Taffer posted:

This is the issue in this conversation. You are expecting people to come up with thorough, robust, and informed proposals detailing how automation should be implemented. You're having this argument on a dead gay comedy forum. Nobody is qualified or interested enough to come up with those proposals. They are half-baked because why the hell would anyone here try to make a full-baked idea?

quote:

Your claims that automation are impossible or too dangerous are short-sighted and ignore basically every example of automation of an industry.

Solkanar512 posted:

No one is saying that this can't be done with redundancy or automation, we already have this. One example are the redundant hydraulic lines that ensure you can still operate flaps and ailerons and so forth. Those used to be routed through the same parts of the plane, but once there were a few engine failures that led to shrapnel cutting those lines, they started being routed through different areas. With sensors you need to be more careful and ensure you have good ways to deal with conflicting information. There are generally three artificial horizons (pilot/copilot/center) that are controlled separately and pilots can switch their own to mirror another if their own fails for some reason. These are improvements that come straight out of the NTSB and related organizations.

quote:

If you want that automated pilot so badly, you need to show that it improves upon the current safety record we have now. On American carriers, we're talking about incidents measured in the single digits per millions of flights and no deaths for seven years straight.

quote:

I don't doubt that more automation is on the way, I just doubt those that think it's a trivial matter to just dump the pilots. It takes time and effort to do this correctly and folks need to take that into consideration.

Owlofcreamcheese posted:

Is that a real thing that really happens though? Are there more than a teeny tiny handful of pilots saving the day with really truely out of the box solutions to issues and not just "instead of pressing THIS button I pressed THAT button" that would be 100% doable by a machine? like physical mcguyvering? You don't even need it to be a robot man that makes up a solution, the ineffable human souls can pre-load "hey, what do you do if these parts fail" modes ahead of time and build hierarchies of "if this breaks, this other part can replace it's function if you use it like this" tables.

Solkanar512 posted:

Actually, I'd just be happy if people accepted that the systems that allow commercial aviation to exist are complicated. I don't think that's too much to ask.

This is pretty much the opposite of saying that it's impossible. I'm not trying to be a luddite here, it's just a different sort of world to work in where a gently caress up could lead to people dying. That changes how you approach things.

Solkanar512 posted:

In absolute terms this is as rare as hens teeth because the redundant human/mechanical/technological systems in place at every step of the process are really good at finding issues and preventing them from happening in the first place. When a serious issue does occur, it's generally because it's either an issue that is completely outside of the normal scope of what is reasonable to consider (say, being shot down by Russians by mistake) or something that has escaped all redundant systems available. In the latter case, you need to come up with a way to fly the plane given the novel situation.

Now, there are systems out there that partially do what you're talking about - one in particular allows a commercial airliner to properly steer without having a horizontal or vertical stabilizers, instead using the ailerons. I can think of two flights where this would have come in handy - JAL123 and there was a 767 freighter out of Iraq that had it's rudder shot out. That's two flights out of millions. Even then, I wouldn't want to remove the pilots - they still provide an important layer of redundancy.

Owlofcreamcheese posted:

The thing is: the improvement on humans coming up with ideas higgly piggly at the moment a disaster happens is having humans come up with those ideas ahead of time and then sitting down and working them out with engineers and working out the physics safely and not in a burning plane and not in 30 seconds.

You can't have a pilot go to school for 400 years to learn every "what if", but if everything stored procedures you can pre calculate, pre test and safely verify what is the right answer in every possible break down. and you can keep adding to that list until you have gone way past what you could expect a human man to be able to come up with on his own while sitting in a chair in a crashing plane and way into "a pilot on the plane couldn't have helped here either".

Malcolm XML posted:

Also while ai will probs replace pilots I don't think it will replace cabin crew since they are needed to deal with passengers directly

StabbinHobo posted:

like it seems to me that long before you get a chef working you can have sold a million dishwasher units that only need to be a tenth as adept.

StabbinHobo posted:

hey if y'all are done dunking on the luddite over autopilot...

check this out: http://www.moley.com/

now i'm sure they're a typical r&d startup overshooting the moon and will never go anywhere... but does anyone know if there are arms/hands/fingers anywhere near that good on the market yet as off the shelf parts?

like it seems to me that long before you get a chef working you can have sold a million dishwasher units that only need to be a tenth as adept.

StabbinHobo posted:

but does anyone know if there are arms/hands/fingers anywhere near that good on the market yet as off the shelf parts?

