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 --> 00:00:22.280 examine the ideas shaping the future of artificial minds. 5 00:00:23.879 --> 00:00:27.839 I want you to visualize something with me for a second. 6 00:00:28.679 --> 00:00:32.600 Picture a really dense patch of a rainforest floor. It 7 00:00:32.679 --> 00:00:37.920 is messy, it has leaf litter everywhere. It just total chaos, exactly. 8 00:00:38.560 --> 00:00:42.039 And right in the middle of this chaos there is 9 00:00:42.079 --> 00:00:45.479 a dead beetle, A big one huge, I mean relative 10 00:00:45.479 --> 00:00:47.759 to the insects around it. It is basically the size 11 00:00:47.759 --> 00:00:50.280 of a minivan, right, Okay, So you see a scout 12 00:00:50.320 --> 00:00:54.439 ant find it, then another ant, yeah, than maybe ten more, 13 00:00:55.200 --> 00:00:58.719 and within a few minutes there are just hundreds of them. 14 00:00:58.759 --> 00:01:02.240 They start dismantling this thing. They are cutting it, they're carrying. 15 00:01:01.880 --> 00:01:04.120 It, forming those little chain right for me, chains. 16 00:01:03.920 --> 00:01:08.359 To drag pieces over obstacles. It is honestly a masterpiece 17 00:01:08.359 --> 00:01:09.040 of logistics. 18 00:01:09.120 --> 00:01:10.799 It really is the ultimate supply chain. 19 00:01:10.959 --> 00:01:11.280 It is. 20 00:01:11.680 --> 00:01:12.079 Yeah. 21 00:01:12.120 --> 00:01:14.959 But here's the thing that absolutely keeps me up at 22 00:01:15.040 --> 00:01:16.640 night when I look at this stuff. What is that 23 00:01:16.879 --> 00:01:20.040 if you look closely there is no foreman. There is 24 00:01:20.159 --> 00:01:22.400 no general ant standing on a pebble with a little 25 00:01:22.439 --> 00:01:25.120 megaphone shouting you three take the left leg and you 26 00:01:25.159 --> 00:01:26.200 five left the wing right. 27 00:01:26.239 --> 00:01:27.680 There is no central command at. 28 00:01:27.560 --> 00:01:29.519 All, exactly, there is no blueprint. 29 00:01:29.799 --> 00:01:32.040 Yeah, if you look for the leader, you are going 30 00:01:32.079 --> 00:01:34.000 to be looking for a very long time because there 31 00:01:34.040 --> 00:01:34.959 just isn't one right. 32 00:01:35.480 --> 00:01:38.719 And yet they are incredibly efficient, they are adaptable, and 33 00:01:38.760 --> 00:01:41.959 they get the job done faster than a human crew probably. 34 00:01:41.560 --> 00:01:43.239 Could relatively speaking. 35 00:01:43.319 --> 00:01:45.840 Yeah, So the core question we are tackling today in 36 00:01:45.920 --> 00:01:49.120 this exploration of our source material is how. 37 00:01:49.040 --> 00:01:49.879 How do they do it? 38 00:01:50.159 --> 00:01:56.079 How do millions of simple, arguably dumb individuals create a 39 00:01:56.239 --> 00:01:58.799 highly complex, intelligent group. 40 00:01:58.879 --> 00:02:02.040 And more importantly for us, how are we stealing their secrets? 41 00:02:02.120 --> 00:02:05.280 Exactly? How are we reverse engineering this to build the 42 00:02:05.280 --> 00:02:06.359 future of robotics? 43 00:02:06.680 --> 00:02:09.439 Is such a fascinating area, it really is. We are 44 00:02:09.520 --> 00:02:12.240 looking at a stack of research today that essentially covers 45 00:02:12.280 --> 00:02:17.400 two twin concepts, swarm robotics and collective intelligence. Okay, and 46 00:02:17.439 --> 00:02:18.960 I want to be clear right off the bat, this 47 00:02:19.000 --> 00:02:22.000 isn't just about making cool robot bugs. This is about 48 00:02:22.000 --> 00:02:24.960 a fundamental shift in how we actually engineer systems. 49 00:02:25.080 --> 00:02:27.599 So let us define our terms before we get into 50 00:02:27.599 --> 00:02:30.520 the weeds here. What is the actual difference between swarm 51 00:02:30.639 --> 00:02:34.680 robotics and collective intelligence or are we just splitting hairs 52 00:02:34.680 --> 00:02:35.360 with those terms. 