WEBVTT 1 00:00:01.199 --> 00:00:06.200 Welcome to the Sentient Code, where intelligence is engineered, autonomy 2 00:00:06.280 --> 00:00:10.439 is emerging, and a line between human and machine grows thinner. 3 00:00:10.800 --> 00:00:15.359 Each episode, we decode the algorithms, explore the robotics, and 4 00:00:15.439 --> 00:00:21.879 examine the ideas shaping the future of artificial minds. 5 00:00:23.960 --> 00:00:26.199 So I want you to just take a second and 6 00:00:26.239 --> 00:00:30.399 imagine a machine. Okay, but like not the sleek, indestructible 7 00:00:30.480 --> 00:00:32.679 kind of android you see in you know, big budget 8 00:00:32.719 --> 00:00:33.560 sci fi movies. 9 00:00:33.679 --> 00:00:33.880 Yeah. 10 00:00:33.920 --> 00:00:37.000 I'm talking about something that is inherently vulnerable. 11 00:00:36.560 --> 00:00:38.840 Right, something grounded in reality exactly. 12 00:00:39.159 --> 00:00:41.759 So let's say this machine is navigating a really rough 13 00:00:41.840 --> 00:00:44.640 environment and it just takes a massive hit, like it 14 00:00:44.679 --> 00:00:48.560 gets crushed by falling debris, or maybe it completely loses 15 00:00:48.600 --> 00:00:49.719 a functional limb. 16 00:00:49.759 --> 00:00:52.280 Which in the normal world of engineering is pretty much 17 00:00:52.280 --> 00:00:52.920 the end of the story. 18 00:00:53.000 --> 00:00:55.960 Right. The machine throws a shower of sparks, the motors whine, 19 00:00:55.960 --> 00:00:59.600 and it just diesoo. But imagine a machine that doesn't 20 00:00:59.640 --> 00:01:02.600 just break and shut down. Oh wow, Okay, imagine it 21 00:01:02.759 --> 00:01:06.120 like quitches. It seems to assess what is actually left 22 00:01:06.120 --> 00:01:09.120 of its shattered body, figures out a completely new way 23 00:01:09.159 --> 00:01:11.719 to distribute its weight, and then it just it just 24 00:01:11.840 --> 00:01:12.840 keeps moving forward. 25 00:01:13.359 --> 00:01:15.799 That is, uh, that's a wild visual. 26 00:01:16.040 --> 00:01:18.680 Yeah, it looks less like a broken household appliance and 27 00:01:18.799 --> 00:01:20.920 you know, much more like a wounded animal that's just 28 00:01:21.000 --> 00:01:22.439 absolutely refusing to give up. 29 00:01:22.680 --> 00:01:25.560 It's an unsettling image for sure, But I mean it 30 00:01:25.680 --> 00:01:27.799 cuts to the absolute core of what we're looking at 31 00:01:27.799 --> 00:01:33.680 today because historically, that specific ability to adapt, to evolve, 32 00:01:33.760 --> 00:01:38.959 to survive sudden catastrophic physical damage in real time, that 33 00:01:39.000 --> 00:01:42.640 has belonged exclusively to biology, right to living things exactly, 34 00:01:42.840 --> 00:01:46.239 for billions of years. Living organisms have refined this capacity 35 00:01:46.480 --> 00:01:50.120 to persist through severe injury, you know, reorganizing their biological 36 00:01:50.120 --> 00:01:53.400 functions to keep operating in totally unpredictable environments. 37 00:01:53.519 --> 00:01:53.719 Right. 38 00:01:54.319 --> 00:01:57.599 But machines, on the other hand, they've always been inherently brittle. 39 00:01:57.879 --> 00:02:00.159 They rely on perfect pristine can. 40 00:02:00.400 --> 00:02:03.599 Up until right now, basically exactly because today we're unpacking 41 00:02:03.640 --> 00:02:08.960 some genuinely mind blowing research coming out of Northwestern University 42 00:02:09.039 --> 00:02:12.800 that is actively dissolving that boundary between biology and machinery. 43 00:02:12.879 --> 00:02:14.719 It's a total paradigm shift, it really is. 44 00:02:14.800 --> 00:02:17.919 We are looking at the development of dynamic modular robots 45 00:02:17.919 --> 00:02:21.120 that are capable of evolving and not just in how 46 00:02:21.159 --> 00:02:23.599 they behave or the software code they run, but in 47 00:02:23.639 --> 00:02:25.639 their actual physical. 48 00:02:25.159 --> 00:02:26.560 Form, yeah, the hardware itself. 49 00:02:26.759 --> 00:02:29.560 Right, So our goal today is to really figure out 50 00:02:29.560 --> 00:02:33.479 how engineers are stepping away from these rigid, pre programmed 51 00:02:33.479 --> 00:02:38.000 tools and moving toward machines that can literally thrive in 52 00:02:38.159 --> 00:02:39.240 unpredictable chaos. 