Solkanar512 posted:

Just remember a few things. Military flight and civilian flight are covered under completely different standards of safety, operating philosophies and so on. The second is that point out that things are complicated is not the same thing as being a technophobe. I don't doubt that more automation is on the way, I just doubt those that think it's a trivial matter to just dump the pilots. It takes time and effort to do this correctly and folks need to take that into consideration. When I see folks like Amazon petition the FAA for exclusion from experimental flight rules (2014 I believe)* because "they're making new parts too quickly" and "they have an astronaut on their team", it comes across like they don't know what they gently caress they're doing. Luckily they've improved since then. Finally, there's a difference in the amount of safety we expect in vehicles that carry humans versus vehicles that don't.

Owlofcreamcheese posted:

Yeah, robot hands are good now.

https://www.youtube.com/watch?v=5kb7U1qzBvc


Also this one isn't realistic or human at all but it's super cool:

https://www.youtube.com/watch?v=-KxjVlaLBmk

Malcolm XML posted:

Humans are really bad at emergencies but currently robots are worse

ElCondemn posted:

They've been cutting back crews for a while now, I'm sure in the next 10 years we'll just have tubes that drop down from where the air masks do. You'll just suck on the end of the tube to get your legally mandated ration as you hurl through the air prone and stacked in cages to maximize space usage.

Half-wit posted:

Please...only the poors will have to take the cattle air-cars. The rich will be able to pay people to fly places for them.

quote:

One objection raised against flight envelope protection is the incident that happened to China Airlines Flight 006, a Boeing 747SP-09, northwest of San Francisco in 1985.[5] In this flight incident, the crew was forced to overstress (and structurally damage) the horizontal tail surfaces in order to recover from a roll and near-vertical dive. (This had been caused by an automatic disconnect of the autopilot and incorrect handling of a yaw brought about by an engine flame-out). The pilot recovered control with about 10,000 ft of altitude remaining (from its original high-altitude cruise). But to do that the pilot had to pull the aircraft with an estimated 5.5 G, or more than twice its design limits.[5] If the aircraft had a flight envelope protection system, this recovery could not have been performed.

quote:

Air France Flight 447, an Airbus A330, entered an aerodynamic stall from which it did not recover and crashed into the Atlantic Ocean killing all aboard. Temporary inconsistency between measured speeds, likely a result of the obstruction of the pitot tubes by ice crystals, caused autopilot disconnection and reconfiguration to alternate law

quote:

US Airways Flight 1549, an Airbus A320, experienced a dual engine failure after a bird strike and subsequently landed safely in the Hudson River. The NTSB accident report[11] mentions the effect of flight envelope protection: "The airplane’s airspeed in the last 150 feet of the descent was low enough to activate the alpha-protection mode of the airplane’s fly-by-wire envelope protection features... Because of these features, the airplane could not reach the maximum AOA attainable in pitch normal law for the airplane weight and configuration; however, the airplane did provide maximum performance for the weight and configuration at that time... The flight envelope protections allowed the captain to pull full aft on the sidestick without the risk of stalling the airplane."

quote:

American Airlines Flight 587, an Airbus A300, crashed when the vertical stabiliser broke off due to large rudder inputs by the pilot. A flight-envelope protection system could have prevented this crash.

Ocrassus posted:

RE computational control of aircraft, specifically commercial airliners, there is actually quite a debate in the flying community that has gone on for a long time, and is pretty pertinent to aviation. The two major manufacturers, Airbus and Boeing, take different approaches to how they design the systems with which a pilot controls the aircraft.

Airbus pilots fly via a joystick, and all of their inputs are sent to a (triple redundancy) system that ensures these actions stay within a flight envelope. There is a decision engine that is context sensitive, and prevents the pilots from making certain manoeuvres depending on the conditions (the system will permit greater freedom of action in what it perceives to be an emergency). In this sense, the design of an airbus aircraft trusts the systems over the pilot.

Boeing aircraft on the other hand, whilst still having programmed flight envelopes that constrain the pilot, are far more permissive, significantly more like a mechanical system. It places more trust in the pilot.

Both systems have encountered criticism and both have been implicated in causing a crash, or being unable to sufficiently prevent it.


Cases against more 'automation' of flight systems



In both these cases, the flight envelope protection system prevented the pilots from pulling up sharply (because in normal flying conditions this would cause a stall). The 'intelligent' systems failed in this respect.

