Hacker News Reader

Comment by Florin_Andrei | Hacker News Reader

This responsibility is to great to leave it to humans.

Automate this stuff.

There's always going to be a human involved at some point (at least until the machines take over, but then they probably won't be flying us around). If it's all automated, then instead of a crazy pilot killing everyone on his airliner, it might be a crazy programmer killing everyone on every airborne airliner.

Flights are almost fully automatic already. They saw data showing the pilot turned off autopilot and set the target cruising altitude to 100m. Putting a block in place unless you're in an area where that is safe (flat).

I was recently flying in a small airplane (SR22) and the terrain detection system was going bonkers when we were landing saying that terrain was getting close to us. They have all the data, they just need to build in some more protections.

Except there are checks and balances in place before software gets installed on a plane.

Those don't always work. There are checks and balances in place before pilots get to fly airplanes, too.

> Those don't always work Of course not, nothing always works. You are implying that checks are worthless because they failed on this co-pilot.

In the context of a pilots mental state verses malicious source code being installed on a set of flight computers, I'd have to say the computer is easier to validate. That many more people have to approve the process. With a pair of pilots the validation process is much smaller and more complex.

We need to stop thinking that machines are flawed and that humans know best. The fact of life is that machines can and do perform better than humans, the only reason flight computers don't currently is that there hasn't been the investment in making them do so. Sensors fail but that is an engineering problem that can be solved.

I think in the near term a possible remote override from air traffic control would make more sense.

That's a nice thought, but how would ATC determine what needs to be done? We are only finding out that this was an intentional act long after the plane has hit the ground. For all ATC knows, opening the door to the cockpit could be the WRONG action. If no one is responding from the cockpit, they have no clue what is going on other than the telemetry they are getting from radar and the plane's systems (which in this case wouldn't tell them much).

In fact, pilots are trained to communicate AFTER they have gotten the plane in a state to where they can either continue the flight or ditch safely if communication would distract them from the task at hand. ATC knows this, so even if they saw the plane dropping like a sack of rocks, they would probably assume that there was a problem and that the pilots are in a position in which a steep dive will help mitigate the problem (which can happen, such as depressurization).

If you add some additional information on the status of the plane (engines, electronics, etc) in addition to the telemetry you could infer much more on the current situation aboard the plane. In this case the simple reversal of the initiation of the descent (as a precautionary measure) might have prevented the unfolding of events. Obviously we can all think of corner cases where the pilots and the ATC would come to different (and potentially disastrous) conclusions while being unable to communicate and resolve these differences.

Ultimately I think that remote control or a comparable safety measure will become a necessity in civil aviation, because passengers will demand it to feel safer.

"In this case the simple reversal of the initiation of the descent (as a precautionary measure) might have prevented the unfolding of events."

I think you might have missed my point. What telemetry do you think would have given ATC enough information in 8 minutes that would tell them reversing the descent is the correct course of action? If anything, the data in this flight would have indicated that the dive was initiated by the pilot, and would suggest allowing him to continue the actions. And how long would they have to discuss what should be done? The descent started 8 minutes before the plane hit the ground. That is a very short amount of time for people on the ground without full information to make the call. They would probably spend at least half that time trying to make contact with the pilot.

The bottom line is that the person best appraised of how to fly the plane is the one behind the controls, and that is how aviation has traditionally treated these situations. If a pilot wants to bring down a plane bad enough, it's mighty hard to stop them.

Anyways, and this is my speculation, I am fairly certain that remote control has been brought up many times before in aviation, and that there is a good reason why it isn't presently implemented. And as for the corner cases, those corner cases matter. This isn't a software engineering problem we are talking about.

Precisely. People are proposing complex algorithmic solutions to a problem where applying such things might potentially make things worse. Generally, civil aviation is remarkably safe, and we should not go with knee-jerk potentially counter-productive reactions.

Airline pilots are a highly selective group of people, well trained, and they are showing every day how they work with integrity. That someone has once gone over and killed himself and others should not obscure our assessment of the situation. We have about 88 000 departures per day in the world. We don't accept even ppm-range risks of disaster if we know how to mitigate them, but when we're this low, we need to be careful that "improvements" don't actually degrade performance.

Now you have two problems: instead of worrying about rogue pilots, you have to worry about rogue pilots and rogue remote control. Oh, and latency in remote control is also a nice thing to have. And signal loss. And jamming. So, five problems (actually, that's just a random sample).

Such a terrible idea in such a short sentence...

The chance that multiple operators in a collective are actively abusing their control (and try to crash the plane) is decreasing in probability. History shows that such actions are usually performed by individuals. Even high latencies would not have prevented a successful intervention in this case (they've had a window of about 8 minutes for an intervention). Jamming is a possibility and an issue with the current communication systems the pilots already rely on.

Multiple? One is quite sufficient: hack the control channel, and you have a plane to play with.

Right, because that signal would never get hacked.

That's not what he's saying. The world isn't as black-and-white as "it can be hacked, so it's not worth it"... He's talking about raising the barrier to entry for this kind of situation via automation, much like TCAS would have prevented the Uberlingen collision.

Preventing accidents and stopping a rogue insider are very different issues. Remote control opens a bunch of new attack surfaces, while not helping with the original problem.

Just because they're "very" different issues doesn't preclude there being a technical solution that helps solve parts of both problems.

No, no, no, no. There is no technical solution to this non-technical issue. I understand the frustration of "but surely there MUST be a scriptable way out of this," but IMNSHO there isn't (short of strong AI).

This issue of humans in the cockpit? You're saying there's no technical solution to this ....... except AI, which would be a technical solution to this issue.

Except that strong AI is not current technology, or even near-future technology. In other words, "hmm, this wouldn't be a problem if only we could do abracadabra and solve it with magic; therefore it's not a problem. We can't do abracadabra, and hence can't solve it this way? La la la, I can't hear you!"

(Not sure I'd call that a technical solution, but let's assume so.)

We could make planes fly themselves today. This is (mostly) how military drones work. On a couple of my tours, we worked with drones that flew themselves almost completely. My best friend in the universe reenlisted just to fly them, and they fly themselves for most of their missions.

Trust me: Planes could fly themselves using today's technology.

Emphasis on almost. I'm not going to reiterate what's been said here: https://news.ycombinator.com/item?id=9289225 , so just the TL;DR version: in case of major trouble, a CFIT of an unoccupied vehicle is a viable safety option, for that 1% that almost works. Try that with an airplane full of people...and you would become a terrorist. Tiny, tiny difference, nothing to worry about.

Reply on news.ycombinator.com