WEBVTT 1 00:00:00.120 --> 00:00:02.600 Okay, let's just jump right into it. We talk a 2 00:00:02.600 --> 00:00:05.559 lot about how fast technology is moving, right like a 3 00:00:06.559 --> 00:00:09.080 you see a new smartphone, or you read about a 4 00:00:09.160 --> 00:00:11.919 chatbot that can write a decent poem, and it feels 5 00:00:11.919 --> 00:00:12.519 like a big deal. 6 00:00:12.640 --> 00:00:14.400 Oh yeah, it always feels like the biggest thing ever. 7 00:00:14.679 --> 00:00:17.800 Right, But today we're looking at a shift that is honestly, 8 00:00:18.160 --> 00:00:20.960 it's arguably as significant as the invention of gunpowder. 9 00:00:21.399 --> 00:00:23.519 I'd go as far as nuclear weapons. It's really on 10 00:00:23.559 --> 00:00:25.320 that historical scale exactly. 11 00:00:26.640 --> 00:00:29.960 We are talking about the so called third revolution in warfare. 12 00:00:30.160 --> 00:00:32.719 We're steering right down the barrel of artificial intelligence and 13 00:00:32.759 --> 00:00:36.359 autonomous weapons systems. And look, before you think we're about 14 00:00:36.399 --> 00:00:39.119 to sit here and recap the plot of some sci 15 00:00:39.159 --> 00:00:41.759 fi movie or you know, talk about a certain famous 16 00:00:41.759 --> 00:00:43.280 cyborg franchise. 17 00:00:43.039 --> 00:00:44.560 I absolutely know you want to bring up. 18 00:00:44.600 --> 00:00:46.079 I mean, it's right there. I always want to bring 19 00:00:46.119 --> 00:00:47.960 it up. But no, this is in fiction. For this 20 00:00:48.000 --> 00:00:50.560 deep dive, we are exploring a very real and very 21 00:00:50.600 --> 00:00:54.640 current analysis. Our source material today is the book Cybersecurity 22 00:00:54.679 --> 00:00:57.880 in the Age of Artificial Intelligence and Autonomous Weapons, edited 23 00:00:57.920 --> 00:01:00.240 by memet emmint Erindor, And. 24 00:01:00.119 --> 00:01:02.600 That's a really crucial distinction to make right out of 25 00:01:02.640 --> 00:01:05.879 the gate here, this isn't just a conversation about big 26 00:01:06.079 --> 00:01:09.920 shiny robots marching down the street. It's about the intersection 27 00:01:10.079 --> 00:01:15.239 where cybersecurity actually meets the battlefield. We're talking about invisible risks, 28 00:01:15.400 --> 00:01:22.000 things like software bugs, hacking vulnerabilities, data poisoning. 29 00:01:22.120 --> 00:01:25.239 Because if you have a weapon that literally thinks for itself, 30 00:01:25.920 --> 00:01:29.040 the scariest thing isn't necessarily the physical weapon, it's the 31 00:01:29.079 --> 00:01:29.920 code running it. 32 00:01:30.040 --> 00:01:31.799 Precisely, the code is the weapon. 33 00:01:31.959 --> 00:01:34.599 So our mission for this deep dive is to explore 34 00:01:34.640 --> 00:01:37.760 this new frontier. We really need to understand what happens 35 00:01:37.799 --> 00:01:39.359 when we take the kill chain and just hand it 36 00:01:39.400 --> 00:01:40.319 over to algorithms. 37 00:01:40.439 --> 00:01:43.519 Yeah, and to really grasp the cybersecurity dilemma that comes 38 00:01:43.560 --> 00:01:46.280 along with that, because think about it, when software is 39 00:01:46.319 --> 00:01:50.040 what pulls the trigger, a simple bug isn't just your 40 00:01:50.079 --> 00:01:52.359 screen freezing, it's a massive catastrophe. 41 00:01:52.560 --> 00:01:54.560 Let's get into the first part of this and define 42 00:01:54.560 --> 00:01:57.760 this new battlefield. The source material makes a fascinating point 43 00:01:57.840 --> 00:02:00.280 right off the bat, which is that security is just 44 00:02:00.280 --> 00:02:01.799 about physical borders anymore. 45 00:02:02.040 --> 00:02:04.959 No, it hasn't been for quite a while. The text 46 00:02:05.040 --> 00:02:09.800 highlights that we've completely moved into a multidisciplinary field. It's 47 00:02:09.800 --> 00:02:12.479 no longer just placing soldiers to guard. 48 00:02:12.240 --> 00:02:14.479 A line on a map, right, It's so much broader. 49 00:02:14.680 --> 00:02:19.159 Exactly, It's about protecting cyberspace, devices, networks and the data 50 00:02:19.159 --> 00:02:22.439 flowing between them all. But the real game changer, the 51 00:02:22.479 --> 00:02:25.599 thing the book focuses on is the hardware that operates 52 00:02:25.639 --> 00:02:27.719 autonomously within that space. 