WEBVTT 00:00:00.000 --> 00:00:04.759 Right now, inside your head, there are 86 billion 00:00:04.759 --> 00:00:08.119 tiny wet cells just shouting at each other in 00:00:08.119 --> 00:00:10.759 the dark. Yeah, it's a completely chaotic electrochemical 00:00:10.759 --> 00:00:14.160 storm. Right. And somehow that exact same biological 00:00:14.160 --> 00:00:16.100 chaos, I mean, that foundational architecture 00:00:16.100 --> 00:00:18.960 of the human mind is currently being used to 00:00:18.960 --> 00:00:22.420 like drive cars and beat chess grandmasters and 00:00:22.420 --> 00:00:25.399 write computer code. It is arguably the most 00:00:25.399 --> 00:00:28.239 profound crossover between biology and computer 00:00:28.239 --> 00:00:30.309 science in human history. We basically looked 00:00:30.309 --> 00:00:32.530 inside ourselves, figured out the wiring, and 00:00:32.530 --> 00:00:35.130 then decided to just recreate it in silicon. 00:00:35.570 --> 00:00:38.289 Which brings us to today's custom tailored deep 00:00:38.289 --> 00:00:41.090 dive designed specifically for you. We're working 00:00:41.090 --> 00:00:43.729 from a really comprehensive Wikipedia article 00:00:43.729 --> 00:00:45.850 on neural network. It's a massive topic. Oh, 00:00:45.950 --> 00:00:49.000 massive. And our mission today is simple. We 00:00:49.000 --> 00:00:51.420 want to cut through the heavy, intimidating jargon 00:00:51.420 --> 00:00:54.140 of artificial intelligence and discover the fascinating 00:00:54.140 --> 00:00:57.259 aha moments that link the swishy biology of the 00:00:57.259 --> 00:01:00.140 human brain to the cutting edge AI shaping our 00:01:00.140 --> 00:01:02.280 world today. Because the reality is the mechanics 00:01:02.280 --> 00:01:04.480 behind all this aren't magic. No, not at all. 00:01:04.719 --> 00:01:07.700 They're logical. They're traceable. And once 00:01:07.700 --> 00:01:10.680 you understand the underlying mechanism, the 00:01:10.680 --> 00:01:12.859 whole landscape of modern technology just makes 00:01:12.859 --> 00:01:15.230 a lot more sense. OK, let's unpack this. Because 00:01:15.230 --> 00:01:17.090 we hear the term neural network thrown around 00:01:17.090 --> 00:01:19.530 constantly, right? It's the ultimate tech buzzword. 00:01:19.950 --> 00:01:23.569 But what are we actually talking about at a foundational 00:01:23.569 --> 00:01:28.170 level? So at its absolute most basic, a neural 00:01:28.170 --> 00:01:31.129 network is really just a group of interconnected 00:01:31.129 --> 00:01:34.170 units. We call them neurons. And they just send 00:01:34.170 --> 00:01:36.349 signals to one another. That's it. That's it. 00:01:36.450 --> 00:01:39.640 That is the foundational concept. Now, Those 00:01:39.640 --> 00:01:41.840 units can be actual biological cells, like the 00:01:41.840 --> 00:01:44.239 ones in your brain right now, or they can be 00:01:44.239 --> 00:01:46.819 purely mathematical models sitting inside a computer 00:01:46.819 --> 00:01:49.489 program. But the architecture is exactly the 00:01:49.489 --> 00:01:51.390 same. You have units, connections, and signals. 00:01:51.430 --> 00:01:53.590 And there's a massive paradox right at the center 00:01:53.590 --> 00:01:56.189 of this that really stood out in the source material. 00:01:56.629 --> 00:01:59.930 Because an individual neuron on its own is incredibly 00:01:59.930 --> 00:02:02.030 simple. Well, very simple. It basically just 00:02:02.030 --> 00:02:04.469 takes in a signal, reaches a threshold, and passes 00:02:04.469 --> 00:02:06.730 a signal along. I mean, it doesn't possess intelligence. 00:02:07.250 --> 00:02:09.310 But you put a bunch of them together in a network. 