WEBVTT 00:00:00.000 --> 00:00:03.040 So I want you to picture something. Picture a 00:00:03.040 --> 00:00:05.719 massive flock of starlings in the evening sky. 00:00:05.940 --> 00:00:08.640 Oh, like a murmuration. Yeah, exactly, a murmuration. 00:00:08.859 --> 00:00:12.980 Thousands of these birds swooping, diving, just 00:00:12.980 --> 00:00:16.039 pulsing together as this one giant fluid shape. 00:00:16.280 --> 00:00:18.399 Usually trying to evade a predator or something. 00:00:18.480 --> 00:00:21.760 Right. Or imagine a colony of ants flawlessly 00:00:21.760 --> 00:00:24.780 navigating a really dense forest floor to find 00:00:24.780 --> 00:00:27.960 the exact shortest route to, like, a discarded 00:00:27.960 --> 00:00:30.379 piece of fruit. It's mesmerizing to watch. It 00:00:30.379 --> 00:00:32.119 really is. And when you see that kind of coordination, 00:00:32.280 --> 00:00:33.920 you naturally assume there's a leader giving 00:00:33.920 --> 00:00:37.079 orders like a CEO bird or a four star general 00:00:37.079 --> 00:00:39.359 and coordinating the logistics. Right. Exactly. 00:00:39.640 --> 00:00:41.659 But there isn't. And that's what we're getting 00:00:41.659 --> 00:00:44.619 into today. Welcome to our deep dive. We're looking 00:00:44.619 --> 00:00:46.880 at sources that prove the smartest systems on 00:00:46.880 --> 00:00:49.859 Earth actually have no boss at all. None. Zero. 00:00:49.960 --> 00:00:52.500 It's called swarm intelligence. And our mission 00:00:52.500 --> 00:00:55.000 today is to unpack how this concept isn't just 00:00:55.000 --> 00:00:58.130 some, you know, biological curiosity it's actually 00:00:58.130 --> 00:01:00.909 a tool that's secretly running your airline flights 00:01:00.909 --> 00:01:04.290 diagnosing your diseases and and even generating 00:01:04.290 --> 00:01:07.469 Hollywood blockbusters which is just crazy to 00:01:07.469 --> 00:01:09.409 think about it really is I mean it completely 00:01:09.409 --> 00:01:12.290 upends how we traditionally view problem -solving 00:01:12.290 --> 00:01:14.290 we're so conditioned to think that high -level 00:01:14.290 --> 00:01:17.349 intelligence requires a massive brain or a master 00:01:17.349 --> 00:01:20.549 planner like a strict top -down hierarchy exactly 00:01:21.390 --> 00:01:24.170 But the research we're diving into today reveals 00:01:24.170 --> 00:01:26.890 that intelligence is often just an emergent property. 00:01:27.659 --> 00:01:29.939 It's simply what happens when a group of individuals 00:01:29.939 --> 00:01:34.500 follows incredibly basic localized rules. OK. 00:01:34.620 --> 00:01:37.260 Let me try to ground this with an analogy, something 00:01:37.260 --> 00:01:39.879 we've all experienced. Think about a massive 00:01:39.879 --> 00:01:42.819 chaotic traffic jam. Oh, the worst. Right. You've 00:01:42.819 --> 00:01:44.900 got hundreds of cars on the highway. Everyone 00:01:44.900 --> 00:01:47.579 is laying on the horn. Nobody's moving. Now imagine 00:01:47.579 --> 00:01:49.700 that traffic jam somehow magically untangles 00:01:49.700 --> 00:01:51.799 itself. Just out of nowhere. Right. The cars 00:01:51.799 --> 00:01:54.230 just start flowing perfectly. But there are no 00:01:54.230 --> 00:01:56.069 traffic lights turning green, no stop signs, 00:01:56.269 --> 00:01:58.590 and absolutely no police officers waving people 00:01:58.590 --> 00:02:01.049 through. Sounds like a dream. Yeah, I know. But 00:02:01.049 --> 00:02:03.269 in this scenario, every driver is just looking 00:02:03.269 --> 00:02:05.349 at the car immediately in front of them and the 00:02:05.349 --> 00:02:07.569 car next to them. They're only interacting locally. 00:02:07.819 --> 00:02:10.300 Right. Boom, the entire miles long system just 00:02:10.300 --> 00:02:12.900 fixes itself. And that visual gets right to the 00:02:12.900 --> 00:02:15.479 core of the mechanics here. To understand how 00:02:15.479 --> 00:02:18.439 a system untangles itself without a boss, we 00:02:18.439 --> 00:02:21.020 have to look at the underlying thesis of swarm 00:02:21.020 --> 00:02:23.740 intelligence, which is a term coined back in 00:02:23.740 --> 00:02:28.099 1989 by Jing Wang and Gerardo Beni. OK, 1989. 00:02:28.360 --> 00:02:30.740 Yeah. And the thesis is you start with a population 00:02:30.740 --> 00:02:33.879 of really simple agents, and those agents only 00:02:33.879 --> 00:02:35.979 interact locally with their immediate neighbors. 00:02:36.099 --> 00:02:38.830 And their environment. Exactly. They aren't looking 00:02:38.830 --> 00:02:41.430 at a map of the whole highway, to use your analogy. 00:02:41.870 --> 00:02:44.330 They follow very basic rules, throw in a little 00:02:44.330 --> 00:02:46.050 touch of randomness so they don't get stuck in 00:02:46.050 --> 00:02:48.990 loops. Not have the randomness. Always. And suddenly, 00:02:49.210 --> 00:02:52.030 you get this highly coordinated global behavior. 00:02:52.270 --> 00:02:54.430 And the crazy part is the individual agents are 00:02:54.430 --> 00:02:56.930 completely clueless about the big picture. I 00:02:56.930 --> 00:02:59.409 mean, the ant dragging a leaf doesn't know it's 00:02:59.409 --> 00:03:01.949 building a structural bridge over a puddle. Right. 