53 00:02:39.319 --> 00:02:40.879 It's going to be a fun one, oh for sure. 54 00:02:41.400 --> 00:02:43.919 So whether you are someone who builds drones in your 55 00:02:43.919 --> 00:02:47.439 garage or you're just intensely curious about the future of automation, 56 00:02:47.960 --> 00:02:51.840 you are about to experience a major aha moment regarding 57 00:02:51.879 --> 00:02:53.240 what a machine can actually be. 58 00:02:53.759 --> 00:02:57.360 I love that, and I think to really grasp how 59 00:02:57.360 --> 00:03:01.360 a robot survives the kind of catastrophic limb crushing damage 60 00:03:01.400 --> 00:03:04.680 you just described, we have to we have to rewind a. 61 00:03:04.599 --> 00:03:06.680 Bit, okay, rewind to wear to look. 62 00:03:06.520 --> 00:03:09.280 At how it's born, I guess you could say, or 63 00:03:09.360 --> 00:03:12.280 rather how it's constructed from day one. At the center 64 00:03:12.319 --> 00:03:15.680 of this Northwestern breakthrough is this concept of modularity. 65 00:03:15.919 --> 00:03:16.639 Modularity. 66 00:03:16.919 --> 00:03:19.800 Yeah, so instead of building a robot as a single 67 00:03:19.879 --> 00:03:24.319 permanent entity, like pouring a solid engine block, these systems 68 00:03:24.360 --> 00:03:26.680 are made out of smaller, interchangeable. 69 00:03:26.080 --> 00:03:28.319 Units, and it like highly advanced construction toys. 70 00:03:28.400 --> 00:03:29.879 Right, It's a great way to think about it. 71 00:03:30.120 --> 00:03:32.199 Like, you don't have a fixed arm or a fixed 72 00:03:32.199 --> 00:03:35.719 torso you just have a bin of active building block exactly. 73 00:03:35.759 --> 00:03:38.759 And each of those modules has its own basic capability. 74 00:03:38.879 --> 00:03:43.000 So maybe a small motor, some sensors, and these connection. 75 00:03:42.639 --> 00:03:44.120 Faces they can snap together. 76 00:03:44.400 --> 00:03:48.000 Right, They have the mechanical ability to connect, detach, and 77 00:03:48.039 --> 00:03:51.879 communicate with the modules immediately adjacent to them. So the 78 00:03:51.919 --> 00:03:53.960 structure itself is incredibly floid. 79 00:03:54.159 --> 00:03:56.800 But I mean people have been tinkering with modular robotics 80 00:03:56.800 --> 00:03:59.599 for a couple of decades. Right, this isn't brand new. 81 00:04:00.000 --> 00:04:02.719 Oh, the concept of modularity isn't new. Yeah, But the 82 00:04:02.800 --> 00:04:06.319 critical pivot here, the thing that changes the entire landscape 83 00:04:06.879 --> 00:04:10.879 is handing the design process over to an AI driven 84 00:04:11.159 --> 00:04:12.439 evolutionary algorithm. 85 00:04:12.479 --> 00:04:14.599 Handing over the design Okay, yeah. 86 00:04:14.280 --> 00:04:16.160 The final shape of the robot and the way it 87 00:04:16.199 --> 00:04:18.680 decides to move, they are not mapped out by a 88 00:04:18.759 --> 00:04:23.160 human sitting at a CAB program. They're discovered through merciless virtual. 89 00:04:22.800 --> 00:04:27.759 Simulation, discovered rather than engineered. That's literally natural selection on 90 00:04:27.839 --> 00:04:28.439 fast forward. 91 00:04:28.519 --> 00:04:29.519 That's exactly what it is. 92 00:04:29.720 --> 00:04:32.920 But thinking about the mechanics of that simulation for a second. 93 00:04:33.680 --> 00:04:36.319 If the AI is just sort of throwing random blocks 94 00:04:36.360 --> 00:04:40.160 together and trying out millions of combinations, aren't the vast 95 00:04:40.160 --> 00:04:44.360 majority of these virtual robots just immediately flailing, collapsing, and 96 00:04:44.439 --> 00:04:47.639 falling on their faces the second the simulation starts. Oh. 97 00:04:47.680 --> 00:04:51.920 Absolutely, the failure rate is astronomical. I figured, yeah, but 98 00:04:52.040 --> 00:04:55.519 that is exactly the engine driving the innovation. In these 99 00:04:55.560 --> 00:04:59.759 simulated physics environments. The AI is compressing what would take 100 00:05:00.040 --> 00:05:04.959 biology millions of years into literally a matter of minutes. 