We have been struggling with this issue for some time over in the fixed wing world as has been alluded to earlier. Regrettably the argument about the level of automation and the maintenance of manual handling skills is something inherent to the introduction of automation. There are several different approaches and positions that can be laid out, most of which have been discussed with reasonable sobriety on this thread. In general my team has opted for the 'use of maximum automation' strategy as defined within the OM-A. This is pretty common I'm the 'other teams' as well. In general 'manual flying' above 10,000' is not allowed unless something stops working - then we cannot go into RVSM airspace and may not get to destination. We cannot disconnect the auto thrust/throttle (unless the MEL permits dispatch!).
So what to do? Well for one, we do something called a 'manual handling sim' where we practice those prohibited skills twice a year for an hour. One of the interesting things in these events is the first ten minutes is usually pretty ropey and then miraculously the sleeping brain seems to recollect the muscle memory required to stay upright, so all is not lost and beaming pilots emerge from the simulator. The limitation with this strategy is the Air France syndrome (btw I suggest someone actually reads the report rather than rely on National Geographic), where they did not have time to 're-discover' their dormant skill. The slightly sad thing about the AF447, is that if the pilots had not been pilots and left the aircraft alone for 2m:30s then they would probably only have had to change their underwear. The natural stability of the aircraft would have generally maintained the flight trajectory until valid anemometric data had returned. It is worth nothing that things happen very quickly in both versions of flying machine.
Now for the additional 'squirrily' bits: the things I have flown since 1999 do not usually have the same level of manual reversion that most helicopters possess - there is too much redundancy. Fly by Wire gives us Normal Law which is like an advanced version of SAS (with the AP out) It is very smooth and remarkably agile, the 575 tonne version flies much the same way in terms of attitude change as the 75 tonne version. The chance of dropping out of this Normal Law is very small - but not so small as to be invisible. One of our pilots on his first operational was bringing his 575 tonne version back from Seoul and en-route suffered a dual independent failure that left him still in Normal Law. However during an ILS approach into the one thunderstorm a year we get in this part of the world, a third independent failure dropped him straight into Direct Law, bypassing the usual pause at the intermediate Alternate Law which is now the closest we get to helicopter AP off. Our man then had to do a go-around from the ILS, reposition for a vectored manually flown raw data ILS at night and IMC with special lighting effects. Now his experience has saved the day, from the FDR trace we could see that there was a distinct ropiness to his flying that once again recovered pretty quickly.
Here we have the conundrum: how do you train the new generation to fly these modern automated machines and yet still expose them to the unexpected events? If you need to train for the improbable, what do you train? In a sense this is what Evidence Based Training is attempting to address, we all know the engine failure shortly after what ever TDP is used, has little application apart from delivering a flight manoeuvre validation 'tick in the box'. The newest versions of conversion training currently in the pipeline emphasise the management of flight trajectory for the first events in the full flight simulator (what it was really designed for). The subsequent course then expands these manual skills to include Upset Recovery.
It is vital however that automation is trained properly and the implication of the different modes fully understood. We have some interesting examples from San Francisco and 'other places' that managing the flight trajectory during an approach with automation still causes extreme confusion. If the automation does not have a sound procedural basis then it would be probably better to remove it and install SAS with a beep trim again - turn the clock back 30 years and say what we had in the past was best (it wasn't). Without proper SOPs that specifically define the roles of both the PF and PM, you are back to 'pilot and helper' or person who flies and person who does the radios. If the PM does not have a comprehensive understanding of the PFs intention and 'what should happen next' then the PM cannot challenge any deviation. Similarly, the role of PF changes dramatically with the disengagement of the auto flight system, their area of responsibility expands to monitor things they were not monitoring as carefully before. When you brave souls decide to do a raw data ILS in IMC 'for the practice' then your PM is probably struggling to catch up.
The key to this training however (IMHO) lies with the attitudes of the trainers and evaluators. We all know it is possible for a pilot to 'fail' in a simulator within 10 minutes, however trainee and experienced pilots need to be allowed to 'fail safely' in the simulator without jeopardy to their careers or licences so they can learn from their own experience of how easy it is to mishandle any flying machine. Hopefully these lessons in the training environment will translate to the line. Personal reflection on personal experience has an enormously powerful influence on the trainees abilities.
TOD