00:03:02.069 --> 00:03:04.409 It has no idea. It just knows it's following 00:03:04.409 --> 00:03:07.120 the ant immediately in front of it. Which brings 00:03:07.120 --> 00:03:09.699 us to how researchers first managed to capture 00:03:09.699 --> 00:03:13.219 this biological magic inside a computer, which 00:03:13.219 --> 00:03:15.240 I find fascinating. Oh, this is the Greg Reynolds 00:03:15.240 --> 00:03:19.180 stuff, right? Yes. Back in 1986, Reynolds created 00:03:19.180 --> 00:03:23.000 this artificial life program called BOIDS. B 00:03:23.000 --> 00:03:26.960 -O -I -D -S. Right. Short for Birdoid Objects. 00:03:27.120 --> 00:03:29.960 He wanted to simulate a flock of birds on a computer 00:03:29.960 --> 00:03:33.300 screen. But instead of trying to write one massive 00:03:33.300 --> 00:03:36.280 complicated algorithm to control the flight paths 00:03:36.280 --> 00:03:38.280 of the entire flock from the top down. Because 00:03:38.280 --> 00:03:40.919 that would be impossible. Pretty much. So he 00:03:40.919 --> 00:03:43.060 decided to just program the individual birds. 00:03:43.659 --> 00:03:46.599 He gave each digital boy just three fundamental 00:03:46.599 --> 00:03:49.530 rules. And these three rules are like the secret 00:03:49.530 --> 00:03:51.229 sauce of the whole concept. Let's break them 00:03:51.229 --> 00:03:53.469 down for everyone. Rule number one is separation. 00:03:53.969 --> 00:03:56.689 Right. Basically, steer to avoid crowding your 00:03:56.689 --> 00:03:58.949 local flock mates. You know, maintain your personal 00:03:58.949 --> 00:04:00.949 bubble so you don't crash into your neighbor. 00:04:01.050 --> 00:04:03.330 It's the don't touch me rule. And then you have 00:04:03.330 --> 00:04:06.430 rule number two, which is alignment. You look 00:04:06.430 --> 00:04:08.610 at the birds immediately around you. and you 00:04:08.610 --> 00:04:10.650 steer towards their average heading. So if the 00:04:10.650 --> 00:04:13.030 three birds next to you bank left, you bank left. 00:04:13.330 --> 00:04:15.530 Exactly. You align your trajectory with your 00:04:15.530 --> 00:04:18.170 local group. And then rule number three is cohesion. 00:04:18.889 --> 00:04:21.670 You steer to move toward the center of mass of 00:04:21.670 --> 00:04:24.750 your local flock mates. So don't wander off alone. 00:04:24.930 --> 00:04:27.009 Try to stick to the densest part of the group 00:04:27.009 --> 00:04:30.850 you can see. Separation, alignment, cohesion. 00:04:31.110 --> 00:04:34.050 Yep. And when Reynolds hit run on the simulation, 00:04:34.379 --> 00:04:36.860 Those individual triangles on his screen didn't 00:04:36.860 --> 00:04:39.680 just vibrate randomly, they actually flocked. 00:04:39.680 --> 00:04:43.459 Wow. They moved to the exact same fluid organic 00:04:43.459 --> 00:04:46.879 grace as a real flock of starlings evading a 00:04:46.879 --> 00:04:50.129 hawk. And later, researchers like Vicksec, with 00:04:50.129 --> 00:04:52.709 his self -propelled particles model, prove that 00:04:52.709 --> 00:04:54.870 these swarming properties occur universally at 00:04:54.870 --> 00:04:57.329 the group level. Regardless of the animal. Exactly. 00:04:57.509 --> 00:04:59.670 The underlying physics are incredibly robust, 00:04:59.730 --> 00:05:01.829 whether you're modeling birds, schools of fish, 00:05:02.250 --> 00:05:04.449 or marching locusts. You know, hearing those 00:05:04.449 --> 00:05:07.129 three specific rules, I kind of have to playfully 00:05:07.129 --> 00:05:09.629 question this whole idea of human beings being 00:05:09.629 --> 00:05:12.050 the pinnacle of complex, independent thought. 00:05:12.129 --> 00:05:14.470 Oh, yeah. How so? Well, think about being at 00:05:14.470 --> 00:05:17.050 a massive, crowded general admission concert. 00:05:17.000 --> 00:05:19.839 OK, picture good. Are we humans basically just 00:05:19.839 --> 00:05:22.399 running the separation, alignment, and cohesion 00:05:22.399 --> 00:05:24.639 algorithm when we try to push our way to the 00:05:24.639 --> 00:05:26.459 front of the stage? That's wow. Yeah. I mean, 00:05:26.819 --> 00:05:29.040 think about it. I separate so I don't spill my 00:05:29.040 --> 00:05:31.480 drink on the guy next to me. I align with the 00:05:31.480 --> 00:05:34.199 general direction the crowd is surging. And I 00:05:34.199 --> 00:05:36.319 cohere to stay near my group of friends. Are 00:05:36.319 --> 00:05:40.240 we all just Boyds in vintage band tees? I love 00:05:40.240 --> 00:05:44.050 that. And honestly, at a macro level, human crowds 00:05:44.050 --> 00:05:47.470 do behave remarkably like fluid dynamics or particle 00:05:47.470 --> 00:05:49.810 swarms. It's kind of humbling. It is. We run 00:05:49.810 --> 00:05:51.709 those localized algorithms constantly without 00:05:51.709 --> 00:05:54.069 realizing it. But the real leap in this field 00:05:54.069 --> 00:05:56.149 wasn't just observing that we act like flocks. 