TR UAV/UCAV Programs | Anka - series | Kızılelma | TB - series

[dBSPL]

Problems that can arise when development engineers are weak in social sciences. You offer the AI a reward every time disables the enemy system. So the AI removes the factors that prevent it from getting more rewards. This actually is a child's way of thinking, and just think how children can bend the rules, growing up is really about understanding social norms and ethical boundaries.

 

[uçuyorum]

My understanding is that there is that there wasn't any real simulation. However this is a likely scenario if you use basic reinforcement learning with a poor reward function. Deciding the reward function is an art itself and a very difficult thing. However you'd be better of with a more structured autonomous robotic control.

 

[dBSPL]

The most interesting point in the story is that when the system was forbidden to "kill" the operator, it "destroyed" the tower that communicated with him So -all candies are mine, I'm sorry.

 

[Afif]

If the SEAD mission was programed to be its highest priority then what happened is not very unexpected outcome.

It is not a glitch in the system nor it suddenly got terminator style free will as some people may think.

AI needs to be programed in a way that, obeying human command is its highest priority under any circumstances.

Clearly this one wasn't.

 

[godel44]

It sounds like a reinforcement learning type of exercise. Having done a fair bit of that myself for commercial applications, I find it a bit far-fetched. In reinforcement learning, you have an agent which observes environment, has a set of actions it can take, outcomes and rewards associated with those outcomes that you use to update the parameters of the model according to some algorithm. Only an extremely poorly specified reward function would give a simple positive reward for every SAM site as you would actually have something like negative infinity for killing the operator. Even more importantly, the set of actions the agents are allowed to take is fixed and would never include killing the operator. It might be configured to possibly perform strikes in a given geographical area but then you would designate that area so that the operator is not in it.

There is a hype around AI risks nowadays and making it sound like we are at the Skynet levels and AI is like people now. Meanwhile I am still dealing with their stupidities daily. It's still just matrices.

 

[Kaan Azman]

As godel said, there was a mistake in priorities. (I was thinking of this while taking a walk ain't that funny?)

Machines don't have the sense of military duty of a soldier, they work linearly when they are tasked with something. What needs to be and will be done is placement of an obstacle on the road to destroying the allied assets, as a way of saying "Be a good boy and don't bite the master no matter what".

 

[GoatsMilk]

It sounds like a reinforcement learning type of exercise. Having done a fair bit of that myself for commercial applications, I find it a bit far-fetched. In reinforcement learning, you have an agent which observes environment, has a set of actions it can take, outcomes and rewards associated with those outcomes that you use to update the parameters of the model according to some algorithm. Only an extremely poorly specified reward function would give a simple positive reward for every SAM site as you would actually have something like negative infinity for killing the operator. Even more importantly, the set of actions the agents are allowed to take is fixed and would never include killing the operator. It might be configured to possibly perform strikes in a given geographical area but then you would designate that area so that the operator is not in it.

There is a hype around AI risks nowadays and making it sound like we are at the Skynet levels and AI is like people now. Meanwhile I am still dealing with their stupidities daily. It's still just matrices.

When i was kid i used to play video games. I always used to wonder why the enemy AI was so basic and dumb. From what i gather AI in computer games is still really shit.

When i read these "skynet" reports i cant help but think these stories are made up. Its a bit like once Turkiye started using and exporting armed drones lots of fear mongering stories in the media came out and talks about needing to create international control measures. I wonder if all this AI fear mongering is about creating international control measures of its use. So that select nations can benefit from it while others are hindered from making their own AI systems.

 

[Mustafa27]

When i was kid i used to play video games. I always used to wonder why the enemy AI was so basic and dumb. From what i gather AI in computer games is still really shit.

When i read these "skynet" reports i cant help but think these stories are made up. Its a bit like once Turkiye started using and exporting armed drones lots of fear mongering stories in the media came out and talks about needing to create international control measures. I wonder if all this AI fear mongering is about creating international control measures of its use. So that select nations can benefit from it while others are hindered from making their own AI systems.

