1 00:00:00,160 --> 00:00:02,919 Speaker 1: Welcome to Thrilling Threads, where we pull in the most 2 00:00:02,960 --> 00:00:06,559 fascinating lines of research and well we try to figure 3 00:00:06,559 --> 00:00:09,199 out what they mean for the future, for your future. 4 00:00:09,320 --> 00:00:15,679 Speaker 2: Specifically, today, we are undertaking a really massive exploration. We're 5 00:00:15,720 --> 00:00:19,399 looking at a concept that is quickly and I mean 6 00:00:19,519 --> 00:00:21,280 very quickly, shedding its science. 7 00:00:20,960 --> 00:00:22,960 Speaker 1: Fiction show and becoming reality. 8 00:00:22,640 --> 00:00:25,679 Speaker 2: Becoming tangible reality. We're talking about the immediate future of 9 00:00:25,760 --> 00:00:27,320 human augmentation. 10 00:00:27,000 --> 00:00:28,359 Speaker 1: And not in the way you might think. 11 00:00:28,839 --> 00:00:33,119 Speaker 2: No, exactly. We're focusing on systems that fundamentally change human 12 00:00:33,159 --> 00:00:36,920 intellectual and physical capability. But this isn't achieved through some 13 00:00:37,000 --> 00:00:41,039 kind of magical biological engineering. It's all about partnership. 14 00:00:40,439 --> 00:00:44,119 Speaker 1: Partnership with artificial intelligence. And I think that partnership idea 15 00:00:44,280 --> 00:00:46,719 is the critical core here. It is because when we 16 00:00:46,759 --> 00:00:49,840 say superhuman, I mean our minds immediately jump to comic books, 17 00:00:49,880 --> 00:00:53,119 right flying super strengths, of course, But what if superhuman 18 00:00:53,200 --> 00:00:57,640 is actually about about instantly processing an impossible amount of 19 00:00:57,679 --> 00:01:01,240 data or communicating by thinking. 20 00:01:01,320 --> 00:01:04,760 Speaker 2: Or creating materials, they can fundamentally reshape our entire global 21 00:01:04,799 --> 00:01:06,640 infrastructure exactly. 22 00:01:06,680 --> 00:01:08,359 Speaker 1: And that's what we've got in front of us. We've 23 00:01:08,359 --> 00:01:12,519 gathered a well substantial stack of sources detailing fifteen new 24 00:01:12,719 --> 00:01:16,760 AI projects that are driving this intellectual and physical leap 25 00:01:17,280 --> 00:01:17,799 right now. 26 00:01:17,840 --> 00:01:20,640 Speaker 2: And our mission really in this deep dive is to 27 00:01:20,799 --> 00:01:23,560 synthesize and structure all of this data. We want to 28 00:01:23,599 --> 00:01:28,400 move past the headlines and understand the underlying mechanisms of 29 00:01:28,439 --> 00:01:31,280 this augmentation, how it actually works, how it works, and 30 00:01:31,319 --> 00:01:35,159 the sources all confirmed this really profound shift. AI is 31 00:01:35,200 --> 00:01:37,760 no longer just a powerful tool. It's moving from being 32 00:01:38,439 --> 00:01:41,000 a basic calculator or a database. 33 00:01:40,560 --> 00:01:41,879 Speaker 1: A glorified search engine. 34 00:01:41,959 --> 00:01:46,599 Speaker 2: Yes, and it's evolving into a true cognitive and physical copilot. 35 00:01:46,760 --> 00:01:49,480 Speaker 1: Okay, let's unpack that term copilot, because I think before 36 00:01:49,519 --> 00:01:51,519 we can even jump into the projects, we have to 37 00:01:51,599 --> 00:01:55,159 draw a really clear line between automation and augmentation. They're 38 00:01:55,200 --> 00:01:56,159 not the same thing. 39 00:01:56,120 --> 00:01:58,560 Speaker 2: Not at all. Automation is the AI doing the job 40 00:01:58,680 --> 00:02:01,760 entirely it replaces the person, and augmentation is the AI 41 00:02:01,879 --> 00:02:03,079 making the human better at. 42 00:02:03,040 --> 00:02:04,680 Speaker 1: The job, sharper, faster. 43 00:02:04,920 --> 00:02:07,599 Speaker 2: Precisely, if you look at the historical fear of AI, 44 00:02:08,240 --> 00:02:11,719 it was always focused on replacement. The robots are coming 45 00:02:11,759 --> 00:02:12,840 for our jobs. 46 00:02:12,919 --> 00:02:16,400 Speaker 1: Right, but for the projects we are examining today, the 47 00:02:16,479 --> 00:02:20,719 human remains the indispensable element, the conductor who sets the 48 00:02:20,759 --> 00:02:22,680 pace and chooses the direction. 49 00:02:23,360 --> 00:02:27,960 Speaker 2: So the AI in this case is managing the complexity, 50 00:02:28,199 --> 00:02:29,599 the sheer volume of it all. 51 00:02:29,800 --> 00:02:32,800 Speaker 1: Yes, the cognitive tasks that normally bogg us down. 52 00:02:32,960 --> 00:02:35,919 Speaker 2: That's the bottleneck, isn't it. It's always been the bottleneck. 53 00:02:36,199 --> 00:02:41,120 The human brain's capacity for working memory, for comparison, it's finite. 54 00:02:41,240 --> 00:02:42,159 Speaker 1: It's very finite. 55 00:02:42,199 --> 00:02:45,800 Speaker 2: If you're a doctor trying to compare fifty documents, twenty 56 00:02:45,919 --> 00:02:50,080 lab reports, and three different scans, the vast majority of 57 00:02:50,080 --> 00:02:53,199 your mental energy is just spent managing that information load, 58 00:02:53,280 --> 00:02:55,000 not making the diagnosis exactly. 59 00:02:55,080 --> 00:02:59,400 Speaker 1: This augmentation specializes in handling that extreme analysis, the comparison, 60 00:02:59,479 --> 00:03:02,360 the pattern spotting, the memory load, all this stuff that 61 00:03:02,400 --> 00:03:05,039 typically consumes our attention, our time, and our focus. 62 00:03:05,120 --> 00:03:07,759 Speaker 2: It's like it clears the administrative burden of intelligence. 63 00:03:07,879 --> 00:03:10,479 Speaker 1: I love that phrasing. It clears the cognitive backlog, and 64 00:03:10,520 --> 00:03:12,400 by doing that, it frees up the human mind for 65 00:03:12,439 --> 00:03:13,599 what it actually does best. 66 00:03:13,639 --> 00:03:18,080 Speaker 2: Critical judgment, ethical reasoning, novel creativity, the things that aren't 67 00:03:18,080 --> 00:03:19,280 about processing power. 68 00:03:19,479 --> 00:03:21,680 Speaker 1: I think that's a perfect frame for this. So we're 69 00:03:21,680 --> 00:03:23,840 going to show you exactly how this is happening across 70 00:03:23,840 --> 00:03:25,560 what three major theaters? 71 00:03:25,599 --> 00:03:29,759 Speaker 2: That's right, accelerating discovery, augmenting the mind, and finally rebuilding 72 00:03:29,759 --> 00:03:30,199 the body. 73 00:03:30,240 --> 00:03:33,039 Speaker 1: Okay, let's jump right in. Where do we start. 74 00:03:33,120 --> 00:03:36,280 Speaker 2: Let's start with discovery. Our first major theme is how 75 00:03:36,319 --> 00:03:42,639 AI is just fundamentally revolutionizing scientific timelines, crushing them completely. 76 00:03:43,199 --> 00:03:46,599 It's turning what used to be years of laborious manual 77 00:03:46,639 --> 00:03:50,560 experimentation into in some cases, mere hours or days. 78 00:03:50,599 --> 00:03:53,319 Speaker 1: And this has achieved. How it's not magic. 79 00:03:53,039 --> 00:03:56,919 Speaker 2: No, it's primarily through the AI's capability for well two things, 80 00:03:57,639 --> 00:03:59,960 unparalleled scale and just sheer persistence. 81 00:04:00,120 --> 00:04:02,120 Speaker 1: Okay, when you say scale, the first thing that jumps 82 00:04:02,159 --> 00:04:04,159 to my mind, it's all over our sources is Project 83 00:04:04,159 --> 00:04:06,159 fifteen Deep minds. Alpha fold. 84 00:04:06,280 --> 00:04:08,919 Speaker 2: Alpha fold is the poster child for this revolution. If 85 00:04:08,919 --> 00:04:11,759 you follow biology or medicine even slightly, you heard it. 86 00:04:12,080 --> 00:04:14,680 This was a genuine seismic shift, not just. 87 00:04:14,680 --> 00:04:17,120 Speaker 1: An improvement, a complete change in the game. 88 00:04:17,399 --> 00:04:20,800 Speaker 2: It was a revolution, not an evolution. So to understand why, 89 00:04:21,399 --> 00:04:23,240 you have to understand protein. 