00:05:56.189 --> 00:05:58.569 Right. It was realizing that if we can simulate 00:05:58.569 --> 00:06:01.350 nature's bottom -up problem -solving in a computer, 00:06:01.730 --> 00:06:04.490 we can actually use it to solve incredibly complex 00:06:04.490 --> 00:06:07.050 human logistics. Which opens the door to using 00:06:07.050 --> 00:06:09.649 these nature -inspired models to solve optimization 00:06:09.649 --> 00:06:12.879 puzzles. The deep dive sources heavily tie this 00:06:12.879 --> 00:06:15.000 to something called ant colony optimization. 00:06:15.259 --> 00:06:18.180 Yes, Marco Dorigo introduced this in 1992. And 00:06:18.180 --> 00:06:20.040 ant colony optimization is just a beautiful piece 00:06:20.040 --> 00:06:22.279 of logic. Walk us through it. So in the physical 00:06:22.279 --> 00:06:25.259 world, when ants leave their nest to search for 00:06:25.259 --> 00:06:28.139 food, they lay down a chemical trail of pheromones. 00:06:28.819 --> 00:06:31.379 If an ant finds a piece of fruit, it turns around, 00:06:31.540 --> 00:06:34.259 walks back to the nest, and leaves more pheromones. 00:06:34.500 --> 00:06:37.819 So the trail gets stronger. Exactly. Other ants 00:06:37.819 --> 00:06:40.899 smell that trail and follow it. But here's the 00:06:40.899 --> 00:06:43.500 real mechanical trick. The shorter the path to 00:06:43.500 --> 00:06:46.279 the food, the faster the ants can run back and 00:06:46.279 --> 00:06:48.720 forth. Oh, I see where this is going. Right. 00:06:49.019 --> 00:06:50.620 Because they're running back and forth faster 00:06:50.620 --> 00:06:53.500 on the short path, that specific trail gets a 00:06:53.500 --> 00:06:55.920 stronger and stronger concentration of pheromones 00:06:55.920 --> 00:06:58.139 before the chemicals have time to evaporate. 00:06:58.319 --> 00:07:02.040 Ah. So the long winding paths literally just 00:07:02.040 --> 00:07:04.519 evaporate into thin air while the quickest route 00:07:04.519 --> 00:07:07.060 becomes this superhighway of chemical signals. 00:07:07.160 --> 00:07:09.600 You got it. The entire colony converges on the 00:07:09.600 --> 00:07:12.339 most efficient route without a single ant knowing 00:07:12.339 --> 00:07:14.810 geometry. That's brilliant. And we figured out 00:07:14.810 --> 00:07:17.610 how to digitize this, right? We have. We basically 00:07:17.610 --> 00:07:21.050 drop simulated ants into a sea of data to locate 00:07:21.050 --> 00:07:23.610 optimal paths for routing puzzles. So when a 00:07:23.610 --> 00:07:26.589 digital ant finds a good solution, it leaves 00:07:26.589 --> 00:07:30.209 a digital pheromone, which attracts future iterations 00:07:30.209 --> 00:07:33.350 of ants to flock toward better and better solutions. 00:07:33.610 --> 00:07:36.490 And this translates to massive financial savings 00:07:36.490 --> 00:07:39.230 in the real world. I mean, Southwest Airlines 00:07:39.230 --> 00:07:43.709 use this exact ant -based routing lot. to solve 00:07:43.709 --> 00:07:46.029 some of their worst logistical nightmare. Oh 00:07:46.029 --> 00:07:48.050 right, the boarding process. Yeah, they built 00:07:48.050 --> 00:07:51.110 a computer simulation with just six simple interaction 00:07:51.110 --> 00:07:53.829 rules to figure out the absolute most efficient 00:07:53.829 --> 00:07:57.060 way to board human passengers onto a plane. What 00:07:57.060 --> 00:07:58.879 I found really interesting in the sources is 00:07:58.879 --> 00:08:00.879 that it's not just the passengers acting like 00:08:00.879 --> 00:08:03.540 a swarm. The Southwest pilots themselves actually 00:08:03.540 --> 00:08:06.180 utilize swarm theory for finding airport gates. 00:08:06.279 --> 00:08:08.259 Yeah, let's look at the mechanism of how that 00:08:08.259 --> 00:08:10.360 actually works. Because traditionally you'd have 00:08:10.360 --> 00:08:13.540 a central boss in a control tower micromanaging 00:08:13.540 --> 00:08:15.939 every single plane's schedule. Like flight 102, 00:08:16.220 --> 00:08:18.839 go to gate four, flight 88, wait for gate nine. 00:08:19.120 --> 00:08:21.839 Exactly. But in a highly dynamic environment, 00:08:22.100 --> 00:08:24.839 if a baggage car breaks down at gate four, the 00:08:24.839 --> 00:08:27.740 central planner's entire schedule collapses into 00:08:27.740 --> 00:08:30.060 this cascading delay. Because it's too rigid. 