101 00:05:05.079 --> 00:05:06.079 That is wild. 102 00:05:06.240 --> 00:05:09.319 It generates thousands of candidate designs. Each one has a 103 00:05:09.319 --> 00:05:12.680 slightly different arrangement of modules and a slightly different internal 104 00:05:12.759 --> 00:05:15.680 rhythm for firing its motors. Okay, they're dropped into a 105 00:05:15.759 --> 00:05:19.519 virtual arena and give it a basic mandate like move 106 00:05:19.560 --> 00:05:22.800 forward or maintain your balance, And like you said, almost 107 00:05:22.800 --> 00:05:23.360 all of them fail. 108 00:05:23.399 --> 00:05:24.519 Incidentally, you just wipe out. 109 00:05:24.600 --> 00:05:27.160 Yeah, they tip over, they spin in useless circles, or 110 00:05:27.199 --> 00:05:29.920 they just vibrate aggressively without going anywhere. 111 00:05:29.959 --> 00:05:32.879 But the AI doesn't just delete them and start over blindly, right, Oh, 112 00:05:33.000 --> 00:05:36.399 none at all. It acts like a roofless evolutionary judge. 113 00:05:36.839 --> 00:05:39.199 It looks at the one bizarre assortment of blocks that 114 00:05:39.279 --> 00:05:42.639 accidentally managed to drag itself half an inch forward before collapsing, 115 00:05:42.680 --> 00:05:46.439 And it's like, I'm keeping that it saves that specific trait. 116 00:05:46.600 --> 00:05:49.319 Yes, it takes the traits from the top one percent 117 00:05:49.399 --> 00:05:54.759 of performers, recombines them, mutates them slightly to introduce new variations, 118 00:05:55.319 --> 00:05:57.160 and then runs the simulation. 119 00:05:56.720 --> 00:05:59.519 Again, generation after generation exactly. Yeah. 120 00:06:00.000 --> 00:06:04.240 System naturally converges on designs that are highly efficient at 121 00:06:04.279 --> 00:06:08.800 moving through that specific virtual environment, and because human brain 122 00:06:08.879 --> 00:06:13.759 isn't guiding the esthetics, the end results are completely alien. 123 00:06:14.240 --> 00:06:18.519 I mean, the AI produces these wildly bizarre, asymmetrical, organic 124 00:06:18.560 --> 00:06:19.759 looking structures, which. 125 00:06:19.600 --> 00:06:22.279 Is so interesting because human engineers are just obsessed with symmetry. 126 00:06:22.360 --> 00:06:25.680 Oh absolutely, we want four equal wheels, two identical legs, 127 00:06:25.680 --> 00:06:29.120 perfectly balanced geometric shapes. We build things that look clean, 128 00:06:29.399 --> 00:06:32.519 you know, always, But the AI doesn't care about looking clean. 129 00:06:32.600 --> 00:06:35.399 It only cares about what survives the harsh math of 130 00:06:35.439 --> 00:06:36.439 the physics engine. 131 00:06:36.680 --> 00:06:39.800 And that lack of human bias is crucial here. Because 132 00:06:39.839 --> 00:06:43.079 these machines are evolved through trial by fire rather than 133 00:06:43.120 --> 00:06:47.279 being traditionally drafted, the relationship between their hardware and their 134 00:06:47.319 --> 00:06:51.879 software changes funomously, oh so well. In a standard robot, 135 00:06:52.439 --> 00:06:56.519 hardware and software are fiercely protected as two completely separate layers. 136 00:06:56.560 --> 00:07:00.319 Okay, So traditional robotics is basically like a person driving 137 00:07:00.360 --> 00:07:00.720 a car. 138 00:07:00.879 --> 00:07:01.319 I like that. 139 00:07:01.480 --> 00:07:04.759 Yeah, you have a fixed rigid body the car, and 140 00:07:04.839 --> 00:07:09.079 it's operated by a completely separate, adaptable intelligence the driver. 141 00:07:09.680 --> 00:07:12.839 You can upgrade the driver's skills without ever changing the 142 00:07:12.879 --> 00:07:13.800 car's engine. 143 00:07:13.839 --> 00:07:16.360 That's a perfect analogy for traditional systems. 144 00:07:16.439 --> 00:07:20.839 But these evolved robots, they seem more like an elite athlete. Okay, 145 00:07:20.920 --> 00:07:23.759 keep gun, like with a gymnast or a sprinter. Their 146 00:07:23.800 --> 00:07:27.399 muscle memory, their nervous system reflexes, and their literal bone 147 00:07:27.439 --> 00:07:30.279 structure are completely inseparable from the way they move. 148 00:07:30.560 --> 00:07:33.600 That athlete analogy captures the exact shift we're seeing. In 149 00:07:33.680 --> 00:07:36.920 traditional engineering. You build a generic robotic arm, and then 150 00:07:36.959 --> 00:07:39.839 you write a software program to tell that specific arm 151 00:07:40.120 --> 00:07:44.240 how to swing. But with these Northwestern robots, the behavioral 152 00:07:44.240 --> 00:07:47.639 strategy is deeply embedded in the physical form. 153 00:07:47.360 --> 00:07:48.839 Itself, embedded in the form. 154 00:07:49.000 --> 00:07:52.720 Yeah, the exact placement of those building block modules, the 155 00:07:52.800 --> 00:07:56.040 asymmetrical shape they form, the rhythm of their movement, it's 156 00:07:56.199 --> 00:08:01.120 all part of a single coevolved solution. Wow. The robot 157 00:08:01.160 --> 00:08:04.759 embodies movement as a property of its structure. Its physical 158 00:08:04.800 --> 00:08:07.879 shape is the literal reason it walks the way it does. 159 00:08:08.439 --> 00:08:10.680 The brain and the body they grew up together in 160 00:08:10.720 --> 00:08:11.399 the simulation. 