53 00:02:27.879 --> 00:02:30.759 Well, let's clarify that because we keep saying autonomous weapons 54 00:02:30.759 --> 00:02:33.719 systems or AWS, we need to be specific about what 55 00:02:33.719 --> 00:02:36.400 the sources mean. Because my nephew has one of those 56 00:02:36.439 --> 00:02:38.879 little drones he flies around the park. But I'm assuming 57 00:02:38.879 --> 00:02:40.479 we aren't talking about that level of tech. 58 00:02:40.599 --> 00:02:44.000 We definitely are not. The book actually cites specific definitions 59 00:02:44.039 --> 00:02:46.759 from the UN and the International Committee of the Red 60 00:02:46.759 --> 00:02:50.719 Cross to clarify this exact point. The primary distinction is 61 00:02:50.759 --> 00:02:54.319 the human element. So your nephew's drone, he's standing there 62 00:02:54.360 --> 00:02:57.199 holding a remote controller. He decides where it flies, he 63 00:02:57.240 --> 00:02:59.520 pushes a button to take a photo or in a 64 00:02:59.520 --> 00:03:01.840 military context, to drop a payload. 65 00:03:01.960 --> 00:03:03.400 So there's always a human in the loop. 66 00:03:03.680 --> 00:03:07.120 Right, That is classified as a remotely controlled platform, but 67 00:03:07.199 --> 00:03:11.319 an autonomous weapons system in AWS, it uses its own 68 00:03:11.400 --> 00:03:15.680 sensors and algorithms to detect, identify, track, and engage targets 69 00:03:15.719 --> 00:03:17.599 completely without human intervention. 70 00:03:17.879 --> 00:03:19.439 Completely without Yes. 71 00:03:19.439 --> 00:03:22.599 Once you switch it on, the system makes the decisions. 72 00:03:22.680 --> 00:03:24.280 The human is totally out of the loop. 73 00:03:24.639 --> 00:03:27.759 That phrase without human invention is really doing the heavy 74 00:03:27.800 --> 00:03:30.360 lifting there, And the book points out this isn't some 75 00:03:30.479 --> 00:03:33.599 far off theoretical concept. We actually have a documented date 76 00:03:33.639 --> 00:03:35.719 for one. This might have crossed over into reality right. 77 00:03:35.719 --> 00:03:38.520 Now, so we do. March twenty twenty, the source material 78 00:03:38.560 --> 00:03:42.000 spends time discussing the Cargo two drone incident in Libya. 79 00:03:42.120 --> 00:03:44.800 Yeah, reading that part gave me absolute chills. Walk us 80 00:03:44.800 --> 00:03:45.599 through what happened there. 81 00:03:45.639 --> 00:03:49.080 So it was during a conflict involving the Haftar armed forces. 82 00:03:49.319 --> 00:03:52.280 A UN expert panel put out a report mentioning that 83 00:03:52.319 --> 00:03:56.520 these lethal autonomous weapons systems were programmed to attack targets 84 00:03:56.719 --> 00:03:59.840 without requiring any data connectivity between the operator and the 85 00:03:59.840 --> 00:04:01.000 new Wait. 86 00:04:00.800 --> 00:04:04.120 So the robot was essentially off the leash entirely effectively. 87 00:04:04.199 --> 00:04:08.639 Yes, it was reportedly hunting down retreating logistics convoys. It 88 00:04:08.759 --> 00:04:11.199 wasn't hovering there waiting for a pilot back at a 89 00:04:11.199 --> 00:04:15.000 base to save fire. It was identifying its own targets 90 00:04:15.000 --> 00:04:18.319 based purely on its programming and then engaging them. That 91 00:04:18.480 --> 00:04:22.079 is the exact threshold we are talking about crossing. 92 00:04:21.959 --> 00:04:26.160 Which brings us to the core motivation. Why are militaries 93 00:04:26.240 --> 00:04:30.199 and nations pursuing this so aggressively? Why remove the human 94 00:04:30.240 --> 00:04:32.920 from the loop. Is it just the appeal of having 95 00:04:32.920 --> 00:04:34.639 the most cutting edge tech? 96 00:04:34.920 --> 00:04:37.639 Hardly? According to the contributors in the book, it really 97 00:04:37.680 --> 00:04:40.759 comes down to speed and efficiency. A human brain can 98 00:04:40.759 --> 00:04:44.720 only process battlefield information so quickly, right, we have limits exactly, 99 00:04:45.199 --> 00:04:48.959 But an ai. An AI can process a chaotic battlefield 100 00:04:48.959 --> 00:04:53.360 of windspeed, thermal signature's, movement patterns of dozens of objects 101 00:04:53.360 --> 00:04:54.879 in milliseconds. 102 00:04:54.199 --> 00:04:55.720 So it acts as a force multiplier. 