53 00:02:35.439 --> 00:02:38.719 No, it is actually a really useful distinction. Swarm robotics 54 00:02:38.800 --> 00:02:43.319 is the engineering field itself. It is the hardware, the code, 55 00:02:43.479 --> 00:02:47.719 actual how to of building large numbers of simple robots 56 00:02:47.719 --> 00:02:48.840 that coordinate. 57 00:02:48.759 --> 00:02:50.360 So the physical application, right. 58 00:02:50.439 --> 00:02:55.439 Collective intelligence is the broader phenomenon. It describes the behavior itself, 59 00:02:55.960 --> 00:02:59.240 how groups of agents, whether they are biological ants or 60 00:02:59.319 --> 00:03:04.560 silicon d zhones, produce intelligent behavior through their interactions. 61 00:03:04.000 --> 00:03:07.360 Rather than through individual reasoning exactly. So collective intelligence is 62 00:03:07.360 --> 00:03:09.879 the way it is the emergent brain power, yes, and 63 00:03:09.960 --> 00:03:12.159 swarm robotics is the tool we used to build it. 64 00:03:12.520 --> 00:03:14.120 That's a very fair way to put it, Okay. 65 00:03:14.159 --> 00:03:17.080 So the mission of this discussion today is to understand 66 00:03:17.159 --> 00:03:19.879 how we are moving from the era of smart robots 67 00:03:20.199 --> 00:03:21.800 to smart systems. 68 00:03:21.479 --> 00:03:24.439 Right, because we are used to a single, very expensive, 69 00:03:24.639 --> 00:03:26.680 very complex machine. 70 00:03:26.159 --> 00:03:29.560 And now we are looking at systems where the intelligence 71 00:03:29.680 --> 00:03:32.240 is in the group, not the individual. 72 00:03:31.960 --> 00:03:36.280 Which really completely flips our traditional understanding of AI upside down. 73 00:03:36.360 --> 00:03:38.879 Doesn't it. It really does. Yeah, because for the last 74 00:03:38.919 --> 00:03:42.039 fifty years, when we thought of smart name, we thought 75 00:03:42.080 --> 00:03:44.960 of a supercomputer. Yeah, we thought of a giant brain 76 00:03:45.039 --> 00:03:48.599 in a box, and those everything like IBM Watson or 77 00:03:49.159 --> 00:03:50.840 HL nine thousand exactly. 78 00:03:51.080 --> 00:03:53.919 We have been obsessed with this idea of God in 79 00:03:53.960 --> 00:03:57.439 a box, God in a box or centralized omniscion intelligence. 80 00:03:57.879 --> 00:04:00.680 But the core premise of swarm theory is that you 81 00:04:00.800 --> 00:04:02.560 do not need a giant brain. 82 00:04:02.759 --> 00:04:04.039 You just need a lot of tiny ones. 83 00:04:04.120 --> 00:04:07.599 Right. You can have thousands of tiny, somewhat limited brains 84 00:04:07.879 --> 00:04:12.120 or simple processors working together, working together, and the intelligence 85 00:04:12.159 --> 00:04:15.879 emerges from the connections between them, not from the agents themselves. 86 00:04:15.960 --> 00:04:18.199 So it is not about the player, It is about 87 00:04:18.199 --> 00:04:21.079 the team dynamic exactly. But let us be real for 88 00:04:21.120 --> 00:04:21.680 a second here. 89 00:04:21.959 --> 00:04:24.639 We didn't invent this, Oh no, we absolutely stole it. 90 00:04:24.680 --> 00:04:25.680 We stole it from nature. 91 00:04:25.839 --> 00:04:31.040 We are essentially reverse engineering evolution. Nature solved these engineering 92 00:04:31.120 --> 00:04:33.079 problems millions of years ago. 93 00:04:33.439 --> 00:04:36.639 So let us talk about the original engineers as the 94 00:04:36.680 --> 00:04:38.680 ants we mentioned them in the intro. But I really 95 00:04:38.680 --> 00:04:41.399 want to get into the mechanics because if I am 96 00:04:41.399 --> 00:04:45.240 an ant and I find that giant dead beetle. I 97 00:04:45.319 --> 00:04:47.360 do not have a two way radio. No, I do 98 00:04:47.439 --> 00:04:50.600 not have a GPS module. How do I actually tell 99 00:04:50.639 --> 00:04:52.560 the rest of the colony where the food is? 100 00:04:52.920 --> 00:04:55.000 You use a process called stigmergy. 101 00:04:55.439 --> 00:04:58.759 Stigmergy. Now that is a word that appears all over 102 00:04:58.800 --> 00:04:59.839 the research stack we went. 103 00:05:00.360 --> 00:05:02.240 It is the foundation break that down force. 