90 00:04:22,800 --> 00:04:24,439 Speaker 1: Folding right, the problem itself. 91 00:04:24,560 --> 00:04:27,199 Speaker 2: So, a protein is a chain of amino acids, and 92 00:04:27,240 --> 00:04:29,959 how that chain arranges itself into a unique three d 93 00:04:30,160 --> 00:04:33,639 shape is well, it's central to all of biology. 94 00:04:33,720 --> 00:04:36,199 Speaker 1: That shape dictates its function, right, whether it fights a 95 00:04:36,240 --> 00:04:39,439 disease or you know, digest your food exactly. 96 00:04:39,680 --> 00:04:43,480 Speaker 2: And before alpha fold solving just one complex structure, it 97 00:04:43,519 --> 00:04:48,120 could require years years of expensive, painstaking lab. 98 00:04:47,920 --> 00:04:50,959 Speaker 1: Work using methods like X ray crystallography, things that are 99 00:04:50,959 --> 00:04:52,519 incredibly difficult and slow. 100 00:04:53,120 --> 00:04:55,959 Speaker 2: I've heard scientists describe the process as being stuck in 101 00:04:56,040 --> 00:04:59,480 a just a massive, impossible maze. 102 00:04:59,279 --> 00:05:02,360 Speaker 1: Where everyw turn costs you six months and ten thousand 103 00:05:02,360 --> 00:05:03,319 dollars at least. 104 00:05:04,000 --> 00:05:07,959 Speaker 2: The difficulty is something called the combinatorial explosion. A protein 105 00:05:08,000 --> 00:05:11,199 with just one hundred amino acids has a staggering number 106 00:05:11,240 --> 00:05:13,000 of possible ways it can fold. 107 00:05:12,879 --> 00:05:14,839 Speaker 1: More than the number of atoms in the universe. 108 00:05:14,879 --> 00:05:17,240 Speaker 2: I've read it's a number that's functionally infinite. You can't 109 00:05:17,279 --> 00:05:20,000 just guess. And this is where the AI provided the 110 00:05:20,000 --> 00:05:20,720 necessary leap. 111 00:05:20,920 --> 00:05:24,399 Speaker 1: It didn't do better chemistry, though, that's the key right correct. 112 00:05:24,759 --> 00:05:28,319 Speaker 2: Alpha fold used a deep learning architecture. It basically modeled 113 00:05:28,360 --> 00:05:31,480 the spatial relationships between all the amino acids at once. 114 00:05:32,079 --> 00:05:36,639 It turned an impossible structural puzzle into a manageable geometry problem. 115 00:05:36,879 --> 00:05:39,399 It's about pattern recognition at a massive scale. 116 00:05:39,480 --> 00:05:43,079 Speaker 1: So the augmentation came through speed and I guess through 117 00:05:43,120 --> 00:05:44,079 put exactly. 118 00:05:44,519 --> 00:05:47,560 Speaker 2: Now a lab can get a highly accurate prediction of 119 00:05:47,600 --> 00:05:51,319 a protein structure in hours. But the scale is what's 120 00:05:51,439 --> 00:05:55,199 really mind blowing. By twenty twenty two, alpha fold had 121 00:05:55,240 --> 00:05:59,000 released predictions for over two hundred million protein structures. 122 00:05:59,040 --> 00:05:59,959 Speaker 1: Two hundred million. 123 00:06:00,160 --> 00:06:03,279 Speaker 2: It effectively mapped the entire known protein universe. 124 00:06:03,360 --> 00:06:07,560 Speaker 1: That's an astonishing amount of foundational science just unlocked overnight. 125 00:06:07,680 --> 00:06:11,079 Speaker 2: It cleared the single biggest time sink in modern drug discovery. 126 00:06:11,480 --> 00:06:14,079 The sources note it's been cited in tens of thousands 127 00:06:14,120 --> 00:06:15,959 of subsequent research papers. 128 00:06:15,879 --> 00:06:19,439 Speaker 1: So it gave human scientists immediate access to structural data 129 00:06:19,480 --> 00:06:22,680 that what would have taken generations to acquire manually easily. 130 00:06:22,720 --> 00:06:25,759 Speaker 2: It's a perfect example of AI clearing the foundational backlog 131 00:06:26,040 --> 00:06:28,000 so humans can get to the interesting workfaster. 132 00:06:28,279 --> 00:06:31,600 Speaker 1: And that idea, that relentless capability, it leads right into 133 00:06:31,639 --> 00:06:32,800 the next point, doesn't it. 134 00:06:32,800 --> 00:06:36,319 Speaker 2: It does. It leads us to autonomous research agents. We've 135 00:06:36,360 --> 00:06:37,759 called this speed running science. 136 00:06:37,800 --> 00:06:40,120 Speaker 1: This is a Project twelve, the relentless assistance. 137 00:06:40,399 --> 00:06:44,399 Speaker 2: Yes, if alpha fold was about solving one known very 138 00:06:44,399 --> 00:06:48,439 hard problem very quickly. These agents are about iterating through 139 00:06:48,560 --> 00:06:50,519 unknowns relentlessly. 140 00:06:50,600 --> 00:06:53,560 Speaker 1: So we're talking about AI systems actually active in labs 141 00:06:53,639 --> 00:06:56,800 right now that can come up with a hypothesis. 142 00:06:56,360 --> 00:06:59,079 Speaker 2: Like molecule X should react with catalyst y okay. 143 00:06:59,079 --> 00:07:02,399 Speaker 1: So it generates the hypothesis. Then it automatically runs a simulation, 144 00:07:02,560 --> 00:07:05,800 analyzes the results. And this is the crucial part. It 145 00:07:05,920 --> 00:07:09,399 uses that feedback to refine the next hypothesis. 146 00:07:08,759 --> 00:07:11,800 Speaker 2: And it repeats that cycle thousands, even millions of times 147 00:07:12,160 --> 00:07:13,639 with zero human intervention. 148 00:07:13,920 --> 00:07:17,199 Speaker 1: The power there is just raw endurance, persistence. 149 00:07:17,439 --> 00:07:21,160 Speaker 2: It's not about superhuman brilliance. It's about massive throughput. These 150 00:07:21,160 --> 00:07:24,639 systems can run literally thousands of experiments overnight. That's an 151 00:07:24,639 --> 00:07:28,079 impossible feat for any human team. You're limited by shifts, 152 00:07:28,079 --> 00:07:31,879 by exhaustion, by just the meticulous nature of setting things up. 153 00:07:31,959 --> 00:07:35,000 Speaker 1: And I saw they're being used in fields that are 154 00:07:35,040 --> 00:07:38,600 really heavy on this kind of iteration, right, chemistry, material. 155 00:07:38,199 --> 00:07:43,399 Speaker 2: Science, software engineering too, Yeah, optimizing molecules, debugging code NonStop. 156 00:07:43,519 --> 00:07:46,319 Speaker 1: I saw fascinating case study in the source material about 157 00:07:46,360 --> 00:07:49,759 one of these agents. It was tasked with optimizing a 158 00:07:49,800 --> 00:07:51,439 new kind of battery electrolyte. 159 00:07:51,560 --> 00:07:53,000 Speaker 2: Oh I remember this one. 160 00:07:53,079 --> 00:07:57,319 Speaker 1: It iterated through something like six thousand simulations, and it 161 00:07:57,399 --> 00:08:00,160 found an optimal compound that was forty percent more more 162 00:08:00,199 --> 00:08:02,160 stable than the best human designed one. 163 00:08:02,199 --> 00:08:05,560 Speaker 2: And the reason why is what's so interesting. It succeeded 164 00:08:05,600 --> 00:08:08,680 because the agent wasn't afraid to test combinations that looked 165 00:08:09,040 --> 00:08:13,759 quote ugly or counterintuitive to human experts. It had no bias. 166 00:08:14,000 --> 00:08:17,600 Speaker 1: That's a perfect illustration of augmentation. The human sets the goalpost, 167 00:08:17,639 --> 00:08:21,120 we need a better electrolyte, but the machine uses tireless, 168 00:08:21,240 --> 00:08:24,560 unbiased iteration to explore the entire possibility space. 