RAF Strategic Bomber

[TheSockPuppet]

How many times have we heard that expression ? And yet we end up having to design new aircraft as we still can't duplicate the ability of the pilot.

And you are seriously over estimating the capability of current AI and it's not going to get much better soon. The issue isn't really the physical but the software. AI's have to be taught and that is both a slow process and currently flawed. We can teach something relatively simple, like chess to an AI. Something as complex as the 4D movement of a system over ground, trying to use a variety of sensor sources to identify a target on the ground that can not be seen clearly whilst determining if the target is valid and can be attacked, whilst avoiding hitting the ground and dodging potential enemy action. That is currently beyond us and will be for some time as we have to learn how we do it first. How can a human pilot, controlling an aircraft over the ground identify a target and decide if it's valid or not, if they can attack it or not and how ?

You are overestimating human capabilities. Humans are good at making complex decisions slowly. Machines can make lots of simple decisions very quickly. The best way to operate is to put both together. You can either put the human in the machine (pilots) or program the machine to make make human-like decisions based on preset criteria. When operating at exceptionally high speeds with a lot going on, humans struggle. Here, I think you are reliant on AI to complete the task.

Putting a pilot in there makes you slow, unmanoeuverable, expensive and inexpendable. Missles are fast, cheap and can be used in high risk situations.

We have self driving cars. Driving a car is a lot more complex than flying a plane. There is nothing to hit in the sky, it's hard to hit other planes and the sky is always almost the same. Why are we so selective with pilots? Because we can be. Remember, we had auto-pilot a long time before we had self driving cars.

 

[Kitmarlowe]

You are overestimating human capabilities. Humans are good at making complex decisions slowly. Machines can make lots of simple decisions very quickly. The best way to operate is to put both together. You can either put the human in the machine (pilots) or program the machine to make make human-like decisions based on preset criteria. When operating at exceptionally high speeds with a lot going on, humans struggle. Here, I think you are reliant on AI to complete the task.

Putting a pilot in there makes you slow, unmanoeuverable, expensive and inexpendable. Missles are fast, cheap and can be used in high risk situations.

We have self driving cars. Driving a car is a lot more complex than flying a plane. There is nothing to hit in the sky, it's hard to hit other planes and the sky is always almost the same. Why are we so selective with pilots? Because we can be. Remember, we had auto-pilot a long time before we had self driving cars.

Again. You are over estimating the capability of current technology. We do have self driving cars, can you pop into your local dealer and buy one ? Can you legally drive on a road in the UK now ?

AI can complement and assist a pilot in some tasks, the USAF has been experimenting with an AI to assist the pilot in a U2 test bed. Which suggests a limitation in the technology available. AI can not yet fly a plane, drone or missile and take autonomous action yet. Nothing wrong with the Armed Forces keeping an eye on development and funding a few projects in case a substantial breakthrough takes place.

But planning and designing a complete replacement of a proven technology for what is frankly Sci Fi at the moment ?

 

[PhotEx]

AI can not yet fly a plane, drone or missile and take autonomous action yet.

that might rather come as a surprise to the Israelis who's high end drones can switch to internal Hunter/killer mode if you jam their datalink and autonomously hunt, select and attack targets of opportunity.

 

[Kitmarlowe]

that might rather come as a surprise to the Israelis who's high end drones can switch to internal Hunter/killer mode if you jam their datalink and autonomously hunt, select and attack targets of opportunity.

That is pretty simple programing and not truly an autonomous action. And a bloody expensive way of doing things. If the Indian purchase of 10 Harop for $100mn in 2009 is an indication of the price target for deploying a system then that's a bit pricy to have a system that kamikaze when comms is lost...

 

[Mattb]

Again. You are over estimating the capability of current technology. We do have self driving cars, can you pop into your local dealer and buy one ? Can you legally drive on a road in the UK now ?

AI can complement and assist a pilot in some tasks, the USAF has been experimenting with an AI to assist the pilot in a U2 test bed. Which suggests a limitation in the technology available. AI can not yet fly a plane, drone or missile and take autonomous action yet. Nothing wrong with the Armed Forces keeping an eye on development and funding a few projects in case a substantial breakthrough takes place.

But planning and designing a complete replacement of a proven technology for what is frankly Sci Fi at the moment ?

