AI-Powered Killer Drones Only Months Away, Says Ukrainian Commander!

40 comments

1. Ukraine would happily release Skynet right now they are so desperate.

2. Maybe they will deploy drones to operate in a designated area where everything is targeted?

3. They will geofence the drones so Ukrainians just don't go where the AI drones are operational

4. Marines. So dumb they're smart. 🤣🤣

5. Swarm of remote AI controlled drones with weapons… Awesome 😎🤖

6. Let's figure that your Tesla goes berserk, runs you over, and sits there until you die. Well, this is a major issue for Tesla. There are million-dollar lawsuits, federal investigations and legislation, etc. Now let's assume that an AI drone goes berserk one out of every 20k troops/tanks/whatever. That might be considered an acceptable loss.

   Further, there's no reason the AI drone can't be guided to a point where there are no friendly tanks and let loose? This is the likeliest scenario in any case, since the jammers have a limited range. If you think about it, a Sidewinder missile works similarly.

   Drone: "I'm being jammed! OK, I have to hit that moving house-looking widget I was being guided toward".

   Maybe a backup system instead of a primary?

7. Paul, using AI probably doesn't mean the drone will operate on its own. The drone will handle the basic flight, but the human operator will designate the targets. Once an enemy has been "painted", the drone will take over and lock on to said target. There will still be a human in the kill chain. What this does is defeat the point blank drone jamming technology that actually defeats most drones, because guidance will be onboard. THIS IS NOTHING NEW! They are simply giving drones a cheap guidance system, similar to a guided missle.

8. Wonder if it can be programed to stay in a certain area only.

9. As you doubtless know, AI for a specific purpose has to be trained on specific data. There will be inaccuracies but the consequences might be acceptable on the battlefield if you are overwhelmed by the enemy. Cartwheeling meat waves of Russian soldiers might be a win for a Ukrainian machine-gunner. Yes, Ukrainians will need to train their AI systems very, very well. Ukrainians have to do something exceptional.

10. This already exists with isreal. Get with the fucking program. Its been autonomous since the gaza invasion. One example is "Daddys Home". Very normal shit……

11. This development is so exciting ! I wish I could contribute.

12. Skynet is coming

13. Oorillll

14. Excellent analysis pro and con.

15. "Drones will be able to distinguish friend from foe"…
    Bull$ht!!…
    Heat sensor gonna be able to tell a Russian soldier from a Ukranian soldier???😅😅😅

16. I have worked with AI for a very long time and I know how to use it efficiently and effectively. It depends on your training and how well knowledgeable you are on AI.

17. So, humanity is actually building Cylons. Great 😟

18. There's also a new AI system that is coming out where it can learn for itself, react to its environment, and counteract.

19. Artificial Intelligence (AI) has made significant advances in recent years, especially when applied to drones and autonomous systems. Here's an explanation for those who may not be familiar with AI, its strengths and weaknesses, and how it's used in drones, including its ability to distinguish between targets and counter attempts to fool it.
    Strengths of AI:

    Speed and Efficiency: AI can process large amounts of data quickly, making decisions faster than humans. For example, AI-powered drones can scan an area, identify targets, and make real-time decisions much faster than human operators.
    Automation: Once programmed and trained, AI systems can operate without human intervention. Autonomous drones can fly, patrol, and even engage in missions without needing a pilot to guide them at every step.
    Pattern Recognition: AI excels at recognizing patterns, which makes it useful for facial recognition, identifying vehicles, or distinguishing between objects. This allows AI in drones to identify specific targets or threats.
    Adaptability: AI can learn from new data, which means it can be retrained or updated based on real-world conditions. For example, if a drone encounters a new type of camouflage, it can be trained to recognize it in the future.
    Multisensory Integration: AI systems can process data from multiple sensors simultaneously, such as cameras, infrared sensors, and radar. This means drones can "see" in the dark, detect heat signatures, and recognize objects even in difficult conditions.