AI in video games are as good as they need to be, if it becomes too good the game will not be enjoyable for the majority of people. You could create an AI that consistently beats a player but what is the point if they will quit because they never win.
A good explanation on the topic here: https://askagamedev.tumblr.com%2Fpost%2F76972636953

 

[Nilgiri]

When i was kid i used to play video games. I always used to wonder why the enemy AI was so basic and dumb. From what i gather AI in computer games is still really shit.

When i read these "skynet" reports i cant help but think these stories are made up. Its a bit like once Turkiye started using and exporting armed drones lots of fear mongering stories in the media came out and talks about needing to create international control measures. I wonder if all this AI fear mongering is about creating international control measures of its use. So that select nations can benefit from it while others are hindered from making their own AI systems.

My reply to Indonesians discussing same matter:

Probably just sandbox testing to see "what if" there were no overrides in the programming (i.e natural state and consequence of AI).

In reality, its not very difficult at all to have robust overrides for humans w.r.t Asimov rules and similar principles etc.

Since that news "leaked", USAF spokesperson denied any such simulation happened.

But then a friend of mine was quick to quip that would be precisely what an AI would say

 

[GoatsMilk]

AI in video games are as good as they need to be, if it becomes too good the game will not be enjoyable for the majority of people. You could create an AI that consistently beats a player but what is the point if they will quit because they never win.
A good explanation on the topic here: https://askagamedev.tumblr.com%2Fpost%2F76972636953

Its not even that, its just really basic dumb stuff. I remember playing half life 1 and coming across the grunts and thinking what a step up. These guys would move as a unit, take cover, flank etc. Since then AI hasnt improved.

Before that enemy AI in games was enemy appears in front of you and fires and has basic movement patterns.

Why don't we get enemies talking to each other, employing real time tactics. Or for example lets say you are playing an RPG. In it you murder someone's father, then 5 hours later the son tries to kill you having spent the rest of the time roaming villages and investigating who was behind his fathers death.

AI is so so basic on a computer game level that i dont think AI will ever be able to think or operate like a human.

We can't even get NPC's in RPG's to operate in any organic way.

Even take a game like civilisation, the rival AI is so so basic, it cannot do anything but a few simply patterns of behaviour based on triggers. To me artificial intelligence is just about how many on/off triggers you can place on it. It operates in a linear sense, frame by frame.

 

[MADDOG]

11th medium altitude system ident test.

https://twitter.com/i/web/status/1666537194058653698

Baykar's revisions so far

 

[dBSPL]

Pay attention to the 19th second in the video. Every time I see this aircraft from the tail, it feels like a Ferrari on LPG. For the AI-25TLT, only seems an additional connection at nozzle from outside (and probably some others inside fuselage) for closing gap. So I have the impression that when the KE's B variant is ready, all the KEs produced so far at that date will probably be able to get install more powerful engines without even returning to the factory, which will be something like AI-322F in the first stage.

 

[Afif]

Air Force Colonel Now Says Drone That Turned On Its Operator Was A "Thought Experiment"

A U.S. Air Force officer says he misspoke when describing a simulation involving a drone going rogue and attacking its controllers.

www.thedrive.com

"Air Force Colonel Now Says Drone That Turned On Its Operator Was A “Thought Experiment"​

Lmao....

@dBSPL @Nilgiri @godel44

 

[Samba]

Any news regarding Anka 3`s first flight?

 

[Agha Sher]

Incredible how effective FPV drones, with an RPG warhead attached, are in the Ukraine war. Both side is using them in a swarm mode taking out infantry, IFV, Tanks, supply convoys at the price of $500 per unit.

Turkiye an other countries need to take notes here.

 

[boredaf]

Incredible how effective FPV drones, with an RPG warhead attached, are in the Ukraine war. Both side is using them in a swarm mode taking out infantry, IFV, Tanks, supply convoys at the price of $500 per unit.

Turkiye an other countries need to take notes here.

We already have that.