00:08:30.319 --> 00:08:33.340 Right. So instead, Southwest pilots act like 00:08:33.340 --> 00:08:36.080 an ant colony. Wait, so what acts as the pheromone 00:08:36.080 --> 00:08:38.299 in a massive commercial airport? I assume they 00:08:38.299 --> 00:08:40.240 aren't spraying chemicals out the window. No, 00:08:40.620 --> 00:08:43.220 no. Digital timestamps and turnaround data. OK, 00:08:43.360 --> 00:08:46.620 gotcha. The pilot's onboard systems look at which 00:08:46.620 --> 00:08:49.159 gates are clearing out the fastest in real time. 00:08:49.879 --> 00:08:51.740 So if gate four has that broken baggage cart, 00:08:52.179 --> 00:08:54.419 its digital pheromone strength plummets. Oh, 00:08:54.480 --> 00:08:56.600 that makes sense. And the pilot system organically 00:08:56.600 --> 00:08:59.399 reads that weakened signal and diverts them to 00:08:59.399 --> 00:09:01.960 gate seven, which is currently emitting a strong 00:09:01.960 --> 00:09:04.659 signal of fast turnaround times. So the pilots 00:09:04.659 --> 00:09:06.700 are adapting locally to the conditions on the 00:09:06.700 --> 00:09:09.159 ground. Completely. And over hundreds of flights, 00:09:09.720 --> 00:09:12.639 the quote -unquote colony of pilots organically 00:09:12.639 --> 00:09:15.399 self -optimizes. They anticipate plane backups 00:09:15.399 --> 00:09:17.860 before they even happen. Entirely without a top 00:09:17.860 --> 00:09:20.700 -down boss directing traffic. Exactly. It proves 00:09:20.700 --> 00:09:22.919 that in chaotic environments, top -down planning 00:09:22.919 --> 00:09:26.080 is just too brittle. Decentralized problem -solving 00:09:26.080 --> 00:09:28.340 absorbs the chaos. Which brings us to another 00:09:28.340 --> 00:09:30.320 variation of this from the sources from around 00:09:30.320 --> 00:09:34.320 1995, I think? Particle swarm optimization. Yes, 00:09:34.480 --> 00:09:37.539 PSO. So, instead of ants looking for physical 00:09:37.539 --> 00:09:40.960 paths, this is used when a best solution exists 00:09:40.960 --> 00:09:43.679 somewhere in a vast multi -dimensional space. 00:09:43.779 --> 00:09:46.919 Right. The digital particles move through the 00:09:46.919 --> 00:09:49.700 data, and at every time step, they check how 00:09:49.700 --> 00:09:53.000 good their current position is. Over time, they 00:09:53.000 --> 00:09:54.960 accelerate toward the particles in their network 00:09:54.960 --> 00:09:57.460 that have the best results. Yep, that's the core 00:09:57.460 --> 00:09:59.419 of it. Okay, let me pause you there, though, 00:09:59.879 --> 00:10:02.769 because I'm trying to picture this. With particle 00:10:02.769 --> 00:10:05.269 swarm optimization, instead of trying to calculate 00:10:05.269 --> 00:10:08.009 the perfect answer yourself from scratch, you 00:10:08.009 --> 00:10:09.990 just look over at your neighbor, who is doing 00:10:09.990 --> 00:10:12.470 slightly better than you, and you accelerate 00:10:12.470 --> 00:10:14.909 toward their answer. Basically, yes. Isn't that 00:10:14.909 --> 00:10:17.350 just mathematically sanctioned cheating? I mean, 00:10:17.350 --> 00:10:19.750 in a classroom, yeah, it's cheating. But in computer 00:10:19.750 --> 00:10:22.629 science, it's radical efficiency. OK, fair enough. 00:10:22.830 --> 00:10:25.129 Think of it like this. Imagine you are a hiker 00:10:25.129 --> 00:10:28.409 trying to find the absolute lowest point in a 00:10:28.409 --> 00:10:31.000 massive rolling mountain range. OK. but it's 00:10:31.000 --> 00:10:33.379 pitch black outside, and you only have a flashlight. 00:10:34.460 --> 00:10:37.000 If you just walk downhill, you might hit the 00:10:37.000 --> 00:10:39.639 bottom of a small valley, and you think, great, 00:10:39.740 --> 00:10:42.000 I've found the lowest point. But you're actually 00:10:42.000 --> 00:10:44.620 just stuck in a shallow crater, while a massive 00:10:44.620 --> 00:10:48.340 canyon exists 10 miles away. Mathematicians call 00:10:48.340 --> 00:10:51.460 that getting trapped in a local minima. Ah, so 00:10:51.460 --> 00:10:53.200 you think you've solved the problem, but you've 00:10:53.200 --> 00:10:56.399 only found a decent solution, not the best global 00:10:56.399 --> 00:10:59.590 solution. Exactly. But now imagine there are 00:10:59.590 --> 00:11:02.049 10 ,000 hikers spread across the mountain range 00:11:02.049 --> 00:11:05.090 and they all have radios. Oh, I see. The moment 00:11:05.090 --> 00:11:07.570 one hiker finds a valley that is deeper than 00:11:07.570 --> 00:11:11.460 yours, they broadcast their elevation. you immediately 00:11:11.460 --> 00:11:13.539 stop searching your shallow crater and start 00:11:13.539 --> 00:11:15.519 walking toward their coordinates. That is so 00:11:15.519 --> 00:11:17.860 smart, because the swarm balances individual 00:11:17.860 --> 00:11:20.759 exploration with collective communication. Right. 00:11:20.919 --> 00:11:23.419 It effortlessly glides toward the best possible 00:11:23.419 --> 00:11:26.320 global solution, avoiding those dead ends entirely. 00:11:26.460 --> 00:11:29.080 So we've got digital ants routing airplanes and 00:11:29.080 --> 00:11:31.039 digital hikers cheating off their neighbors' 00:11:31.399 --> 00:11:34.039 math tests, but... What really fascinates me 00:11:34.039 --> 00:11:36.580 is how this scales. Like, the technology didn't 00:11:36.580 --> 00:11:39.179 just stay in logistics. It went straight to Hollywood. 