161 00:08:12.040 --> 00:08:14.759 If the body and the brain are that tightly woven together, 162 00:08:15.120 --> 00:08:18.000 then what happens when half of that integration gets smashed 163 00:08:18.000 --> 00:08:18.879 by a falling rock? 164 00:08:19.040 --> 00:08:20.240 Right the catastrophe? 165 00:08:20.399 --> 00:08:23.240 Yeah, this brings us back to that wounded animal concept 166 00:08:23.240 --> 00:08:26.079 from the start. Let's look at how conventional robots handle 167 00:08:26.120 --> 00:08:30.360 catastrophe first, because normally, damage to a highly engineered system 168 00:08:30.439 --> 00:08:31.360 is fatal. 169 00:08:31.120 --> 00:08:34.559 Totally fatal. In a conventional robot, the control code relies 170 00:08:34.600 --> 00:08:39.039 on rigid mathematical models, specifically something called inverse kinematics. The 171 00:08:39.120 --> 00:08:42.799 central computer is constantly running equations that assume the physical 172 00:08:42.799 --> 00:08:47.080 body matches the original blueprints perfectly. It calculates that joint 173 00:08:47.120 --> 00:08:50.159 A is exactly twenty centimeters away from joint B. 174 00:08:50.600 --> 00:08:54.080 So if a heavy piece of debris falls and cleanly 175 00:08:54.200 --> 00:08:57.480 snaps off one of the robot's four legs, the computer 176 00:08:57.639 --> 00:08:58.279 just panics. 177 00:08:58.320 --> 00:08:59.799 Oh, it completely freaks out. 178 00:09:00.000 --> 00:09:02.840 Because the code is still trying to send movement commands 179 00:09:02.919 --> 00:09:05.879 to a motor that literally doesn't exist anymore. And the 180 00:09:05.879 --> 00:09:08.919 matrix math just returns an endless string of errors. 181 00:09:09.039 --> 00:09:13.320 Exactly, the robot freezes or crashes because its mental map 182 00:09:13.320 --> 00:09:16.200 of its body no longer matches physical reality. 183 00:09:16.320 --> 00:09:20.559 But the Northwestern robots operate under a totally different paradigm. 184 00:09:20.159 --> 00:09:23.600 Completely different because their control isn't centralized into one rigid 185 00:09:23.600 --> 00:09:27.080 mathematical model. Damage doesn't cause a systemic crash. So what 186 00:09:27.120 --> 00:09:29.879 does it do when a module is forcibly removed or 187 00:09:29.919 --> 00:09:33.519 an entire appendage becomes non functional? The robot doesn't throw 188 00:09:33.559 --> 00:09:36.639 an air code and shut down. It reinterprets its own 189 00:09:36.639 --> 00:09:37.879 physical form on the fly. 190 00:09:38.159 --> 00:09:41.360 See the idea of a machine reinterpreting its own body 191 00:09:41.559 --> 00:09:45.240 feels so incredibly sci fi. It does, But there is 192 00:09:45.279 --> 00:09:48.480 a specific example from the research that illustrates us perfectly. 193 00:09:48.559 --> 00:09:49.679 Oh the four legged one. 194 00:09:49.799 --> 00:09:52.840 Yes, so they had this physical robot evolved in the 195 00:09:52.840 --> 00:09:57.399 simulation to walk on four somewhat awkward appendages. It's moving 196 00:09:57.399 --> 00:10:00.960 across the floor and the researcher is intentional remove one 197 00:10:01.000 --> 00:10:03.879 of the appendages, s rip it right off. Yeah. Now, 198 00:10:03.919 --> 00:10:07.480 a normal machine would tip over and helplessly spin its 199 00:10:07.480 --> 00:10:13.360 remaining wheels, but this robot, it momentarily stumbles and then 200 00:10:13.399 --> 00:10:17.320 smoothly shifts its internal rhythm to adopt a three point gate. 201 00:10:17.440 --> 00:10:18.600 It's incredible to watch. 202 00:10:18.720 --> 00:10:21.639 And then when they damaged it even further, literally destroying 203 00:10:21.679 --> 00:10:24.759 its ability to walk upright at all, it didn't give up. 204 00:10:25.080 --> 00:10:28.399 It simply started exploiting whatever degrees of freedom were left 205 00:10:28.399 --> 00:10:30.960 in its surviving modules to drag itself. 