AI making threats at ATC AND PILOTS?
 

https://www.forbes.com/sites/christopherhelman/2019/01/25/as-shutdown-slows-flights-into-laguardia-maybe-its-time-to-let-artificial-intelligence-handle-air-traffic-control/#5268aecb2bcd

He is more talking about using private plane cars.

People can’t even drive normal cars on roads. What makes anyone think people can drive flying cars. What a disastrous idea. Unless they’re saying fully automatic flying cars, and in that case why not just make fully automated normal driving cars?

This is almost the same tripe I read @1990.

Everyone thinks AI (back then called neural networks) is sooooo smart. People love to assign super-fancy capabilities to technology without ever pausing to look into how that technology works-and what it’s limitations really are.

Those people don’t understand AI. The capabilities of AI are exactly what they were back then (as far as their ability to replace stuff like ATC.). And those capabilities will be pretty much the same 30 years from now.

AI is good at certain very limited things, but something like ATC is many decades off.

Stories like that are designed to get investors to throw more money at companies to support technology they don’t understand. This story is almost verbatim what was written almost 30 years ago-and all the claims that there would be no humans in ATC in a handful of years.

Back in the early 90s, the FAA threw many billions into the AI rathole and were strung along by contractors who always promised this stuff was “right around the corner,” but whom I’ve always suspected knew it wasn’t going to work.

Stupid regulators plus stupid politicians plus stupid investors always yield stupid results.

The issue with autonomous cars isn't the autonomous ones, it's getting them to interact with the human driven, unpredictable ones.

Fast forward about 30 secs in

https://m.youtube.com/watch?v=YQIMGV5vtd4

The figure 8 pattern is interesting (last sequence).

A computer controlled system controlling computer flown aircraft is a perfect combo. About the best thing we have going for us is small autonomous aircraft will still be extremely expensive on a CASM basis even without paying the pilot. Hopefully larger manned aircraft will compete economically for a while.

But, save your pennies. For those under 40 you'll probably be out of a job before you retire.

For sure, the smaller Caravan sized guys will be gone first, already happening in China.

There is absolutely no way that anyone actually able to read this today will be impacted. Don't scare people, management will come looking for concessions to compete with (non-existent) automation, and people might fall for it.

Even if suitable AI existed (it doesn't), there are massive regulatory, infrastructure, and economic hurdles. There will be a very, very long and very, very, very expensive road from initial investment to profitability. Nobody can even guess how long or how expensive. The airlines will not invest in anything until it's certified AND the infrastructure exists (ground handling and ATC). The government is simply not going to embark on a manhattan project to put a paltry 100,000 well-paid union members out of work. So that leaves the airframers. They can take a longer view than the airlines, but still need to show some rational logic and a timeline before spending shareholder money on something which cannot be certified, cannot be insured, and for which no market exists. ONLY the airlines stand to benefit from this, not the airframers, so they will only go there when the airlines are ready to buy. For the government it's all downside (risk) with no upside. And of course you know how much politicians and bureaucrats love risk... :rolleyes:

On top of that, add whatever delays will occur due to government inertia and public acceptance. Maybe by the end of the century.

The technical challenge is just the tip of iceberg. Social and economic issues are the real long poles in that tent. Anyone who understands the real world knows that.


Yes it will start with smaller aircraft.... cheaper to experiment with and easier to certify (or get waivers). Also short-range missions with VTOL aircraft don't require the kind of human judgment that longer (or over-water) ops require. If you get in trouble you can just auto-land almost anywhere.

You can start to guess a timeline for automated airlines when uber is turning a profit with autonomous UAMs and the military is moving cargo with automated heavies.

Clearly you've never seen the Terminator.

Let's pretend for a moment....

Let's pretend that AI is good enough to fly a Caravan filled with 1000 lbs of jet fuel and 3000 lbs of cargo over little Billy and Mary's elementary school without crashing.