00:02:09.490 --> 00:02:11.189 And that's the power of immersions right there. 00:02:11.449 --> 00:02:13.870 A single neuron is just a biological switch. 00:02:14.189 --> 00:02:16.909 But when you link millions of these switches 00:02:16.909 --> 00:02:20.449 together, they can perform astoundingly complex 00:02:20.449 --> 00:02:23.949 tasks. It's crazy. But to really grasp how the 00:02:23.949 --> 00:02:26.610 artificial versions pull this off, we have to 00:02:26.610 --> 00:02:28.189 understand the original blueprint. We have to 00:02:28.189 --> 00:02:30.729 look at the biology first. Right, because before 00:02:30.729 --> 00:02:32.590 neural networks lived in computers, they lived 00:02:32.590 --> 00:02:35.430 inside us. And they still do. A biological neural 00:02:35.430 --> 00:02:38.860 network is a very real physical structure. It's 00:02:38.860 --> 00:02:41.520 a population of biological nerve cells, the neurons, 00:02:42.120 --> 00:02:43.900 chemically connected to each other by structures 00:02:43.900 --> 00:02:46.719 called synapses. And the sheer density of this 00:02:46.719 --> 00:02:49.479 web, it's just hard to overstate. Yeah, a single 00:02:49.479 --> 00:02:51.819 neuron in your brain might be connected to hundreds 00:02:51.819 --> 00:02:54.180 of thousands of other neurons through these synapses. 00:02:54.300 --> 00:02:58.009 It's a phenomenally dense, tangled web. And across 00:02:58.009 --> 00:03:01.210 that web, they are constantly firing electrochemical 00:03:01.210 --> 00:03:04.449 signals, which the source calls action potentials, 00:03:04.550 --> 00:03:07.349 to their connected neighbors. Right. But a common 00:03:07.349 --> 00:03:09.449 misconception is that this is just a blind chain 00:03:09.449 --> 00:03:12.310 reaction, like a row of dominoes falling over. 00:03:12.389 --> 00:03:14.810 Like one hits the next, hits the next. Exactly. 00:03:14.870 --> 00:03:17.550 But it's far more dynamic than that. A biological 00:03:17.550 --> 00:03:20.669 neuron generally serves one of two primary roles 00:03:20.669 --> 00:03:23.590 when it sends a signal. It's either excitatory 00:03:23.590 --> 00:03:26.759 or it's inhibitory. Wait, if an individual neuron 00:03:26.759 --> 00:03:29.539 is simple, how do we actually get complex thoughts 00:03:29.539 --> 00:03:33.319 from that? Is it like a stadium wave? A stadium 00:03:33.319 --> 00:03:36.020 wave? Yeah, or like a massive game of telephone 00:03:36.020 --> 00:03:38.020 where some people are shouting to keep the message 00:03:38.020 --> 00:03:40.800 going and others are trying to quiet it down? 00:03:41.139 --> 00:03:43.460 Actually, that analogy works perfectly with the 00:03:43.460 --> 00:03:45.919 source material. It's exactly like that. The 00:03:45.919 --> 00:03:48.639 excitatory role is the people shouting, right? 00:03:48.639 --> 00:03:50.699 Yeah. Amplifying and propagating the signals. 00:03:51.120 --> 00:03:53.500 They push the receiving neuron closer to its 00:03:53.500 --> 00:03:55.900 firing threshold. OK, so the inhibitory role. 00:03:55.960 --> 00:03:57.639 Those are the people trying to quiet it down. 00:03:57.979 --> 00:04:00.500 They actively suppress the receiving neuron, 00:04:00.919 --> 00:04:03.300 making it less likely to fire. So the receiving 00:04:03.300 --> 00:04:05.599 neuron is just sitting there tallying up all 00:04:05.599 --> 00:04:07.159 the shouting and all the shushing it's getting 00:04:07.159 --> 00:04:10.219 from its thousands of neighbors. Right. And all 00:04:10.219 --> 00:04:13.469 complex thought. All memory, all brain activity 00:04:13.469 --> 00:04:16.170 is simply the result of this constant balancing 00:04:16.170 --> 00:04:19.649 