104 00:05:02.319 --> 00:05:03.360 What does that actually mean? 105 00:05:03.480 --> 00:05:07.720 So stigmag comes from the Greek words stigma meaning mark, 106 00:05:07.759 --> 00:05:09.439 and ergon meaning work. 107 00:05:09.720 --> 00:05:11.160 Okay, mark and work. 108 00:05:11.240 --> 00:05:15.519 It essentially means indirect coordination through environmental modification. 109 00:05:15.720 --> 00:05:18.519 Okay, that is a great textbook definition. Well what does 110 00:05:18.560 --> 00:05:19.639 it look like in practice? 111 00:05:19.800 --> 00:05:23.040 Imagine you are leaving notes for your roommates, right, but 112 00:05:23.160 --> 00:05:24.680 you cannot speak to them. 113 00:05:24.519 --> 00:05:26.680 Directly, so we are on opposite schedules. 114 00:05:26.759 --> 00:05:29.759 Exactly. You can only leave sticky notes on the fridge. Okay, 115 00:05:29.839 --> 00:05:32.560 that is stigmagy. In the case of ants, the sticky 116 00:05:32.560 --> 00:05:34.319 note is a pheromone. 117 00:05:33.720 --> 00:05:35.759 Trail, right, the centrail they leave behind. 118 00:05:35.959 --> 00:05:39.319 Yes, but the mechanics are much more subtle than just 119 00:05:39.759 --> 00:05:42.160 follow the smell. How So this is where the math 120 00:05:42.199 --> 00:05:46.040 gets really interesting. Let us say there are two paths. 121 00:05:45.639 --> 00:05:48.279 To a food source, Okay, Path A and path B. 122 00:05:48.519 --> 00:05:51.160 Right, Path A is short. Path B is long. 123 00:05:51.480 --> 00:05:51.839 Got it. 124 00:05:51.920 --> 00:05:54.959 The ants start out exploring completely randomly. Some take A 125 00:05:55.120 --> 00:05:57.680 and some take B. Okay, and they are laying down 126 00:05:57.759 --> 00:06:01.120 pheromones as they walk. The ants on path A the 127 00:06:01.160 --> 00:06:04.199 short path are going to make a round trip faster 128 00:06:04.319 --> 00:06:04.959 than the ants on. 129 00:06:05.000 --> 00:06:07.839 Path B because it is shorter. It is just basic 130 00:06:07.920 --> 00:06:08.920 physics exactly. 131 00:06:09.000 --> 00:06:11.279 So, in the span of say ten minutes, an ant 132 00:06:11.360 --> 00:06:14.160 on the short path might make three trips. 133 00:06:14.040 --> 00:06:16.399 Laying down three layers of pheromones. 134 00:06:15.920 --> 00:06:18.360 Right, but the ant on the long path only. 135 00:06:18.120 --> 00:06:21.199 Makes one trip laying down one layer. Yes, so the 136 00:06:21.240 --> 00:06:24.319 short path literally smells stronger to the other ants. 137 00:06:24.439 --> 00:06:27.000 Yes it does. But here is the critical part that 138 00:06:27.079 --> 00:06:30.120 most people miss about this. What is that evaporation. 139 00:06:30.319 --> 00:06:33.360 The scent disappears, has to disappear. Why think about it. 140 00:06:33.639 --> 00:06:36.519 If pheromones stayed forever, the forest floor would be a 141 00:06:36.519 --> 00:06:39.480 confusing mess of old, useless trails leading to food that 142 00:06:39.560 --> 00:06:40.800 has already gone. 143 00:06:40.560 --> 00:06:43.519 Oh right, it would just be background noise exactly. 144 00:06:43.920 --> 00:06:46.720 The system relies on the signal decaying. Okay, on the 145 00:06:46.759 --> 00:06:49.800 long path, the pheromones might evaporate almost as fast as 146 00:06:49.800 --> 00:06:50.680 they are laid down. 147 00:06:50.720 --> 00:06:52.959 Because the ants takes so long to get back right. 148 00:06:53.120 --> 00:06:56.959 But on the short path, the deposition rate exceeds the 149 00:06:57.000 --> 00:06:58.399 evaporation rate, so the. 150 00:06:58.360 --> 00:07:02.319 Signal amplifies itself on the efficient route and naturally dies 151 00:07:02.360 --> 00:07:03.519 out on the inefficient route. 