103 00:04:55.800 --> 00:04:58.399 Right, You significantly reduce the physical risk to your own 104 00:04:58.439 --> 00:05:00.480 soldiers because you just don't need them phys on the 105 00:05:00.480 --> 00:05:04.279 front line, and you reduce costs. Autonomous machines don't need 106 00:05:04.360 --> 00:05:06.399 pensions or long term medical care. 107 00:05:06.920 --> 00:05:10.040 But the text argues this creates that cybersecurity dilemma we 108 00:05:10.079 --> 00:05:10.720 mentioned earlier. 109 00:05:11.040 --> 00:05:12.639 Yes, and it's a huge point. 110 00:05:12.800 --> 00:05:15.360 How is that dilemma different from just a standard arms 111 00:05:15.439 --> 00:05:17.040 race like during the Cold War. 112 00:05:17.240 --> 00:05:19.560 It's related to an arms race, but it's more specific 113 00:05:19.680 --> 00:05:22.560 to the nature of the technology. It's this rapid escalation 114 00:05:22.800 --> 00:05:27.399 spiral driven by deep insecurity. If Nation A builds a 115 00:05:27.439 --> 00:05:31.279 swarm of autonomous drones that can react faster than human reflexes, 116 00:05:31.759 --> 00:05:34.160 Nation B feels they have absolutely no choice but to 117 00:05:34.160 --> 00:05:37.000 build their own swarm just to survive, just to keep 118 00:05:37.079 --> 00:05:40.439 up right. But by doing so, both nations end up 119 00:05:40.480 --> 00:05:44.000 feeling vastly less secure because the sheer speed of a 120 00:05:44.040 --> 00:05:48.519 potential conflict has now been accelerated completely beyond human control. 121 00:05:48.600 --> 00:05:50.560 It's like two people pointing guns at each other, but 122 00:05:50.600 --> 00:05:53.120 the triggers are these insane hair triggers that might just 123 00:05:53.160 --> 00:05:54.720 go off if the wind blows too hard. 124 00:05:54.800 --> 00:05:57.519 That is a highly accurate analogy, and that hair trigger 125 00:05:57.560 --> 00:06:00.639 reality brings us to the Achilles heel of this entire concept, 126 00:06:01.000 --> 00:06:03.959 because here is the profound irony the book continually comes 127 00:06:04.000 --> 00:06:07.319 back to. We are pouring billions into building these super advanced, 128 00:06:07.399 --> 00:06:11.240 terrifying weapons systems, but at their absolute core, they are 129 00:06:11.279 --> 00:06:12.079 just computers. 130 00:06:12.439 --> 00:06:15.480 And computers can be hacked. I mean, my email account 131 00:06:15.480 --> 00:06:18.399 gets hacked, and it's a massive headache. If an autonomous 132 00:06:18.399 --> 00:06:20.959 tank gets hacked, that is globally catastrophic. 133 00:06:21.480 --> 00:06:25.800 The source material references a concept called normal accidents theory. 134 00:06:26.240 --> 00:06:29.600 It's an academic framework, but it is absolutely critical for 135 00:06:29.720 --> 00:06:30.839 understanding this risk. 136 00:06:31.040 --> 00:06:33.000 Can you unpack that theory for us a bit? 137 00:06:33.160 --> 00:06:37.160 Sure? It's a theory used to describe complex technological systems 138 00:06:37.199 --> 00:06:40.639 that are what they call tightly coupled tightly cut. 139 00:06:40.800 --> 00:06:41.000 Okay. 140 00:06:41.079 --> 00:06:43.720 Think of a loosely coupled system first, like a line 141 00:06:43.720 --> 00:06:47.040 of human soldiers marching. If one soldier trips over a rock, 142 00:06:47.079 --> 00:06:48.920 the others can just stop, help him up, or simply 143 00:06:48.959 --> 00:06:52.040 walk around him. There is natural slack in the system, 144 00:06:52.199 --> 00:06:56.120 human buffer room basically exactly. But in a tightly coupled system, 145 00:06:56.120 --> 00:06:59.319 like a network of autonomous weapons, everything is digitally integrated 146 00:06:59.360 --> 00:07:03.720 and reactions happen instantaneously. If just one part fails, maybe 147 00:07:03.720 --> 00:07:06.480 a single sensor of glitches or one line of code 148 00:07:06.560 --> 00:07:09.360 has a logic error, that failure cascades through the entire 149 00:07:09.399 --> 00:07:12.560 network before a human operator can even blink, let alone intervene. 150 00:07:12.720 --> 00:07:14.959 So if I'm an adversary, I don't necessarily need to 151 00:07:14.959 --> 00:07:17.519 blow up your million dollar robot with a missile. I 152 00:07:17.560 --> 00:07:18.800 just need to mess with its brain. 153 00:07:19.040 --> 00:07:22.600 That's exactly it. You attack the software. The book details 154 00:07:22.759 --> 00:07:26.360 several specific attack vectors that are highly concerning. Let's walk 155 00:07:26.399 --> 00:07:28.199 through a few of those, starting with spoofing. 