act between excitatory and inhibitory signals 00:04:19.649 --> 00:04:22.730 across massive populations of cells. And it scales 00:04:22.730 --> 00:04:25.009 up too. The source mentions you start with small 00:04:25.009 --> 00:04:27.970 groups called neural circuits. Yes, very small 00:04:27.970 --> 00:04:29.870 local connections. And then you link those up 00:04:29.870 --> 00:04:32.670 into large scale brain networks. And eventually 00:04:32.670 --> 00:04:34.889 those networks have to actually interact with 00:04:34.889 --> 00:04:37.689 the physical world, right? They do. The signals 00:04:37.689 --> 00:04:40.889 generated in the brain travel out through the 00:04:40.889 --> 00:04:43.410 nervous system, eventually crossing what are 00:04:43.410 --> 00:04:46.189 called neuromuscular junctions. Neuromuscular 00:04:46.189 --> 00:04:47.889 junctions. Basically where the nerve meets the 00:04:47.889 --> 00:04:50.430 muscle. The final signal tells the muscle cells 00:04:50.430 --> 00:04:53.139 to contract, resulting in physical motion. So 00:04:53.139 --> 00:04:55.339 for you listening right now, just think about 00:04:55.339 --> 00:04:58.060 that. Every single time you've ever moved a muscle 00:04:58.060 --> 00:05:01.100 or taken a breath or formed a thought, this exact 00:05:01.100 --> 00:05:03.920 microscopic chemical communication network is 00:05:03.920 --> 00:05:07.120 what made it happen. Millions of excitatory and 00:05:07.120 --> 00:05:09.579 inhibitory signals tallied up just so you could 00:05:09.579 --> 00:05:11.660 reach out and grab your phone. It's the biological 00:05:11.660 --> 00:05:14.410 engine of everything you do. It is completely 00:05:14.410 --> 00:05:17.529 mind -boggling. But knowing how the human brain 00:05:17.529 --> 00:05:21.050 works brings up a huge question. How do we cross 00:05:21.050 --> 00:05:23.269 the gap? Right, from biology to machines. Yeah, 00:05:23.310 --> 00:05:25.910 from biological biology to mathematical models 00:05:25.910 --> 00:05:28.930 in machines. Because honestly, I always assumed 00:05:28.930 --> 00:05:31.949 neural networks were a 21st century Silicon Valley 00:05:31.949 --> 00:05:34.329 invention. You know, a couple of engineers in 00:05:34.329 --> 00:05:37.550 a garage cracking a new algorithm. That is such 00:05:37.550 --> 00:05:40.350 a common assumption, but the historical bridge 00:05:40.350 --> 00:05:43.089 is much longer. and honestly much older than 00:05:43.089 --> 00:05:45.430 most people realize. We actually have to go all 00:05:45.430 --> 00:05:47.550 the way back to the 19th century. Right, the 00:05:47.550 --> 00:05:52.949 1800s. Yes, the 1800s. In 1873, a thinker named 00:05:52.949 --> 00:05:56.069 Alexander Bain, and then later the psychologist 00:05:56.069 --> 00:05:59.029 William James in 1890, they independently proposed 00:05:59.029 --> 00:06:02.420 the foundational theory. Wow. Yeah. Long before 00:06:02.420 --> 00:06:04.920 we had the tools to map the brain on a microscopic 00:06:04.920 --> 00:06:07.500 level, they theorized that human thought must 00:06:07.500 --> 00:06:09.920 emerge from the complex interactions among massive 00:06:09.920 --> 00:06:12.480 networks of neurons. So the conceptual framework, 00:06:12.620 --> 00:06:14.379 the whole idea that thought is a network effect 00:06:14.379 --> 00:06:16.300 was there before we even really had light bulbs? 