152 00:07:03.560 --> 00:07:08.600 Precisely, the colony calculates the shortest route, but no single 153 00:07:08.639 --> 00:07:12.199 ant actually did the math. That is wild, the environment 154 00:07:12.279 --> 00:07:15.680 plus the physics of evaporation did the calculation for them. 155 00:07:15.959 --> 00:07:19.040 That is the part that I find so fascinating. That 156 00:07:19.160 --> 00:07:22.959 forgetting is actually a crucial part of the intelligence. Yes, 157 00:07:23.079 --> 00:07:25.720 because if the system remembered everything, it would actually be stupid. 158 00:07:25.839 --> 00:07:28.199 Exactly. You need the noise to filter at the signal. 159 00:07:28.399 --> 00:07:28.720 Wow. 160 00:07:28.879 --> 00:07:32.279 And it is not just ants. Termites are arguably even 161 00:07:32.360 --> 00:07:35.519 more impressive architects we have. We've seen those massive termite mounds, the. 162 00:07:35.439 --> 00:07:37.680 Ones in Africa and Australia right that looks like these 163 00:07:37.720 --> 00:07:41.040 towering Gothic cathedrals made of mud. Yes, some of them 164 00:07:41.079 --> 00:07:43.959 are what ten or fifteen feet tall easily. 165 00:07:44.000 --> 00:07:48.079 And inside they have these incredibly complex ventilation systems for 166 00:07:48.199 --> 00:07:52.399 airflow right that keep the core temperature completely constant for 167 00:07:52.480 --> 00:07:54.959 the queen and their fungus gardens. 168 00:07:54.639 --> 00:07:57.120 Even when the weather outside is changing, Even when. 169 00:07:57.000 --> 00:08:00.600 The outside temperature swings from forty degrees we elseiaus in 170 00:08:00.639 --> 00:08:02.920 the day to near freezing at night. 171 00:08:03.079 --> 00:08:04.160 That is incredible. 172 00:08:04.199 --> 00:08:08.879 It is an absolute marvel of passive HVAC engineering. 173 00:08:09.000 --> 00:08:12.639 But again, no blueprint, no blueprint. There is no master 174 00:08:12.839 --> 00:08:16.240 termite architect holding a schematic in a tiny hard hat. 175 00:08:16.480 --> 00:08:17.480 No none at all. 176 00:08:17.519 --> 00:08:20.319 So how do they ensure a ventilation shaft actually lines up? 177 00:08:20.480 --> 00:08:22.160 What do you mean like, if I am building a 178 00:08:22.160 --> 00:08:24.079 tunnel from the left side and you are building a 179 00:08:24.079 --> 00:08:26.480 tunnel from the right side, how do we guarantee we 180 00:08:26.560 --> 00:08:27.920 meet in the middle without talking? 181 00:08:28.240 --> 00:08:31.800 They follow local cues, local cues right. A termite has 182 00:08:31.800 --> 00:08:35.320 a very simple rule script that is genetically hardwired into it. 183 00:08:35.360 --> 00:08:36.559 Give me an example of a rule. 184 00:08:36.720 --> 00:08:39.679 It might be something like, if I detect a high 185 00:08:39.720 --> 00:08:43.279 concentration of carbon dioxide right here, I place a pellet 186 00:08:43.320 --> 00:08:46.080 of mud right here. Or if the air current flows 187 00:08:46.120 --> 00:08:49.559 this specific way, I build a wall that curves like this. 188 00:08:49.919 --> 00:08:51.759 So the current state of the building tells the builder 189 00:08:51.799 --> 00:08:52.200 what to do. 190 00:08:52.200 --> 00:08:56.200 Next, exactly, the structure itself dictates the next move. 191 00:08:56.360 --> 00:08:57.679 That is another form of stigma G. 192 00:08:57.799 --> 00:09:01.240 Then yes, it is as the airflow naturally changes because 193 00:09:01.240 --> 00:09:04.399 of the wall you just built. That new airflow triggers 194 00:09:04.440 --> 00:09:06.799 the next termite to come along and build the ceiling. 195 00:09:07.080 --> 00:09:09.480 So it is just a chain reaction. 196 00:09:09.399 --> 00:09:12.519 A chain reaction of simple local rules that leads to 197 00:09:12.639 --> 00:09:15.080 a massive, complex global structure. 198 00:09:15.320 --> 00:09:17.679 It is mind bending because it implies that the plan 199 00:09:17.720 --> 00:09:20.799 for the mound doesn't actually exist anywhere the termite's heads, right, 200 00:09:20.919 --> 00:09:23.080 The plan only exists in the physical interaction. 