00:11:39.279 --> 00:11:42.149 Oh yeah, massive impact there. Right. Anyone 00:11:42.149 --> 00:11:44.370 who has watched those sprawling battle scenes 00:11:44.370 --> 00:11:46.850 in the Lord of the Rings has probably wondered 00:11:46.850 --> 00:11:49.769 how they animated tens of thousands of orcs crashing 00:11:49.769 --> 00:11:52.090 into each other. It would take a lifetime to 00:11:52.090 --> 00:11:54.610 hand animate every single soldier in a battle 00:11:54.610 --> 00:11:57.769 of that scale. Exactly. So they used a swarm 00:11:57.769 --> 00:12:00.750 software called Massive. Right. And earlier versions 00:12:00.750 --> 00:12:03.750 were used for things like the Bats and Tim Burton's 00:12:03.750 --> 00:12:07.090 Batman Returns, or even flocks in this 1987 film, 00:12:07.090 --> 00:12:09.169 Breaking the Ice. Oh, I didn't know it went back 00:12:09.169 --> 00:12:11.809 that far. But for Lord of the Rings... the animators 00:12:11.809 --> 00:12:14.830 gave each digital orc a localized vision field 00:12:14.830 --> 00:12:17.250 and a set of simple rules. Just like the voids. 00:12:17.429 --> 00:12:20.529 Yep. An orc basically looks at its immediate 00:12:20.529 --> 00:12:23.309 radius and asks, you know, is the terrain steep 00:12:23.309 --> 00:12:25.570 here? Is the guy next to me a friend or an enemy? 00:12:25.769 --> 00:12:28.850 If enemies swing sword. Right. They drop 10 ,000 00:12:28.850 --> 00:12:31.190 of these agents onto a digital battlefield and 00:12:31.190 --> 00:12:33.429 the battle basically fights itself. It generates 00:12:33.429 --> 00:12:36.490 incredibly realistic, chaotic, emergent behavior 00:12:36.490 --> 00:12:39.690 because every single orc is reacting locally. 00:12:39.899 --> 00:12:42.460 And what's wild in the deep dive sources is that 00:12:42.460 --> 00:12:45.240 this technology scales perfectly from the massive 00:12:45.659 --> 00:12:48.279 all the way down to the microscopic. Because 00:12:48.279 --> 00:12:51.240 while Hollywood was simulating orcs, researchers 00:12:51.240 --> 00:12:55.460 like Lewis and Bickey back in 1992 were proposing 00:12:55.460 --> 00:12:58.259 using swarm intelligence to control nanobots 00:12:58.259 --> 00:13:00.700 inside the human body. And the medical application 00:13:00.700 --> 00:13:03.519 is where this becomes truly life saving. The 00:13:03.519 --> 00:13:06.240 traditional idea of medical robotics was injecting 00:13:06.240 --> 00:13:09.600 one highly complex robot to navigate the bloodstream. 00:13:09.639 --> 00:13:11.419 Which sounds incredibly difficult to pull off. 00:13:11.639 --> 00:13:14.259 Right. Swarm theory says no, inject a swarm of 00:13:14.259 --> 00:13:17.059 microscopic incredibly simple agents. They follow 00:13:17.059 --> 00:13:19.600 basic local rules to seek out and destroy cancer 00:13:19.600 --> 00:13:21.559 cells. The sources mention a researcher named 00:13:21.559 --> 00:13:24.100 Derafaye using an ant -based algorithm called 00:13:24.100 --> 00:13:27.019 Stochastic Diffusion Search to locate real tumors 00:13:27.019 --> 00:13:29.740 in human bone scans. Yes. But how does an ant 00:13:29.740 --> 00:13:32.700 algorithm even see a tumor on an x -ray? That 00:13:32.700 --> 00:13:34.279 sounds like science fiction. Well, you have to 00:13:34.279 --> 00:13:36.320 think of a medical scan not as a picture, but 00:13:36.320 --> 00:13:38.340 as a digital landscape made of black, white, 00:13:38.440 --> 00:13:41.679 and gray pixels. OK. The algorithm releases thousands 00:13:41.679 --> 00:13:45.080 of digital ants onto this pixel landscape. And 00:13:45.080 --> 00:13:47.419 the ants are programmed to search for specific 00:13:47.419 --> 00:13:50.759 jagged edges or textures or dense clusters of 00:13:50.759 --> 00:13:53.200 white pixels that indicate an anomaly. Oh, I 00:13:53.200 --> 00:13:55.740 see. When an ant crawls over a suspicious texture, 00:13:56.259 --> 00:13:59.269 it drops a digital pheromone. So as the ants 00:13:59.269 --> 00:14:02.169 randomly explore the x -ray, the areas with normal 00:14:02.169 --> 00:14:04.830 smooth bone tissue just get ignored. Exactly. 00:14:05.049 --> 00:14:07.610 But the moment a few ants stumble onto the rough 00:14:07.610 --> 00:14:10.509 texture of a tumor, their pheromones attract 00:14:10.509 --> 00:14:12.830 the rest of the swarm, lighting up the exact 00:14:12.830 --> 00:14:14.929 perimeter of the cancer for the doctor to see. 00:14:15.090 --> 00:14:17.909 It's the universal language of decentralized 00:14:17.909 --> 00:14:20.809 problem -solving. The digital orcs fighting on 00:14:20.809 --> 00:14:23.570 the battlefield in Middle Earth, and the microscopic 00:14:23.570 --> 00:14:26.049 digital ants hunting for cancer cells in a bone 00:14:26.049 --> 00:14:28.440 scan. They're fundamentally running on the exact 00:14:28.440 --> 00:14:30.779 same underlying mechanics. The exact same math. 00:14:31.019 --> 00:14:34.019 That is just, it's wild. Okay, so we've proven 00:14:34.019 --> 00:14:36.519 that mindless code can mimic ants to solve math, 00:14:36.759 --> 00:14:39.720 route planes, and find tumors. But humans aren't 00:14:39.720 --> 00:14:41.679 mindless code. Last time I checked, yeah. Right, 00:14:41.899 --> 00:14:45.279 we have egos, biases, and highly complex brains. 00:14:45.679 --> 00:14:48.519 Does imposing these rigid swarm rules on human 00:14:48.519 --> 00:14:52.120 beings stifle our intelligence? Or amplify it. 00:14:52.120 --> 00:14:54.960 It amplifies it exponentially. Oh yeah. This 00:14:54.960 --> 00:14:58.000 is the frontier of artificial swarm intelligence 00:14:58.000 --> 00:15:01.259 or ASI, which is often called human swarming. 