Looking at the Mountain Air Cargo (if not the largest employers of 208 pilots, certainly in the top 5) pay scale for 208 pilots start at 45k and rise to 50k at five years. (I stop at five years because not that many folks make a career of flying the Beast)

What do you think the insurance cost will be for an AI piloted Caravan vs. a human piloted one? Remember, insurance companies always think the plane will crash into Billy and Mary's school.

I'll give you a bit of a clue....

When I flew a 208B for a family who had a 40 year old son with a ppl, what they payed me was about the same as the difference in cost of the insurance if he had flown the plane. They were paying wages which were at that time equal to about a 3 or 4 year Mountain Air rate, so roughly average.

Insurance companies will need more than an Elon Musk promise that everything will work perfectly. Those companies will charge a huge premium for a computer piloted plane ... at least until someone else takes the risks to prove the concept.

Companies will pay for a pilot because he or she a cheaper to keep than eliminate.

Oh, and to give another insurance perspective, Lloyds of London insured at least the early Space Shuttle missions. The rate was roughly equal to a third the cost of a launch. In other words, Lloyds figured to break even if one flight out of three crashed. Considering that there were 135 missions of which 133 were successful, they probably made a ton of money.

This is typical of when insurance companies (who live and die on actuarial table statistics) have to insure something they don't understand. Actually they don't *have* to insure anything like this at all, but if they do it will be at a vast premium... until they have some long-term data.

Another of several real-world hurdles.

There's no question that automation can do that.

It's what it can't do which is far more interesting.

Yes, it can do that.

My original "let's suppose" statement was not as clear as it should have been. Sorry, I failed to realize how literal some folks are.

Yes, it can be done. The Buran, the Soviet equivalent to the US space shuttle, had only one flight, and it was fully controlled by self automation and ground control.

One at bat, and they batted 1.000. Well done!

But what I was trying to point out was the failure rate of autonomous aircraft is too high for the public to accept.

I submit that crash rates back in 1919 may have been acceptable, or at least tolerable, those same crash rates in 2019 would have put an end to then industry.

We live in a society that is 100% knee jerk. A produce farm that has operated safely for 40 or 50 years lets out a batch of e.coli tainted lettuce which makes a dozen folks sick and kills one 110 year old immediately starts a call for new regulations.

So, yes, AI can fly a Caravan. And if this were 1919, a crash or three would be considered growing pains. But in 2019, two crashes would be the end of the industry.

Also, as the automated car industry will learn soon enough, while juries will have a degree of tolerance for humans being humans, I believe they will have zero tolerance for AI killing people... especially since the only person to sue is the big corporation who recklessly sent under-developed, untested, and unsafe AI out into society with no regard for who it kills.

This stuff will be ambulance chasers wet dream, and mark my words the automation industry will end up asking congress for liability waiver laws.

https://en.m.wikipedia.org/wiki/Iran–U.S._RQ-170_incident

I posted this same example in a similar thread a while back regarding AI and aircraft- public acceptance is the biggest hurdle to overcome, and if this can happen to advanced defense contractor technology, then...

Hi there,


A quick thread-jacking from me here



Are US fighters and heavies insured ? Maybe only when flying at home ?

Are fighter aircraft insured?

Yes. By taxpayers.

Military equipment is not insurable on the commercial market. People will actually be trying to destroy it, and it would be very hard to asses and quantify their odds of success.

Also operational procedures can be adjusted to allow for more risk as needed.

Airline pilot tired? Flight cancelled, pilot goes to hotel.

Combat pilot tired? Here, have another amphetamine...

There are many things that technology cannot solve in 121 Flying. Everything from the managerial decisions that a captain makes to get a flight out on time to manually lowering the landing gear to smelling fumes before the gauges give you the bad news to the walk around to see if ice is on top of the wings etc. Some of these issues aren’t technological and can’t be solved with technology. This won’t happen anytime soon.

