Not a bad stab at what it might look like.

But the cost of redundancy still far outweighs the cost of a second "PoB".

Redundancy for the pilot requires one of...

a) 99.99999999% secure and 99.9999999% reliable ground-air communication system, and it needs to be broadband with streaming HD video, not just a few bytes of acars data. Also the USAF found they cannot safely land drones via SATCOM, the latency is too high for split-second reactions, so they moved contract pilots into theater so they can control them via line-of-sight at the airfield. The mission pilots stayed in CONUS, SATCOM is good enough for that.

b) An AI suitable to completely replace a human, which does not exist. Nor does anyone have an idea how to make it. And if they could make it, they have no idea how to certify it since you cannot predict exactly what it will do, it could respond differently to nearly identical situations on different days. The best idea they have so far is to use a non-deterministic model to learn how to do something (ex be a pilot), and then freeze that configuration and use it as a deterministic box which would have to be ops tested for many years to be certifiable. You could probably never be completely confident in what it will do, but maybe after millions of flight hours you can say it's good for the 10(-9) civil aviation standard. Maybe. The risk is if you go all in on that box, and then it fails, the industry is hosed... there would be no way to "tweak" such a box like MCAS, swat it on the arse, and send it back up. The entire process would start all over, with millions of hours of "IOE" required to develop confidence (ie certification) in the new configuration.

And it might need to be done for each aircraft variant, since you really have no idea what effect underlying circumstances and experiences affect the final model.