201 00:09:23.320 --> 00:09:25.799 And that is the very definition of emergence. 202 00:09:25.960 --> 00:09:26.919 Emergence right. 203 00:09:27.000 --> 00:09:30.000 And we see it in decision making too, especially with honeybees. 204 00:09:30.120 --> 00:09:32.639 A yes, the waggle dance, the famous waggle dance. 205 00:09:32.679 --> 00:09:34.360 This always sounds like a joke when you first hear 206 00:09:34.360 --> 00:09:37.279 about it, but the research papers treat it with absolute 207 00:09:37.399 --> 00:09:38.159 dead seriousness. 208 00:09:38.200 --> 00:09:39.440 Oh, it is not a joke at all. It is 209 00:09:39.480 --> 00:09:43.200 a rigorous, distributed, democratic decision making process. 210 00:09:43.480 --> 00:09:45.679 So set the scene for us. When does this happen? 211 00:09:46.000 --> 00:09:48.679 When a hive gets too big, they need to split. Okay, 212 00:09:48.759 --> 00:09:51.600 a swarm leaves the main hive and they go hang 213 00:09:51.679 --> 00:09:53.039 on a tree branch somewhere. 214 00:09:53.120 --> 00:09:54.799 Right, you see those big clumsy bees. 215 00:09:54.600 --> 00:09:58.679 Sometimes exactly now they are exposed and they need a 216 00:09:58.679 --> 00:10:01.639 new home quickly, right, they have to find a tree hollow. 217 00:10:01.679 --> 00:10:04.679 That is the exact right size, the right hide off 218 00:10:04.720 --> 00:10:06.679 the ground, and has the right humidity. 219 00:10:07.039 --> 00:10:09.559 It is basically insect real estate hunting. 220 00:10:09.720 --> 00:10:13.360 It is so they send out scouts. They might send 221 00:10:13.399 --> 00:10:15.879 out hundreds of scouts in all directions. 222 00:10:16.000 --> 00:10:16.320 Okay. 223 00:10:16.559 --> 00:10:19.879 The scouts find potential sites, evaluate them, and come back 224 00:10:19.919 --> 00:10:22.320 to the cluster on the branch, and then they dance. 225 00:10:22.440 --> 00:10:25.120 And then they dance. The angle of their dance relative 226 00:10:25.200 --> 00:10:27.840 to the sun tells the other bees the direction of 227 00:10:27.879 --> 00:10:31.879 the site. The duration of the dance tells them the distance. Okay, 228 00:10:32.039 --> 00:10:36.039 But the intensity how vigorously they shake their bodies tells 229 00:10:36.120 --> 00:10:37.399 them the quality of the site. 230 00:10:37.480 --> 00:10:39.879 So if I fly out and find a mediocre hole, 231 00:10:40.080 --> 00:10:42.320 I come back and do a really lazy dance exactly. 232 00:10:42.399 --> 00:10:44.639 But if I find an absolute mansion of a tree hollow, 233 00:10:44.799 --> 00:10:46.600 I come back and dance my legs off. 234 00:10:47.000 --> 00:10:49.879 You give it everything you have. And the other scouts 235 00:10:49.879 --> 00:10:53.960 are watching this, okay. If they see a really enthusiastic dance, 236 00:10:54.960 --> 00:10:58.000 they fly out to verify the site themselves. 237 00:10:58.519 --> 00:11:00.559 Oh so they don't just take the first bees word 238 00:11:00.600 --> 00:11:00.799 for you. 239 00:11:00.759 --> 00:11:02.480 It is essentially pure review. 240 00:11:02.759 --> 00:11:03.720 That is amazing. 241 00:11:03.919 --> 00:11:06.120 If they agree that the site is good, they come 242 00:11:06.120 --> 00:11:08.879 back and dance for it too, and this leads. 243 00:11:08.639 --> 00:11:10.759 To something called quorum sensing, right. 244 00:11:10.639 --> 00:11:14.919 Yes, quorum sensing. As more bees visit the good site, 245 00:11:15.320 --> 00:11:16.559 more bees come back. 246 00:11:16.440 --> 00:11:18.440 And dance for it, so it's snowballs. 