169 00:08:24,800 --> 00:08:27,199 Speaker 2: It fundamentally changes the speed at which we reach those 170 00:08:27,240 --> 00:08:28,319 scientific outcomes. 171 00:08:28,399 --> 00:08:31,680 Speaker 1: And this idea of exploring vast possibility spaces, it connects 172 00:08:31,680 --> 00:08:35,120 directly to another project right, project seven Yes. 173 00:08:35,000 --> 00:08:37,879 Speaker 2: The use of machine learning to find new therapeutics. The 174 00:08:37,919 --> 00:08:41,159 classic example here is the discovery of the antibiotic candidate 175 00:08:41,399 --> 00:08:43,399 Hellison by researchers at MIT. 176 00:08:43,840 --> 00:08:46,320 Speaker 1: The story of Hallison is powerful because it highlights a 177 00:08:46,360 --> 00:08:49,600 concept you mentioned earlier coverage exactly. 178 00:08:49,919 --> 00:08:54,399 Speaker 2: Traditional drug discovery is often based on existing knowledge. Chemists 179 00:08:54,480 --> 00:08:57,399 look for derivatives of compounds that have already been successful. 180 00:08:57,559 --> 00:08:58,240 It's logical. 181 00:08:58,360 --> 00:09:00,519 Speaker 1: You look where you think the answer should be you 182 00:09:00,600 --> 00:09:01,559 follow the bread crumbs. 183 00:09:01,759 --> 00:09:05,279 Speaker 2: But the chemical space, the total number of possible stable 184 00:09:05,399 --> 00:09:11,080 molecules is unimaginably huge, far beyond our capacity to search manually. 185 00:09:11,399 --> 00:09:13,279 Speaker 1: So the AI doesn't follow the bread. 186 00:09:13,120 --> 00:09:16,039 Speaker 2: Crumbs, it eats the whole loaf. Yeah, the machine learning 187 00:09:16,039 --> 00:09:18,840 model they use screened over one hundred million compounds in a. 188 00:09:18,799 --> 00:09:21,320 Speaker 1: Matter of days, a task that would take a human 189 00:09:21,399 --> 00:09:23,720 team years, decades. 190 00:09:23,360 --> 00:09:26,840 Speaker 2: Years of effort and resources. Absolutely, but the core insight 191 00:09:27,000 --> 00:09:29,840 isn't just the speed. It was the model's ability to 192 00:09:29,919 --> 00:09:34,440 explore compounds that were chemically dissimilar to any existing antibiotics. 193 00:09:34,679 --> 00:09:37,559 Speaker 1: So it didn't just find a better version of penicillin. 194 00:09:37,679 --> 00:09:40,720 It found something humans had completely overlooked because it didn't 195 00:09:40,720 --> 00:09:42,120 fit the established patterns. 196 00:09:42,159 --> 00:09:46,559 Speaker 2: That's augmentation through scope. The AI operates without the inherent 197 00:09:46,639 --> 00:09:50,519 biases or the learned intuition of a human expert. It 198 00:09:50,600 --> 00:09:53,360 can search the entire chemical map, and it often finds 199 00:09:53,399 --> 00:09:57,519 these highly effective discoveries in corners that researchers had, you know, 200 00:09:57,600 --> 00:09:58,720 sort of ignored. 201 00:09:58,480 --> 00:10:01,480 Speaker 1: Because there was no signposts there saying look here, no. 202 00:10:01,600 --> 00:10:02,840 Speaker 2: Prior success markers. 203 00:10:02,919 --> 00:10:03,039 Speaker 1: YEA. 204 00:10:03,320 --> 00:10:07,120 Speaker 2: This ability to provide limitless, unbiased scope. It's one of 205 00:10:07,159 --> 00:10:09,759 the biggest multipliers we have in foundational signs right now. 206 00:10:09,840 --> 00:10:12,039 Speaker 1: Okay, so let's shift gears a bit. Let's talk about 207 00:10:12,080 --> 00:10:16,720 foundational infrastructure breakthroughs. This isn't about finding a specific drug, 208 00:10:17,000 --> 00:10:20,519 but enabling the very tools that future breakthroughs will depend. 209 00:10:20,360 --> 00:10:24,840 Speaker 2: On, and that starts with Project six. Stabilizing fragile systems specifically, 210 00:10:24,879 --> 00:10:27,279 this is absolutely crucial for tuantum error correction. 211 00:10:27,440 --> 00:10:32,200 Speaker 1: Right quantum computing, the promise is ultimate computational power, but 212 00:10:32,240 --> 00:10:35,159 the reality is that they are well. They're notoriously volatile, 213 00:10:35,440 --> 00:10:36,440 incredibly fragile. 214 00:10:36,639 --> 00:10:40,559 Speaker 2: The fundamental units, the kubits. They exist in this delicate 215 00:10:40,639 --> 00:10:47,279 superposition state. But any tiny environmental interaction, a bit of noise, vibration, heat. 216 00:10:47,399 --> 00:10:50,480 Speaker 1: No thing collapses. The process is called decoherence, and it 217 00:10:50,559 --> 00:10:52,360 ruins the calculation almost instantly. 218 00:10:52,600 --> 00:10:55,279 Speaker 2: You can't build the future of chemistry simulations if your 219 00:10:55,320 --> 00:10:58,720 computer keeps crashing every millisecond. This instability has been the 220 00:10:58,759 --> 00:11:02,879 single biggest engineering bottleneck in quantum mechanics for decades. 221 00:11:03,320 --> 00:11:06,159 Speaker 1: So how on earth does AI solve a fundamental physics 222 00:11:06,159 --> 00:11:06,919 problem like that? 223 00:11:07,080 --> 00:11:11,200 Speaker 2: It acts as a sophisticated real time stabilization system. Research 224 00:11:11,240 --> 00:11:14,840 from Google, from IBM, from various university labs. It all 225 00:11:14,919 --> 00:11:17,720 shows that machine learning is now being used to accelerate 226 00:11:17,759 --> 00:11:19,360 something called syndrome decoding. 227 00:11:19,600 --> 00:11:22,159 Speaker 1: Syndrome decoding, what does that mean? In practice? 228 00:11:22,240 --> 00:11:25,240 Speaker 2: It means the AI is constantly monitoring the quibits and 229 00:11:25,279 --> 00:11:29,919 their environment. It's detecting specific types of decoherence, these tiny errors, 230 00:11:30,240 --> 00:11:33,080 and then it's issuing immediate correction signals. And it does 231 00:11:33,080 --> 00:11:36,519 this faster and more accurately than any traditional algorithm could 232 00:11:36,519 --> 00:11:37,240 ever manage. 233 00:11:37,440 --> 00:11:41,799 Speaker 1: So the AI is basically a tireless, hypervigilant technician that 234 00:11:41,879 --> 00:11:44,200 works almost at the speed of light to prevent the 235 00:11:44,279 --> 00:11:45,759 quantum state from collapsing. 236 00:11:46,159 --> 00:11:50,399 Speaker 2: Is a perfect analogy. Yeah, it provides the necessary infrastructural stability. 237 00:11:50,960 --> 00:11:54,879 The AI isn't performing the quantum calculation itself. It's just 238 00:11:55,080 --> 00:11:58,360 making sure that calculation can run long enough and accurately 239 00:11:58,480 --> 00:12:00,000 enough to actually be useful. 240 00:12:00,200 --> 00:12:02,960 Speaker 1: And that's why we're finally starting to see these breakthroughs 241 00:12:02,960 --> 00:12:07,360 in material science and chemistry coming out of quantum simulations. 242 00:12:06,799 --> 00:12:10,159 Speaker 2: Because the AI has provided the persistent, high speed cleanup 243 00:12:10,200 --> 00:12:10,720 crew and. 244 00:12:10,679 --> 00:12:13,960 Speaker 1: Those stabilized computations. They feed directly into our next project, 245 00:12:13,960 --> 00:12:17,720 don't they Project three. AI materials discovery at scale they. 246 00:12:17,600 --> 00:12:21,879 Speaker 2: Do, and this area might sound abstract, but it affects everything. 247 00:12:21,919 --> 00:12:25,080 We rely on, your phone, battery, the global energy grid. 248 00:12:24,919 --> 00:12:28,720 Speaker 1: Absolutely everything. Deep Mind's genome system, for instance, has predicted 249 00:12:28,720 --> 00:12:32,919 the stability of hundreds of thousands of new inorganic materials. 