156 00:07:28.399 --> 00:07:31.439 Right. Spoofing, this is where you feed the system fake 157 00:07:31.560 --> 00:07:34.920 signal data to confuse it. The classic example, the text 158 00:07:34.920 --> 00:07:37.839 springs up is from twenty twelve with a research team 159 00:07:37.879 --> 00:07:39.360 from the University of Texas. 160 00:07:39.439 --> 00:07:43.160 Yes, Todd Humphreys and his research team, they successfully used 161 00:07:43.279 --> 00:07:48.439 fake GPS signals to completely hijack an unmanned aerial drone. 162 00:07:48.480 --> 00:07:49.680 And they didn't even hack it right. 163 00:07:49.879 --> 00:07:53.120 Nope, they didn't break into the drone's actual computer system. 164 00:07:53.199 --> 00:07:56.839 They didn't crack any encryption passwords. They literally just convinced 165 00:07:56.879 --> 00:07:59.240 the drones navigation system that it was somewhere else. 166 00:07:59.319 --> 00:08:01.600 They gas at the robot in a manner. 167 00:08:01.319 --> 00:08:03.839 Of speaking, Yes they did. They fed it false data, 168 00:08:03.879 --> 00:08:06.639 so the drone thought it was suddenly gaining too much altitude, 169 00:08:06.920 --> 00:08:09.759 so its automated systems tried to correct that by diving 170 00:08:09.800 --> 00:08:12.600 down right into the ground or in this specific test case, 171 00:08:12.839 --> 00:08:14.920 right into the flight path of researchers wanted. 172 00:08:15.160 --> 00:08:18.240 That is wild, and that experiment was over a decade ago. 173 00:08:18.439 --> 00:08:21.920 Correct, the tech has only gotten more sophisticated. Then. Another 174 00:08:22.000 --> 00:08:25.399 major vector is data poisoning and This one is much 175 00:08:25.439 --> 00:08:28.199 more insidious because it happens during the machine learning phase, 176 00:08:28.519 --> 00:08:31.199 long before the weapon is ever deployed to a battlefield. 177 00:08:31.279 --> 00:08:33.519 This is where you mess with the massive data sets 178 00:08:33.519 --> 00:08:36.879 the AI trains on right, because these vision systems learn 179 00:08:37.200 --> 00:08:39.039 by looking at millions and millions. 180 00:08:38.759 --> 00:08:42.159 Of pictures exactly. You feed it data saying this is 181 00:08:42.200 --> 00:08:45.039 a tank, this is a civilian vehicle, this is a tree. 182 00:08:45.679 --> 00:08:48.639 If a bad actor can infiltrate that data set early 183 00:08:48.679 --> 00:08:50.720 on and subtly manipulate. 184 00:08:50.240 --> 00:08:53.080 It, like telling the AI that a civilian bus is 185 00:08:53.120 --> 00:08:54.519 actually a hostile tank. 186 00:08:54.480 --> 00:08:58.759 Yes, or training it so a specific camouflage pattern is 187 00:08:58.799 --> 00:09:00.759 ignored as just background noise. 188 00:09:00.519 --> 00:09:02.960 Then the AI essentially learns the wrong lesson, and it. 189 00:09:03.000 --> 00:09:05.559 Learns it perfectly. When you finally deploy that system into 190 00:09:05.559 --> 00:09:08.879 a live combat zone, it honestly thinks it's functioning one 191 00:09:08.960 --> 00:09:11.720 hundred percent correctly when it targets that silly and bus. 192 00:09:12.159 --> 00:09:15.039 The internal logic is perfectly sound based on what it learned, 193 00:09:15.279 --> 00:09:19.759 but the foundational data was poisoned, and because deep learning 194 00:09:19.799 --> 00:09:22.399 models often operate as a black box, you might not 195 00:09:22.480 --> 00:09:25.240 even realize that data was poisoned until after the smoke 196 00:09:25.360 --> 00:09:27.720 clears and the tragedy has already happened. 197 00:09:27.759 --> 00:09:31.440 That is genuinely terrifying, because machine doesn't experience doubt, It 198 00:09:31.480 --> 00:09:34.600 doesn't know it's making a mistake. It is absolutely confident 199 00:09:34.639 --> 00:09:35.240 in its error. 200 00:09:35.360 --> 00:09:39.159 Highly confident. Yes, and then you have jamming. This is 201 00:09:39.279 --> 00:09:42.759 the more brute force method of cyber attack, simply cutting 202 00:09:42.799 --> 00:09:45.360 off the communication link between the weapon and the base. 203 00:09:45.679 --> 00:09:49.159 Now, wait a second. If these things are totally autonomous, 204 00:09:49.240 --> 00:09:51.960 why does jamming even matter. Isn't the whole selling point 205 00:09:52.000 --> 00:09:54.639 that they don't need a human signal to keep fighting. 