247 00:11:18.679 --> 00:11:22.200 The number of dancers for the best site grows exponentially, 248 00:11:22.720 --> 00:11:25.720 and once a certain threshold is reached the quorum, the 249 00:11:25.879 --> 00:11:28.840 entire swarm just lifts off as one and moves to 250 00:11:28.879 --> 00:11:30.000 the new home, so. 251 00:11:29.919 --> 00:11:33.480 They completely avoid the bad leader problem exactly. Like if 252 00:11:33.480 --> 00:11:36.039 one scout is just crazy and loads a terrible damp pole, 253 00:11:36.759 --> 00:11:38.679 they might come back and dance for it, right, But 254 00:11:38.720 --> 00:11:40.600 when the other scouts go to check it out, they 255 00:11:40.679 --> 00:11:43.600 realize it is garbage and they won't dance when they. 256 00:11:43.480 --> 00:11:45.919 Get back, So the bad idea naturally dies out. 257 00:11:46.159 --> 00:11:48.600 It pulls the wisdom of the group and filters out 258 00:11:48.639 --> 00:11:49.759 the individual errors. 259 00:11:50.120 --> 00:11:52.679 It does unless, of course, everyone makes the exact same error, 260 00:11:52.720 --> 00:11:54.799 which is something we can talk about a bit later, where. 261 00:11:54.679 --> 00:11:56.559 The dark side of the swarm. We will definitely get 262 00:11:56.600 --> 00:11:58.519 to that good but before we move on, we have 263 00:11:58.559 --> 00:11:59.440 to talk about movement. 264 00:12:00.759 --> 00:12:02.039 Ah Yes, birds and fish. 265 00:12:02.519 --> 00:12:05.440 Because reading through the sources, this seemed to be the 266 00:12:05.440 --> 00:12:09.200 real breakthrough that let us start simulating this stuff on computers. 267 00:12:09.480 --> 00:12:12.240 It was Craig Reynolds in nineteen eighty. 268 00:12:12.000 --> 00:12:14.480 Six the Boyd's simulation. 269 00:12:14.200 --> 00:12:16.919 Boyds not birds, birdoid objects exactly. 270 00:12:17.320 --> 00:12:20.639 Reynolds was a computer graphics researcher, and he wanted to 271 00:12:20.679 --> 00:12:24.639 animate a flock of birds for a film or a simulation. Okay, 272 00:12:24.799 --> 00:12:28.360 before him, animators literally had to manually plot the path 273 00:12:28.440 --> 00:12:31.639 of every single bird on the screen, frame by frame, 274 00:12:31.720 --> 00:12:34.279 frame by frame. It was an absolute nightmare. 275 00:12:33.960 --> 00:12:34.799 I can imagine. 276 00:12:34.879 --> 00:12:38.159 So Reynolds stepped back and asked a biological question, how 277 00:12:38.200 --> 00:12:39.360 do real birds do it? 278 00:12:39.519 --> 00:12:39.759 Right? 279 00:12:39.960 --> 00:12:42.559 He realized they are not following a leader. They are 280 00:12:42.600 --> 00:12:46.679 just watching their immediate neighbors. Okay, So he wrote a 281 00:12:46.720 --> 00:12:51.320 software simulation with just three simple rules that every Boyd followed. 282 00:12:51.759 --> 00:12:55.159 Just three rules to get that incredibly complex swirling liquid 283 00:12:55.200 --> 00:12:58.320 motion you see in Starling murmurations. Just three, what are they? 284 00:12:58.559 --> 00:13:02.480 Rule one is separations separation, which basically means do not 285 00:13:02.799 --> 00:13:03.919 crash into your neighbors. 286 00:13:03.960 --> 00:13:06.000 Okay, maintain personal space. That makes sense. 287 00:13:06.000 --> 00:13:08.559 If you are too close, you steer away. Rule two 288 00:13:08.759 --> 00:13:09.519 is alignment. 289 00:13:09.759 --> 00:13:10.279 Alignment. 290 00:13:10.440 --> 00:13:12.799 Steer in the same average direction. 291 00:13:12.559 --> 00:13:17.200 As your neighbors, so peer pressure basically match their velocity exactly. 292 00:13:16.720 --> 00:13:20.440 And rule three is cohesion. Cohesion, steer toward the average 293 00:13:20.480 --> 00:13:23.480 position of your neighbors, stay close to the center of 294 00:13:23.519 --> 00:13:26.639 the group, do not drift off and become a loner. 295 00:13:26.879 --> 00:13:28.440 And that is literally it. 296 00:13:28.639 --> 00:13:31.320 That is it. You program those three mathematical rules into 297 00:13:31.360 --> 00:13:34.360 dots on a screen and suddenly you have this complex, 298 00:13:34.440 --> 00:13:38.399 choreographed flocking behavior that looks exactly like nature. 299 00:13:38.720 --> 00:13:42.120 So if you remove one of those rules, what happens, 300 00:13:42.159 --> 00:13:43.879 Like if I delete the separation. 301 00:13:43.519 --> 00:13:46.200 Rule, the flock just collapses into a single point. They 302 00:13:46.279 --> 00:13:47.480 all crash into each other. 