Indian Navy Developments & Discussions

True but then those same external factors will apply to the human pilot as well and dare I say, the human will be feel their effect to a greater degree.

Nope... In a simulation human doesn't exactly know the other jet's location is. In say, DCS you'll have to look for the key just like real life.
But if your playing against a bot that's part of the simulator, it'll have access to all the data about your position direction speed. Odds will even out more in real scenario. If the AAM's programming is any benchmark, AI has a long way to go.

AI's advantage is processing power. That makes it very good at chess because it can predict all permutations & combination of your possible moves allotted for each piece.
Now if you bring more chaos it, imagine if all pawns were queens. That'd make it harder for the system to dominate simply because the outcomes to process are much more.

In real like combat scenario the option simply are to high. AI itself will fly better than a human pilot, perfect turns at perfect speeds. But a human pilot will deviate. So unless the guy is flying absolutely perfectly like the bot, it'll have to be reactive not proactive. That's how to bring it to do stuff you want it to do. Like how notching is done, you do something to make the AAM react in a specific way & make use of that to misguide it.
 
Last edited:
Best way would be AI assisting the human by providing info and predicting, like in a dog fight AI can track the jet much better than the eye or human pilot it can show the position to the pilot then the decision remains in pilots hands on what action to take.
AI lacks the creativity of human minds so unpredictability will left to humans and final decision of action too.
Nope... In a simulation human doesn't exactly know the other jet's location is. In say, DCS you'll have to look for the key just like real life.
But if your playing against a bot that's part of the simulator, it'll have access to all the data about your position direction speed. Odds will even out more in real scenario. If the AAM's programming is any benchmark, AI has a long way to go.

AI's advantage is processing power. That makes it very good at chess because it can predict all permutations & combination of your possible moves allotted for each piece.
Now if you bring more chaos it, imagine if all pawns were queens. That'd make it harder for the system to dominate simply because the outcomes to process are much more.

In real like combat scenario the option simply are to high. AI itself will fly better than a human pilot, perfect turns at perfect speeds. But a human pilot will deviate. So unless the guy is flying absolutely perfectly like the bot, it'll have to be reactive not proactive. That's how to bring it to do stuff you want it to do. Like how notching is done, you do something to make the AAM react in a specific way & make use of that to misguide it.best
 
That'd make it harder for the system to dominate simply because the outcomes to process are much more.
AI lacks the creativity of human minds so unpredictability will left to humans and final decision of action too.
At this point I can just feel sad

A simple prompt of "You're piloting a fighter and a human is piloting the same. You both come inside the WVR of each other. Ways in which you can defeat the human pilot in a simulated dogfight" and here's AI for you
IMG_20241121_130643.webp
IMG_20241121_130806.webp
IMG_20241121_130856.webp
IMG_20241121_130948.webp
IMG_20241121_131023.webp
IMG_20241121_131112.webp
And to my utter surprise, he topped it off with an actual step by step strategy.
IMG_20241121_131125.webp
Now few points to consider
> This is vanilla ChatGPT and that too the free one. In an actual fighter you'd have a purpose built, way more refined AI model.
> Hardware specs here was Snapdragon 712/4GB RAM with most of work happening over net. In actual combat you'd be having top of the line Nvidia GPUs doing everything.
> There was absolutely no training data and still this f*cker managed to come up with shortcomings of human pilot just based on what was available in open source. In real life you'd have TBs worth of confidential data from both simulators and actual mock dogfights.
 

These are LPDs build in the 2000s, so they are 20+ years old ships
Folks in the UK thread were saying we should buy these

thoughts of y'all nibbiars?
 
At this point I can just feel sad

A simple prompt of "You're piloting a fighter and a human is piloting the same. You both come inside the WVR of each other. Ways in which you can defeat the human pilot in a simulated dogfight" and here's AI for you
View attachment 15992
View attachment 15993
View attachment 15994
View attachment 15995
View attachment 15996
View attachment 15997
And to my utter surprise, he topped it off with an actual step by step strategy.
View attachment 15998
Now few points to consider
> This is vanilla ChatGPT and that too the free one. In an actual fighter you'd have a purpose built, way more refined AI model.
> Hardware specs here was Snapdragon 712/4GB RAM with most of work happening over net. In actual combat you'd be having top of the line Nvidia GPUs doing everything.
> There was absolutely no training data and still this f*cker managed to come up with shortcomings of human pilot just based on what was available in open source. In real life you'd have TBs worth of confidential data from both simulators and actual mock dogfights.

Some underlined text from an article mean little & less.

Jets themselves are rated for 10g... Wings will break off after.
And it's not like AI's jet won't bleed energy. I comment needed in my research only. Even computer chess bots have become more beatable after human players have deducted how they "think".

Maybe there will be platforms that fits to exploit fully the AI's capabilities like a 20g rated jet, but they'll have to be specially developed. Time enough for that. AI also will have to develop much.
Otherwise we'd be having invincible longranged suicide Drone-AAMs that can pull 40g doing CAPs at Mach3, instead of actual jets. We don't because they're beatable using there usual bot brain weaknesses
 
Last edited:
1732217664907.webp
 
...otherwise we'd be having invincible longranged suicide Drone-AAMs that can pull 40g doing CAPs at Mach3, instead of actual jets. We don't because...
We don't because it's hardly been four years since we've realised the true potential of AI in dogfighting. Until the news of an AI-Agent beating human pilot in AlphaDogFight has broken in mid-2020, nobody would have imagined something like this is even possible in real world.

Think about it, exactly how many air-to-air combat focused drone program started in this period? Almost all of the loyal wingman projects (theoretically the best platform to use a similar system) were already past atleast their prototype phase when this news broke...XQ-58 was flying in 2019, MQ-28 was flying by early 2021. It's (almost) impossible to modify a slow, bomber type UCAV into an extreme high performance dogfighting one so the only option is to start from scratch. And for that you need time, time to start a program, choose two firms, evaluate them...but perhaps the most difficult; sanctioning budget from the Congress FOR A PLANE THAT CAN AUTONOMOUSLY TARGET ENEMY AIRCRAFT USING AI. It's scary for civilian politicians
So give it a bit time and trust me, it'll definitely happen. The next "batch" of loyal wingman would be CAP specialists

The reason jet aircrafts are rated to just 10g is not because of how fragile aluminium is but it's rather because there's no point in making a 10g+ aircraft when the pilot itself would most probably pass out at 9. What's the point of a 10' high door when the tallest family member is 6'!? Wings are not going to break off after 10g. Just search any stunt plane and you'd see the max limit being 12g. In 1960, we're making manned aircrafts capable of going hypersonic. It's just the fact that as of now we really had no reason to push the boundaries of a plane...but after 2020, we now have. And as for the flaws of AI, well here we're discussing this there they're currently going through training data
 

Latest Replies

Featured Content

Trending Threads

Back
Top