Could a fully autonomous aircraft have "done a Sully"? Unlikely.
Despite this undoubted fact, it is also a fact that losses are caused by pilots.
It is a fact, that not all losses could be prevented by pilots, we all know that. And in the course of events, some losses are caused by pilots by making the wrong decisions due to several reasons, from limited information due to failed sensors, by overload while working with downgraded systems, or just by wrong judgement to name a few.
Not all sensors necessary for an autonomous operating aircraft will fail and cause the loss of the aircraft, but at the moment they do fail now and then and will continue to do so. When we are able to produce those sensors failsafe, then lets do so, there is no need to wait for an computerized aircraft.
Not all computer programs will have false codes and cause loss of aircraft, but some will. If we can make them safer, lets start with them now.
Car drivers are not only causing accidents, they also prevent them by acting with situational awareness. If there are systems in the pipeline which increase their situational awareness in a reliable way like it will be necessary for autonomous vehicles in daily traffic, lets implement them. But there are none tested and certified for daily use by the daily driver now.
All those systems first have to be tested and introduced in piloted aircraft, and they will improve the accident rates if they work ok like other computerized systems did. At that point we may look again what effect the removal of humans from the flight deck may have, a further improvement or a greater risk.
To remove one ccontroling entity (the pilot) and replace it with another one (computer) removes one statistical failure source, but creates other possible failure sources. At the present concept the human is the last redundancy when everything else fails, is unreliable or is unable to make decisions. As the accident numbers turn out this concept works fairly well. It does not prevent all accidents, and sometimes it even leads to accidents, but to oversee the functioning of computers by another computer dependent on the same or a different set of sensors didn't work until now that good. How many ADIRU's will we add and how will the decision making take place which ones to reject and which ones to use? Will we add another sixpack of FCPc's to compensate for the failure possibility of the present ones? Will we add how much more rate and load factor sensors and how will be the decision made which ones are good and which ones are failed? How much cameras IR and Radar sensors do we have to add to the airframe to get the necessary redundant informations? It is not a question of computer power, it is a question of information reliability and redundancy.
If we like to talk science fiction ( what will be in 50 or 100 years), than go ahead. We all do not know. All the necessary input devices, sensors, software and hardware might get that failproof, that complicated systems can and will work autonomously. But all the present developements for the military or for some specialised niche systems are no indicators, that such systems are operational and implemented within the next time (20years) for civil air traffic.