00:06:16.639 --> 00:06:19.379 Precisely. And then, if you fast forward to the 00:06:19.379 --> 00:06:23.100 1930s, a movement called connectionism started 00:06:23.100 --> 00:06:26.439 taking hold. Researchers began actively trying 00:06:26.439 --> 00:06:29.860 to use rudimentary artificial networks to model 00:06:29.860 --> 00:06:32.319 biological ones. But there had to be a breakthrough 00:06:32.319 --> 00:06:35.399 on how they actually learn, right? Yes. The monumental 00:06:35.399 --> 00:06:38.079 breakthrough, the moment we figured out the mechanism 00:06:38.079 --> 00:06:41.079 of how a network actually learns over time, came 00:06:41.079 --> 00:06:44.540 in 1949 from psychologist named Donald Hebb. 00:06:45.040 --> 00:06:47.180 Ah, Hebbian learning. I saw this in the notes. 00:06:47.560 --> 00:06:50.420 Break down the actual mechanism for us. How does 00:06:50.420 --> 00:06:53.459 a web of cells learn a new trick? So Hebbian 00:06:53.459 --> 00:06:55.779 learning operates on a surprisingly intuitive 00:06:55.779 --> 00:06:59.449 idea. It's often summarized as Cells that fire 00:06:59.449 --> 00:07:01.350 together, wire together. Fire together, wire 00:07:01.350 --> 00:07:04.149 together, catchy. Right. Hebb realized that neural 00:07:04.149 --> 00:07:07.009 networks learn over time by physically strengthening 00:07:07.009 --> 00:07:09.870 a synapse, the connection point, every time a 00:07:09.870 --> 00:07:11.990 signal successfully travels along it. So it's 00:07:11.990 --> 00:07:14.610 like walking through a dense field of tall grass. 00:07:15.209 --> 00:07:16.850 The first time you walk through, it's really 00:07:16.850 --> 00:07:18.569 difficult. There's a lot of resistance. But if 00:07:18.569 --> 00:07:21.089 you walk that exact same path every single day, 00:07:21.430 --> 00:07:23.639 you eventually trample the grass down. You pack 00:07:23.639 --> 00:07:26.180 the dirt, it becomes a clear trail, and the next 00:07:26.180 --> 00:07:28.959 time you need to cross, that path offers the 00:07:28.959 --> 00:07:31.480 path of least resistance. What's fascinating 00:07:31.480 --> 00:07:34.040 here is how accurately that physical metaphor 00:07:34.040 --> 00:07:36.939 translates to the early psychological theories, 00:07:37.519 --> 00:07:40.339 which then directly pave the way for hardware. 00:07:40.980 --> 00:07:43.600 Because if you understand the biological rules 00:07:43.600 --> 00:07:46.639 of learning, you can try to build a machine that 00:07:46.639 --> 00:07:49.500 plays by those same rules. Exactly. And around 00:07:49.500 --> 00:07:52.439 the same time, we were getting proof of how complex 00:07:52.439 --> 00:07:57.040 biological processing could get. In 1956, Sveijetingen 00:07:57.040 --> 00:07:59.339 discovered the function of second -order retinal 00:07:59.339 --> 00:08:01.540 cells. Okay, second -order retinal cells. What 00:08:01.540 --> 00:08:03.860 does that mean for the network? Basically, he 00:08:03.860 --> 00:08:06.379 proved that biological networks process information 00:08:06.379 --> 00:08:09.220 in distinct layers. The eye doesn't just send 00:08:09.220 --> 00:08:11.759 raw light straight to the brain. There are intermediate 00:08:11.759 --> 00:08:14.279 layers of cells that process and refine the image 00:08:14.279 --> 00:08:16.579 before passing it along. Layered processing. 00:08:16.839 --> 00:08:18.240 Okay, keep a pin in that because that becomes 00:08:18.240 --> 00:08:20.720 huge for the math later. So the biologists are 00:08:20.720 --> 00:08:22.620 mapping this out. When does the computer science 00:08:22.620 --> 00:08:24.959 actually kick in? That brings us to the first 00:08:24.959 --> 00:08:30.060 machine. In 1943, a neurophysiologist named Warren 00:08:30.060 --> 00:08:33.389 McCulloch and a logician named Walter Pitts teamed 00:08:33.389 --> 00:08:35.990 up to invent the perceptron. Well, the first 00:08:35.990 --> 00:08:38.389 mathematical model of an artificial neural network. 