But to take a different tack...remember that technology and its exponential growth will affect everyone, not just a specific job. Doctors, accountants, pharmacists, even models and actors I’ve read. I’ve read stuff saying coding will eventually fall to AI. Each revolution takes shorter and shorter. Right now we spend most of our time at a ‘job’ to pay for the necessities in life. This is a relatively recent phenomena. How we occupy our time and pay for things will probably change in the future, if we don’t annhilate ourselves first.

They will be able to justify lower pilot standards after a few accident free years. Then we will be back to 90 day wonders making 17k.

Here's how I see it going, at least initially. A hybrid system that keeps a human on board for flexibility to react to the unexpected and the robot will make sure the human doesn't do any of the typical boneheaded moves that are usually associated with mishaps. This will cut the number of pilot's by half and allow maturation of the technology. We only have to worry if the robot asks the human, "Just what do you think you are doing Dave?":eek: https://www.youtube.com/watch?v=om18cOWFL3Q Or even worse when the robots start posting on APC about how incompetent the humans are they are having to fly with :)

"AI controlled aircraft will NEVER happen because I don't want it to."


There, I've summarized the opposing conversation.



I'm amazed at the number of software and neural net experts that reside on this forum :rolleyes:

I actually have real professional experience and education in that area, plus systems engineering. I also still talk to folks on the cutting edge in the valley. It's not happening any time soon (replacing pilots). They simply don't know how to to build it. Once they do (or think they do), it will have to prove it's stability in other arenas first. Traditional software can be tested almost perfectly, since you know all of the boolean logic and calculations that it's supposed to perform. That's lengthy and expensive but is routinely done in safety-sensitive applications. But AI software which can learn and evolve cannot be tested conclusively... you simply have to turn it loose in the wild and see how it does over time, like a very long period of time.

Once somebody thinks they can make an economic case, it will still have to get pre-approved by legislators and regulators, and also by insurance companies. The last is a real biatch, since they like to use statistical tables to quantify their risk and that will be impossible until some real world data is acquired (likely need some of that to come from other applications such as automated cars/trucks). I doubt any airline would go out on a limb and self-insure since one or two accidents could put them under... shareholders might not be interested in that kind of risk.

I suspect when airframers think they can do it, they'll build a new generation of "unmanned capable" airliners, which will be flown by probably two, maybe one pilot for a long time until enough operational data exists so that regulators, legislators, insurers, shareholders, and the travelling public accept the idea. Since we seem to be able to go a decade without a fatal accident in the US, that's probably the absolute minimum demo period. Of course the airframers and airlines will have to pony up for the new technology up front, with an uncertain timeframe on the ROI (elimination of pilots). Also there's the risk that all the (rather massive) investment could fail, if the unmanned mode isn't reliable.

Part of the problem is airframers and airlines have to invest money in something that cannot possibly pay off anytime soon. Business executives and shareholders typically don't think on very long (decades) timelines, unless there's some predictability involved. They're greedy, not really interested in making sacrifices on their watch so that future generations might reap a windfall. That's a big part of the problem... the long timeline, required large upfront investment, and the uncertainty. Economics and human nature are just as much factors as technology.


Oh yeah, and you have to get the government to revamp ATC as well. If you think automating pilots jobs is hard, try to getting the government to automate some lucrative civil service union jobs... :rolleyes:

Most of the "experts" said that SpaceX would never work or is decades off.

Most of the "experts" said that electric cars would never work or are decades off.

All you need is one company to interrupt the status quo and the entire industry shifts, sometimes very quickly.

There are plans for large cargo aircraft with a single pilot and other plans for unmanned buddy aircraft. Unmanned large aircraft will follow then single pilot large passenger aircraft eventually be approved.

It will happen. The technology is easy. The procedures are easy. The software is not necessarily easy, but known.

Most modern air carrier aircraft are already built as unmanned capable with minimal upgrade required.

Once a single airline makes single pilot and/or unmanned systems work, the rest will follow.

When was NexGen/ADS-B announced? How long has it taken?