250 00:12:32,399 --> 00:12:36,279 Speaker 2: Which is a huge deal. Traditional material science often follow 251 00:12:36,279 --> 00:12:38,000 what's called an Edisonian. 252 00:12:37,519 --> 00:12:40,320 Speaker 1: Approach, trial and error. Mix some stuff together in the 253 00:12:40,360 --> 00:12:43,000 lab and on a hunch, and see what happens exactly. 254 00:12:43,080 --> 00:12:47,000 Speaker 2: It was incredibly slow, incredibly expensive, and gets heavy. You 255 00:12:47,080 --> 00:12:50,559 synthesize material, try to figure out its crystal structure, test 256 00:12:50,600 --> 00:12:52,600 its conductivity, and maybe. 257 00:12:52,440 --> 00:12:54,799 Speaker 1: Years down the line, find out it wasn't stable enough 258 00:12:54,799 --> 00:12:57,480 for any real world application, a dead end. 259 00:12:57,679 --> 00:13:01,600 Speaker 2: AI has completely inverted that entire manathodology. How so, by 260 00:13:01,639 --> 00:13:05,720 analyzing massive data sets of existing crystal structures, chemical bonds, 261 00:13:05,759 --> 00:13:09,879 and thermodynamic properties, the AI predicts the stability and the 262 00:13:09,919 --> 00:13:12,240 potential function of hypothetical new. 263 00:13:12,120 --> 00:13:14,399 Speaker 1: Materials before they're even made, before. 264 00:13:14,200 --> 00:13:17,559 Speaker 2: A single gram is synthesized in a lab. It turns 265 00:13:17,600 --> 00:13:21,200 material science from a craft into a pure search problem. 266 00:13:20,879 --> 00:13:22,759 Speaker 1: And the sources know that a lot of these predictions 267 00:13:22,799 --> 00:13:26,559 have already been validated experimentally. The real world implications here 268 00:13:26,559 --> 00:13:27,480 are just huge. 269 00:13:27,519 --> 00:13:30,799 Speaker 2: They are faster material prediction means we accelerate the development 270 00:13:30,840 --> 00:13:35,600 of things like high density battery cathodes, novel superconductors, high 271 00:13:35,600 --> 00:13:37,000 efficiency semiconductors. 272 00:13:37,080 --> 00:13:41,240 Speaker 1: This is a quiet foundational transformation. It's changing the fundamental 273 00:13:41,279 --> 00:13:44,519 limits of our electrical infrastructure and our energy storage. 274 00:13:44,600 --> 00:13:47,039 Speaker 2: And finally, within this theme of discovery, we have to 275 00:13:47,039 --> 00:13:51,440 look at the incredibly data heavy challenge of aging project. 276 00:13:51,080 --> 00:13:54,759 Speaker 1: For longevity research, this has always struggled because the biological 277 00:13:54,840 --> 00:13:58,840 data sets are just they're vast, they're high dimensional, and 278 00:13:58,879 --> 00:14:02,559 they're full of complex, often contradictory signals, and the human. 279 00:14:02,360 --> 00:14:06,679 Speaker 2: Life span is so long. The biological changes from cellular 280 00:14:06,679 --> 00:14:10,840 senessence to epigenetic drift are so nuanced that simply collecting 281 00:14:10,840 --> 00:14:13,480 and correlating the data is a massive barrier to entry. 282 00:14:13,600 --> 00:14:16,480 So what changed the ability of AI to handle that 283 00:14:16,559 --> 00:14:20,360 sheer volume and complexity. Labs are now deploying machine learning 284 00:14:20,399 --> 00:14:25,600 to analyze patterns across these massive data sets single cell sequencing, data, proteomics, 285 00:14:25,879 --> 00:14:27,559 all kinds of epigenetic clocks. 286 00:14:27,720 --> 00:14:31,000 Speaker 1: So the augmentation here is the AI's ability to find 287 00:14:31,000 --> 00:14:34,480 these really subtle patterns, patterns that might indicate longevity markers 288 00:14:34,559 --> 00:14:37,799 or specific reprogramming signals that a human eye or even 289 00:14:37,840 --> 00:14:40,879 a traditional statistical analysis would just miss because of all 290 00:14:40,879 --> 00:14:41,320 the noise. 291 00:14:41,440 --> 00:14:45,320 Speaker 2: Yes, but crucially we have to maintain perspective here. There 292 00:14:45,360 --> 00:14:48,919 is no proven human life span extension yet, right. 293 00:14:48,840 --> 00:14:50,120 Speaker 1: We need to be very clear about that. 294 00:14:50,360 --> 00:14:53,279 Speaker 2: But what AI has done is revolutionize the speed of 295 00:14:53,279 --> 00:14:58,200 the research pipeline. The models help scientists decouple complex correlation 296 00:14:58,679 --> 00:15:01,120 from true causation in these longevity markers. 297 00:15:01,200 --> 00:15:05,080 Speaker 1: So hypotheses that would have taken say, five years of animal. 298 00:15:04,759 --> 00:15:08,360 Speaker 2: Studies, they can now be modeled and tested much much earlier. 299 00:15:08,639 --> 00:15:12,080 It allows scientists to pursue these really complex avenues much 300 00:15:12,120 --> 00:15:13,799 faster than they ever could before. 301 00:15:13,919 --> 00:15:17,159 Speaker 1: Okay, so from accelerating discovery in the lab, let's pivot, 302 00:15:17,240 --> 00:15:19,799 let's talk about augmenting the mind itself. This is about 303 00:15:19,799 --> 00:15:21,639 cognitive copilots. 304 00:15:21,120 --> 00:15:24,639 Speaker 2: Exactly, systems that work in parallel with our brains, reducing 305 00:15:24,639 --> 00:15:28,720 our mental load and synthesizing complex information to speed up 306 00:15:28,759 --> 00:15:29,960 expert decision making. 307 00:15:30,159 --> 00:15:33,159 Speaker 1: This is where we see that clearing the cognitive backlog 308 00:15:33,279 --> 00:15:35,360 idea in its most direct form. 309 00:15:35,200 --> 00:15:38,960 Speaker 2: And it really starts with Project fourteen multimodal AI models. 310 00:15:39,080 --> 00:15:41,879 These are models that can see, here and read all 311 00:15:41,919 --> 00:15:42,919 at the same time. 312 00:15:42,840 --> 00:15:45,360 Speaker 1: Which is so important because that's how the world actually 313 00:15:45,360 --> 00:15:48,120 presents information to us, right, and it's messy. We don't 314 00:15:48,120 --> 00:15:50,600 just get a text document never you. 315 00:15:50,559 --> 00:15:53,360 Speaker 2: Get a spreadsheet, You read the company email, you hear 316 00:15:53,399 --> 00:15:57,600 the voice note, you watch the video presentation. Human cognition 317 00:15:57,679 --> 00:15:59,720 has to integrate all of those. 318 00:15:59,519 --> 00:16:03,200 Speaker 1: Signals, and that's why models like Google deepminds Gemini are 319 00:16:03,320 --> 00:16:04,120 such a big deal. 320 00:16:04,279 --> 00:16:08,039 Speaker 2: They can reason across text, images, audio, and video all 321 00:16:08,080 --> 00:16:11,759 at once. Real world problems are never neatly formatted into 322 00:16:11,759 --> 00:16:13,080 a single column of data. 323 00:16:13,559 --> 00:16:15,879 Speaker 1: You could have a situation where I don't know. A 324 00:16:15,960 --> 00:16:19,000 text document says one thing, but a chart embedded in 325 00:16:19,039 --> 00:16:21,440 an image file completely contradicts. 326 00:16:20,960 --> 00:16:23,240 Speaker 2: It, right, and a human has to spot that contradiction, 327 00:16:23,879 --> 00:16:26,759 process the visual data from the chart, and then integrate 328 00:16:26,799 --> 00:16:30,399 that new context back into their textual understanding. It takes 329 00:16:30,440 --> 00:16:31,559 time and focus, and. 330 00:16:31,519 --> 00:16:33,759 Speaker 1: A multimodal system does this instantly. 