206 00:09:55.320 --> 00:09:58.440 You've hit on the ultimate double edged sword there. Proponents 207 00:09:58.480 --> 00:10:01.279 of these systems argue exactly that. They say AWS is 208 00:10:01.320 --> 00:10:04.360 superior because it works even if the enemy jams all 209 00:10:04.399 --> 00:10:07.440 your signals. Yeah, you know, the enemy cut our calms, 210 00:10:07.480 --> 00:10:09.919 No problem, our robots will just keep executing the mission. 211 00:10:10.000 --> 00:10:12.440 But I'm guessing there's a huge downside. 212 00:10:11.879 --> 00:10:14.960 A massive one. The text raises a critical counterpoint here. 213 00:10:15.480 --> 00:10:19.440 If the communication link is jammed, you completely lose the 214 00:10:19.519 --> 00:10:23.639 kill switch. If the AI suddenly starts malfunctioning or targeting 215 00:10:23.639 --> 00:10:27.600 the wrong facilities, or in a true nightmare scenario, turns 216 00:10:27.639 --> 00:10:31.519 on your own friendly forces, the human operator is totally 217 00:10:31.600 --> 00:10:34.879 locked out, you cannot send the signal to shut it down. 218 00:10:34.919 --> 00:10:38.080 You've built a monster you can't leash, and that naturally 219 00:10:38.200 --> 00:10:42.559 leads to what the book categorizes as the ultimate nightmare scenario, 220 00:10:42.720 --> 00:10:44.120 which is repurposing. 221 00:10:44.279 --> 00:10:47.440 Yes, repurposing. This is where a hacker doesn't just want 222 00:10:47.440 --> 00:10:50.440 to disable the weapon or jam its calms, they want 223 00:10:50.519 --> 00:10:54.639 administrative control. Imagine a scenario where a swarm of autonomous 224 00:10:54.679 --> 00:10:58.440 drones is hacked mid flight and suddenly their targeting protocols 225 00:10:58.480 --> 00:11:01.080 are rewritten to attack their own launch site. 226 00:11:01.360 --> 00:11:05.159 You've essentially handed your enemy a highly advanced weapon system 227 00:11:05.159 --> 00:11:07.000 that they didn't even have to spend a dime to 228 00:11:07.039 --> 00:11:08.000 build exactly. 229 00:11:08.159 --> 00:11:10.960 It really highlights how the modern battlefield isn't just physical 230 00:11:11.080 --> 00:11:13.919 terrain anymore. It's lines of code. But let's actually shift 231 00:11:13.960 --> 00:11:16.000 gears a bit. We've talked a lot about the shooting aspect, 232 00:11:16.320 --> 00:11:18.159 but before a military can shoot, they have. 233 00:11:18.080 --> 00:11:19.799 To see right the intelligence side. 234 00:11:19.919 --> 00:11:25.279 Yes, the text outlines how AI is completely revolutionizing intelligence gathering. 235 00:11:25.240 --> 00:11:28.879 And this is where AI is arguably already having its 236 00:11:28.919 --> 00:11:32.799 most massive real world impact. We're moving from just automating 237 00:11:32.799 --> 00:11:34.519 the trigger to automating the eyes. 238 00:11:34.759 --> 00:11:38.240 It's the sheer unmanageable scale of the data. The source 239 00:11:38.279 --> 00:11:42.000 material throws out a staggering figure eight hundred million global 240 00:11:42.039 --> 00:11:46.679 connected devices plus thousands of satellite images, plus endless social 241 00:11:46.720 --> 00:11:47.399 media feeds. 242 00:11:47.519 --> 00:11:50.320 It's an ocean of nois No building full of human 243 00:11:50.360 --> 00:11:52.559 analysts could possibly read or watch all of that. 244 00:11:52.799 --> 00:11:55.360 It's the classic needle in a haystack problem. But the 245 00:11:55.399 --> 00:11:58.320 haystack is the size of the entire planet. Humans literally 246 00:11:58.399 --> 00:12:02.360 cannot do it, but AI can. The book discusses Project 247 00:12:02.360 --> 00:12:03.799 Maven as a prime example of. 248 00:12:03.720 --> 00:12:06.159 This shift right. Project Maven this was the US military 249 00:12:06.200 --> 00:12:08.480 initiative that caused quite a massive stir in the tech 250 00:12:08.519 --> 00:12:09.519 sector a few years ago. 251 00:12:09.679 --> 00:12:13.200 It certainly did. It involved the military using AI algorithms 252 00:12:13.240 --> 00:12:16.120 to analyze drone footage. Specifically, they were using it to 253 00:12:16.279 --> 00:12:19.320 identify objects and find radical terrorist targets in places like 254 00:12:19.360 --> 00:12:22.639 Syria and Iraq. It effectively automated the stair surveillance. 