• brucethemoose@lemmy.world
    link
    fedilink
    English
    arrow-up
    10
    arrow-down
    2
    ·
    edit-2
    19 hours ago

    Eh. I see this as an “flying landmine,” not an existential Terminator kind of thing.

    A computer is not making a “decision” to kill. Its a machine. It’s bomb exploding when triggered, just with extra steps of flying to a specific spot.


    It’s morally problematic. Like landmines.

    But I think people are falsly attributing anything resembling sentience to this system, like they do with conversation-trained LLMs. Humans made the “decision” way ahead of time, just like when they set up land mines or perform a bombing run.

  • tgirlschierke@lemmy.blahaj.zone
    link
    fedilink
    English
    arrow-up
    4
    ·
    19 hours ago

    Headline is somewhat misleading, article is based off an admission from a Ukrainian military officer about a test that happened two years ago.

  • Feathercrown@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    edit-2
    23 hours ago

    Even setting aside any apocalyptic scenarios, this is bad. You used to have to look your enemy in the eyes and stab them. Then you just had to see them from afar and shoot them. Now you don’t have to acknowledge them at all.

    • Nalivai@lemmy.world
      link
      fedilink
      English
      arrow-up
      10
      ·
      edit-2
      21 hours ago

      Most of the casualties were from bombers for a long time already, this is that but one step removed.
      It’s bad for a different reason, you in this case don’t exist anymore. We used to have people killing people, now there are autonomous robots doing this, and that’s a new, completely different class of warfare.

      • bthest@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        18 hours ago

        Even before bombers existed for every person killed on a battlefield a ton of people who didn’t see a single enemy soldier died because that army had destroyed or used up the region’s food supplies or introduced a new disease or diverted a river’s water.

        • k0e3@lemmy.ca
          link
          fedilink
          English
          arrow-up
          1
          ·
          12 hours ago

          Yeah, and also, many political leaders just say, “kill your enemies,” and the deed gets done nowadays. I’m pretty sure that’s been the case for a long time too.

  • wuffah@lemmy.world
    link
    fedilink
    English
    arrow-up
    31
    ·
    edit-2
    1 day ago

    The one-off test involved 10 AI-controlled “Terminator” drones on the front line of the Ukraine war.

  • Pickleideas@lemmy.world
    link
    fedilink
    English
    arrow-up
    14
    arrow-down
    3
    ·
    edit-2
    1 day ago

    I’m glad they, allegedly, stopped after the one test. This sounds an awful lot like a war crime on Ukraine’s part. Enemy combatants (especially since we know most of Russia’s army are being trafficked) need the opportunity to surrender peacefully.

  • Hueristic_Autistic@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    1 day ago

    No video? They did not know what the Ai chose until it sent a manned vehicle to check it out. That’s extraordinarily concerning.

      • Hueristic_Autistic@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        6
        ·
        1 day ago

        In the current landscape of military technology, what the public conceptually calls “terminator drones” refers to Lethal Autonomous Weapon Systems (LAWS), One-Way Attack (OWA) uncrewed aerial systems, and Collaborative Combat Aircraft (CCA).

        Because modern electronic warfare (EW) can instantly sever radio control and jam GPS, these systems cannot rely on a human pilot or cloud computing. They are designed as self-contained, edge-computing robotic hunters.

        The physical and technological anatomy of a modern autonomous combat drone is categorized into five core systems:


        1. The Neural Architecture (Edge AI Compute Core)

        The “brain” is no longer a simple autopilot board; it is an onboard AI accelerator optimized for computer vision and localized decision-making.

        • The System-on-Chip (SoC): Military-grade edge AI processors (such as ruggedized variants of the Nvidia Jetson Orin series or custom ASICs) capable of executing hundreds of Trillions of Operations Per Second (TOPS).
        • The Local Model Stack: Instead of connecting to an external server, the drone carries local, highly compressed convolutional neural networks (CNNs). These models are trained on massive synthetic datasets to immediately recognize, classify, and track military hardware (tanks, radar dishes, infantry) even when camouflaged or partially obscured.

        2. The Sensor Suite (The Perception Layer)

        To operate in “denied environments” where GPS is jammed, the drone relies on a fused sensory array to build an internal map of the world.

        • Electro-Optical/Infrared (EO/IR) Gimbals: High-resolution thermal and visual cameras that feed raw video directly into the AI computer vision core.
        • Optical Flow & Visual Inertial Odometry (VIO): Downward-facing cameras that track the movement of the ground pixel-by-pixel. Combined with an Inertial Measurement Unit (IMU), VIO allows the drone to navigate with pinpoint accuracy by “looking” at the terrain, rendering GPS entirely unnecessary.
        • Solid-State LiDAR / Micro-Radar: Used for low-altitude obstacle avoidance and capturing the 3D geometry of a target during the terminal attack phase.

        3. The EW-Resistant Communications & Swarm Mesh

        When drones operate collectively, they utilize decentralized mesh networking.

        • Software-Defined Radios (SDR): Radios that dynamically hop across thousands of frequencies per second to evade electronic jamming.
        • Inter-Drone Mesh Networking: The drones communicate directly with each other rather than a ground station. If Drone-A detects an air defense system, it instantly updates the target coordinates across the entire swarm mesh. If Drone-A is destroyed, the remaining swarm automatically re-allocates mission roles dynamically.

        4. Modular Airframe & Propulsion

        Modern mass-production initiatives (like the Pentagon’s Drone Dominance program) prioritize cost-effective, modular structures over exquisite, expensive aerospace frames.

        • Materials: Carbon fiber composites or high-density 3D-printed polymers designed for rapid assembly.
        • Propulsion: High-KV brushless electric motors powered by solid-state or high-capacity Lithium-Polymer (LiPo) batteries for small, tactical, low-signature loitering. Larger variants utilize small gas turbines or hybrid-rotary engines for extended range.
        • Signatures: The geometry is explicitly shaped to minimize both Radar Cross-Section (RCS) and acoustic signatures, allowing them to approach targets completely undetected until the final seconds.

        5. The Integrated Kinetic Payload (The Warhead)

        Modern military philosophy dictates that an attack drone is not just a vehicle carrying a bomb—the drone is the weapon. New architectures utilize highly specialized, plug-and-play modular payloads (such as the Terminus or Common UAS Payload designs).

        • Electronic Safe and Arm Devices (ESAD): Microprocessor-controlled safety systems that keep the warhead completely inert until the onboard AI confirms a positive target lock and enters the terminal dive.
        • Directional Frag/Shaped Charges: Optimized to direct the explosive energy entirely forward into the target upon impact, maximizing lethality while minimizing the structural weight the drone has to carry.

        The Functional Workflow (The Autonomy Loop)

        [ Launch ][ Visual Navigation (No GPS) ][ Onboard AI Target Detection ][ Terminal Engagement ] 🗲 [ Local Target Classification & Tracking ] ◄────┘
        
        

        When deployed, the drone is launched into a designated hunt-zone. It navigates purely via visual landmarks. The onboard AI constantly screens the video feed. When an object matches its classification matrix (e.g., a specific mobile missile launcher), the system locks onto the pixel coordinate, arms the ESAD, and executes a terminal dive completely independent of human input.

        This tightly integrated anatomy of Edge Compute + Computer Vision + Modular Lethality is what defines the reality of autonomous robotic warfare today.