There’s also the fundamental problem - AI strike fighters and intercontinental missiles are all well and good in a near-peer or peer conflict - but for every one of those we have, there’ll be a dozen deployments where “the gloves aren’t off”.

BVR capability is great, but of little use if you have ROE which say you need to have visual confirmation of the target for example.

Optional manning is the way to go for future combat aircraft.

 

[PhotEx]

That is pretty simple programing and not truly an autonomous action. And a bloody expensive way of doing things. If the Indian purchase of 10 Harop for $100mn in 2009 is an indication of the price target for deploying a system then that's a bit pricy to have a system that kamikaze when comms is lost...

That’s the start up costs included, once you have the support stuff, it’s just munitions.

Fly to point ‘x’ if comms lost, enter search mode and locate targets in your library, prioritise targets, que cheaper suicide munitions if in your area, if not, select and destroy, or if no HVT found, return home - is fairly autonomous.

 

[Kitmarlowe]

That’s the start up costs included, once you have the support stuff, it’s just munitions.

Fly to point ‘x’ if comms lost, enter search mode and locate targets in your library, prioritise targets, que cheaper suicide munitions if in your area, if not, select and destroy, or if no HVT found, return home - is fairly autonomous.

And that sounds incredibly simple to us as a human. Now break that down onto every single step required to achieve that. And I mean every single step.

Just so you can understand how complex this is. List every single step required to make a cup of tea. From selecting the cup/mug you want to use to drinking the tea. Every single step you take without thinking every time you do it. Every single step. Don't forget things like picking the kettle up, removing the lid, moving the kettle to a tap, placing the kettle under the tap and turning the water on. Observe the flow of water into the kettle, decide at what water level you will turn the tap off.

Can you see what steps were missed ? Is this an electric kettle ? If it is, did I switch the electrical supply off for safety reasons and unplug the kettle ? Is that tap in the same place as the kettle ? Can I fit the kettle under the tap ? How did I move the kettle from where it is to the tap ?

Did you notice the missing word ? Most of us will have filled a word in to make the sentence work without thinking because that is how our brains have been taught. An AI wouldn't understand that.

Can you know see just how insanely complex this is ?

 

[TheSockPuppet]

Again. You are over estimating the capability of current technology. We do have self driving cars, can you pop into your local dealer and buy one ? Can you legally drive on a road in the UK now ?

AI can complement and assist a pilot in some tasks, the USAF has been experimenting with an AI to assist the pilot in a U2 test bed. Which suggests a limitation in the technology available. AI can not yet fly a plane, drone or missile and take autonomous action yet. Nothing wrong with the Armed Forces keeping an eye on development and funding a few projects in case a substantial breakthrough takes place.

But planning and designing a complete replacement of a proven technology for what is frankly Sci Fi at the moment ?

It isn't Sci Fi. And missiles do not need to be able to replicate pilots, they just need to provide an effect.

Pilots spend most of their mental capacity trying to compensate for the limitations that humans in aircraft provide - slow, intolerant to G forces, fragile, need oygen, need controlled pressure and temperature etc.

Take the shackles away from the system and all of a sudden the missile only needs to worry about hitting the right target.

 

[TheSockPuppet]

I think we're having two different arguments here and you've missed my point. My argument is that we don't need expensive and obsolete manned aircraft because missiles can replicate the effect, better and cheaper.

You are arguing that missiles aren't as good at being manned aircraft than manned aircraft.

I'm done here.

 

[Cold_Collation]

I think we're having two different arguments here and you've missed my point. My argument is that we don't need expensive and obsolete manned aircraft because missiles can replicate the effect, better and cheaper.

You are arguing that missiles aren't as good at being manned aircraft than manned aircraft.

I'm done here.

You might be in a decade or three.

 

[Graculus]

That’s the start up costs included, once you have the support stuff, it’s just munitions.

Fly to point ‘x’ if comms lost, enter search mode and locate targets in your library, prioritise targets, que cheaper suicide munitions if in your area, if not, select and destroy, or if no HVT found, return home - is fairly autonomous.

Peace breaks out...then what?

 

[Graculus]

I think we're having two different arguments here and you've missed my point. My argument is that we don't need expensive and obsolete manned aircraft because missiles can replicate the effect, better and cheaper.

They can't. See my comments on USN and USAF NG a/c

You are arguing that missiles aren't as good at being manned aircraft than manned aircraft.