255 00:12:23.000 --> 00:12:25.639 So instead of a human intelligence analyst having to sit 256 00:12:25.679 --> 00:12:28.919 there and watch twelve hours of grainy video of a 257 00:12:29.000 --> 00:12:32.039 random dirt road just to see if one specific truck 258 00:12:32.279 --> 00:12:34.519 drives by, the AI just watches it exactly. 259 00:12:34.639 --> 00:12:37.279 The AI watches the full twenty four hours of footage 260 00:12:37.279 --> 00:12:39.799 and just flags the three seconds that actually matter for 261 00:12:39.840 --> 00:12:44.480 the human interview. It handles signal intelligence and geospatial intelligence 262 00:12:44.639 --> 00:12:46.159 at a scale we've never seen. 263 00:12:46.720 --> 00:12:49.919 And it's not even just classified military footage anymore. The 264 00:12:49.960 --> 00:12:54.440 text mentions a specific software called Orbital Insight. This one 265 00:12:54.559 --> 00:12:56.519 really fascinated me because it's a commercial product. 266 00:12:56.559 --> 00:13:01.360 Right, it is entirely commercial. Orbital Insight analyzes satellite imagery 267 00:13:01.399 --> 00:13:04.919 on a massive global scale. It literally counts cars in 268 00:13:05.000 --> 00:13:08.480 retail parking lots. It monitors global supply chains. It spots 269 00:13:08.519 --> 00:13:11.159 anomalies and construction sites halfway around the world. 270 00:13:11.360 --> 00:13:14.039 So wait, an algorithm can tell if a rival Nations 271 00:13:14.080 --> 00:13:16.720 factory is running at full capacity. Yeah, just by counting 272 00:13:16.799 --> 00:13:18.399 the employee cars in the parking. 273 00:13:18.159 --> 00:13:21.840 Lot from space exactly. And if you know the factory's output, 274 00:13:22.000 --> 00:13:25.279 you know the economic health of your potential enemy. The 275 00:13:25.440 --> 00:13:28.120 immense benefit here, as the experts of the book point out, 276 00:13:28.159 --> 00:13:30.879 is that it moves intelligence from simply reacting to things 277 00:13:30.879 --> 00:13:33.879 that happen to predicting things before they happen. 278 00:13:34.240 --> 00:13:36.320 Predictive analytics for global conflict. 279 00:13:36.480 --> 00:13:40.120 Right, if you see the supply chain abruptly shifting in 280 00:13:40.159 --> 00:13:43.639 a certain way, or troop transports gathering at rail yards, 281 00:13:43.720 --> 00:13:46.960 or oil reserves suddenly being topped up. The AI can 282 00:13:47.039 --> 00:13:50.080 predict the conflict before the first shot is ever fired. 283 00:13:50.480 --> 00:13:53.720 But and in this topic, there is always a bail. 284 00:13:54.320 --> 00:13:57.399 But this level of technology isn't exclusive to the traditional 285 00:13:57.440 --> 00:14:00.399 good guys. This brings us to the human element and 286 00:14:00.799 --> 00:14:03.559 the incredibly dark side of this tech, which the book 287 00:14:03.600 --> 00:14:05.440 explores regarding radicalization. 288 00:14:05.919 --> 00:14:08.200 This is a really crucial chapter. We generally tend to 289 00:14:08.200 --> 00:14:11.320 think of AI as the exclusive domain of major superpowers 290 00:14:11.480 --> 00:14:14.360 the US, China, Russia, but the barrier to entry is 291 00:14:14.440 --> 00:14:18.480 dropping rapidly. Extremist groups like ISIS or Hezbola are actually 292 00:14:18.519 --> 00:14:20.360 early adopters of this technology. 293 00:14:20.720 --> 00:14:23.080 How exactly are they using it? Because I assume they 294 00:14:23.080 --> 00:14:27.120 aren't manufacturing autonomous tank swarms and secret facilities. 295 00:14:26.799 --> 00:14:30.799 Not hardware, no, but they are absolutely weaponizing the software. 296 00:14:31.120 --> 00:14:34.799 The text explains how these groups use AI driven algorithms 297 00:14:34.840 --> 00:14:38.759 for mass propaganda. They use it to create highly tailored 298 00:14:38.799 --> 00:14:42.440 information bubbles to recruit people much more efficiently than ever. 299 00:14:42.279 --> 00:14:45.120 Before, like using social media algorithms against us. 300 00:14:45.240 --> 00:14:49.000 Exactly, they use these tools to find psychologically vulnerable people 301 00:14:49.000 --> 00:14:52.399 online and then feed them a constant stream of highly 302 00:14:52.440 --> 00:14:54.120 specific radicalizing content. 