331 00:16:34,120 --> 00:16:36,559 Speaker 2: It can analyze the data points in the chart, explain 332 00:16:36,600 --> 00:16:39,399 a visual error in an image, or interpret a video 333 00:16:39,480 --> 00:16:42,080 scene while simultaneously processing. 334 00:16:41,559 --> 00:16:43,960 Speaker 1: The text, so its function is less like a. 335 00:16:44,000 --> 00:16:48,519 Speaker 2: Chatbot and more like a tireless integrated pattern spot. It 336 00:16:48,639 --> 00:16:52,440 synthesizes all those disparate signals into a single, coherent interpretation 337 00:16:52,519 --> 00:16:53,480 for the human to review. 338 00:16:53,720 --> 00:16:58,279 Speaker 1: So, if you listening right now are constantly battling information overload, 339 00:16:58,759 --> 00:17:03,200 wishing you could just synthesize these complex perspectives faster, this 340 00:17:03,320 --> 00:17:06,160 is the technology designed to give you that advantage. 341 00:17:06,200 --> 00:17:10,079 Speaker 2: It performs that initial high speed synthesis that is required 342 00:17:10,160 --> 00:17:11,839 for any expert. 343 00:17:11,440 --> 00:17:15,640 Speaker 1: Decision, which brings us directly to Project thirteen, the use 344 00:17:15,640 --> 00:17:19,839 of these frontier models as cognitive copilots for expert level work. 345 00:17:20,119 --> 00:17:23,640 Speaker 2: And this is arguably the most pervasive and immediately impactful 346 00:17:23,680 --> 00:17:27,480 form of augmentation available today. It's happening across almost every industry. 347 00:17:27,519 --> 00:17:29,519 Speaker 1: Oh, we have to really emphasize this again. This is 348 00:17:29,519 --> 00:17:32,400 about augmenting the expert, not replacing the novice. 349 00:17:32,480 --> 00:17:35,720 Speaker 2: Absolutely, this is about reducing friction for people who already 350 00:17:35,720 --> 00:17:38,440 know what they're doing. These models are being studied in law, 351 00:17:38,640 --> 00:17:42,920 in advanced engineering, in medicine, always as support systems, not 352 00:17:43,039 --> 00:17:44,480 as final decision makers. 353 00:17:44,519 --> 00:17:46,519 Speaker 1: And the statistics are pretty compelling. 354 00:17:46,440 --> 00:17:49,920 Speaker 2: Extremely and they're consistent across multiple studies from places like 355 00:17:50,000 --> 00:17:54,160 MIT and Stanford experienced professionals who use AI to assist 356 00:17:54,160 --> 00:17:57,039 them with research and drafting. They not only completed their 357 00:17:57,079 --> 00:17:59,519 tasks roughly twenty to forty percent faster. 358 00:17:59,480 --> 00:18:02,559 Speaker 1: Which is all already a huge gain, but they also 359 00:18:02,599 --> 00:18:06,799 significantly reduced factual and clerical errors. And that distinction the 360 00:18:06,839 --> 00:18:10,119 expertise of the user. It's everything, isn't it It. 361 00:18:10,119 --> 00:18:13,839 Speaker 2: Is The AI isn't doing the final judgment. It's offloading 362 00:18:13,839 --> 00:18:16,119 the immense cognitive burden of working. 363 00:18:15,799 --> 00:18:20,119 Speaker 1: Memory, so it allows the expert to instantly compare dozens 364 00:18:20,119 --> 00:18:24,079 of legal precedents, or review thousands of lines of code. 365 00:18:24,240 --> 00:18:28,200 Speaker 2: Or synthesize a massive patient medical history, all before their 366 00:18:28,279 --> 00:18:31,519 human brain even begins the actual process of judgment. 367 00:18:31,680 --> 00:18:33,279 Speaker 1: It clears the decks so they can think. 368 00:18:33,480 --> 00:18:37,200 Speaker 2: Think of a researcher trying to optimize a new drug candidate. Traditionally, 369 00:18:37,240 --> 00:18:42,039 they spend countless hours manually cross referencing databases to predict critical. 370 00:18:41,640 --> 00:18:46,960 Speaker 1: Properties, things with acronyms like eightma day, absorption distribution. 371 00:18:46,799 --> 00:18:51,240 Speaker 2: Metabolism, excretion, and toxicity. If the AI can instantly review 372 00:18:51,279 --> 00:18:53,920 all the relevant literature and flag molecules that are predicted 373 00:18:53,920 --> 00:18:57,200 to have poor outcomes, the human researcher's energy is preserved 374 00:18:57,319 --> 00:19:00,440 entirely for solving the truly difficult chemical press. 375 00:19:00,720 --> 00:19:03,599 Speaker 1: The human still decides what molecule to pursue, but the 376 00:19:03,640 --> 00:19:06,680 AI makes sure that decision is made with the maximum 377 00:19:06,680 --> 00:19:11,599 possible context and with minimal effort wasted on just data retrieval. 378 00:19:11,759 --> 00:19:15,319 Speaker 2: And this copilot concept it extends surprisingly well into creative 379 00:19:15,359 --> 00:19:18,279 workflowse too, beyond these traditional expert fields. 380 00:19:18,519 --> 00:19:21,559 Speaker 1: Right the sources mentioned an example called overseer os for 381 00:19:21,640 --> 00:19:22,759 content creation. 382 00:19:22,759 --> 00:19:27,480 Speaker 2: Exactly, And the AI isn't writing the final blockbuster script 383 00:19:27,559 --> 00:19:30,279 or making the final artisticated on a film now. 384 00:19:30,440 --> 00:19:33,799 Speaker 1: Its function is pattern analysis on a massive scale. 385 00:19:33,920 --> 00:19:37,119 Speaker 2: It analyzes existing content in the market. It breaks down 386 00:19:37,160 --> 00:19:41,319 structural patterns across thousands of successful projects, and it highlights 387 00:19:41,319 --> 00:19:45,599 specific thematic or pacing elements that statistically resonate with an audience. 388 00:19:45,680 --> 00:19:48,599 Speaker 1: So it's providing the creator with synthesized context on what 389 00:19:48,720 --> 00:19:50,160 structurally works, which is. 390 00:19:50,160 --> 00:19:53,319 Speaker 2: Something creators normally figure out only through years of painful 391 00:19:53,359 --> 00:19:57,119 trial and error or just intense time consuming market study. 392 00:19:57,279 --> 00:20:01,200 Speaker 1: So by managing that information complexity and revealing the underlying patterns, 393 00:20:01,480 --> 00:20:05,119 the AI gives the creator a massive efficiency advantage. It 394 00:20:05,200 --> 00:20:09,880 speeds up their path to creative success. Okay, so next, 395 00:20:10,039 --> 00:20:12,799 let's explore the most direct forms of augmentation. This is 396 00:20:12,799 --> 00:20:18,839 where gets really interesting, rebuilding the body interfaces, robotics, and this. 397 00:20:18,799 --> 00:20:22,759 Speaker 2: Really highlights the shift from fixed, hard coded machine interaction 398 00:20:23,119 --> 00:20:26,480 to real time, adaptive learning based physical assistance. 399 00:20:26,960 --> 00:20:30,720 Speaker 1: Let's start by closing the communication gap. Project ten. Non 400 00:20:30,720 --> 00:20:34,960 invasive brain computer interfaces or BCIs. 401 00:20:34,559 --> 00:20:37,480 Speaker 2: The fact that we can achieve such high functionality without 402 00:20:37,480 --> 00:20:41,039 the enormous barrier of invasive surgery. That's a massive social 403 00:20:41,079 --> 00:20:42,119 and medical lead board. 404 00:20:42,240 --> 00:20:45,799 Speaker 1: It's a total game changer for accessibility. Researchers at Stanford 405 00:20:45,799 --> 00:20:49,240 and Meta Reality Labs have demonstrated these non invasive BCIs 406 00:20:49,279 --> 00:20:50,640 that use AI trained on. 407 00:20:50,640 --> 00:20:54,839 Speaker 2: EEG data right electron cepholography the cap with all the sensors. 