Implications of A.I. to the Fighter Pilot and Modern Combat

a reply to: ridgerunner

I think that like a HUD it will be an assistant at first. Then like Samm is saying, it will be your "wingman". But the algorithms are set for "attack" and will assist a fighter pilot (until they are replaced) then either it is a race to the death to see who wins or we get our sh1t together and become human beings to each other. Just like all races after the last world wars... and I hope we do not need another one to advance, we will have to find common ground.

(I am not a liberal, just a realist with a knowledge of history! You guys here in the Aviation Threads, keep it up!)

Nice to see BASSPLYR back after some hardship. And yes, his AI will be the best!

Will the AI have "instincts" and have dirty tricks in his inventory along with "flying by the seat of your pants"?
A big part of combat flying is "survival".

a reply to: TEOTWAWKIAIFF

Oh, I have a feeling there will be some forthcoming expose' further detailing the astounding capabilities of the D.U.D.E. AI system.

As for dirty tricks...for sure....."aaaaand booted!"

a reply to: Blackfinger

A genetic fuzzy algorithm is a decision based system that clusters possible results together and then "chooses" the best next set of solutions. The computer does not "think" or even really make a decision it just chooses from the set of answers for Situation A the course of action so bad results: A does not happen. Then it goes on to the next problem it is facing. The weighted options are what the human programmers add in. The "fuzzy" part is that there are ways to modify which answer sets can be added (e.g., you do not need "deploy landing gear" option so replace it with one you want).

So it does this 250 times in the blink of an eye. Each blink of an eye. In 8 simulated dog fights it "won" each one. That means that the weighted values for that decision tree to produce "winning results" has been very well calculated.

Programming "dirty tricks" is probably in the set of solutions to various situations (if it is not I am sure it soon will be).

Wikipedia: Genetic algorithm

a reply to: Blackfinger

I imagine a major use for air-combat AI would be to decide upon proper maneuvers against incoming missiles and defensive weaponry. That's pretty non-controversial.

A good AI would be trained (via simulator) against a variety of expected missile threats, and could more accurately compute likely successful tactics if appropriate parameters of opponent missiles were available to the simulations. I.e. it could know better than a human that a missile going a certain speed for a certain time at a certain altitude probably has "X" amount of fuel left.

Also, what about anti-missile defensive missiles? remember Pye Wacket?

en.wikipedia.org...

It would be great to have a GTFO button.

Advances in "reinforcement learning" training technology, like the tech which resulted in the astonishingly powerful Alpha Go machine, might help here.

It's not surprising really as there are lots of examples where software can outperform a human in situations where a limited number of optimal strategies exist.

This is great in a situation where the AI has all the data to hand (i.e the simulation environment in the article where the AI is just crunching numbers generated within it's own program ) but this is a country mile away from having it fly against a real pilot in the physical world.

Significant amounts of raw data from multiple sensors, comm links, databases, parallel processing sources etc need to be constantly ingressed, processed and then the outcome verified against what is actually happening.

It will happen eventually but if the Google Car/Boston Dynamics level of real time environmental processing and spatial awareness is indicative of current capabilities- there's a long way to go....

AI vs AI is going to be interesting.Russia and China will have it too,gonna be some really fast developing tech coming to stay ahead of the other guys.

AI is a bit of a misnomer for what I would call either remotely piloted combat vehicles. The AI is not relevant in those types of vehicles, but it may be a bit more important for cruise missiles and other types of missiles. That's exactly where China/Russia excel. The fact that a defensive usage of missiles is the best response to incoming attacks is very well known since the soviet times. NATO instead did not focus very much on defensive SAMs but mostly on strike weapons, while using piloted fighter jets for CAP and Intercept. In these types of weapons, AI can play an important role, but mostly in pattern recognition in various spectrums. The most important scientific and technological field to develop during the coming years though has rather little to do with AI. It's all about quantum mechanics, quantum computing and quantum bit teletransportation (No, It's not about Star Trek)

If the other side obtains the AI software they will be able to predict what it will do in a given situation and should be able to create a situation to defeat it.

a reply to: Flanker86
I completely and strongly disagree on the AI part, the other stuff is not in the field of my expertise and may well be true.

a reply to: JIMC5499
Only if you think about static instances, this is crucial to know.

a reply to: verschickter

Hey good sir.

What do you mean by static instances?

I dont know much about coding. My best friends a big time java programmer for some major corporations. Id ask him but hes part of the self absorbed "me" movement and cant return messages for months at a time cause..cant be bothered. So id otherwise ask him.

So whats a static instance ?

a reply to: BASSPLYR
Where do I start... the definition depends. Let´s use object oriented programing as an example and take it to the real world but very simple.

I create a class called "Cat". This cat has properties like fur color, height, weight... It has functions like, sleep, hiss, miau, run and so on... You can call those functions and they´ll execute what you defined in the class before that was the template for your object.

Using c# for example because I deem it more readable:

Cat Garfield = new Cat(Garfield,orange);

while Garfield would be both the name of the object and the propertie "Name" and the orange would be the propertie for fur color.

the part where I call Cat(...) is the constructor and while I call it, the class creates the object for me. This object is called an instance.

Garfield.Miau(3);

This would execute the function "Miau" and I pass a value to it. In this case, our cat would execute the function "Miau" and would make your computer beep three times or play a "miau.wav" 3 times or whatever you want it to do.