206 00:10:30.639 --> 00:10:32.960 Forward across the floor, just refusing to quit. 207 00:10:33.159 --> 00:10:37.639 It accepted its new mangled condition instantly. But okay, I 208 00:10:37.679 --> 00:10:39.120 have to push back on the mechanics of this a 209 00:10:39.159 --> 00:10:43.559 little bit. Is the robot actually pausing recognizing it lost 210 00:10:43.559 --> 00:10:46.919 a leg and calculating a massive new physics equation for 211 00:10:46.960 --> 00:10:49.879 how to balance on three points? Or is it more 212 00:10:50.000 --> 00:10:53.240 like a dog that steps on a thorn, hurts its paw, 213 00:10:53.720 --> 00:10:56.679 and instantly just knows how to hobble without doing any 214 00:10:56.679 --> 00:10:57.360 conscious math. 215 00:10:57.720 --> 00:11:00.840 It operates much much closer to the dog. Really, Yeah, 216 00:11:00.879 --> 00:11:04.159 it is absolutely not freezing in place to run complex 217 00:11:04.240 --> 00:11:07.919 kinematic equations or recalculate a new center of gravity from scratch. 218 00:11:08.279 --> 00:11:11.120 It doesn't have to do that heavy computational lifting because 219 00:11:11.120 --> 00:11:13.320 of the way it was born in that simulated environment. 220 00:11:13.399 --> 00:11:17.039 The evolution part, right, Remember those millions of spectacular failures 221 00:11:17.080 --> 00:11:18.799 we talked about during its evolution. 222 00:11:18.559 --> 00:11:22.159 All those generations of flailing and collapsing, exactly. 223 00:11:21.919 --> 00:11:26.759 Those failures actually mapped out an incredibly vast landscape of possibilities. 224 00:11:27.320 --> 00:11:31.039 This robot does not possess one single rigid correct way 225 00:11:31.080 --> 00:11:34.480 to function. Oh interesting, because it evolved through endless variations. 226 00:11:34.759 --> 00:11:38.399 It exists within a massive space of possible behaviors. So 227 00:11:38.440 --> 00:11:41.720 when it takes catastrophic damage, that physical trauma simply forces 228 00:11:41.720 --> 00:11:44.559 the system out of its current behavioral state and into 229 00:11:44.600 --> 00:11:46.799 a different region of that pre mapp space. 230 00:11:47.120 --> 00:11:49.480 Okay, wait, so it's less like a computer trying to 231 00:11:49.519 --> 00:11:53.039 recalculate a blocked route on a GPS and more like 232 00:11:53.240 --> 00:11:56.879 dropping a marble into an uneven bowl when the shape 233 00:11:56.919 --> 00:11:59.879 of the bowl changes when it loses a leg, the 234 00:12:00.159 --> 00:12:03.320 marble just naturally rolls down into whatever the next stable 235 00:12:03.399 --> 00:12:03.919 valley is. 236 00:12:04.159 --> 00:12:05.840 That is a brilliant way to put it. 237 00:12:05.840 --> 00:12:09.559 It just falls into a new behavioral rhythm that basically says, okay, 238 00:12:09.720 --> 00:12:12.440 we are a three legged entity now, and we already 239 00:12:12.559 --> 00:12:14.679 roughly know how this shape interacts with gravity. 240 00:12:14.879 --> 00:12:18.720 Yes, because the control is highly distributed, it doesn't even 241 00:12:18.799 --> 00:12:21.440 need perfect information about what just happened to its body. 242 00:12:21.519 --> 00:12:25.240 It doesn't no, it just needs enough remaining mechanical structure 243 00:12:25.279 --> 00:12:29.600 to keep pushing forward. Resilience isn't some extra subroutine they 244 00:12:29.639 --> 00:12:31.120 programmed in for emergencies. 245 00:12:31.200 --> 00:12:31.879 It's built in. 246 00:12:32.120 --> 00:12:35.759 Resilience is an emergent property of having a sufficiently rich, 247 00:12:36.120 --> 00:12:39.200 deeply evolved repertoire of movement strategies. 248 00:12:39.639 --> 00:12:43.120 That is wild, but I mean watching a strange asymmetric 249 00:12:43.200 --> 00:12:47.320 robot drag itself across the laboratory floor is a fascinating 250 00:12:47.399 --> 00:12:50.399 parlor trick. But the reason we are really dissecting this 251 00:12:50.519 --> 00:12:54.440 today is what happens when you take this technology outside. 252 00:12:54.480 --> 00:12:58.879 This concept of emergent resilience is going to completely revolutionize 253 00:12:59.320 --> 00:13:03.279 how we operate rate in the most extreme vital industries 254 00:13:03.320 --> 00:13:05.120 on Earth and off it for that matter. 