STM - BOYGA - Autonomous Multi-Rotor Combat UAS with Morter Payload

Click for detailed information about BOYGA, Autonomous Multi-Rotor Combat UAS with Morter Payload produced with STM's national engineering skills.

www.stm.com.tr

Edit: Not to mention we also have Kargu to fill in the roll of rotary wing kamikaze drone that Ukraine and Russia also uses

STM - KARGU - Combat Proven Rotary Wing Loitering Munition System

Click to learn more about the rotary-wing attack drone KARGU®, its capabilities, and technical features!

www.stm.com.tr

 

[Agha Sher]

We already have that.

STM - BOYGA - Autonomous Multi-Rotor Combat UAS with Morter Payload

Click for detailed information about BOYGA, Autonomous Multi-Rotor Combat UAS with Morter Payload produced with STM's national engineering skills.

www.stm.com.tr

Edit: Not to mention we also have Kargu to fill in the roll of rotary wing kamikaze drone that Ukraine and Russia also uses

STM - KARGU - Combat Proven Rotary Wing Loitering Munition System

Click to learn more about the rotary-wing attack drone KARGU®, its capabilities, and technical features!

www.stm.com.tr

I know about these - however, Kargu is in the rather expensive category compared to these $500 per unit solutions.

Turkiye needs a cheaper solution for mass use by infantry on enemy positions, personnel, armor and vehicles.

Just look at the below. Trading a $500 dollar drone with an RPG warhead attached for a L2 A6.

https://twitter.com/i/web/status/1667554661899935748

 

[Khagan1923]

I know about these - however, Kargu is in the rather expensive category compared to these $500 per unit solutions.

Turkiye needs a cheaper solution for mass use by infantry on enemy positions, personnel, armor and vehicles.

Just look at the below. Trading a $500 dollar drone with an RPG warhead attached for a L2 A6.

https://twitter.com/i/web/status/1667554661899935748

The best solution is this

https://twitter.com/i/web/status/1667516486410469376

We have various 30mm cannons to use, newest one is the Venom LR. You can slap those on everything from 4x4 APC to MBT. Its time for the Armed Forces to abandon their love for the puny 25mm which couldn't even penetrate walls in Syria and move to serious firepower.

30mm for 4x4 APC/MRAP/MBT and 35/40mm for IFV wheeled and tracked.

Another operation in Syria with the equipment we have right now and we will see dozens of turkish tanks and apcs taken out by the same kind of drones.

This is just the beginning in the near future you will have groups of 20-30 drones equipped with rpg warheads swarming whole armored groups taking out dozens of tanks and armored vehicles for literally the cost of a single 4x4 if even that. And with AI there won't even be a operator on the other hand. A UGV equipped with a launcher will launch dozens of drones and AI will lead them to their target.

 

[Oublious]

The best solution is this

https://twitter.com/i/web/status/1667516486410469376

We have various 30mm cannons to use, newest one is the Venom LR. You can slap those on everything from 4x4 APC to MBT. Its time for the Armed Forces to abandon their love for the puny 25mm which couldn't even penetrate walls in Syria and move to serious firepower.

30mm for 4x4 APC/MRAP/MBT and 35/40mm for IFV wheeled and tracked.

that will make the tank expensive and heavy....

A unit of SPAAG like Korkut is needed in modern battlefield.

 

[Era_shield]

I know about these - however, Kargu is in the rather expensive category compared to these $500 per unit solutions.

Turkiye needs a cheaper solution for mass use by infantry on enemy positions, personnel, armor and vehicles.

Just look at the below. Trading a $500 dollar drone with an RPG warhead attached for a L2 A6.

https://twitter.com/i/web/status/1667554661899935748

We can do a lot better than those FPV drones. They have low payload weight, short range, and they lose radio contact when they approach their target. Instead of that, we should be making drones which are either quadcopter or fixed wing which can release mortars/grenades with automatic ballistic targeting. This way the per-sortie cost is even lower than those crappy FPV drones, optics are much better, they don't lose radio contact due to having to drop down to ground level, they can do their own post-engagement damage reporting, etc.