303 00:14:54.240 --> 00:14:56.879 So it's essentially hyper targeted marketing, but. 304 00:14:56.879 --> 00:14:59.960 For terrorism precisely, and deep fakes play a massive role 305 00:15:00.279 --> 00:15:00.919 in the strategy. 306 00:15:01.000 --> 00:15:02.519 Oh absolutely defikes her everywhere. 307 00:15:02.519 --> 00:15:07.120 Now they're using AI to create incredibly realistic fake videos 308 00:15:07.159 --> 00:15:10.919 of political leaders saying things they never actually said, or 309 00:15:11.000 --> 00:15:14.360 fabricating video evidence of war crimes to incite immediate anger 310 00:15:14.399 --> 00:15:18.039 and violence. It entirely blurs the line between truth and. 311 00:15:18.000 --> 00:15:20.080 Fiction, which is a weapon in itself. 312 00:15:20.279 --> 00:15:23.159 Exactly, if a population can't trust its own eyes, they 313 00:15:23.159 --> 00:15:25.200 can be easily manipulated into conflict. 314 00:15:25.480 --> 00:15:28.840 And this brings up a huge legal headache that the 315 00:15:28.879 --> 00:15:32.559 book calls the accountability gap, because if you think about it, 316 00:15:33.679 --> 00:15:37.080 let's say an autonomous weapon or even a radicalized AI 317 00:15:37.200 --> 00:15:41.120 bought commits a documented war crime. Who actually goes to 318 00:15:41.200 --> 00:15:41.960 jail for that? 319 00:15:41.960 --> 00:15:45.440 That is the multi billion dollar legal question of our era. 320 00:15:46.159 --> 00:15:48.679 Is it the software programmer who wrote the original code 321 00:15:48.720 --> 00:15:51.919 three years ago in an office, is it the military 322 00:15:51.960 --> 00:15:55.480 commander who authorized the deployment of the unit, or is 323 00:15:55.480 --> 00:15:56.679 it the machine itself? 324 00:15:56.759 --> 00:15:59.000 Well, you obviously can't put a machine in a jail cell. 325 00:15:59.000 --> 00:16:01.080 You can't court martial, you can't. 326 00:16:01.240 --> 00:16:03.639 And because of that, Black box problem we mentioned earlier, 327 00:16:03.679 --> 00:16:06.639 the fact that deep learning algorithms often cannot explain the 328 00:16:06.639 --> 00:16:09.440 why behind a specific decision. You might not even be 329 00:16:09.440 --> 00:16:12.480 able to legally prove negligence by the programmer or the commander. 330 00:16:12.600 --> 00:16:12.919 Wow. 331 00:16:13.039 --> 00:16:17.000 It creates this terrifying legal vacuum where absolute atrocities could 332 00:16:17.000 --> 00:16:20.080 happen on the battlefield without any single human being held 333 00:16:20.120 --> 00:16:22.639 legally responsible under current international law. 334 00:16:22.840 --> 00:16:25.919 The book also gets a little philosophical here, which I appreciated, 335 00:16:25.960 --> 00:16:28.960 referencing Karlvon Clausovits. I know, Class of Its is kind 336 00:16:28.960 --> 00:16:32.120 of the founding father of modern war theory, but how 337 00:16:32.200 --> 00:16:35.919 on earth does an eighteenth century Prussian general fit into 338 00:16:35.960 --> 00:16:37.639 the age of artificial intelligence. 339 00:16:37.799 --> 00:16:40.799 It's a great connection the authors make, so Clausibitz famously 340 00:16:40.840 --> 00:16:43.039 talked about the trinity of war. He said war is 341 00:16:43.080 --> 00:16:46.399 driven by three forces. The government, which provides the rational 342 00:16:46.440 --> 00:16:51.240 political objective, the military, which provides the operational skill and chance, 343 00:16:51.600 --> 00:16:54.240 and the people who provide the raw passion and hostility. 344 00:16:54.399 --> 00:16:56.399 Okay, so government, military, people right. 345 00:16:56.720 --> 00:16:59.679 The expert contributors in the book argue that autonomous weapon 346 00:16:59.679 --> 00:17:03.919 system fundamentally disrupt this trinity. AI removes the passion and 347 00:17:04.000 --> 00:17:07.480 it attempts to eliminate the chance war suddenly becomes this 348 00:17:07.720 --> 00:17:10.400 very cold, calculated, mathematical exercise. 349 00:17:10.480 --> 00:17:12.920 I mean, from a purely tactical standpoint, that almost sounds 350 00:17:12.920 --> 00:17:13.079 of it. 351 00:17:13.200 --> 00:17:19.680 YEA efficient maybe, but incredibly dangerous Strategically. If war becomes 352 00:17:19.759 --> 00:17:23.079 cold and calculated and a leader doesn't have to risk 353 00:17:23.160 --> 00:17:25.559 the lives of their own human citizens, they just send 354 00:17:25.559 --> 00:17:29.559 in the robots, it drastically lowers the political threshold for 355 00:17:29.599 --> 00:17:31.200 starting a conflict in the first place. 