无人机蜂群将AI推向现代战争前沿

Bloomberg News cybersecurity and tech reporter Katrin Hanson visited a farm in Maryland to get a firsthand look at how these swarms work. I started by asking her what she saw. Well, we drew up at this farm in Maryland. So imagine a field of young green corn, but the thing that immediately catches your eye is there's a Cold War era tank in the middle of it. And at the other end, there's a khaki tent and an operator who demonstrated for me a drone swarm attack.

And at the touch of a button, five quite small drones went up from the ground into the air, and you instantly get that buzzing sound, which for any of your viewers who have been following the war in Ukraine, that is the haunting sound of of this huge land war. And very quickly, the operator chooses pixels on the side of the tank for where they want these drones to attack. And at one touch of a button, all five drones essentially turn into missiles and, autonomously attack.

And so very quickly from just from the side, we saw all five drones attack that tank. They were actually loaded with not with plastic explosives as they've done before in a live experiment, but with chalk. So red chalk that exploded on the side of this tank simulate. Right. To simulate that they were on target and that the five drones were talking to each other. And Here's a tank. Yes. And we'll see. The the the company bought that for just over a $100,000 and then trailed it from the Midwest.

So the idea is that would be this coordinated attack, and it would blow up the tank. Yes. And the technology I mean, people have seen drones before. These drones have been around for a while. Yeah. What is it different from the drone shows that we see that are entertainment? So those entertainment ones, each drone is centrally controlled and given a precise precise GPS, coordinate exactly where to go, and they're extraordinary feats of imagine synchronized swimming.

Right. Everyone is preprogrammed to do exactly the same thing. The technology about drone swarms has been hard because although you can put a lot of different drones up in the air at the same time, the thing that operators in Ukraine and elsewhere have been trying to work out is GPS jamming. Now that means that the communication to the drone might be interrupted by the adversary, by the enemy, and so you lose control of your drone.

In the Ukraine, what they did was they've discovered, actually keeping that connection to the drone with with a with a cable with a fiber optic cable. So when the controller is controlling it, they have direct control over one drone. Now in a scenario like a Chinese attack on Taiwan and defending that, cables aren't gonna cut it. Right. But that could easily be GPS tramming. So one of the things that people are trying to develop is autonomous drone swarming.

So the operator would control multiple drones with one controller. And if that cuts, the drones can also talk to each other and attack either a preloaded target or go find a target using AI, which really is this very futuristic and highly unreliable tech at the moment. You know, there are a lot of concerns about that when it comes to autonomous drones. You know? Because what if they get the target wrong, or what if there's a misfunction?

How do how are they figuring out that? I mean, that's gotta be one of the problems or one of the obstacles they still have to figure out. I spoke to a US defense official about exactly this, and they have this effort that they launched at the beginning of this year called Swarm Forge. So the company I visited, Orterian, and 26 others are competing now in the third heat in September. Mhmm. But at the previous heat, you know, these are experiments that are expected to go wrong.

They're learning from each one. But the defense official explained to me this key thing, which I think is giving them pause and everybody else. Right. The drone was sent out to look in in what is called a boundary box area for targets for military vehicles, and it was supposed to distinguish between military and nonmilitary civilian vehicles. Mhmm. Instead, what happened is the AI on those drones started looking in the car park far away from where they were supposed to be looking and found white Chevy trucks, civilian Mhmm.

And suggested those as targets. Now that's exactly not what should be happening. And so the defense official thought that what had happened is the AI has been trained previously on previous data of, white Toyota Hiluxes in the desert, all the wars The US has previously been fighting. Mhmm. And the AI is reading off that old data, doing pattern recognition, which is what it's meant to do, but then not sticking to the new rules.

And so that is something those companies have to go back and work on. And I think it's key, you know, as people are so concerned about autonomous warfare. For me, it was an insight that the defense department is spending time distinguishing between military and civilian vehicles and not being satisfied. We just have a couple of minutes left, and I think about this company that you write about, Arterion. I mean, they started out doing delivery drones for Walmart, pivoted, right, when it came to the Russian invasion in Ukraine, the the the the the founder of the company, I believe it was.

So it's like warfare has definitely shifted in terms of the type of vehicles people wanna use or the type of weapons. And it does sound like it's not just a US thing of making them increasingly more autonomous. That's a global thing, though, and we've already seen it. Yeah. It's a global thing. Says these strikes, have been used in Ukraine autonomously. The first swarms, the Pentagon is trying to get to they're already experimenting with teens.

They hope to get to hundreds, but, obviously, The US is concerned what China is able to do, what Russia is able to do. Israel is many of the companies actually developing this tech are Israelis, so there are multiple countries that are ahead and experimenting with this tech. Ortyrian was doing defense contracts right from the beginning, but they've really scaled up and shifted and and no longer do, deliveries for Walmart.

So it is it is a changing, feature of war, but also a changing feature of, how tech is used. Katrina, when you see this, what just really strikes you? This is what you do this world. You, like, dig into whether it's artificial intelligence, generative AI, like, just how technology is advancing and how it's impacting how we fight wars. What do you think? I think, really, it's the mismatch. You have the secretary general of the United Nations saying machines that have the discretion, to to kill people without human control are morally repugnant.

And then you have this technology essentially existing today. And so that debate, that discussion is happening often without an understanding of sometimes the proponents of it would argue the conditions of war, and the campaigners against it would argue the real risks that this whole technology goes tremendously wrong. And watching that and watching the experiments always for me encourages me to look for the concrete examples of what goes right and what goes wrong.

Would you say, though, bottom line, that I mean, at this point, they everyone seems to say there's gonna be a human element to it. Yes. US defense policy is the commander verifies the target. That is still part of US policy. That is still the way these drones are working. Today. Today. But the same defense official I spoke to who explained how that process works said that if there are human lives at risk, they should carry on using human verification per target.

But there is a scenario, the official told me, where you would have drone on drone attacks. Mhmm. And there will be too many targets for single human to select. And at that point, the drones need to do the attacking. That's their view. As I said, new era of warfare in a big way.