So if we package up AI into an autonomous loop, so that it can file for a patent unprompted, that is sufficient precondition to allow patents to be assigned to such a loop?
Move this to the "Autonomous vehicle" line of thought. IF we accept the risks of a vehicle without a driver, how do we accomodate the _culpability_ of that risk?
Same thing in patent law. Someone needs to hold that "culpability". Maybe there'll be a robust future systems of laws that assigns the correct weightings, but it's not today.
One fundamental problem with such approach would of course be money. The AI would need to have a means of paying the fees and the USPTO declaration (37 CFR 1.63) may also prove difficult for an AI - "wilfulness" w.r.t. to false statements comes to mind. But that's a whole new can of worms I don't want to open.