00:15:01.580 --> 00:15:03.559 A researcher named Louis Rosenberg developed 00:15:03.559 --> 00:15:06.240 this around 2015. Okay. He looked at natural 00:15:06.240 --> 00:15:08.980 swarms and realized that human collective intelligence 00:15:08.980 --> 00:15:12.220 is usually limited by our tools. So he asked 00:15:12.519 --> 00:15:15.240 What would happen if we networked human minds 00:15:15.240 --> 00:15:18.019 together dynamically, just like a flock of birds, 00:15:18.320 --> 00:15:20.679 to make complex decisions? Wait, I need to push 00:15:20.679 --> 00:15:22.360 back on this a little. Go for it. Because when 00:15:22.360 --> 00:15:24.460 you say network humans together to make a decision, 00:15:24.580 --> 00:15:26.320 I'm just picturing a bunch of doctors in a Zoom 00:15:26.320 --> 00:15:28.100 chat shouting over each other, or maybe just 00:15:28.100 --> 00:15:30.179 sending out a digital poll. Right. How are they 00:15:30.179 --> 00:15:32.960 actually swarming a medical scan without it devolving 00:15:32.960 --> 00:15:35.240 into complete chaos? Like, how is this actually 00:15:35.240 --> 00:15:36.940 different from just taking a simple majority 00:15:36.940 --> 00:15:39.820 vote in a room? That's the key question. A simple 00:15:39.820 --> 00:15:42.720 majority vote is a static process. Everyone casts 00:15:42.720 --> 00:15:44.860 their vote in total isolation. You tally them 00:15:44.860 --> 00:15:48.039 up, and the majority wins. Right. But it doesn't 00:15:48.039 --> 00:15:51.679 capture conviction or doubt. Human swarming, 00:15:51.740 --> 00:15:55.419 on the other underhand, is a real -time dynamic 00:15:55.419 --> 00:15:58.019 physical interface. Think of it like a giant 00:15:58.019 --> 00:16:01.379 digital Ouija board. A Ouija board! Or a massive 00:16:01.379 --> 00:16:04.159 multi -directional tug -of -war. Okay, a digital 00:16:04.159 --> 00:16:06.500 Ouija board. Paint that picture for me. How does 00:16:06.500 --> 00:16:09.059 that work? Imagine a group of 50 radiologists 00:16:09.059 --> 00:16:11.419 all looking at their own computer screens. Okay. 00:16:11.929 --> 00:16:15.429 On the screen is a medical scan and an interface 00:16:15.429 --> 00:16:18.269 with a glass puck in the center. Around the edges 00:16:18.269 --> 00:16:21.070 of the screen are different potential diagnoses, 00:16:21.330 --> 00:16:24.049 pneumonia, tuberculosis, benign, malignant. Gotcha. 00:16:24.350 --> 00:16:26.769 Every doctor uses their mouse to grab the puck 00:16:26.769 --> 00:16:28.429 and pull it toward what they think is the right 00:16:28.429 --> 00:16:30.850 answer. So everyone is pulling at the exact same 00:16:30.850 --> 00:16:32.970 time. Right. And you can see the puck moving 00:16:32.970 --> 00:16:35.149 in real time based on the collective pull of 00:16:35.149 --> 00:16:37.669 the entire group. So if you're pulling the puck 00:16:37.669 --> 00:16:40.669 toward benign tumor, but you see the puck accelerating 00:16:40.669 --> 00:16:43.090 rapidly toward malignant, you are feeling the 00:16:43.090 --> 00:16:46.070 momentum of the swarm. Oh, wow. In real time, 00:16:46.269 --> 00:16:48.909 you might reevaluate the scan, realize you missed 00:16:48.909 --> 00:16:51.649 a crucial detail, and dynamically shift your 00:16:51.649 --> 00:16:54.429 mouse to help the swarm converge. Or, I guess 00:16:54.429 --> 00:16:56.409 if you are absolutely certain it's benign, you 00:16:56.409 --> 00:16:59.029 can dig in your heels and pull harder. Exactly. 00:16:59.409 --> 00:17:01.389 Signaling your high conviction to the rest of 00:17:01.389 --> 00:17:04.309 the group. It's not a rigid vote. It's a real 00:17:04.309 --> 00:17:07.380 -time negotiation of thousands of micro adjustments, 00:17:07.880 --> 00:17:09.880 just like birds adjusting their wings second 00:17:09.880 --> 00:17:13.069 by second to stay with the flock. That is fascinating. 00:17:13.190 --> 00:17:15.470 And the data in the sources on how well this 00:17:15.470 --> 00:17:17.750 works is staggering. Oh, the results are incredible. 00:17:18.170 --> 00:17:21.130 Yeah, a 2018 study from Stanford University School 00:17:21.130 --> 00:17:23.829 of Medicine took groups of human radiologists 00:17:23.829 --> 00:17:26.549 and connected them using these real -time swarming 00:17:26.549 --> 00:17:29.809 algorithms to diagnose chest x -rays. And what 00:17:29.809 --> 00:17:32.410 happened? This human swarm reduced diagnostic 00:17:32.410 --> 00:17:35.650 errors by 33 % compared to traditional methods. 