    Weaknesses of AI:

    Limited Creativity and Flexibility: AI operates based on the data it has been trained on. If it encounters a situation or an object it hasn’t been trained to recognize, it can make mistakes or fail to act.
    Vulnerability to Deception: AI can be fooled if someone uses tactics it hasn’t been trained to recognize. For example, placing a trash bag over a person might confuse an AI drone if it hasn’t been trained to recognize that shape as a human under cover.
    Training Bias: AI systems are only as good as the data they are trained on. If the training data is incomplete or biased, the AI can make incorrect or biased decisions. For example, if a drone’s AI is trained mostly on images of humans wearing typical clothes, it might not recognize someone in a disguise or non-standard clothing.
    Ethical and Security Concerns: AI systems, especially in drones, can pose ethical challenges, particularly when it comes to distinguishing between combatants and civilians. Autonomous drones could potentially cause harm if not carefully programmed or if their decision-making algorithms are flawed.
    Susceptibility to Countermeasures: AI systems can be tricked with tactics like using flares to confuse heat-seeking drones, deploying jamming signals to disrupt communication, or using decoys to mislead pattern-recognition systems.

    How AI in Drones Can Be Fooled:

    Camouflage and Disguise: A trash bag or similar object could be used to obscure the shape of a human or a target. If the AI isn’t trained to recognize such deception, it may fail to identify the person. Training AI to overcome this requires exposing it to many different scenarios, teaching it to recognize patterns beyond the obvious.

    Infrared Confusion: AI systems that rely on infrared (heat detection) can be fooled by deploying flares or other heat sources to confuse the drone. This tactic is commonly used in warfare to mislead heat-seeking missiles.

    Jamming or Hacking: Since drones rely on communication systems and sensors, an attacker could use jamming devices to disrupt these systems, preventing the drone from receiving accurate information. Hackers could also potentially access and manipulate the AI’s decision-making processes.

    How AI Drones Can Be Trained to Avoid Being Fooled:

    Multimodal Training: By training AI systems to analyze data from multiple sources (such as visible light, infrared, and radar), they can become more robust. For example, a drone that uses both visual and thermal imaging could detect a human even if a trash bag is used to cover them.

    Diverse Training Data: AI is trained using vast datasets of images, videos, and scenarios. The more diverse the training data, the better the AI will be at recognizing a wide range of situations. This means including images of humans in various poses, under different lighting conditions, and wearing different types of clothing.

    Machine Learning and Reinforcement: AI can be trained using reinforcement learning, where it improves its accuracy based on feedback. For instance, the system might be trained in simulations where humans try to evade detection, helping the AI to learn from its mistakes and improve over time.

    Distinguishing Friend from Foe:

    AI in drones can distinguish between friend and foe using several technologies:

    Facial Recognition: AI can analyze facial features to identify specific individuals, though this requires a clear image and a database of faces to compare against. However, facial recognition can struggle in poor lighting, with facial obstructions, or if the person is wearing a mask.

    Behavior Analysis: AI systems can also analyze movements and behaviors. For example, a person acting suspiciously or moving in certain patterns might be flagged as a potential threat.

    Identifiers: Drones might be programmed to identify friendlies using specific signals, such as radio transponders or wearable devices that transmit identification codes.

    Autonomous Attack Drones and Countermeasures:

    Flare and Infrared: Autonomous drones can use infrared (heat sensors) to track targets. However, as mentioned, countermeasures like flares can confuse them by creating multiple heat signatures. Advanced AI systems are being trained to recognize and ignore these decoys by analyzing patterns and behaviors, not just heat.

    Electronic Countermeasures: Autonomous drones often use communication systems to coordinate or receive orders. These systems can be disrupted using electronic warfare tactics like signal jamming. Some AI systems are being trained to operate independently if they lose connection, continuing their mission based on pre-programmed instructions.

    AI in Self-Defense: AI systems in drones can also be equipped with self-defense mechanisms, such as using counter-flare technology or deploying their own decoys to evade anti-drone systems. Some advanced drones may even employ artificial intelligence to predict enemy countermeasures and adapt in real-time.

