Removing the human factor from potential aircraft accidents?

alioth wrote:

This is the sticking point: a computer can only do what it was programmed to do, a human crewmember can be inventive and use judgement.

Machines can learn as well. Their strength is in consistency. Question is whether such a complex artificial inteligence capable of fully replacing human won’t gain some human weaknesses. When a human makes a briliant save, the decisions made often aren’t based in hard facts.

alioth wrote:

We can see the pitfalls in automation – e.g. AF447 – denied of sensor information, and with certain assumptions now being faulty, would a computer have done any better? (I doubt it).

Actually, that incident was rather trivial. Assuming all SOPs were programmed in, I think it would have been a non-event. Plus a computer could presumably keep track of hundreds of parameters, unlike humans, so it could work the raw data directly. The issue was the human crew was never trained for this scenario and I recall there was some issue with where the SOP was in the manual. In the end what really crashed them was one pilot losing it and the crew not working properly as a team. Another thing is that a completely different weather avoidance strategy might have been chosen (they cocked that up as well) – IIRC most crews flew a different way around that night.

Peter wrote:

And not just in the cockpit. Within ATC too – they have been talking about digital messages from ATC loading autopilot preselects etc (but only upon a manually executed pilot confirmation, to prevent data link hacking) for many years.

And it works.

Actually, that incident was rather trivial. Assuming all SOPs were programmed in, I think it would have been a non-event. Plus a computer could presumably keep track of hundreds of parameters, unlike humans, so it could work the raw data directly. The issue was the human crew was never trained for this scenario and I recall there was some issue with where the SOP was in the manual

But, if the software was written by Airbus, with the help of someone who developed the syllabus for Air France training, you would have got the same problem except instead of being trained into the crew it would be implemented in software

Looks at the totally stupid Airbus inhibit on the stall warner if the IAS falls below 60kt. That was one of the holes in the cheese in AF447. That inhibit should have been done as weight on wheels, or the wheels spinning up (as is used to implement the interlock on the thrust reverser on the DC10, according to one pilot I spoke to).

And it works.

Is it currently used and where? Whenever I fly I hear ATC giving headings, levels, etc, etc – all the usual stuff.

I didn’t say it doesn’t work but it is a very long way from remote control.

Michael wrote:

I believe that there is a minimum Cabin Crew head count that needs to be observed mostly for safety reasons in the unlikely event of having to exit the acft during an emergency.

Yes. It is unlikely that the number of CC required for passenger service during the flight would be higher than the number required for safety reasons.

LeSving wrote:

ATC is not the only way, and far from the most efficient way to organize a flock.

But aircraft are not birds and don’t fly in flocks. I imagine the certification would be a nightmare, especially if different manufacturers could implement different algorithms. Would you really expect them to solve competing interests without outside mediation? Every company would want an aggressive AI that gets the job done on time, not some wuss of a computer that yields to competition all the time. And how would it work with planes piloted by humans? Or would you give up flying so computers can? I have a hard time imagining a plane talking to other planes it’s likely to meet to work out what is the best time to push back to minimize the fuel used.

172driver wrote:

Has it ever occurred to you that that is exactly what our brain does – all the time?

Not like a computer. We don’t have the capacity to analyze, in principle, an indefinite numbers of possible outcomes when there are 100s or 1000s of variables, not within a second. We are good at finding the few most likely good alternatives, given experience and training, we are good at filtering out all the stupid options, but so is a computer. It’s like GPS navigation by car. Google maps will find the fastest route in no time, and it will change it continuously in real-time based on traffic information received from other units.

LondonMike wrote:

I hope this isn’t too demeaning to the professional pilots here, but wouldn’t it save almost exactly the same amount of money to get rid of a couple the cabin crew and have the pilots serving drinks during cruise?

With ships, the problem is getting enough crew to stay for weeks isolated on a ship. This has lead to faster and larger (cargo) ships, when the most economical and most efficient would be smaller and slower ships, and more of them. Going from 16 to 11 knots will reduce fuel cost by 50% . I have no idea if this is transferable to aviation, but there seem to be a tendency of increased “hubbing” and less direct routes, which is very inefficient for travelers (you use a total of 4-6 h or more for a trip that could take 1-2 h). A direct route would not be economically viable due to too few passengers on each flight. Autonomous, smaller and slower aircraft would offset this, when there is no crew cost.

Peter wrote:

But, if the software was written by Airbus, with the help of someone who developed the syllabus for Air France training, you would have got the same problem except instead of being trained into the crew it would be implemented in software

Of course, if they chose not to program that scenario, it wouldn’t know what to do. There is a chance it would be programmed to work with more raw data inputs and the algorithms used to process and combine it would still arrive at a usable solution. For example, the pilot never seemed to acknowledge high AoA which they have available, albeit not on any “primary” display I think.

Peter wrote:

Looks at the totally stupid Airbus inhibit on the stall warner if the IAS falls below 60kt. That was one of the holes in the cheese in AF447.

That’s a question of the data looking unbelievable to the computer (and WoW might have been used to determine they probably are in the air) hence disregarding it. I would have to look into the report but I think an issue was a too high AoA. They were so far outside the envelope computers couldn’t believe it.

Peter wrote:

Is it currently used and where? Whenever I fly I hear ATC giving headings, levels, etc, etc – all the usual stuff.

And what would you expect to hear? I think there are routes (like the so called silk road) where this is mandatory but I’m not aware of any in the EU. You still have voice traffic, only less of it. For example Maastricht (upper airspace) has been using it since the nineties AFAIK and reportedly about 5 % of traffic uses it.

LeSving wrote:

Autonomous, smaller and slower aircraft would offset this, when there is no crew cost.

But smaller aircraft would have worse fuel economy. Also, you’d have more airframes and engines to maintain.

Of course, if they chose not to program that scenario, it wouldn’t know what to do.

And this is the biggest problem. Humans will be writing the code, so they need to decide on these things. Airbus made the with-hindsight stupid decision to suppress the stall warner. There is no guarantee that a pilotless aircraft will also not be full of stupid decisions. Mankind has not got more clever in roughly the last 100,000 years. Software quality moves along at a glacial pace. Software whose complexity is above the “totally trivial” level cannot be formally proved to be bug-free.

There is a view that one day humans will be able to generate amazingly clever software. I don’t think it will ever happen.

Peter wrote:

Looks at the totally stupid Airbus inhibit on the stall warner if the IAS falls below 60kt.

Totally stupid? Certainly not. There has been a good reason for that. But unfortunately, there was also the assumption involved, that a well trained and competent crew would be on board, which is able to use the multiply-redundant information provided to them from a multitude of sources to tell a stall from a normal flying condition. That assumption (expecting a minimum amount of competence from the crew) was stupid, not the software. Get rid of the crew ASAP and accidents like this one will be history.