=================================================================
now, I preset the path to the file "miau.wav" forever by making it a static variable, means while instantiated, this variable will not be changeable. If I make it non-static (eg. dynamic) I could give you the option to pass a filepath into my Miau function or other internal function to change this value.

Garfield.Miau(3,uservar);

the variable uservar would be set by the user before calling the function, with a textbox for example and then pressing a button that calls the function (a textbox is what I´m typing in currently).
=================================================================

So what I ment, when I said, that a AI can not be static, was a bit different but you have to know the definition the get what I´m writing now. Since we talk about an AI, that has to learn, from the getgo, where you instantiate two AIs like this:

AI intelligence1 = new AI();
AI intelligence2 = new AI();

intelligence1.Birth();
intelligence2.Birth();

This would of course give you the exact same data on first glance. But because the object AI does not posses data it has to compute it. This is what I will call learning. So both instances are instantiated and they begin to do analyze (in reality, this is so complicated to explain, I´ll just call it that).

Or in a short, if two identical twins are born(instantiated) and grow up(learn) they will have different outcomes. That´s why JIMC5499´s post is factual wrong. You can´t just copy the "program", start it and then find out countermeasures because you´d start from zero.

Now, I left out much, like for example, how the hell is it possible that a program can learn? Let me use the google textbox above in your browser. while you type, it constantly checks your input and looks it up in a database. If you start to type in the second word, the function looks up what others have typed in (and pressed enter) and tries to predict what you will type. It has learned from other users that after typing "cat", 87% of the people typed "memes".

Now ask yourself what happens if you solve those graphic captcha´s that asks you to mark "all pictures that contain apples". Of course those captchas have been preset to a certain degree. For example the program knows that picture2 and picture6 contain apples. The third one, was singled out by an internal algorithm that determined "this contains the object "apple"". Now it asks you, to verify it.

Any questions left?
I must note, there are some factual errors in this explanation, for the sake of "simplicity".

a reply to: verschickter

Actually thst made sense and helped.

thank you.

a reply to: Flanker86

You couldn't have that more backwards if you tried. The only way to have a truly autonomous UAV capable of air to air combat is with an AI, or near AI.

Think about the post above for a while, before you read the next paragraph. I´ve posted this in another thread (you´ll soon discover I bet).

Most people forget that since AI is not static but dynamic, a learning factor is involved and also reflection. I don´t know how much you´re into the topic so I´ll give you a very crude example. Probably you´ve heard about compilers and what they do (convert human readable code into machine executeable). What would happen if you(the programm in machine code) got a blueprint(the human readable) and a 3D printer (the compiler/linker) and you´re just smart(parallelism) enough to reflect(analyze) yourself. Maybe you would start making several sub copies of yourself and tell them to analyze each other, while you analyze how they perform, while, at the same time passing, down information without the need to learn but reflect since the underlying platform has changed slightly...

If you´ve digested the above, again take your time understanding this. I´m not mocking you or anyone, but you need to have an overview. Keep in mind, I´m generalizing VERY much this is just the tip of the iceberg.

What you´ve read in the quote above is considered an evolution. One evolution can spawn several sub-evolutions. Some might develop quickly, whole others degrade over time because you know, it´s not capable of understanding the underlying code and it´s variations yet, but goes by trial and error. And that´s where you start to run into problems. YOU as a human, will not be able to pick the one that is best. How would you know? you would have to test it. So you automate this process, depending on what your AI needs to be specialized. Is it shape recognition? Or playing chess?

Now, I hope you sit steady... All those evolutions of AIs, can be integrated into each other to some degree. So you have your specialist in shape recognition, voice analysis and then you eventually merge it, hoping the integrety of your data is not lost and you get the benefits of shape recognition in the voice analysis and vice versa. This has to be tested against other similar ones and after a huge rise of numbers of AI, you eventually get back to one or even zero AIs.

Diggest for a moment please.

One of the countless things you do is that you gather statistics of course. Those statistics can be plotted or visualized into waveforms. Utilizing fourier transformations, differential and integral calculus is where the "magic" starts and this is where I have to stop explaining.

a reply to: BASSPLYR

^--- read this, this is something you won´t read on the internet, the way I explain it.

Edit: To be clear, there are several ways to define what artifical intelligence is. Even if this AI get´s on par with humans in terms of decisive and "intuitive", does not mean it´s self aware.

originally posted by: Zaphod58
a reply to: Flanker86

You couldn't have that more backwards if you tried. The only way to have a truly autonomous UAV capable of air to air combat is with an AI, or near AI.

Yep... and even considering a human "pilot" directing it towards targets. It will need the ability to "protect" itself, meaning some ability to enact autonomous flight route modification, evasive maneuverability and retaliation against random unanticipated threats. This will require a higher "software" priority than its offensive role, unless the two are one in the same.

a reply to: paradoxious
And instead of one single AI, make it 20 and report to the main AI, each one has it´s own range of freedom. The weapons officer AI, pilot AI, each single one communicating with each other. Make each of them capable of overtaking functions of other ones.

edit: and metadata accessible, because during flight, AI will improve eventually, although not going through an evolutional stage, that can be done back in the hangar, with the help of the metadata.

Adding to all this, you need machine power, plenty of it.

a reply to: paradoxious

The Navy is planning an autonomous helicopter that will be able to detect and avoid pop-up threats along its planned route by the early 2020s.