356 00:17:31.279 --> 00:17:34.279 Oh, I see, it makes going to war politically easier exactly. 357 00:17:34.920 --> 00:17:37.240 If the cost of waging a war is just measured 358 00:17:37.279 --> 00:17:40.279 in tax dollars and silicon ships rather than human blood 359 00:17:40.279 --> 00:17:43.839 and coffins coming home, leaders might be far quicker to 360 00:17:43.839 --> 00:17:46.759 pull the trigger on an invasion, and that is a 361 00:17:46.920 --> 00:17:50.599 deeply terrifying prospect for overall global stability. 362 00:17:50.720 --> 00:17:54.039 It's making war completely clean for the aggressor, while it 363 00:17:54.200 --> 00:17:57.880 obviously remains horrifically dirty for the victim. Exactly. Let's shift 364 00:17:57.920 --> 00:18:00.920 to our final major topic. Here' talked extensively about the 365 00:18:00.960 --> 00:18:02.640 software in the code, but at the end of the day, 366 00:18:03.079 --> 00:18:06.559 these autonomous machines are physical objects made of metal and silicon. 367 00:18:07.119 --> 00:18:08.960 We have to talk about the global supply chain. 368 00:18:09.039 --> 00:18:13.559 This is an incredibly critical and often completely overlooked vulnerability 369 00:18:13.640 --> 00:18:17.119 that the text highlights. An autonomous weapon, whether it's a 370 00:18:17.240 --> 00:18:21.480 drone or a submarine, is just a massive collection of microchips, 371 00:18:21.559 --> 00:18:25.400 sensors and processors. And where do all those physical components 372 00:18:25.440 --> 00:18:25.880 come from. 373 00:18:25.960 --> 00:18:27.160 They come from all over the world. 374 00:18:27.160 --> 00:18:31.640 They're manufactured globally, right, The global electronics supply chain is 375 00:18:31.680 --> 00:18:35.640 incredibly vast and complex. The book points out that malicious 376 00:18:35.680 --> 00:18:38.400 actors don't actually need to hack your robot while it's 377 00:18:38.400 --> 00:18:41.480 on the battlefield. They can just infiltrate the manufacturing process 378 00:18:41.880 --> 00:18:45.240 years earlier by tampering with the hardware itself exactly. They 379 00:18:45.240 --> 00:18:49.279 can introduce a compromised hardware component, a tiny back door 380 00:18:49.400 --> 00:18:52.400 baked right into the silicon of a processor before the 381 00:18:52.400 --> 00:18:54.279 weapon system is even fully assembled. 382 00:18:54.440 --> 00:18:57.480 So a military could spend millions buying the absolute latest, 383 00:18:57.519 --> 00:19:00.920 greatest autonomous tank, but its secretly has an off switch 384 00:19:01.039 --> 00:19:03.839 that an adversary installed at a factory. 385 00:19:03.440 --> 00:19:06.160 In another country, or it's a switch that quietly makes 386 00:19:06.200 --> 00:19:10.720 the tank broadcast its exact GPS coordinates constantly, or maybe 387 00:19:10.759 --> 00:19:13.039 a logic bomb hard coded into the chip that just 388 00:19:13.160 --> 00:19:16.200 permanently disables the weapon systems on a highly specific date. 389 00:19:16.440 --> 00:19:19.880 That's insidious, and we actually have a massive historical precedent 390 00:19:20.039 --> 00:19:23.799 proving that cyber weapons can physically destroy hardware. The book 391 00:19:23.799 --> 00:19:24.799 brings up Stucksnet. 392 00:19:24.839 --> 00:19:28.079 Stucks Net, right, that is the absolute classic example of this. 393 00:19:28.160 --> 00:19:30.960 It is the ultimate proof of concept. Stuck's Net was 394 00:19:31.000 --> 00:19:34.839 a highly advanced computer worm that targeted Iran's nuclear enrichment 395 00:19:34.839 --> 00:19:38.279 facilities back in the early twenty tens. But the key 396 00:19:38.319 --> 00:19:41.079 thing is it didn't just steal data or crash computers. 397 00:19:41.480 --> 00:19:45.720 It physically destroyed the actual heady machinery the centrifuges by 398 00:19:45.799 --> 00:19:49.519 manipulating their motors to spin entirely out of control, all. 399 00:19:49.319 --> 00:19:53.319 While actively feeding fake data to the monitoring software, so 400 00:19:53.359 --> 00:19:56.599 the human operators thought everything was running totally fine exactly. 401 00:19:56.680 --> 00:20:00.039