303 00:13:47.559 --> 00:13:49.399 And if I delete the cohesion rule, they. 304 00:13:49.360 --> 00:13:52.639 Drift apart like gas molecules in a room. The flock dissolves. 305 00:13:53.320 --> 00:13:56.759 You absolutely need the dynamic tension between those three rules 306 00:13:56.759 --> 00:13:57.799 to create the structure. 307 00:13:58.279 --> 00:14:01.360 That is a huge takeaway from me on the biology side. Yeah, 308 00:14:01.480 --> 00:14:04.799 complexity is essentially an illusion. It looks like there's a 309 00:14:04.840 --> 00:14:08.480 complex master plan, but it is really just simple local 310 00:14:08.559 --> 00:14:09.480 rules interacting. 311 00:14:09.600 --> 00:14:13.320 And that realization is the entire blueprint for swarm robotics. 312 00:14:13.399 --> 00:14:16.440 So let us cross that bridge from biology to engineering. 313 00:14:16.519 --> 00:14:17.039 Let's do it. 314 00:14:17.120 --> 00:14:19.799 We see how nature does it. But building a robot 315 00:14:19.879 --> 00:14:22.600 is obviously not the same as hatching an ant far 316 00:14:22.639 --> 00:14:25.320 from it. What are the core principles when we actually 317 00:14:25.360 --> 00:14:29.000 try to build this stuff the source nodes highlight. Decentralization 318 00:14:29.120 --> 00:14:29.720 is the big one. 319 00:14:29.799 --> 00:14:31.879 Decentralization is the golden rule. 320 00:14:31.960 --> 00:14:32.840 Okay, break that down. 321 00:14:32.919 --> 00:14:36.879 In a traditional robot system, say a car manufacturing factory 322 00:14:36.879 --> 00:14:39.360 in the nineteen nineties, you have a central brain. 323 00:14:39.600 --> 00:14:41.120 Right, I'm a massive server. 324 00:14:41.039 --> 00:14:43.720 A massive server that dictates everything. It tells robot ARM 325 00:14:43.840 --> 00:14:46.360 A to move left, and it tells robot ARM B 326 00:14:46.759 --> 00:14:47.720 to weld a joint. 327 00:14:47.759 --> 00:14:49.320 It is the puppet master, exactly. 328 00:14:49.919 --> 00:14:53.799 But that design creates a massive single point of failure. 329 00:14:54.000 --> 00:14:57.240 Oh I see if that server crashes or loses power 330 00:14:57.440 --> 00:15:00.679 or gets hacked, the entire factory freezes. 331 00:15:00.279 --> 00:15:04.200 Or even if just the communication line gets cut, right, exactly. 332 00:15:04.519 --> 00:15:07.000 But in a swarm, there is no central brain. 333 00:15:07.360 --> 00:15:07.600 Right. 334 00:15:07.840 --> 00:15:11.120 Every single robot is its own autonomous agent. If you 335 00:15:11.159 --> 00:15:14.399 step on the leader ant, well you can't because there 336 00:15:14.440 --> 00:15:15.039 is no leader. 337 00:15:15.120 --> 00:15:17.679 So the colony just does not stop. It keeps going, 338 00:15:17.759 --> 00:15:21.480 which naturally leads to the next core principle. Yes, scalability. 339 00:15:21.559 --> 00:15:22.480 Yes, scalability. 340 00:15:22.559 --> 00:15:24.200 This is the one that seems to matter the most 341 00:15:24.320 --> 00:15:26.639 for the big commercial applications you read about. 342 00:15:26.720 --> 00:15:31.480 It is huge. Think about a centralized system. Adding more 343 00:15:31.559 --> 00:15:34.159 robots makes the math exponentially harder. 344 00:15:33.879 --> 00:15:36.519 For the central computer because it has to track everything. 345 00:15:36.720 --> 00:15:39.399 Right, If the central brand has to track five robots, 346 00:15:39.440 --> 00:15:41.639 that is easy, sure, But if it has to track 347 00:15:41.679 --> 00:15:45.080 five thousand robots, calculate all their individual paths and prevent 348 00:15:45.120 --> 00:15:48.039 them from colliding, you need a supercomputer. 349 00:15:48.240 --> 00:15:49.600 The system just chokes on the. 350 00:15:49.639 --> 00:15:52.360 Data it does. But in a swarm, robot number four 351 00:15:52.399 --> 00:15:55.240 thousand only cares about its three immediate neighbors. 