255 00:13:05.200 --> 00:13:08.679 Oh, without a doubt, the most immediate and profound impact 256 00:13:08.799 --> 00:13:12.080 will be in environments where human intervention is either too 257 00:13:12.159 --> 00:13:16.679 dangerous or physically impossible, like where the deep ocean floor, 258 00:13:16.759 --> 00:13:20.320 for one, or navigating the unstable, shifting rubble of an 259 00:13:20.320 --> 00:13:23.559 earthquake zone. And of course, extraterrestrial exploration. 260 00:13:23.799 --> 00:13:26.600 Space exploration is the ultimate use case for this, isn't 261 00:13:26.600 --> 00:13:29.440 it Absolutely You send a multi billion dollar rover to 262 00:13:29.480 --> 00:13:32.480 the surface of Mars. The environment is brutal. If a 263 00:13:32.480 --> 00:13:35.399 sharp rock tears off a vital wheel, or a Martian 264 00:13:35.440 --> 00:13:38.399 sandstorm jams a primary joint, the mission is effectively over. 265 00:13:38.519 --> 00:13:39.240 You just out of luck. 266 00:13:39.519 --> 00:13:41.919 Right. You cannot dispatch a technician with a wrench to 267 00:13:41.960 --> 00:13:44.559 go fix it, and the time delay in communications means 268 00:13:44.840 --> 00:13:47.240 engineers on Earth can't joys stick it out of trouble 269 00:13:47.240 --> 00:13:49.159 in real time. The machine just has to endure on 270 00:13:49.240 --> 00:13:49.639 its own. 271 00:13:49.799 --> 00:13:52.360 And a rover built on these Northwestern principles wouldn't need 272 00:13:52.360 --> 00:13:53.120 a rescue mission. 273 00:13:53.279 --> 00:13:54.960 It would just adapt exactly. 274 00:13:55.200 --> 00:13:59.159 It would adapt its gate, reconfigure its remaining functional modules, 275 00:13:59.360 --> 00:14:02.960 and just continue you gathering data despite the degradation. It 276 00:14:03.039 --> 00:14:07.679 fundamentally changes the risk calculus of deep space exploration when 277 00:14:07.720 --> 00:14:11.320 your hardware can dynamically heal its own behavioral strategies. 278 00:14:11.840 --> 00:14:14.320 But it's not just the dramatic stuff like Mars or 279 00:14:14.360 --> 00:14:18.759 deep ocean trenches. Right. Think about the everyday technology that 280 00:14:18.840 --> 00:14:21.559 runs our world and how incredibly frustrating it is. 281 00:14:21.600 --> 00:14:22.639 Oh, I know where you're going with this. 282 00:14:22.960 --> 00:14:26.679 If a single millimeter of silicon brace inside a modern smartphone, 283 00:14:26.879 --> 00:14:31.080 the entire device totally bricks itself. It becomes an incredibly 284 00:14:31.080 --> 00:14:33.960 expensive paperweight over one tiny failure. 285 00:14:34.240 --> 00:14:35.159 It's infuriating. 286 00:14:35.200 --> 00:14:38.240 Now and apply that fragility to industry. Could this modular 287 00:14:38.279 --> 00:14:40.960 adaptive approach mean the end of the factory out of 288 00:14:41.080 --> 00:14:44.360 order sign? That's the dream, because imagine a multi million 289 00:14:44.399 --> 00:14:47.960 dollar automotive assembly line right now. If one tiny gear 290 00:14:48.000 --> 00:14:50.960 strips on one robotic arm, the entire line might have 291 00:14:51.000 --> 00:14:53.879 to shut down for hours while a human replaces the part, 292 00:14:54.039 --> 00:14:59.120 costing millions exactly. But with this technology, the robot recognizes 293 00:14:59.240 --> 00:15:03.320 the stripped gear, reconfigures its movement path to avoid using 294 00:15:03.399 --> 00:15:07.320 that specific joint, and just keeps working at say, eighty 295 00:15:07.360 --> 00:15:09.519 percent capacity until the shift is over. 296 00:15:09.799 --> 00:15:13.600 You've just described a really profound industrial engineering concept known 297 00:15:13.600 --> 00:15:15.039 as graceful degradation. 298 00:15:15.360 --> 00:15:17.600 Graceful degradation I like the sound of that. 299 00:15:17.840 --> 00:15:22.519 It's crucial. Modern infrastructure is deeply, entirely dependent on automation. 300 00:15:23.159 --> 00:15:26.799 Yet most of those systems demand perfectly sterile, highly controlled 301 00:15:26.919 --> 00:15:29.879 environments to function. They fail abruptly, one. 302 00:15:29.759 --> 00:15:32.639 Thing goes wrong and boom, the whole system halts right. 