Hacker News Reader

Comment by amix | Hacker News Reader

I think this could be prevented by removing humans from the cockpit and let computers take over. Computers are already controlling most of the things inside an airplane. If this isn't possible, then the flight software should at least have protection against malicious pilots (this would protect against the mentally ill and also against terrorism (e.g. 9/11)). Basically, the planes could have failsafes to prevent malicious pilots from crashing them down or crashing them into things.

View on HN

I think this could be prevented by removing humans from the cockpit and let computers take over.

I think we are not there yet:

http://www.spiegel.de/panorama/lufthansa-airbus-computerpann...

tl;dr: frozen sensors put the plane on a crash course. Plane is saved by pilots completely shutting down the computer system.

The question is at what point flight software can be demonstrated to have a small fraction of the failure rate that humans have.

We could certainly have software flying a plane tomorrow if we wanted to, at the cost of failure rates several orders of magnitude higher. Humans have the capacity look around, intuit what's going on and what doesn't make sense, and work around it. Software that can detect and understand what to do when 3 things that each occur only one in 1,000 flights, happen simultaneously while 3 out of 10 of the sensor networks are down for maintenance and rare weather patterns are occurring on the route.

We have 100,000 flights per day to cover. Right now, we have lethal accidents roughly monthly, but we have contingency conditions that might plausibly cause a lethal accident a thousand times more often, conditions that are caught, troubleshooted, and successfully mitigated by a human.

Unfortunately we're about 40-50 years out before an automated flight control system can reproduce the critical activities of a well-trained human team of pilots.

But yes, we'll reach a point where the liability of having human beings fly planes vs the sophistication/reliability of an automated system will reach the point where the answer is a foregone conclusion. It's just a half-century away.

we're about 40-50 years out

More like 10-15 years.

Airliners already spend most of their life on autopilot today. Many (including the Airbus that crashed here) have auto-landing systems[1], too.

Fully autonomous passenger aircraft are being tested[2] in shared airspace since 2012.

[1] http://en.wikipedia.org/wiki/Autoland

[2] http://www.baesystems.com/magazine/BAES_051920/look-no-hands

A comple of months ago I experienced the auto-landing system (as a passenger of course). The captain made the announcement before landing as meteorological conditions weren't that good (fog). I must admit I felt a little uneasy...and I am an engineer (maybe that's the reason :))

Landing is (as you've already noticed) the least significant reason as to why we have a human piloting team in an airplane.

If I could afford it, I would love to make a $10,000 long bet as to when we will see a major airline routinely (1/day) fly a route with > 200 passengers.

I'm saying it won't be later than 2065, but also won't be sooner than 2050.

I would take you up on that bet, and go for around 2025. Technology is progressing pretty fast in this area.

100% agree that the tactical technologies are racing forward, and it's apparent to everyone that things like self driving cars on freeways will be available within the next two-three years, and that sensor and response technologies will make a lot of the typical flying events a lot safer - takeoffs and landings are just two of them.

Where I don't think the technology is moving is on the executive level, which is where having a highly trained and knowledgeable flight crew in the plane is invaluable.

I'm guessing that any trained and qualified pilot of a large commercial airliner (which I am not), can immediately come up with 100 different scenarios for which there is no technology apparent in the next 15 years.

Put another way, in 10 years, 99.99%+ of the time, automated systems will be doing all of the actual takeoffs/flying/landing, but that 0.01% of the time is why we'll keep a flight crew in the cockpit.

Even the amount of automation we have today leads to the very real danger of pilot boredom: what do they do to keep busy when the flight is on auto for most of the trip and they are stuck in the cockpit just monitoring things? At least if they could fly the plane, they would have something to do! Imagine if your job was just to monitor the computer program itself, you'd get paid (yeh!), but not having anything to do would take a toll!

Check out:

http://www.newyorker.com/science/maria-konnikova/hazards-aut...

If we automate even more, the problem just becomes worse. We could pull back on automation to keep the pilots alert, but we are increasingly relying on it to optimize resources (air space, fuel, airport take off and landing slots, and ya safety).

And...that's just it. Computers don't get bored. I suffer from some depression also (who doesn't!), but I find that my job keeps me busy enough that I can keep it in check. Imagine being a bit depressed and sitting in a cockpit doing nothing for a few hours at a time.

The utility of autopilot to a pilot is comparable to that of cruise control to a driver, it takes away some of the drudgery. It is not robust enough to be unmonitored by a human. It can not make the many decisions that arise every day in aviation, such as if and how to avoid a thunderstorm.

Autoland needs ground equipment that is expensive to install and has stringent requirements on the surrounding topography that means many airports can not install it. It also takes 2 people's full attention to make the autoland happen in a consistently safe manner, it's not a case of pressing the LAND button and sitting back sipping tea. This video shows a bit of that https://www.youtube.com/watch?v=BMydKAcqKCg

It is not robust enough to be unmonitored by a human.

I'm no aviation expert, but isn't that only because the systems in civilian aircraft are decades old?

The state of the art in military drones seems to be autonomous landings on aircraft carriers[1]. The safety requirements for an Airliner are of course much higher than for an UAV, but the underlying technologies seem to be rather mature already.

[1] http://www.theverge.com/2013/7/10/4511476/autonomous-drone-f...

Ya Get rid of the persons involved...i.e. Self flying planes. It will happen eventually, it is a much easier problem than self driving cars, and there are already planes that can take off and land themselves in certain conditions.

But this isn't really a huge frequent problem, just one that gets a lot of press.

Why stop there? The best way to get rid of persons involved is not flying the passengers.

Until then, having the human pilot capable of overriding the computer is actually a good thing, see danieldk's comment. That the computers should help the humans as much as possible is indisputable, but it's a gradual and hard process to bring the real improvements.

In the long term it's ineveitable, juist like self ddiving cars are. Computers are actually very suitable for this.

Today, ya, maybe not yet, but we are definitely close. Flying is a hard job on humans, the hours suck, the job can be very monontomous. The problem isn't that the pilots are evil/bad, just that humans are unsuited to these kinds of tasks. Like driving.

This is very easy to imagine but needs quite a bit of testing. Human sensors are taken care of 16/7 approximately, not so with mechanical ones. A pilot can always say "I feel sick." A frozen sensor cannot.

Reply on news.ycombinator.com