need an AI that can deal with a huge range of potential environments, build extraction, processing, and assembly machinery, and do all of this with perfect error correction and oversight. Pretty much eternally.

It's an interesting thought experiment. But as a practical engineering project, it really doesn't sound even remotely credible. Not without appealing to "Yes, but one day..." magic unobtanium technology.

I wouldn't be surprised if there's some kind of hard entropy limit on how long complex systems can remain functional and stable. Sooner or later you're going to run out of memory, and if you restart a build process from scratch at each iteration without accumulating information, what have you actually achieved - beyond a rather pointless machine infestation?

[ncmncm]

You understand that we, personally, are von Neumann machines? Not very well-tuned ones, because we happened by accident. Bacteria did, too, and are more efficient, just limited in scope.

So, we have proof it is possible; it only needs refinement.

[LargoLasskhyfv]

https://www.baen.com/chapters/W200203/0743435265___0.htm