303 00:15:32.960 --> 00:15:36.480 But graceful degradation means introducing systems that tolerate wear and 304 00:15:36.519 --> 00:15:40.200 tear physical damage and the messy variability of the real world. 305 00:15:40.240 --> 00:15:44.080 Without needing constant maintenance, they don't break, They just slowly 306 00:15:44.120 --> 00:15:46.679 degrade in performance while maintaining their core emission. 307 00:15:46.840 --> 00:15:50.200 It totally changes the economics of automation. Think about global 308 00:15:50.240 --> 00:15:54.399 logistics networks, massive construction sites, mining operations. They would all 309 00:15:54.440 --> 00:15:57.799 benefit immensely from robotic labor that just stubbornly finds a 310 00:15:57.799 --> 00:15:59.759 way to get the job done. Even as the machinery 311 00:15:59.799 --> 00:16:02.600 it's self slowly wears down over years of use. 312 00:16:02.799 --> 00:16:07.279 It dramatically expands where robotics can be reliably deployed. But 313 00:16:07.360 --> 00:16:10.799 I will say bringing this into the industrial sphere forces 314 00:16:10.879 --> 00:16:14.519 a rather uncomfortable realization for the people who actually build 315 00:16:14.559 --> 00:16:19.639 these things. Uncomfortable because if machines are now evolving themselves 316 00:16:19.639 --> 00:16:23.399 in simulations, determining their own physical forms and figuring out 317 00:16:23.399 --> 00:16:26.759 their own survival strategies, we have to really rethink what 318 00:16:26.840 --> 00:16:28.559 it actually means to be an engineer. 319 00:16:28.720 --> 00:16:31.440 Well, yeah, because the human didn't build that three legged, 320 00:16:31.480 --> 00:16:34.759 hobbling robot. The human just set up the digital sandbox it. 321 00:16:34.679 --> 00:16:38.159 Grew up in exactly. It is a massive philosophical shift. 322 00:16:38.840 --> 00:16:42.240 Humans are stepping away from being traditional designers and becoming 323 00:16:42.240 --> 00:16:43.879 what we call meta designers. 324 00:16:43.960 --> 00:16:45.679 Meta designers. Okay, break that down for me. 325 00:16:45.919 --> 00:16:50.519 Historically, an engineer dictates the exact, granular solution. They decide 326 00:16:50.519 --> 00:16:52.480 the torque of the motor, the length of the lever, 327 00:16:52.840 --> 00:16:56.240 the exact placement of every single screw, very top down right. 328 00:16:56.679 --> 00:16:59.600 But a meta designer doesn't dictate any of that. They 329 00:16:59.600 --> 00:17:02.840 shape the environment. They define the physical rules of the simulation. 330 00:17:03.240 --> 00:17:06.000 They set the ultimate goal like moving from point A 331 00:17:06.079 --> 00:17:08.880 to point B, and then they step back and let 332 00:17:08.920 --> 00:17:12.640 the AI discover the solution within those parameters. 333 00:17:12.680 --> 00:17:15.400 But this totally blurs the lines of authorship, doesn't it? 334 00:17:15.440 --> 00:17:16.200 Oh completely? 335 00:17:16.279 --> 00:17:19.279 If the algorithm runs for three days and spits out 336 00:17:19.400 --> 00:17:24.920 a brilliant, highly capable, completely bizarre design that no human 337 00:17:25.000 --> 00:17:28.319 brain would have ever dreamed up, ye, who actually gets 338 00:17:28.359 --> 00:17:31.759 to say they invented the robot? The great question does 339 00:17:31.799 --> 00:17:34.200 the credit go to the human who set the gravity 340 00:17:34.240 --> 00:17:37.119 parameters or the algorithm that did the heavy lifting of 341 00:17:37.160 --> 00:17:38.119 discovering the form? 342 00:17:38.200 --> 00:17:42.759 Authorship becomes completely diffuse. The final robot isn't invented in 343 00:17:42.799 --> 00:17:45.960 the conventional historical sense of an inventor sitting at a 344 00:17:46.039 --> 00:17:49.599 drafting table. The researchers themselves describe the robot as being 345 00:17:49.839 --> 00:17:53.079 found within a space of possibilities. The human guided the 346 00:17:53.079 --> 00:17:57.079 search party, but no one explicitly imagined the final product beforehand. 347 00:17:57.160 --> 00:17:58.079 That is so poetic. 348 00:17:58.359 --> 00:18:01.680 It feels much closer to discovering a new species of 349 00:18:01.720 --> 00:18:04.319 animal in the wild than it does to building a 350 00:18:04.359 --> 00:18:05.200 grandfeather clock. 