Cases for more 'automation' of flight systems




This is relevant to the discussion because an AI pilot is likely to be a significant extension of current flight envelope systems, narrowing down the range of inputs and actions in a context sensitive way, until it is the one ultimately making the decisions. The jury is very much out, atleast in the aviation community, on the efficacy of autopilots and automated systems being able to handle unexpected events, but it is a discussion that has been going on for a fair while.

Owlofcreamcheese posted:

The thing is: the machine and the person both messed up. In version 8.2.123 of the machine next year they can fix the machine and never have it happen again, if the pilot messes up they can maybe mention the issue in a text book and hope that in 10 years some random pilot happens to be in that situation that can recall a one in a million thing he read once in a text book one time but probably he won't.

Ocrassus posted:

In both these cases, the flight envelope protection system prevented the pilots from pulling up sharply (because in normal flying conditions this would cause a stall). The 'intelligent' systems failed in this respect.

quote:

The airspeed continued to decrease, while the autopilot rolled the control wheel to the maximum left limit of 23 degrees. As the speed decreased even further, the plane began to roll to the right, even though the autopilot was maintaining the maximum left roll limit. By the time the captain disconnected the autopilot, the plane had rolled over 60 degrees to the right and the nose had begun to drop. Ailerons and flight spoilers were the only means available to the autopilot to keep the wings level as the autopilot does not connect to the rudder during normal flight. To counteract the asymmetrical forces created by the loss of thrust from the No. 4 engine, it was essential for the pilot to manually push on the left rudder. However, the captain failed to use any rudder inputs at all, before or after disconnecting the autopilot. The resulting uncontrolled flight path is depicted in the diagram.

Paradoxish posted:

Aren't these both cases of the autopilot just specifically encountering situations that it wasn't designed for and wasn't equipped to handle? I'm not at all familiar with China Airlines Flight 006 so I'm only basing this on a cursory googling, but everything I can find seems to indicate that the autopilot just wasn't built to handle this kind of failure. From the wikipedia article on it:


This doesn't sound like a case of too much automation causing an issue to me, but if you're more familiar with the incident maybe you can explain why you think it is. From a layman's point of view there's nothing here to make me think that this was something that was overlooked in the design of the autopilot so much as it was something that was specifically outside of its design parameters.

Ocrassus posted:

Actually flying is one of those cases where learning is very iterative and lessons are distributed across the world. Black boxes are meticulously studied and catalogued, with these lessons applied frequently to pilots who, with most major airlines, go in for simulator every few months.

Ocrassus posted:

Airbus pilots fly via a joystick, and all of their inputs are sent to a (triple redundancy) system that ensures these actions stay within a flight envelope. There is a decision engine that is context sensitive, and prevents the pilots from making certain manoeuvres depending on the conditions (the system will permit greater freedom of action in what it perceives to be an emergency). In this sense, the design of an airbus aircraft trusts the systems over the pilot.

Boeing aircraft on the other hand, whilst still having programmed flight envelopes that constrain the pilot, are far more permissive, significantly more like a mechanical system.

Senor Tron posted:

My hunch has always been that AI pilots will take over until there is a crash that would have been easily solved with a human in the cockpit. At that point even if AI has made flying statistically safer there will be the desire to have a warm body sitting behind the controls.

Mr Chips posted:

It's not even that much of a difference, despite the Boeing marketing. Both will automatically throttle back in over-speed, both will take action to correct a stall. The Airbus system will limit Gs, the Boeing one won't. Airbus pilots can still flick a couple of switches to turn off the flight envelope protections, their Boeing flying counterparts have to fight the stick to do it.

On the A380/787 generation, nearly everything is automated to an extent that the comparison is almost meaningless.

I'm not sure the China Airlines & Air France examples are really that instructive as negative cases for automation. Air France Flight 447 was controlled flight into terrain, it's a textbook example of crew breakdown in the case of equipment/systems failure. The QF72 incident was similar, but the crew handled it better (and in clear daylight were able to orient against the true horizon). The 747SP in the China Airlines incident is two generations older than the planes with heavy automation, and it's built like a tank* compared to newer planes. Pulling 5.5G in a 787/A380 after overriding the computers has a much lower likelihood of being successful. (FAA regulations for transport aircraft G tolerance specify a max of 3.8G, for some perspective.)

*and the entire 747 series was derived from a military transport designed to carry actual tanks

Owlofcreamcheese posted:

People in lots of places ride trains with no drivers. In not too long people will be in a place they've spent years riding in cars that no one is driving. People will be well used to it before planes. It won't be their first step to the concept.