408 00:20:54,400 --> 00:20:58,160 Speaker 1: And it decodes those brain signals directly into actions. Typing text, 409 00:20:58,200 --> 00:20:59,160 moving a cursor on a. 410 00:20:59,160 --> 00:21:04,000 Speaker 2: Screen mechanism is really sophisticated pattern translation. The AI learns 411 00:21:04,039 --> 00:21:08,559 to correlate specific, recognizable neural signatures with a person's intent. 412 00:21:09,160 --> 00:21:11,359 For instance, it might focus on mollar rhythms in the 413 00:21:11,359 --> 00:21:14,599 motor cortex, which signify a planned movement. 414 00:21:14,440 --> 00:21:17,359 Speaker 1: Or P three hundred waves which indicate that a decision 415 00:21:17,400 --> 00:21:18,200 has been made, and. 416 00:21:18,160 --> 00:21:20,279 Speaker 2: It learns in individual's unique brain patterns. 417 00:21:20,279 --> 00:21:22,960 Speaker 1: It's personalized, and the speed is what really caught my 418 00:21:23,000 --> 00:21:26,559 attention in the data. In twenty twenty three, Stanford researchers 419 00:21:26,680 --> 00:21:30,519 enabled a paralyzed patient to generate text at sixty words 420 00:21:30,559 --> 00:21:30,920 per minut. 421 00:21:31,119 --> 00:21:33,279 Speaker 2: Sixty words per minut it just by thinking. 422 00:21:33,920 --> 00:21:37,400 Speaker 1: That is approaching a natural conversational typing rate, and. 423 00:21:37,400 --> 00:21:42,119 Speaker 2: That speed achieved non invasively, fundamentally changes the upper limit 424 00:21:42,160 --> 00:21:45,559 of communication for individuals who can't speak or type. And 425 00:21:45,599 --> 00:21:48,079 we have to be clear, this is not mind reading now. 426 00:21:48,200 --> 00:21:51,400 It is a learned, personalized dictionary of neural activity that 427 00:21:51,519 --> 00:21:54,839 translates intent into digital action at conversational speed. 428 00:21:54,920 --> 00:21:57,319 Speaker 1: Now, if we look at project IE the neuralink style 429 00:21:57,400 --> 00:22:00,720 brain machine interfaces or BMIs, we move to the most 430 00:22:00,720 --> 00:22:02,519 direct form of augmentation. 431 00:22:02,400 --> 00:22:05,599 Speaker 2: And the most bandwidth rich, but it does require that 432 00:22:05,720 --> 00:22:07,920 highly regulated invasive surgery. 433 00:22:08,279 --> 00:22:12,200 Speaker 1: With these systems, they're surgically inserting electrode rays directly into 434 00:22:12,240 --> 00:22:13,680 the brain tissue. 435 00:22:13,240 --> 00:22:16,200 Speaker 2: Which provides a much much higher resolution signal. You get 436 00:22:16,279 --> 00:22:19,440 richer data transfer both in and out. The mechanism is 437 00:22:19,440 --> 00:22:23,319 a closed loop, neural activity is captured, it's translated by 438 00:22:23,359 --> 00:22:27,160 these learning algorithms, and that produces precise digital actions in 439 00:22:27,200 --> 00:22:27,759 real time. 440 00:22:28,240 --> 00:22:31,680 Speaker 1: This really represents the physical limit of augmentation, doesn't it 441 00:22:32,039 --> 00:22:37,039 Communication control sensory input all mediated directly by computational power. 442 00:22:37,160 --> 00:22:41,039 Speaker 2: It does. We've seen patients with severe paralysis use thought 443 00:22:41,079 --> 00:22:46,400 alone to control cursors, complex robotic arms, computer interfaces, and 444 00:22:46,440 --> 00:22:47,880 with incredibly high fidelity. 445 00:22:48,240 --> 00:22:51,559 Speaker 1: And while this technology is and should be tightly regulated 446 00:22:51,599 --> 00:22:53,960 and is nowhere near being consumer. 447 00:22:53,519 --> 00:22:56,960 Speaker 2: Tech, conceptually, it proves the viability of a complete high 448 00:22:56,960 --> 00:23:02,079 bandwidth fusion between biological signaling and digital computation. If the 449 00:23:02,079 --> 00:23:04,640 non invasive methods are chipping away at our limits, the 450 00:23:04,680 --> 00:23:07,519 invasive methods are showing us how to remove those limits entirely. 451 00:23:07,599 --> 00:23:10,599 Speaker 1: Okay, So moving from communication to physical control, this is 452 00:23:10,599 --> 00:23:14,279 where AI is transforming movement itself. Project nine AI guided 453 00:23:14,319 --> 00:23:16,920 prosthetics and exoskeletons. 454 00:23:16,400 --> 00:23:20,119 Speaker 2: And this represents a radical break from the older fixed technologies. 455 00:23:20,319 --> 00:23:24,279 Speaker 1: Older prosthetics were well, they were basically mechanical. They operated 456 00:23:24,319 --> 00:23:26,039 on simple, hard coded rules. 457 00:23:26,119 --> 00:23:29,160 Speaker 2: If muscle signal A fires, then move joint B. That 458 00:23:29,240 --> 00:23:30,680 sort of thing, right, and. 459 00:23:30,599 --> 00:23:34,079 Speaker 1: It resulted in movement that felt heavy, clunky, mechanical. It 460 00:23:34,119 --> 00:23:37,559 required intense conscious effort from the user just to manage 461 00:23:37,599 --> 00:23:38,079 the device. 462 00:23:38,480 --> 00:23:42,319 Speaker 2: Modern AI guided devices learn. They are trained on rich 463 00:23:42,480 --> 00:23:47,839 EMG signals, electromiography, analyzing the nuance variants in muscle contraction. 464 00:23:47,559 --> 00:23:51,119 Speaker 1: Patterns, so they can predict the user's intent faster and 465 00:23:51,200 --> 00:23:53,599 more accurately than any fixed rule set, and. 466 00:23:53,559 --> 00:23:57,599 Speaker 2: They adapt in real time based on subtle muscle signals, 467 00:23:57,759 --> 00:24:00,720 on changes in gait, on the specific time the person 468 00:24:00,799 --> 00:24:01,440 is walking. 469 00:24:01,200 --> 00:24:05,160 Speaker 1: On that responsiveness is everything. The clinical trials show users 470 00:24:05,160 --> 00:24:08,960 of these modern devices expend significantly less metabolic energy. 471 00:24:08,839 --> 00:24:12,160 Speaker 2: Because the device is assisting the movement intuitively, rather than 472 00:24:12,200 --> 00:24:14,880 the user having to fight against the device's fixed programming. 473 00:24:15,119 --> 00:24:18,279 Speaker 1: We're seeing this most vividly in rehab centers. These machine 474 00:24:18,359 --> 00:24:22,119 learning powered exoskeletons are helping patients with spinal injuries walk. 475 00:24:21,960 --> 00:24:26,319 Speaker 2: Again, and they're providing this adaptive support that mimics the 476 00:24:26,400 --> 00:24:30,440 natural compensatory movements a healthy person would subconsciously make. The 477 00:24:30,559 --> 00:24:35,000 quality of movement restoration is just it's profoundly improved. 478 00:24:35,160 --> 00:24:37,799 Speaker 1: It's smoother, more intuitive exactly, and. 479 00:24:37,720 --> 00:24:42,880 Speaker 2: That concept of intuitive learned physical control, it's vital for 480 00:24:42,920 --> 00:24:47,200 our next project, project eight. Learning based robotic manipulation. 481 00:24:47,599 --> 00:24:53,039 Speaker 1: For decades, industrial robots have been incredibly powerful, but also incredibly. 482 00:24:52,400 --> 00:24:55,920 Speaker 2: Clumsy because the real world is messy. Objects are rarely 483 00:24:55,920 --> 00:24:59,799 placed in the exact same spot. Their properties, weight, texture, 484 00:25:00,119 --> 00:25:02,160 shape they constantly vary. 485 00:25:02,079 --> 00:25:05,319 Speaker 1: And traditional AI, the kind based on hand coded rules, 486 00:25:05,599 --> 00:25:07,680 failed miserably in those dynamic environments. 487 00:25:07,680 --> 00:25:10,319 Speaker 2: Oh completely. If you dropped a slightly crumpled box, the 488 00:25:10,359 --> 00:25:12,880 classic robot was lost. It had no rule for that 489 00:25:12,920 --> 00:25:14,559 specific contour orientation. 