RR, I find myself without adequate time to do your topic justice, but I will initially say this; in my mind two differing interfaces need discussion. the "man - machine" interface and the "human intelligence - computer intelligence" interface.
The man machine interface discussion is easy, at least for me. I want the machine to give me the information in the least confusing manner, in the most usable manner. And I want the machine to allow my control inputs to be made from the perspective of and in the manner that the average human achieves control in abnormal situations.
For example, I have a talk radio station playing on my computer as I type this and I realize that even though I am interested in listening to the topic being discussed. I have NO idea what was said during the time I focused on typing this reply.
IOW, humans only have limited attention, or focus, and it is fairly well accepted that humans have limited in focus when in stressful situations (tunnel vision) ; so why does an aircraft designer build a interface in which multiple inputs are offered? Why does the operator build a process that requires attention to peripheral issues? The Airbus cockpit, and other modern airliners as well I assume, virtually explode with data in an abnormal situation. Bells go off, lights flash, clicks sound, horns sound, ECAM messages appear, more ECAM messages appear, etc. Operators train pilot enslavement to the machine, AB procedures would convince a pilot that the proper technique in clearing an ECAM message is more important than actually flying the airplane.
But that is only a tangential issue. Back to the topic.
The intelligence interface becomes the problem, especially in the AF447 example, and that is the what I think you really desire to discuss.
The Airbus has a sort of AI, or at least a memory bank of programmed data/intelligence. The problem I see is that the "machine" was designed to be flown by reference to the programmed intelligence with oversight of/management by the human intelligence. The problem with the AB approach is simply that the machine interfaces with the human through the AI interface. When the AI is unable to fly the machine, the human is forced to interface with the machine using the same interface that just failed. When one ponders this, one wonders why the designers did not think it necessary to provide a "last resort" interface that allows the human to interface directly with the machine.