Compared to that, autonomous airliners are a much, much larger undertaking. It's not happening any time soon except as perhaps a technology demonstration or a demo at the Paris Air Show.

Really? Give me a list. I'll wait.

I am no expert, but I can do the math.

In the US, we burn about 140 Billion gallons of gas per year (https://www.eia.gov/tools/faqs/faq.php?id=23&t=10 note that is per day)

A gallon of gas has about the equivalent of 36 kwh of electricity. https://www.convertunits.com/from/kW...otive+gasoline

If we replace ALL gasoline with electricity, it would be over 5 Trillion kwh.

Current US annual needs for electricity are a bit under 4 trillion kwh. https://www.statista.com/statistics/...on-since-1975/

So we would only need to increase our electrical production by about 125%.

While possible, I doubt it will happen in any of our lifetimes.

Well actually in this case you need a generalized artificial intelligence, which does not exist. And none of the experts know how to make one.

The one thing that the smartest people in the world do mostly agree on is that actually turning a general AI loose has very serious potential hazards, of the existential sort.

A buddy aircraft works for military operations where the buddy is expendable, and so are all the people under the flight path. For non-wartime ops, it will need to be fully capable of taking care of itself in an emergency.

The "software" as you call it is most certainly NOT known. Frankly you're talking about things which you don't understand. It can follow the magenta line and autoland, but if you think about it, on any given trip you (or your CA) make some judgement calls. And correct some DX errors. And W&B issues. And dodge some weather. Occasionally you deal with a systems failure or emergency. Software can't do any of that in a flexible manner... divert, RTB.

Sure. As long as everything goes smoothly.

The biggest problem, that most of the uneducated folks don't grasp, is the economics. Doesn't mater how much you save on pilots if your unmanned airliner diverts or returns to gate every time there's any sort of glitch. It has to be able to think in order to deal with the unexpected.. if you stump the chump, it will divert/RTB. Can't do that too often and stay in business.

Also can't crash more than once or twice. All that trillions of dollars in R&D will be for naught if the public and politicians get scared... and scared they'll be, far more so than if it was pilot error.

The driverless car advocates have already learned this hard way. At this point they have pretty much given up on the near term, they didn't make big public announcements to that effect, but it's true. A handful of fatalities and governments and underwriters have pulled the plug on testing. What you'll see are greatly enhanced automation to aid the driver, not replace him. Liability now insists that the driver have his hands on the wheel and eyes open.

Again it's as much people as technology. I think the only way ahead for autonomous cars in the near term is dedicated roads. That will eliminate most of the real-world factors which confuse automation, plus you can build in guidance aids.

This is true, you only have to do it once. But that's far more difficult than most folks comprehend.

I remember you from ten years ago on this forum when you were spouting doom and gloom about peak oil and the end of civilization. That didn’t pan out, so now you’ve found the next boogie man. Must be fun going through life with such a negative perspective!

Progress for automation...

https://www.npr.org/2019/03/06/70080...ar-says-prosec


Uber is not criminally liable for the person their automated car killed. Looks like the blame will be shifted to the safety monitor, who was streaming video when the crash occured. So all they need are minimum wage safety monitors who can resist the urge to look at their phones for hours on end :rolleyes:

They're still going to pay in the civil suit though...

Interesting topic. What if the first step to autonomous flight is to transition to drones? In other words, what if you can have a reliable communication channel between the plane and the ground? Then you can still have two humans flying the plane from the ground just like they would if they were in the cockpit. The shift to fly-by-wire facilitates this as all you do it tell the commuter what you want to do rather than pulling actual cables, etc. So why do you have to communicate with the computer from the cockpit when you can be on the ground doing the same, again assuming a reliable communication channel? I know the technology is not there yet, and I am deliberately simplifying, e.g. you still need to design a way to pull a circuit breaker if you want to do that, etc.