Whit????

I'm done here.

Promise?

 

[PhotEx]

You might be in a decade or three.

GCI fast jet intercepts were a thing 50 years ago.

 

[Not a Boffin]

Here's the clincher as to why it's not yet possible.

How do you certify as "safe" an AI system incorporating machine learning? Particularly one that is controlling a highly complex, high speed vehicle that may or may not be armed?

Complex software certification is an emerging issue. There are plenty of IEC standards about that help with deterministic software, but AI (non-deterministic) in which the reaction of the system can't be reliably predicted? Not so much. With humans, you get to walk through the logic of component reactions in relatively slow time and in a comprehensible (mostly!) fashion. Even if the end reaction is wrong you can work out whether it was the result of an inherent flaw or merely misinterpretation and do something about it. Not so simple with millions of lines of code and 1's and 0's.

 

[Lindermyer]

Here's the clincher as to why it's not yet possible.

How do you certify as "safe" an AI system incorporating machine learning? Particularly one that is controlling a highly complex, high speed vehicle that may or may not be armed?

Complex software certification is an emerging issue. There are plenty of IEC standards about that help with deterministic software, but AI (non-deterministic) in which the reaction of the system can't be reliably predicted? Not so much. With humans, you get to walk through the logic of component reactions in relatively slow time and in a comprehensible (mostly!) fashion. Even if the end reaction is wrong you can work out whether it was the result of an inherent flaw or merely misinterpretation and do something about it. Not so simple with millions of lines of code and 1's and 0's.

At this point I would point to Watchkeeper for emphasis, a sizeable chunk of whos problems related to the requirement to operate in controled airspace and the certification thereof**.

Thats a man in the loop RPS , AI will only be magnitudes more complicated.

**Its not just the Avionic systems that need to meet the requisite standard - its the ground station as well eg the "pilots" controls instruments etc need to meet the equivelant aircrafts systems MTBF.

Then theres data links and power supply quality and redundency - Whats that your ground station is going to be mobile and hooked up to a Genny and from here on in it gets a little bit complicated.

 

[Alamo]

At this point I would point to Watchkeeper for emphasis, a sizeable chunk of whos problems related to the Royal Artillery

FOC

 

[Gravelbelly]

We have self driving cars. Driving a car is a lot more complex than flying a plane. There is nothing to hit in the sky, it's hard to hit other planes and the sky is always almost the same. Why are we so selective with pilots? Because we can be. Remember, we had auto-pilot a long time before we had self driving cars.

We don't have self-driving cars. We have cars that can, under limited circumstances, operate for short periods without driver intervention. Those cars rely on high-quality GPS, active sensors, and the absence of visual countermeasures. The current crop are still quite happy to run over innocent civvies or drive into the central reservation, if the driver doesn't pay attention. Ramp up the CSURV measures and throw in some jamming, they're dead metal.

Insisting that there's nothing to hit in the sky, for aircraft whose flight regimes are often nap-of-the-earth, does seem a slight oversimplification.

 

[Cold_Collation]

We don't have self-driving cars. We have cars that can, under limited circumstances, operate for short periods without driver intervention. Those cars rely on high-quality GPS, active sensors, and the absence of visual countermeasures. The current crop are still quite happy to run over innocent civvies or drive into the central reservation, if the driver doesn't pay attention. Ramp up the CSURV measures and throw in some jamming, they're dead metal.

Insisting that there's nothing to hit in the sky, for aircraft whose flight regimes are often nap-of-the-earth, does seem a slight oversimplification.

If I may:

We have cars that can operate on roads and, to a greater or lesser extent, cooperatively with the infrastructure around them. Other than the things roaming around on Mars, there's arguably nothing out there that is 'autonomous'.

The military has looked at logistics vehicles that can operate off-highway but the technology content and cost are magnitudes different to what we'll see on road. We're a long way from a driverless vehicle for th private user that can happily roam around in a field.

A driverless vehicle will have to, as you note, take cues from the infrastructure around it. That includes guidance technology, onboard vision systems, in the future, 'Here I am'-type apps on pedestrians' and cyclists' smart devices, and so on.

The idea that vehicles can up and go in the way that those with a human driver can is fallacious.

Oh, and I'd add, potentially, electronic countermeasures to visual. Things get dicey if someone can jam or spoof GPS/GNSS.