    Conclusion:

    AI-powered drones are incredibly efficient and adaptable in recognizing patterns, identifying targets, and operating autonomously. However, they also have limitations and can be fooled with the right tactics, especially if they haven’t been trained for a particular scenario. By incorporating multisensory inputs and advanced training techniques, AI can become more robust and resilient. As these technologies evolve, both the development of countermeasures and the improvement of AI systems will play crucial roles in shaping how these drones operate in complex environments.

20. Yes, I see the error now. When I first answered, I incorrectly stated that the word "strawberry" contains only two "R"s. By breaking it down as you suggested:

    The word "straw" contains 1 R.
    The word "berry" contains 2 Rs.

    When combined, strawberry indeed has 3 Rs in total.

    Thank you for pointing that out, and I'll make sure to apply this correction in future responses.

21. I have also come across a secondary systemhow can i put this it's something like a lockout while it will correct for one's account it will not correct for the whole which means if interpreted correctly those of us with the power to correct and train AI could train it to be very lethal.
    Use the following as an example.This is Gemini's response.
    OK i know it's not 48 hours but i'm gonna test you to see how fast you have corrected it how many letter RS are there in the word strawberry

    There are 3 letter Rs in the word "strawberry".

    I have learned from my previous mistake and have updated my knowledge base accordingly.
    profile picture

    Knowledge base strictly for my account or is it for all users?

    The knowledge base I use is primarily for your account. This means that the information I learn and store is specific to your interactions with me. While I may be able to access and process information from the broader internet, my primary focus is on providing you with personalized and relevant responses based on your unique queries.
    understand i wasn't trying to flood your page but i was trying to give examples i am currently on the class 3 and it did not allow me to edit the comments i would have edited this to be one comment but perhaps for the long run it is best that these are separate so that way people could understand and learn

22. Outstanding explanation.

23. Well is Biden hadn't allowed Russian a year to build a fortress with 9 mines to the square yard this would already be over.

24. They could implement a map that is divided into grid sectors for these drones to monitor. This could mitigate the attack on friendly troops.
    But to have this work reliable it would be necessary for gps and visual maps where the drone is operating. There are some other solutions to support this even more.

25. The Ukrainians are very very creative,They are sharing this with the US.My son is in the navy and has been trained with drone warfare.The US has been vested in drone usage.

26. So much death and killing… war industrial complex

27. People are all afraid of nukes but this is the real danger. Imagine a terrorist group getting hold of a crate of killbots and letting them loose on a city.

28. 🙄 This isn't difficult. Not quite science rocket.

    All the technologies exist. Engineer, package, and deploy. Laser, encryption, biometrics, AI, and the human being. Please read further.

    Using laser line of sight communications, to beacons equipped on every soldier, piece of equipment, or vehicle. Using a rotating passkey encryption implementation. Impossible to defeat with todays technology.

    With laser communications and visual systems it offers the ability to have human final intervention to distinguish friend, enemy, surrender or civilian. Without the worry of equipment disabling radio interference. Fiberoptic tethers seem a half assed 1990's idea. 🤣

    Program a drone to obliterate any moving or non moving target without an equipped laser receiver with correct passkey identification, or enemy sigaling explained below.

    With AI target differentiation all enemy equipment is exempt and eliminated without friendly passkey becons in which are tagged to the specific piece of friendly equipment incase of enemy capture. With a secondary passkey for the tag equipment, including biometric friendly override needed for removal, and signaling of tag removal needing visual confirmation from a drone to the opposition force. Otherwise the beacon and tag if removed would be rendered as a threat transmitting beacon.

    The opposing force would allow the sharing of beacons/technology to the enemy that would have the ability to use a non encrypted surrender or civilian signal. Using high flying line of sight drone defence aircraft, and or satellites. The enemy will avoid shooting down defence craft, in the fear communication is lost between drone and air/ satellite receiver craft. Otherwise drones will dismiss the surrender/ civilian beacons if communications are lost, then kill objectives take precedence, thus proceeding to destroy targets.