00:08:38.549 --> 00:08:40.129 They didn't build a physical machine yet, right? 00:08:40.129 --> 00:08:42.590 No, just a mathematical proof showing how a simplified 00:08:42.590 --> 00:08:44.789 artificial neuron could act as a logic gate. 00:08:44.870 --> 00:08:49.230 OK. But then in 1957, Frank Rosenblatt took that 00:08:49.230 --> 00:08:51.990 math and actually built a physical perceptron 00:08:51.990 --> 00:08:54.529 in hardware. He actually wired up a physical 00:08:54.529 --> 00:08:57.559 learning machine in the 50s. He did? using motors 00:08:57.559 --> 00:09:00.259 and photocells. It was the first true translation 00:09:00.259 --> 00:09:02.779 from biology to artificial hardware. That is 00:09:02.779 --> 00:09:05.240 an incredible bridge. We go from 19th century 00:09:05.240 --> 00:09:07.899 philosophy to biological observation to mathematical 00:09:07.899 --> 00:09:12.059 proofs to a physical machine in 1957. But obviously 00:09:12.059 --> 00:09:14.679 those early physical machines evolved, right? 00:09:14.720 --> 00:09:17.980 We moved entirely into software. We did. Today, 00:09:18.379 --> 00:09:20.539 artificial neural networks are almost universally 00:09:20.539 --> 00:09:23.340 implemented in software to approximate what we 00:09:23.340 --> 00:09:25.820 call nonlinear functions. Here's where it gets 00:09:25.820 --> 00:09:28.279 really interesting, because I want to visualize 00:09:28.279 --> 00:09:31.759 this modern architecture. If the brain is a tangled 00:09:31.759 --> 00:09:34.919 web of cells, the artificial version sounds much 00:09:34.919 --> 00:09:37.940 more structured, almost like a strict corporate 00:09:37.940 --> 00:09:40.700 hierarchy. That is a very helpful way to frame 00:09:40.700 --> 00:09:43.740 it. Let's build that out. In an artificial network, 00:09:44.299 --> 00:09:47.700 the mathematical neurons are arranged into very 00:09:47.700 --> 00:09:51.789 specific sequential layers. Right. So at the 00:09:51.789 --> 00:09:54.049 very front, you have the input layer. These are 00:09:54.049 --> 00:09:55.929 the frontline workers. They just take in the 00:09:55.929 --> 00:09:58.850 raw data, like the pixels of a photograph, and 00:09:58.850 --> 00:10:01.690 pass it up the chain. Exactly. Then that data 00:10:01.690 --> 00:10:04.029 moves into the hidden layers. This is your middle 00:10:04.029 --> 00:10:06.509 management. The hidden layers. Right. And finally, 00:10:06.509 --> 00:10:08.269 at the very end of the line, you have the output 00:10:08.269 --> 00:10:10.710 layer. The executives who take all the process 00:10:10.710 --> 00:10:12.929 reports from middle management and make the final 00:10:12.929 --> 00:10:15.289 decision. Like, this is a picture of a cat. OK. 00:10:15.350 --> 00:10:18.490 So information flows sequentially. Input, hidden, 00:10:18.700 --> 00:10:21.679 Output. But let's look at the actual math happening 00:10:21.679 --> 00:10:24.200 inside one of those hidden layer neurons. It's 00:10:24.200 --> 00:10:26.360 not a chemical pulse anymore. It's just numbers. 00:10:26.759 --> 00:10:29.940 Yes. The signal input to each neuron is what 00:10:29.940 --> 00:10:32.740 we call a linear combination of the outputs from 00:10:32.740 --> 00:10:35.039 the previous layer. Okay, stop right there. A 00:10:35.039 --> 00:10:37.500 linear combination. That sounds incredibly dense. 00:10:37.659 --> 00:10:41.379 It sounds worse than it is. Basically every connection 00:10:41.379 --> 00:10:44.059 between two artificial neurons has a specific 00:10:44.059 --> 00:10:46.879 weight assigned to it. Just a number acting as 00:10:46.879 --> 00:10:50.059 a multiplier. The receiving neuron takes every 00:10:50.059 --> 00:10:52.639 incoming number, multiplies it by its specific 00:10:52.639 --> 00:10:54.779 connection weight, and adds them all together. 