Playing devil's advocate: you don't have to. It just has to be better than a pair of experienced airplane pilots working together. Which is still very hard, and there's still a good chance we won't see it in our lifetimes, but at least it's not impossible.
Surely, it depends on the nature of the emergency. As I understand it, in this Qantas example, the pilots did not need to fly the plane with real-time responses, just to make good decisions.
Let's not completely discount remote pilots, while recognising they are not a universal panacea.
A partial mitigation of these issues could be high bandwidth / low latency networks just in take-off / landing corridors?
They may still need help at landing, or by then it could be relatively normal.
But if you can’t provide that level of help everywhere (including over oceans) the design of the system is choosing to lose planes in a trade off for needing fewer human pilots.
Let's do completely discount remote pilots, please.
For AI to replace pilots, you don’t need to prove that sometimes humans fuck up. You need to demonstrate that AI would fuck up less often and in a more acceptable way. This requires looking at the big picture, not only bad cases.
But I reckon single pilot operation with emergency autoland will happen. The tech already exists for general aviation.