00:17:35.829 --> 00:17:38.789 33 %? I know. And it didn't just beat individual 00:17:38.789 --> 00:17:40.930 human doctors. The human swarm actually beat 00:17:40.910 --> 00:17:43.730 traditional machine learning AI by 22%. Which 00:17:43.730 --> 00:17:46.170 is huge. For something as critical as diagnosing 00:17:46.170 --> 00:17:50.089 pneumonia, a 33 % reduction in errors is a monumental 00:17:50.089 --> 00:17:53.170 leap. Seriously. And then there was a later study 00:17:53.170 --> 00:17:57.569 in 2021 out of UC San Francisco that showed a 00:17:57.569 --> 00:18:01.430 23 % increase in diagnostic accuracy for MRI 00:18:01.430 --> 00:18:04.230 images when doctors swarmed, compared to just 00:18:04.230 --> 00:18:06.390 taking a traditional majority vote. Right. Because 00:18:06.390 --> 00:18:09.109 by allowing the doctors to deliberate dynamically, 00:18:09.670 --> 00:18:11.410 feeling the push and pull of their colleagues' 00:18:11.970 --> 00:18:14.109 convictions, they achieved a level of insight 00:18:14.109 --> 00:18:17.829 that no individual genius and no static poll 00:18:17.829 --> 00:18:20.319 could ever match. And the sources show this works 00:18:20.319 --> 00:18:22.339 outside of medicine, too. I mean, the United 00:18:22.339 --> 00:18:25.099 Nations Food and Agriculture Organization uses 00:18:25.099 --> 00:18:27.900 human swarming to forecast global famines. Oh, 00:18:27.960 --> 00:18:30.240 wow. And on a lighter note, sports fans have 00:18:30.240 --> 00:18:33.059 used ASI platforms to consistently outperform 00:18:33.059 --> 00:18:36.599 Vegas betting markets. Really, Vegas? Yep. It 00:18:36.599 --> 00:18:38.180 fundamentally proves that if you want to predict 00:18:38.180 --> 00:18:40.700 the future or solve an impossible problem, you 00:18:40.700 --> 00:18:42.200 don't look for the smartest person in the room. 00:18:42.599 --> 00:18:45.220 You network the room into a swarm. That's incredible. 00:18:45.359 --> 00:18:47.259 So we've covered biological survival, logistics, 00:18:47.519 --> 00:18:50.119 Hollywood, and medical diagnoses. But we have 00:18:50.119 --> 00:18:51.940 to look at the edge case. Oh, my favorite part. 00:18:52.500 --> 00:18:55.460 Can a decentralized swarm operating entirely 00:18:55.460 --> 00:18:58.279 without a master planner generate creativity? 00:18:58.819 --> 00:19:02.869 Like, can algorithms create art? Swarmic art. 00:19:03.269 --> 00:19:05.730 I love the mechanics of this. So researchers, 00:19:05.950 --> 00:19:08.529 like Al -Rufay, again asked if algorithms could 00:19:08.529 --> 00:19:11.549 sketch, and they ended up creating swarm grammars. 00:19:11.609 --> 00:19:14.089 Which is a completely novel integration of two 00:19:14.089 --> 00:19:15.369 different systems we've already talked about 00:19:15.369 --> 00:19:17.609 today. Right, they basically merged the ants 00:19:17.609 --> 00:19:19.990 and the birds. Exactly. The process starts with 00:19:19.990 --> 00:19:22.410 the ant foraging algorithm, the stochastic diffusion 00:19:22.410 --> 00:19:25.509 search. The researchers feed a blank digital 00:19:25.509 --> 00:19:28.349 canvas and an input photograph into the system. 00:19:28.430 --> 00:19:31.160 Okay. And the job of the ants is to control the 00:19:31.160 --> 00:19:33.660 system's attention. The digital ants scatter 00:19:33.660 --> 00:19:36.240 across the canvas, specifically searching for 00:19:36.240 --> 00:19:39.799 areas of high detail, sharp contrast, or visual 00:19:39.799 --> 00:19:42.799 complexity in the input image. They are the explorers. 00:19:43.019 --> 00:19:45.779 And once the ants lock onto a focal area, say, 00:19:45.839 --> 00:19:47.940 like the sharp lines of a person's eye in a portrait, 00:19:48.279 --> 00:19:50.779 they drop their digital pheromones. And that's 00:19:50.779 --> 00:19:53.180 when the second algorithm, particle swarm optimization, 00:19:53.299 --> 00:19:55.980 the birds takes over. The bird algorithm senses 00:19:55.980 --> 00:19:58.519 the pheromones left by the ants, swoops into 00:19:58.519 --> 00:20:01.079 that specific area of the canvas, and actually 00:20:01.079 --> 00:20:03.819 sketches the physical lines. Wow. It creates 00:20:03.819 --> 00:20:06.480 this fascinating artistic tension. The system 00:20:06.480 --> 00:20:09.640 balances exploration. The ants restlessly seeking 00:20:09.640 --> 00:20:13.359 out new complex regions of the canvas with exploitation. 00:20:14.279 --> 00:20:16.980 The birds intensely focusing on drawing the lines 00:20:16.980 --> 00:20:19.339 in those specific spots. You know, if we think 00:20:19.339 --> 00:20:22.900 about this like a human painter. The ant algorithm 00:20:22.900 --> 00:20:25.740 is essentially the artist's eye, constantly darting 00:20:25.740 --> 00:20:27.640 around the subject, looking for inspiration, 00:20:28.180 --> 00:20:30.460 noticing the way the light hits a cheekbone. 00:20:30.559 --> 00:20:32.759 Beautiful analogy, yes. And the bird algorithm 00:20:32.759 --> 00:20:35.119 is the artist's hand swooping in with the charcoal 00:20:35.119 --> 00:20:37.319 to actively commit those lines to the paper. 