00:10:55.059 --> 00:10:57.500 Okay, so it's mathematically tallying the votes, 00:10:57.580 --> 00:10:59.899 just like the excitatory and inhibitory signals 00:10:59.899 --> 00:11:02.820 in biology. Exactly. And once it has that total 00:11:02.820 --> 00:11:05.980 sum, it calculates the final output via an activation 00:11:05.980 --> 00:11:08.320 function. It's just a rule book that says, based 00:11:08.320 --> 00:11:10.480 on this sum, here's the exact number I'm passing 00:11:10.480 --> 00:11:12.779 to the next layer. OK, so we have layers of neurons 00:11:12.779 --> 00:11:15.059 multiplying inputs by weights, summing them up, 00:11:15.080 --> 00:11:16.919 and passing them through activation functions. 00:11:17.460 --> 00:11:19.460 But how does this corporate hierarchy actually 00:11:19.460 --> 00:11:22.799 learn anything? The source uses terms like empirical 00:11:22.799 --> 00:11:25.820 risk minimization and backpropagation to modify 00:11:25.820 --> 00:11:29.159 those connection weights during training. That 00:11:29.159 --> 00:11:30.960 sounds... I mean, can we just think of this as 00:11:30.960 --> 00:11:33.360 high -tech trial and error? High -tech trial 00:11:33.360 --> 00:11:35.759 and error is exactly what it is. If we connect 00:11:35.759 --> 00:11:38.399 this to the bigger picture, adjusting these weights 00:11:38.399 --> 00:11:41.679 to fit a pre -existing data set is exactly how 00:11:41.679 --> 00:11:44.460 the network learns. So if it guesses wrong... 00:11:44.220 --> 00:11:47.759 It figures out how wrong it was and sends a signal 00:11:47.759 --> 00:11:51.019 backwards through the network. Back propagation. 00:11:51.460 --> 00:11:54.059 Right. Adjusting every single weight slightly 00:11:54.059 --> 00:11:56.399 so it makes a better guess next time. It repeats 00:11:56.399 --> 00:11:59.159 this millions of times until the empirical risk, 00:11:59.379 --> 00:12:02.009 the error. is minimized. And that's what training 00:12:02.009 --> 00:12:04.490 in AI is. That's the entirety of it. And by the 00:12:04.490 --> 00:12:07.009 way, officially, a deep neural network is just 00:12:07.009 --> 00:12:09.370 a network with more than three layers. So an 00:12:09.370 --> 00:12:11.929 input layer, an output layer, at least two hidden 00:12:11.929 --> 00:12:14.289 layers. That's what deep learning means. Just 00:12:14.289 --> 00:12:16.750 more than two hidden layers. That's it. Wow. 00:12:16.950 --> 00:12:19.049 OK, so now that we understand both the biological 00:12:19.049 --> 00:12:21.649 and the artificial, there's a glaring point of 00:12:21.649 --> 00:12:23.809 divergence that we have to talk about. Because 00:12:23.809 --> 00:12:25.830 if artificial neural networks were originally 00:12:25.830 --> 00:12:29.059 designed to mimic our biology, Why does the source 00:12:29.059 --> 00:12:31.080 say they are increasingly different from their 00:12:31.080 --> 00:12:33.379 biological counterparts today? It comes down 00:12:33.379 --> 00:12:36.399 to efficiency. While biology was the initial 00:12:36.399 --> 00:12:39.480 inspiration, machine learning applications required 00:12:39.480 --> 00:12:42.779 different approaches. The goal shifted from perfectly 00:12:42.779 --> 00:12:45.820 modeling a human brain to just solving specific 00:12:45.820 --> 00:12:48.299 artificial intelligence problems. Like, biological 00:12:48.299 --> 00:12:50.580 brains are amazing, but they have constraints, 00:12:50.779 --> 00:12:53.360 right? Exactly. Human brains run on very little 00:12:53.360 --> 00:12:55.480 power and are limited by chemical diffusion. 