490 00:25:14,720 --> 00:25:16,720 Speaker 1: So the breakthrough isn't a better robotic claw. 491 00:25:17,000 --> 00:25:20,519 Speaker 2: Breakthrough is patients its persistence. Labs at Google and open 492 00:25:20,559 --> 00:25:22,279 AI are using reinforcement. 493 00:25:21,839 --> 00:25:24,440 Speaker 1: Learning, so the robots improve their dexterity through millions of 494 00:25:24,480 --> 00:25:26,039 cycles of trial and feedback. 495 00:25:26,319 --> 00:25:29,759 Speaker 2: They use self directed attempts to learn how to grasp, move, 496 00:25:29,960 --> 00:25:33,759 and manipulate unfamiliar or irregularly shaped objects. 497 00:25:34,039 --> 00:25:37,079 Speaker 1: The machine can repeat a task, say picking up a 498 00:25:37,160 --> 00:25:40,599 unique pile of clutter, millions of times, without getting tired 499 00:25:40,759 --> 00:25:44,480 or bored or frustrated. It just constantly refines its approach 500 00:25:44,559 --> 00:25:46,880 based on these minimal feedback signals. 501 00:25:47,119 --> 00:25:50,119 Speaker 2: That sheer volume of trial and error creates a kind 502 00:25:50,160 --> 00:25:54,720 of physical dexterity and precision that humans simply cannot sustain 503 00:25:54,880 --> 00:25:56,000 over long periods. 504 00:25:56,119 --> 00:25:59,200 Speaker 1: So its augmentation through tireless physical. 505 00:25:58,759 --> 00:26:01,880 Speaker 2: Practice, and it's achieving a robustness in handling our messy 506 00:26:01,880 --> 00:26:04,839 world that human workers have always struggled to match. 507 00:26:05,039 --> 00:26:07,960 Speaker 1: Okay, that brings us to our final major theme, precision 508 00:26:08,240 --> 00:26:11,920 and prediction. These are the AI applications that offer better timing, 509 00:26:12,079 --> 00:26:15,839 reduced errors, and optimized results in well high stakes, everyday 510 00:26:15,839 --> 00:26:16,680 life scenarios. 511 00:26:16,720 --> 00:26:19,039 Speaker 2: This is really where the AI moves from the research 512 00:26:19,119 --> 00:26:23,200 lab into immediate practical application for millions of people. 513 00:26:23,000 --> 00:26:26,720 Speaker 1: Starting with Project eleven predictive health AI from our everyday wearables. 514 00:26:27,039 --> 00:26:30,160 Speaker 2: This application might not sound as flashy as a brain implant, 515 00:26:30,599 --> 00:26:35,359 but it's profoundly impactful. It brings augmentation directly into preventative 516 00:26:35,400 --> 00:26:37,519 care through continuous monitoring. 517 00:26:37,720 --> 00:26:41,799 Speaker 1: We're all basically walking around with these incredibly dense longitudinal 518 00:26:41,920 --> 00:26:45,200 data sets being generated every second by our smart watches. 519 00:26:45,240 --> 00:26:50,279 Speaker 2: Heart rate, heart rate, variability, sleep stages, movement, blood oxygen, 520 00:26:50,359 --> 00:26:50,720 and the. 521 00:26:50,640 --> 00:26:55,319 Speaker 1: Power here lies in that continuous longitudinal data analysis. A 522 00:26:55,400 --> 00:26:58,440 normal doctor's visit is a single snapshot in time. 523 00:26:58,559 --> 00:27:00,079 Speaker 2: The wearable provides the whole movie. 524 00:27:00,160 --> 00:27:04,079 Speaker 1: And these machine learning systems trained on vast physiological data 525 00:27:04,079 --> 00:27:07,880 sets can detect subtle early signals that a human or 526 00:27:07,960 --> 00:27:11,039 even a single clinical test would almost certainly miss. 527 00:27:11,279 --> 00:27:13,640 Speaker 2: There are peer reviewed studies out there published by the 528 00:27:13,680 --> 00:27:16,440 major tech companies to show these systems can detect early 529 00:27:16,480 --> 00:27:19,480 indicators of things like atrial fibrillation. 530 00:27:19,319 --> 00:27:22,799 Speaker 1: Or subtle shifts in metabolic risk markers days or even 531 00:27:22,880 --> 00:27:25,119 weeks in advance of serious symptoms showing up. 532 00:27:25,200 --> 00:27:27,240 Speaker 2: But again, to be very clear, these systems do not 533 00:27:27,319 --> 00:27:28,279 diagnose right. 534 00:27:28,359 --> 00:27:32,759 Speaker 1: They flag anomalies. They're a sophisticated personal alert system, but in. 535 00:27:32,759 --> 00:27:36,400 Speaker 2: Health, better timing often accounts for the most significant improvements 536 00:27:36,400 --> 00:27:40,400 in outcomes. If an early non invasive signal prompts someone 537 00:27:40,519 --> 00:27:44,000 to seek care earlier, the intervention is almost invariably easier 538 00:27:44,000 --> 00:27:44,359 and more. 539 00:27:44,279 --> 00:27:50,039 Speaker 1: Successful its augmentation through perpetual non invasive internal monitoring exactly. 540 00:27:50,359 --> 00:27:53,240 Speaker 2: Now, let's look at achieving foundational accuracy with Project five 541 00:27:53,720 --> 00:27:56,200 AI assisted gene editing Optimization. 542 00:27:56,680 --> 00:28:01,920 Speaker 1: Gene editing, especially Crisper style methods, is revolutionary therapy, but 543 00:28:02,119 --> 00:28:05,359 its success relies on absolute, uncompromising precision. 544 00:28:05,680 --> 00:28:08,680 Speaker 2: The danger is something called the off target effect. An 545 00:28:08,799 --> 00:28:12,799 unintended cut somewhere else in the vast complex genome. 546 00:28:12,480 --> 00:28:15,799 Speaker 1: Which can range from just reducing the therapy's effectiveness to 547 00:28:16,119 --> 00:28:19,039 well causing catastrophic unintended mutations. 548 00:28:19,079 --> 00:28:22,759 Speaker 2: So the AI provides the necessary surgical precision. AI models 549 00:28:22,799 --> 00:28:26,400 are now indispensable for predicting those off target cleavage. 550 00:28:26,000 --> 00:28:29,640 Speaker 1: Sites, So researchers use machine learning to screen thousands of 551 00:28:29,640 --> 00:28:32,079 potential guide RNA sequences. 552 00:28:31,640 --> 00:28:34,599 Speaker 2: And identify which ones will maximize the on target activity 553 00:28:34,920 --> 00:28:38,559 while simultaneously minimizing the risk of unintended cuts elsewhere in 554 00:28:38,559 --> 00:28:38,960 the genome. 555 00:28:39,240 --> 00:28:42,240 Speaker 1: This is a beautiful example of what you call quiet accuracy. 556 00:28:43,000 --> 00:28:46,920 This work is mostly preclinical, it's foundational, but it's absolutely 557 00:28:47,039 --> 00:28:47,839 mission critical. 558 00:28:48,200 --> 00:28:50,920 Speaker 2: Precision in this field is the difference between a safe, 559 00:28:51,079 --> 00:28:54,240 curative therapy and one that is simply too dangerous to 560 00:28:54,279 --> 00:28:58,759 even pursue. The augmentation here isn't about speed. It's about 561 00:28:58,799 --> 00:29:00,960 providing the confidence to move forward. 562 00:29:00,880 --> 00:29:03,920 Speaker 1: And finally that brings us to project too. This feels 563 00:29:03,920 --> 00:29:06,680 like the culmination of everything we've been talking about human 564 00:29:06,720 --> 00:29:08,039 AI decision systems. 565 00:29:08,359 --> 00:29:12,279 Speaker 2: This represents the ultimate partnership model, especially in high stakes environments. 566 00:29:12,519 --> 00:29:14,880 Speaker 1: This is where AI really earns its keep as a 567 00:29:14,880 --> 00:29:20,039 true colleague and feels that require maximum judgment. Medicine, finance, 568 00:29:20,160 --> 00:29:21,359 climate science. 