So once you transition to unmanned aircraft flown from the ground, then the threat to the pilot profession will be there, because you can then assign a crew to do the departures and arrivals in the terminal environment and then let the automation fly the rest. Perhaps a computer can monitor thousands of aircraft in cruise, and if there is any alarm coming from any of them, then a ready crew on the ground is immediately assigned to it. You can then fly a lot of planes with a fraction of the pilots it takes today.

And the threshold for any alarm can be as high or as low as you like
- weather ahead, turbulence level, smell (maybe there are sensors to detect the smell of burning wire as well as a human can), etc. Maybe even the FA can press a button to raise an alarm if something doesn't feel right if that is what it takes to make the flying public more comfortable with the idea.

Once you transition to this, then the introduction of AI can be incremental. Maybe you can trust AI to deal with some alarms and not others. As the AI technology gets better and better you have more alarms being dealt with by a computer and therefore fewer humans will be required to keep the system operational.

So the first step is to replicate the cockpit on the ground and have reliable communication as if the pilot is in the aircraft. If you can do that, the rest will follow rather quickly.

You could do some of this.

But you won't. Because of the economics. Any discussion about replacing pilots with automation (or remote control) is fundamentally an economic one.

People used to look at the moon and dream of going there. Nobody looks up in the sky at a contrail and thinks "God we have to get those guys out of the cockpit".

Just to put the pilots on the ground would take a MASSIVE investment in equipment, which is almost certainly going to require integration with a clean-sheet design, ie it would not be economical to retrofit to existing airliners due to certification challenges. And for security and reliability purposes we are looking at satellite comm systems... likely dedicated and certainly redundant. Not just text message data rates either, but streaming high-res video data rates $$$$$$$$$$$$$,

Initially, there is no way you could certify this without at least one backup pilot on board. Regulatory and political caution would see to that. So you'd have to build a new plane, with extra (heavy and expensive) equipment PLUS a cockpit, and then pay not two but three guys to fly it for some indeterminate trial period.

Airlines are publicly traded commodity vendors... they CANNOT take the long view at the expense of short-term stock prices and dividends.

The air-framers aren't going to invest in something unless they know they have a market, and that their product can be certified.

The regulators have no incentive to push for something like which has... they don't get paid or rewarded to take risks, all downside, no upside.


And any airline who even dreams of going there has to worry after all that, the public might just prefer its' manned competitors. Really need the public to get accustomed to driver-less cars first.

Also... ground-based pilots makes the most sense for ultra-long haul. Assume you need the pilots' full attention from gate to TOC and from TOD to the gate. Short-haul flight would have a high ratio of "pilot required" time to drone time. Might as well just leave them in the cockpit. Long haul flights are a small percentage of all flights, and the larger planes have more pax to dilute the pilot costs anyway.

Very interesting opinions all around. Definitely food for thought.

Will fully autonomous scheduled passenger flights going to happen in my lifetime? Probably not. The next generation - maybe, maybe not. Ever? Barring catastrophic extinction or near extinction of humanity and given infinite time, I think most likely, yes.

I still think that the likely first step will be ground based crews. Then technology will gradually take on more and more of the flying, until humans deal with emergencies only, and then eventually no humans.

It will be a slow and painful death of the profession, if you think about it - it will turn into a 9-5 job office job, the pilot pool will keep shrinking as technology takes on more and more of the flying, the morale will be low as a result, you will deal with more and more emergency type situations, all stress. Eventually there will only be a handful of pilots dealing with dire emergencies only. Then none.

Self-driving cars will be a precursor to that and it may happen in our lifetime. A lot less risk, regulation and cost involved. I haven't followed this closely, but if and when it happens this will kill millions of jobs in the USA and a lot more globally. And the jobs will go the way of the switchboard operators - very quickly - maybe within a decade. Goldman and Sachs estimate 300K job loses per year, and there are about 4 million driving jobs in the USA.

https://www.cnbc.com/2017/05/22/gold...-job-loss.html

Scary to think of the social upheaval this may cause.