352 00:15:55.279 --> 00:15:58.480 Oh right. It does not need the big picture exactly. 353 00:15:58.559 --> 00:16:01.200 It does not care if there are fifty robots or 354 00:16:01.399 --> 00:16:03.639 five million robots in the rest of the swarm. 355 00:16:03.840 --> 00:16:06.720 So the computational load per robots stays exactly the same. 356 00:16:06.799 --> 00:16:09.759 It stays constant. So you can just dump more units 357 00:16:09.759 --> 00:16:13.360 into the system without having to upgrade the main processor. 358 00:16:12.879 --> 00:16:15.480 Because there is no main processor exactly. 359 00:16:15.519 --> 00:16:17.639 It scales infinitely in theory. 360 00:16:17.720 --> 00:16:19.399 But there is a real trade off here, isn't there 361 00:16:19.519 --> 00:16:23.000 There is? We have to talk about autonomy versus simplicity. 362 00:16:23.279 --> 00:16:26.919 We are so used to robots being these high tech, 363 00:16:27.360 --> 00:16:28.879 multimillion dollar marveles. 364 00:16:28.960 --> 00:16:31.799 Yeah, Boston Dynamics, dogs and things like that, right. 365 00:16:31.799 --> 00:16:34.840 But swarm robots, from what I am reading, are usually 366 00:16:34.919 --> 00:16:35.519 kind of dumb. 367 00:16:35.559 --> 00:16:38.159 They are very dumb, and that is by design. The 368 00:16:38.159 --> 00:16:42.360 whole philosophy is quantity over quality. Individual robots in a 369 00:16:42.399 --> 00:16:46.159 swarm are often cheap, simple, and very sensor poor. 370 00:16:46.200 --> 00:16:48.679 Sensor poor that feels like a very polite way of 371 00:16:48.679 --> 00:16:50.799 saying they are practically blind sometimes. 372 00:16:50.840 --> 00:16:53.399 Yes. Take the kilobots at Harvard for example. 373 00:16:53.480 --> 00:16:54.200 The kilobots. 374 00:16:54.279 --> 00:16:56.360 They are these tiny little robots about the size of 375 00:16:56.360 --> 00:16:58.720 a quarter. They sit on little stick legs. They do 376 00:16:58.799 --> 00:17:01.679 not even have wheels. They literally just vibrate to move 377 00:17:01.720 --> 00:17:04.400 across a table. Ah, that is simple, and they absolutely 378 00:17:04.480 --> 00:17:06.759 cannot see the room. They don't have cameras. They can 379 00:17:06.799 --> 00:17:10.359 only detect infrared light from about ten centimeters away. 380 00:17:10.400 --> 00:17:12.240 That is it. Ten centimeters, that is it. 381 00:17:12.400 --> 00:17:14.279 But the point is they do not need to see 382 00:17:14.279 --> 00:17:17.079 the map. They just need to see their neighbor because 383 00:17:17.119 --> 00:17:20.960 the rules are local, right. And this extreme simplicity is 384 00:17:21.000 --> 00:17:24.279 what allows for the next principle, which is robustness. 385 00:17:23.680 --> 00:17:25.720 The disposable hero concept. 386 00:17:25.880 --> 00:17:27.279 I love that phrase to me too. 387 00:17:28.160 --> 00:17:29.920 Explain how that works in practice. 388 00:17:30.039 --> 00:17:35.400 It means the system degrades gracefully rather than failing catastrophically. Okay, 389 00:17:35.480 --> 00:17:39.400 if you have one giant, multimillion dollar robot doing a 390 00:17:39.480 --> 00:17:41.519 dangerous rescue mission and it breaks. 391 00:17:41.279 --> 00:17:43.720 A tread, the mission is over. He failed. 392 00:17:44.240 --> 00:17:46.359 But if you send in a swarm of a thousand 393 00:17:46.359 --> 00:17:49.799 incredibly cheap robots and ten percent of them break or 394 00:17:49.799 --> 00:17:51.960 fall down a hole or run out of battery. 395 00:17:51.680 --> 00:17:53.319 The job just gets a tiny bit slower. 396 00:17:53.440 --> 00:17:56.799 Exactly, it does not fail. The swarm as an entity 397 00:17:57.160 --> 00:17:59.160 survives the loss of the individual. 398 00:17:59.480 --> 00:18:02.839 So we have hardware philosophy down, cheap, dumb, and many. 399 00:18:03.119 --> 00:18:05.480 Now let us talk about the software side. How do 400 00:18:05.599 --> 00:18:08.960 they actually think the algorithm right? Because local rules. It's 401 00:18:09.039 --> 00:18:11.519