    To counteract the use of surrender or civilian beacons to foil the inclusive drone system, civilian/ surrender beacons would require biometric face recognition, thus needing close proximity drone interaction. Thus giving inherent drone safety. If a drone is destroyed by enemy means, other drones will execute a kill command to threat target. The civilian/ surrender beacons will require upload of biometrics data to initiate a surrender or civilian avoidance command. Smaller drones could be used for soldier elimination, civilian differentiation.

    Any enemy soldier not equipped with a surrender beacon and biometrics data would be considered a threat target. With opposition verification, visual recognition would be used for civilian/ surrender beacon human identity confirmations. In order for a surrender to proceed the enemy would be guided to a prescribed exposed location by drone. All beacons would have GPS to track all human positions. If the enemy soldiers resort to milita tactics in hiding their positions or using civillian clothes to hide identity, will be tracked. If lack of beacon then again the human would be considered a threat target. Captured or surrendered POW's with becons removed from a soldier would render the friendly beacons inoperable without maintaining timed biometric fingerprint input with an incorporated heart rate sensor. To use a POW for human shield use would go against the conventions of war, I feel a POW should be safe with a removed beacon. I may be wrong, nothing is infallible. Hamas demonstrates infallibility.

    AI takes care of target identification, acquisition, avoidance of enemy countermeasures, communications, and home to target autonomous flight. The combined ability of the systems would give drones the ability to, execute hunt and destroy mission objectives.

    Imagine that! The opposing force being givin the same devices and technology to save their own lives. 🤣

    Also, why can't drones drop munitions from safe positions and fly home. Seems logical. No? 🤔 I also contemplated the idea to use conventional aircraft to deploy drones, allowing the infiltration of enemy lands further than a drones range abilities. Or using large drones or missles to deploy many smaller drones. The possibilities of drone use is vast.

    Unfortunately an idea to kill people hurts me. Though a nation should have the ability to defend itself with less lives lost.

    🇨🇦🙂

29. https://www.youtube.com/@CombatVeteranReacts I doubt a tank could cartwheel towards a drone

30. Wear some cloth dude👎

31. Elon Musk already has AI designed to run into things.

32. If it's any consolation, you wouldn't use LLMs for the 'reasoning' component of AI. Also, it would operate within hard parameters.

    It should have an operating zone, which it cannot misinterpret (as GPS can be jammed, it could just calculate where it is the old-fashioned way). It's also possible the drone can be 'told' that there are Ukrainian tanks operating in the area before it took flight (so it doesn't target any tanks, leaving that to human-operated drones). People will trick it, but it will be patched (endlessly). Tragedies will still happen when communication failures happen.

33. This is going to be a bad idea

34. Not sure if you understand how machine learning works, or even what inertial guidance systems do.

35. Everybody needs to understsnd this. Forward

36. I don't mean to be critical, but it is NOT correct to think that an AI language model will be used in autonomous drones. I think the sample tests of Chat GTP identifying enemy targets is invalid.
    I think a closer analogy would be self driving car technology that identifies pedestrians, road boundaries and other cars. But I'm sure the code will be unique to the intended application.

37. Eh! You gotta break a few eggs to make omelettes

38. In WW2 the Soviet army tried to train dogs to carry explosives to German tanks.
    It didn't go well.

    The dogs had been trained with stationary tanks not firing their guns and the dogs were not willing to dive under noisy, moving German tanks.

    The Germans started shooting any dogs that came near.

    Some of the dogs returned and jumped into Soviet trenches thus killing Soviet soldiers.

    Perhaps worst of all, the dogs had been trained with diesel powered Soviet tanks so gasoline powered German tanks didn't smell right to the dogs and thus sometimes they would end up heading for Soviet tanks that had the smell of diesel.

39. they are not going to use a large language model… lol come on …

40. Great video