00:12:56.140 --> 00:12:58.580 Computers don't have those limits. So engineers 00:12:58.580 --> 00:13:01.139 stop trying to build a perfectly accurate biological 00:13:01.139 --> 00:13:04.539 replica and just optimize the math for data processing. 00:13:04.879 --> 00:13:06.700 And it's a good thing they did because that shift 00:13:06.700 --> 00:13:09.539 is what made this technology so incredibly useful 00:13:09.539 --> 00:13:12.679 in our daily lives. The real -world applications 00:13:12.679 --> 00:13:15.460 mentioned in the source are everywhere. Oh absolutely 00:13:15.460 --> 00:13:17.659 everywhere. Artificial neural networks are the 00:13:17.659 --> 00:13:19.860 backbone of modern predictive modeling. They 00:13:19.860 --> 00:13:22.419 power adaptive control systems, like in manufacturing 00:13:22.419 --> 00:13:25.399 robots or autonomous vehicles. And pattern recognition, 00:13:25.700 --> 00:13:27.620 right? Like facial recognition to unlock your 00:13:27.620 --> 00:13:30.120 phone, or handwriting recognition. Yes. There 00:13:30.120 --> 00:13:32.320 are also the engines behind general game playing 00:13:32.320 --> 00:13:35.279 AIs that beat human champions. And of course, 00:13:35.600 --> 00:13:38.179 generative AI. Right. The big one right now. 00:13:38.620 --> 00:13:40.759 So what does this all mean for you, the listener? 00:13:41.100 --> 00:13:44.460 Well, it means that when you use a generative 00:13:44.460 --> 00:13:47.840 AI to write an email, or the facial recognition 00:13:47.840 --> 00:13:50.440 to unlock your phone, you are interacting with 00:13:50.440 --> 00:13:53.840 the direct descendant of an 1873 theory about 00:13:53.840 --> 00:13:56.279 human thought, filtered through complex math 00:13:56.279 --> 00:13:58.600 layers. It's an incredible evolution. It really 00:13:58.600 --> 00:14:01.340 is. So just to recap our journey today, we started 00:14:01.340 --> 00:14:04.659 with the biological synapses and excitatory signals 00:14:04.659 --> 00:14:06.980 happening in our heads right now. Then we crossed 00:14:06.980 --> 00:14:09.340 the historical bridge, looking at Hebbian learning 00:14:09.340 --> 00:14:12.320 and the perceptron. And finally, we dove into 00:14:12.320 --> 00:14:15.379 the hidden layers of modern software -based deep 00:14:15.379 --> 00:14:17.399 neural networks. And this raises an important 00:14:17.399 --> 00:14:19.399 question, I think. Yeah. I want to leave you 00:14:19.399 --> 00:14:21.740 with a final provocative thought to mull over. 00:14:22.159 --> 00:14:24.120 The source material notes that artificial neural 00:14:24.120 --> 00:14:26.620 networks started by copying biology, but eventually 00:14:26.620 --> 00:14:28.580 diverged to become better at machine learning. 00:14:29.139 --> 00:14:31.419 If artificial intelligence continues to evolve 00:14:31.419 --> 00:14:34.059 and tackle even more complex human -like tasks 00:14:34.059 --> 00:14:36.759 in the future, will we need to start making these 00:14:36.759 --> 00:14:39.120 mathematical networks look more like biological 00:14:39.120 --> 00:14:41.259 brains again? Right, with all that messy chemical 00:14:41.259 --> 00:14:44.279 complexity. Exactly. Or will they just evolve 00:14:44.279 --> 00:14:47.340 into a completely alien structure that we can 00:14:47.340 --> 00:14:50.179 no longer even comprehend? It's a fascinating 00:14:50.179 --> 00:14:52.360 thing to consider as the technology advances. 00:14:52.580 --> 00:14:54.700 It really is. Well, thank you so much for sharing 00:14:54.700 --> 00:14:56.759 your sources and coming along on this deep dive 00:14:56.759 --> 00:14:59.139 with us. Keep learning and stay curious.