00:20:37.539 --> 00:20:39.339 And the tension between those two, the restless 00:20:39.339 --> 00:20:42.220 eye and the focused hand, is what fulfills the 00:20:42.220 --> 00:20:44.920 core philosophical requirement of computational 00:20:44.920 --> 00:20:47.380 creativity. Which is what? Balancing freedom 00:20:47.380 --> 00:20:49.990 and constraints. The ants provide the freedom 00:20:49.990 --> 00:20:52.789 to endlessly explore the image, and the birds 00:20:52.789 --> 00:20:55.230 provide the physical constraint of actually rendering 00:20:55.230 --> 00:20:57.900 the sketch. And because the ants are constantly 00:20:57.900 --> 00:21:01.140 reacting to the evolving drawing, the swarm never 00:21:01.140 --> 00:21:04.900 takes the exact same path twice. Never. It generates 00:21:04.900 --> 00:21:08.319 a unique, non -identical, highly stylized sketch 00:21:08.319 --> 00:21:11.380 every single time you hit run. It proves that 00:21:11.380 --> 00:21:14.279 local rules don't just solve cold, hard math 00:21:14.279 --> 00:21:17.559 problems. They can actually generate unpredictable 00:21:17.559 --> 00:21:20.779 beauty. It's just wild. Let's take a step back 00:21:20.779 --> 00:21:22.400 for a second and look at the massive journey 00:21:22.400 --> 00:21:24.779 we just took through these sources. We started 00:21:24.779 --> 00:21:27.660 with three Simple rules, separation, alignment, 00:21:27.920 --> 00:21:31.420 cohesion coded into digital voids. We explored 00:21:31.420 --> 00:21:34.019 how airline pilots use digital pheromones to 00:21:34.019 --> 00:21:36.240 act like an ant colony and prevent cascading 00:21:36.240 --> 00:21:39.259 flight delays. We saw networked human minds pulling 00:21:39.259 --> 00:21:42.640 a digital puck to diagnose diseases 33 % better 00:21:42.640 --> 00:21:44.900 than a room full of isolated experts. And we 00:21:44.900 --> 00:21:48.359 finished with algorithmic ants and birds collaborating 00:21:48.359 --> 00:21:51.019 to paint original artwork. It's a lot. It is. 00:21:51.420 --> 00:21:53.500 But the through line connecting all of this is 00:21:53.500 --> 00:21:56.490 profoundly humbling. The most complex, brilliant 00:21:56.490 --> 00:21:59.390 outcomes in our world rarely require a master 00:21:59.390 --> 00:22:01.450 planner. They don't require a boss. They really 00:22:01.450 --> 00:22:05.750 don't. Accurate and creative solutions emerge 00:22:05.750 --> 00:22:08.609 organically when a group simply follows the right 00:22:08.609 --> 00:22:12.170 local rules and trusts the constant dynamic interactions 00:22:12.170 --> 00:22:14.549 between its members. So, for everyone listening, 00:22:14.750 --> 00:22:16.450 I want you to reflect on your own life for a 00:22:16.450 --> 00:22:19.529 second. Next time you're stuck on an incredibly 00:22:19.529 --> 00:22:22.190 complex problem at work, or maybe you are feeling 00:22:22.190 --> 00:22:24.829 totally lost in a chaotic, top -down corporate 00:22:24.829 --> 00:22:27.450 hierarchy. We've all been there. Ask yourself, 00:22:27.829 --> 00:22:30.269 what local rules are you following? Instead of 00:22:30.269 --> 00:22:32.089 sitting around waiting for top -down orders from 00:22:32.089 --> 00:22:34.630 the CEO, could your team benefit from acting 00:22:34.630 --> 00:22:37.250 a bit more like a swarm? Just interacting locally, 00:22:37.730 --> 00:22:40.190 sharing your real -time data dynamically and 00:22:40.190 --> 00:22:42.349 trusting the emergent intelligence to find the 00:22:42.349 --> 00:22:45.130 best path forward. Exactly. Moving away from 00:22:45.130 --> 00:22:47.049 the myth of the hero individual and learning 00:22:47.049 --> 00:22:49.170 to trust the collective intelligence of the network 00:22:49.170 --> 00:22:51.710 is a massive paradigm shift for how we organize 00:22:51.710 --> 00:22:53.890 our lives. It really is. And we want to leave 00:22:53.890 --> 00:22:56.730 you with one final kind of mind -bending thought 00:22:56.730 --> 00:22:59.269 to mull over. We've seen that if you take human 00:22:59.269 --> 00:23:01.450 minds, which are already incredibly smart, and 00:23:01.450 --> 00:23:04.430 network them into a swarm, they become 33 % better 00:23:04.430 --> 00:23:06.970 at diagnosing life -threatening diseases. Right. 00:23:07.210 --> 00:23:10.230 And we've seen that if you take mindless digital 00:23:10.230 --> 00:23:13.640 ants and apply swarm rules, They become brilliant 00:23:13.640 --> 00:23:17.900 at solving global logistical puzzles. So what 00:23:17.900 --> 00:23:20.099 happens in the very near future when we connect 00:23:20.099 --> 00:23:22.599 artificial intelligence agents into a real -time 00:23:22.599 --> 00:23:25.880 swarm? Oh, wow. If simple rules can make mindless 00:23:25.880 --> 00:23:28.380 code brilliant, what could swarm intelligence 00:23:28.380 --> 00:23:31.400 achieve with AI agents that are already highly 00:23:31.400 --> 00:23:33.940 intelligent to begin with? That's, I mean, the 00:23:33.940 --> 00:23:35.700 diagnostic muddy waters we talked about at the 00:23:35.700 --> 00:23:38.140 beginning. The places where individual human 00:23:38.140 --> 00:23:41.200 precision hits a wall might just require a completely 00:23:41.200 --> 00:23:43.359 different kind of vision. Not a smarter individual. 00:23:43.559 --> 00:23:46.039 No, but a murmuration. A murmuration of minds. 00:23:46.539 --> 00:23:47.900 Something to think about the next time you look 00:23:47.900 --> 00:23:49.599 up at a flock of birds in the evening sky.