Malcolm XML posted:

And lmao at someone "in the industry" not knowing that airbus and Inmarsat are trialling real time FDRs, I a dead gay comedy forums poster, googled this and expected it to be another decade while they flight qualify it and was pleasantly surprised that they are in testing for end of decade deployment


Also while ai will probs replace pilots I don't think it will replace cabin crew since they are needed to deal with passengers directly

Owlofcreamcheese posted:

The thing is: the improvement on humans coming up with ideas higgly piggly at the moment a disaster happens is having humans come up with those ideas ahead of time and then sitting down and working them out with engineers and working out the physics safely and not in a burning plane and not in 30 seconds.

You can't have a pilot go to school for 400 years to learn every "what if", but if everything stored procedures you can pre calculate, pre test and safely verify what is the right answer in every possible break down. and you can keep adding to that list until you have gone way past what you could expect a human man to be able to come up with on his own while sitting in a chair in a crashing plane and way into "a pilot on the plane couldn't have helped here either".

Senor Tron posted:

That's why I specified the crash. People are bad at judging risk in general. Even if AI pilots make flying safer, there will eventually be a crash that kills hundreds of people, and there will at some point be one that would have been prevented if there were humans at the controls. At that point we will see people demanding to have humans at the controls, even if it statistically makes things a bit more unsafe.

Owlofcreamcheese posted:

People eventually gave up on drivers for elevators and in some places for trains without demanding them back.

SeaWolf posted:

I'm not doubting that there's very different safety standards between military and civilian aviation. But the overwhelming vast majority of development in both military and civilian aviation is done by the same handful of companies, so they are going to be the ones best positioned to understand how to translate the work they've done in one realm to the more rigorous standards required in the other. There's also a convenient overlap between military transport and civil freight. I could certainly see for example FedEx being very interested in working with Boeing using their experience as contracted through the military to automate their air fleet. If anything it's going to come out of civil freight and be proven mature there before it's ever used to move people.

Amazon isn't looking to move people (yet), they're concentrating on more local delivery with their cockamamie drones and airships. And with the rate that they still need to crank out new giant distribution centers here on the ground, I'd say they think it's still a pretty far off goal. So they're still going to need freight carriers to haul everyone's packages around until then. While they're certainly going about it pushing towards the do it first ask forgiveness later method like most other SV companies working with AI and general transportation, the big push isn't going to come out of SV. These companies pushing dangerous and what I hope are just pie in the sky personal giant quadcopters and similar projects aren't going to really go anywhere just like there aren't the flying cars we were promised in the year 2000. It will generate interest and headlines, but they are not going to be the ones to change transportation as we know it.

Ocrassus posted:

Ocrassus posted:

I mean a lot of this was explained to me in training and by assorted veterans. The flight envelope characteristics are significantly different the two.

The 787 and the a350 xwb (ie the latest gen composite fuselage aircraft) are also far more capable of handling sustained g-forces, so I'm not sure where you're getting the idea that the heavier and less flexible SP, which might as well be made of pig iron, is more resilient.

the SP certainly isn't able to perform feats like this.

Owlofcreamcheese posted:

People eventually gave up on drivers for elevators and in some places for trains without demanding them back.

Solkanar512 posted:

Come on man, elevators and trains move in one dimension along a rail, cars move in two and aircraft moves in three. The former are going to be much, much easier to automate than the latter.

Senor Tron posted:

What automated elevator accident has occurred that would have been avoided with a human at the controls?

Nevvy Z posted:

Someone already did an effort post explaining why this is a bad analogy, but in short airplanes have a much larger margin of error to work with so some aspects are going to be easier.

It's almost like they are different modes of transit with different issues affecting the automation.

Beowulfs_Ghost posted:

Elevators, trains, and cars can all fall back to just stopping and waiting for help if the AI can't handle the situation.

Coming to a dead stop may be inconvenient, but is rarely life threatening. Every day, cars stall in freeway lanes without fatalities.

Planes can't do that.

GEMorris posted:

The fact that these SV true believers can't accept this or acknowledge that the difficulty this poses makes the problem orders of magnitude more difficult to solve is the crazy part. Just gotta say "disrupt" and "innovate" enough and the magic will happen.

Bates posted:

We can argue over the time frame but there's really nothing to indicate it's a problem that can't be solved. Do you think we will need pilots in say 100 years?

GEMorris posted:

The idea that some SV wunderkids are going to automate passenger flight travel in the next 10 years is loving laughable though.

Tasmantor posted:

Did anyone say in the next ten years? Isn't the point that it doesn't matter when your job will be automated but that it will be and then what?