569 00:29:21,200 --> 00:29:23,960 Speaker 2: And the division of labor is crystal clear. The AI 570 00:29:24,079 --> 00:29:27,759 highlights the patterns, the risks, the possible outcomes, and the 571 00:29:27,839 --> 00:29:31,039 humans apply context and judgment to decide what action to take. 572 00:29:31,240 --> 00:29:34,440 Speaker 1: Let's take radiology diagnostics in a hospital. An AI system 573 00:29:34,440 --> 00:29:37,599 can analyze hundreds of complex patient scans in minutes. It 574 00:29:37,640 --> 00:29:40,720 can perform this rapid feature extraction to flag subtle high 575 00:29:40,839 --> 00:29:42,279 risk anomalies. 576 00:29:42,000 --> 00:29:45,519 Speaker 2: Anomalies that a tired human eye might miss. After reviewing 577 00:29:45,559 --> 00:29:47,799 dozens and dozens of cases in a row. 578 00:29:47,519 --> 00:29:51,599 Speaker 1: And the studies confirm the synergistic benefit. AI assisted reviews 579 00:29:51,759 --> 00:29:54,559 significantly reduce these kinds of oversight errors. 580 00:29:54,720 --> 00:29:57,720 Speaker 2: The machine has sharpened the perception by managing the throughput, 581 00:29:58,680 --> 00:30:03,720 but the human radiologist still brings years of training, contextual understanding, 582 00:30:04,039 --> 00:30:06,839 and final medical judgment to that patient's bedside. 583 00:30:06,880 --> 00:30:09,200 Speaker 1: It's the same in climate science, right, very similar. 584 00:30:09,279 --> 00:30:14,240 Speaker 2: The models are constantly processing these vast chaotic global data sets, 585 00:30:14,480 --> 00:30:17,839 camperature records, oceanic currents, atmospheric composition. 586 00:30:18,160 --> 00:30:22,279 Speaker 1: The AI improves forecasting resolution, it improves scenario testing by 587 00:30:22,279 --> 00:30:25,960 modeling hundreds of these interacting variables simultaneously, which. 588 00:30:25,759 --> 00:30:29,640 Speaker 2: Gives the human climate scientists much clearer, higher resolution data 589 00:30:29,759 --> 00:30:34,079 on potential outcomes, say the probability of a specific extreme 590 00:30:34,119 --> 00:30:35,519 weather event five years from now. 591 00:30:35,680 --> 00:30:37,799 Speaker 1: The AI doesn't dictate the policy, No. 592 00:30:38,319 --> 00:30:41,960 Speaker 2: It provides the robust computational context that allows human policy 593 00:30:41,960 --> 00:30:44,759 makers and researchers to apply their ethical frameworks and their 594 00:30:44,759 --> 00:30:46,279 political judgment more effectively. 595 00:30:46,440 --> 00:30:50,799 Speaker 1: So the core function, again and again is augmentation. AI 596 00:30:50,960 --> 00:30:54,839 doesn't replace judgment. It provides the depth of context, the 597 00:30:54,920 --> 00:30:58,759 synthesis of data, the relentless pattern spotting that is necessary 598 00:30:58,799 --> 00:31:00,240 to sharpen human judgment. 599 00:31:00,079 --> 00:31:03,000 Speaker 2: Especially in environments where the complexity is so high that 600 00:31:03,319 --> 00:31:05,759 without it, failure would otherwise be inevitable. 601 00:31:05,880 --> 00:31:08,720 Speaker 1: Okay, So as we pull all of this together synthesizing 602 00:31:08,759 --> 00:31:12,119 these fifteen projects, it feels like we can distill this 603 00:31:12,240 --> 00:31:15,880 down into four major ways AI is making us, for 604 00:31:15,960 --> 00:31:17,559 lack of a better term, superhuman. 605 00:31:17,680 --> 00:31:19,720 Speaker 2: I think so four major themes emerge. 606 00:31:19,880 --> 00:31:21,759 Speaker 1: First, AI provides persistence. 607 00:31:22,000 --> 00:31:25,400 Speaker 2: Yes, this is those relentless research agents running thousands of 608 00:31:25,400 --> 00:31:29,200 experiments overnight. It's the learning based robotics that refine physical 609 00:31:29,240 --> 00:31:33,400 precision through millions of cycles of self directed attempts. Sheer endurance. 610 00:31:33,599 --> 00:31:35,799 Speaker 1: Second, AI provides scope. 611 00:31:35,400 --> 00:31:38,960 Speaker 2: The capacity to search those chemical, material and genomic spaces 612 00:31:39,000 --> 00:31:42,039 that are just too vast for human intuition or manual exploration. 613 00:31:42,920 --> 00:31:45,519 This leads to foundational breakthroughs that we would simply never 614 00:31:45,599 --> 00:31:46,640 locate on our own. 615 00:31:46,759 --> 00:31:48,680 Speaker 1: Third AI provides friction reduction. 616 00:31:49,200 --> 00:31:54,759 Speaker 2: This is the cognitive copilot, synthesizing multimodal data, offloading working 617 00:31:54,799 --> 00:31:59,200 memory so that human experts can make faster, better informed decisions, 618 00:32:00,000 --> 00:32:03,079 deserves their mental energy for complex judgment. 619 00:32:03,240 --> 00:32:06,720 Speaker 1: And Fourth AI provides interface. 620 00:32:06,720 --> 00:32:09,839 Speaker 2: From the non invasive BCIs enabling high speed thought to 621 00:32:09,920 --> 00:32:14,200 text to adaptive prosthetics that learn your gait. AI is 622 00:32:14,279 --> 00:32:18,319 directly managing and stretching our biological limits of communication and 623 00:32:18,359 --> 00:32:20,559 physical dexterity in real time. 624 00:32:20,680 --> 00:32:23,519 Speaker 1: I think the overarching concept here is just so powerful. 625 00:32:23,640 --> 00:32:26,720 These projects. They're not aimed at creating artificial humans that 626 00:32:26,799 --> 00:32:29,960 at all, They are meticulously engineered to create enhanced humans. 627 00:32:30,240 --> 00:32:33,160 We're entering an era where are innate biological limits. The 628 00:32:33,160 --> 00:32:35,920 speed of our research, the capacity of our memory, the 629 00:32:35,960 --> 00:32:37,640 physical precision we can sustain. 630 00:32:37,799 --> 00:32:39,640 Speaker 2: All of these things are now being managed and even 631 00:32:39,720 --> 00:32:42,799 transcended by these tireless computational partners. 632 00:32:42,920 --> 00:32:44,880 Speaker 1: So if the goal of these ais is to handle 633 00:32:44,920 --> 00:32:47,839 all the high volume comparison, the data analysis, the memory 634 00:32:47,839 --> 00:32:50,319 management that normally eats up eighty percent of our attention, 635 00:32:50,519 --> 00:32:50,799 they are. 636 00:32:50,799 --> 00:32:54,400 Speaker 2: Essentially freeing us from the administrative burden of intelligence. 637 00:32:54,039 --> 00:32:57,559 Speaker 1: Which raises the truly provocative question. I think, if we 638 00:32:57,640 --> 00:32:59,480 no longer have to worry about the memory load or 639 00:32:59,519 --> 00:33:04,400 the speed of processing, what aspect of pure human thinking, 640 00:33:04,799 --> 00:33:08,680 the non analytical, non enduring part, what becomes the most 641 00:33:08,759 --> 00:33:10,839 valuable trait that we must preserve and cultivate. 642 00:33:10,960 --> 00:33:13,920 Speaker 2: Is it pure unstructured creativity. 643 00:33:13,720 --> 00:33:16,960 Speaker 1: Or maybe that uniquely human capacity for ethical reflection and 644 00:33:17,000 --> 00:33:18,119 emotional intelligence. 645 00:33:18,519 --> 00:33:22,279 Speaker 2: So, considering all of these immense technological advancements, where do 646 00:33:22,359 --> 00:33:25,000 you think the human element will be most indispensable in 647 00:33:25,039 --> 00:33:27,279 the next five years? What is the one thing that 648 00:33:27,400 --> 00:33:28,880 AI cannot augment? 649 00:33:29,160 --> 00:33:30,480 Speaker 1: We really want to hear what you think. 650 00:33:30,559 --> 00:33:32,559 Speaker 2: Head over to our comments section and leave us your 651 00:33:32,599 --> 00:33:35,640 perspective on the future of the superhuman mind. We'll see 652 00:33:35,640 --> 00:33:37,039 you next time on Throlling Threads.