351 00:18:05.359 --> 00:18:08.720 Okay, wait, we need a reality check here. Sure, simulating 352 00:18:08.799 --> 00:18:12.400 millions of generations of evolution in a pristine digital environment 353 00:18:12.519 --> 00:18:15.960 sounds incredible, But a virtual robot doesn't have to deal 354 00:18:16.000 --> 00:18:18.440 with real gravity. Sure it doesn't have to deal with 355 00:18:18.519 --> 00:18:21.279 a sudden gust of wind pushing it over, or a 356 00:18:21.400 --> 00:18:24.400 change in humidity that totally alters the friction of its 357 00:18:24.480 --> 00:18:29.720 rubber footpads. How much of this simulated evolution actually survives 358 00:18:29.720 --> 00:18:31.640 when you try to build it out of physical metal 359 00:18:31.640 --> 00:18:32.240 and plastic. 360 00:18:32.480 --> 00:18:35.519 That is the single greatest bottleneck in the field right now. 361 00:18:35.880 --> 00:18:38.079 It is known as crossing the reality gap. 362 00:18:38.200 --> 00:18:39.000 The reality gap. 363 00:18:39.119 --> 00:18:43.319 Yeah, the transition from a perfect virtual simulation to the dirty, 364 00:18:43.519 --> 00:18:47.759 unforgiving physical world is incredibly difficult. In a physics engine, 365 00:18:47.759 --> 00:18:51.079 you can calculate friction, but it is an idealized version 366 00:18:51.119 --> 00:18:51.839 of friction, and. 367 00:18:51.759 --> 00:18:55.200 A virtual motor never overheats. And melts its own wiring. 368 00:18:55.000 --> 00:18:59.640 Exactly that real world materials have complex constraints. Metals bend 369 00:18:59.640 --> 00:19:03.920 and fitt over time, plastic snap under unexpected torsion. Energy 370 00:19:03.960 --> 00:19:05.960 efficiency is a massive hurdle. 371 00:19:05.839 --> 00:19:08.759 Because a physical robot has to carry its own battery right. 372 00:19:09.039 --> 00:19:12.279 And power drain is highly unpredictable when a machine is 373 00:19:12.319 --> 00:19:17.039 improvising new ways to walk. Ensuring that an intricate asymmetrical 374 00:19:17.119 --> 00:19:21.640 design that evolved in the digital utopia remains viable when 375 00:19:21.759 --> 00:19:24.880 it is constructed from physical atoms it is a massive 376 00:19:24.960 --> 00:19:26.519 ongoing engineering challenge. 377 00:19:26.559 --> 00:19:29.839 But even with that reality gap, even acknowledging the friction 378 00:19:30.039 --> 00:19:33.079 and the battery issues, the broader trajectory of this technology 379 00:19:33.079 --> 00:19:34.279 is undeniable, isn't it. 380 00:19:34.279 --> 00:19:37.200 It really is. We are watching the entire field of 381 00:19:37.319 --> 00:19:41.200 robotics move fundamentally away from rigid, pre defined tools. 382 00:19:41.359 --> 00:19:46.079 Right by combining modular, interchangeable hardware with evolutionary AI, we 383 00:19:46.119 --> 00:19:48.519 are staring at a future where machines are no longer 384 00:19:48.559 --> 00:19:52.759 static objects. They are active participants in a continuous, open 385 00:19:52.880 --> 00:19:54.759 ended process of physical change. 386 00:19:54.799 --> 00:19:58.319 And in that future, a robot's identity isn't defined solely 387 00:19:58.359 --> 00:20:00.319 by how it looked when it rolled off the embly 388 00:20:00.359 --> 00:20:02.640 line on day one, it will be defined by its 389 00:20:02.640 --> 00:20:06.039 capacity to transform on day one hundred. Well, physical damage 390 00:20:06.240 --> 00:20:09.160 no longer marks the end of its usefulness. Damage just 391 00:20:09.200 --> 00:20:12.759 becomes another piece of environmental data. It's just another condition 392 00:20:12.880 --> 00:20:16.480 to adapt to another puzzle for the distributed network of 393 00:20:16.519 --> 00:20:17.480 modules to solve. 394 00:20:17.839 --> 00:20:21.119 Function is no longer tied to one perfect symmetrical shape. 395 00:20:21.319 --> 00:20:25.279 It is distributed across countless possible forms. And that means 396 00:20:25.440 --> 00:20:29.079 our basic definition of machine intelligence has to evolve as well, 397 00:20:29.119 --> 00:20:33.079 doesn't it. Absolutely, intelligence isn't just a supercomputer calculating a 398 00:20:33.119 --> 00:20:37.400 massive math problem in a sterile room anymore. In this context, 399 00:20:37.440 --> 00:20:40.680 intelligence is defined by the physical ability to persist through 400 00:20:40.759 --> 00:20:41.599 violent change. 401 00:20:42.079 --> 00:20:46.440