1 00:00:00,000 --> 00:00:03,040 All right, let's jump into this paper about Psi Agents. 2 00:00:03,040 --> 00:00:03,540 Okay. 3 00:00:03,540 --> 00:00:05,760 The title is, kind of long. 4 00:00:05,760 --> 00:00:06,260 Yeah. 5 00:00:06,260 --> 00:00:12,380 Psi Agents, automating scientific discovery through bio-inspired, multi-agent, intelligent 6 00:00:12,380 --> 00:00:13,380 graph reasoning. 7 00:00:13,380 --> 00:00:13,880 Right. 8 00:00:13,880 --> 00:00:16,800 But the idea is really cool. 9 00:00:16,800 --> 00:00:22,000 Using AI to speed up scientific breakthroughs, especially with bio-inspired materials. 10 00:00:22,000 --> 00:00:25,120 And we're going straight to the source for this deep dive. 11 00:00:25,120 --> 00:00:25,620 Right. 12 00:00:25,620 --> 00:00:27,440 Excerpts from the research paper itself. 13 00:00:27,440 --> 00:00:27,940 Okay. 14 00:00:27,940 --> 00:00:29,600 So get ready for some cutting-edge stuff. 15 00:00:29,600 --> 00:00:33,960 It is fascinating work when you think about all the scientific data out there. 16 00:00:33,960 --> 00:00:36,680 It's a lot, even for really smart people. 17 00:00:36,680 --> 00:00:37,180 Right. 18 00:00:37,180 --> 00:00:42,240 Psi Agents tries to deal with that by going through all this data and finding hidden connections 19 00:00:42,240 --> 00:00:44,040 to suggest new research. 20 00:00:44,040 --> 00:00:46,040 So it's like a super-powered research system. 21 00:00:46,040 --> 00:00:46,540 Exactly. 22 00:00:46,540 --> 00:00:48,960 Psi Agents builds what's called a knowledge graph. 23 00:00:48,960 --> 00:00:49,460 Okay. 24 00:00:49,460 --> 00:00:54,880 It's basically a map of scientific concepts taken from a ton of research papers. 25 00:00:54,880 --> 00:00:57,560 And then it uses specialized AI agents. 26 00:00:57,560 --> 00:00:58,060 Hold on. 27 00:00:58,060 --> 00:00:59,960 Like secret agents. 28 00:00:59,960 --> 00:01:00,960 Not exactly. 29 00:01:00,960 --> 00:01:02,600 But it's a good analogy. 30 00:01:02,600 --> 00:01:04,600 These AI agents have different roles. 31 00:01:04,600 --> 00:01:07,320 Some map out interesting connections. 32 00:01:07,320 --> 00:01:10,320 Some look into the specifics of some research. 33 00:01:10,320 --> 00:01:12,120 And others are like critics. 34 00:01:12,120 --> 00:01:13,120 Oh, yeah. 35 00:01:13,120 --> 00:01:15,100 Evaluating the research proposals. 36 00:01:15,100 --> 00:01:17,560 Just like a tough peer reviewer would. 37 00:01:17,560 --> 00:01:18,560 That's pretty wild. 38 00:01:18,560 --> 00:01:22,600 So it's like a whole team of AI researchers all working together. 39 00:01:22,600 --> 00:01:23,600 Exactly. 40 00:01:23,600 --> 00:01:29,400 And the researchers behind Psi Agents tested two versions of this system. 41 00:01:29,400 --> 00:01:34,720 The first one is what they call pre-programmed AI teamwork. 42 00:01:34,720 --> 00:01:37,280 Imagine a play that's been rehearsed a lot. 43 00:01:37,280 --> 00:01:40,680 Each agent knows its part and what to do. 44 00:01:40,680 --> 00:01:44,760 So it's reliable, but maybe not very flexible. 45 00:01:44,760 --> 00:01:46,480 And what about the second version? 46 00:01:46,480 --> 00:01:48,080 That's where it gets really interesting. 47 00:01:48,080 --> 00:01:51,160 It's called Autonomous AI Collaboration. 48 00:01:51,160 --> 00:01:54,040 Here the AI agents actually figure out their own workflow. 49 00:01:54,040 --> 00:01:57,640 It's more like a jazz band improvising. 50 00:01:57,640 --> 00:02:02,960 They listen to each other, adjust how they play, and create something new and exciting. 51 00:02:02,960 --> 00:02:05,840 They even check their ideas against existing research. 52 00:02:05,840 --> 00:02:07,560 So they make sure they're on the right track. 53 00:02:07,560 --> 00:02:09,360 That's really cool. 54 00:02:09,360 --> 00:02:13,960 So the paper mentioned this example with silk and energy intensive. 55 00:02:13,960 --> 00:02:14,960 What happened there? 56 00:02:14,960 --> 00:02:16,800 That's one of the most exciting parts. 57 00:02:16,800 --> 00:02:19,520 Psi Agents came up with a new material idea. 58 00:02:19,520 --> 00:02:25,520 A silk composite material strengthened with pigments extracted from dandelions. 59 00:02:25,520 --> 00:02:27,800 Dandelions, that's not what I would have guessed. 60 00:02:27,800 --> 00:02:33,920 It's amazing Psi Agents realized that traditional silk processing uses a lot of energy. 61 00:02:33,920 --> 00:02:39,320 So it suggested this new composite that could be made using a low energy process. 62 00:02:39,320 --> 00:02:40,320 And what's the result? 63 00:02:40,320 --> 00:02:45,200 A material that could be even stronger than regular silk. 64 00:02:45,200 --> 00:02:48,400 Beautifully colored from the pigments and much more sustainable. 65 00:02:48,400 --> 00:02:50,000 Like a super silk. 66 00:02:50,000 --> 00:02:51,000 But eco-friendly. 67 00:02:51,000 --> 00:02:52,000 Exactly. 68 00:02:52,000 --> 00:02:57,040 And there's more Psi Agents even speculated about other potential properties. 69 00:02:57,040 --> 00:02:59,200 Like self-healing capabilities. 70 00:02:59,200 --> 00:03:04,040 It's amazing how it goes beyond just combining ideas and starts to think about completely 71 00:03:04,040 --> 00:03:05,400 new possibilities. 72 00:03:05,400 --> 00:03:06,560 That's incredible. 73 00:03:06,560 --> 00:03:09,360 So how does Psi Agents come up with these ideas? 74 00:03:09,360 --> 00:03:11,480 Does it just randomly put concepts together? 75 00:03:11,480 --> 00:03:12,480 Not at all. 76 00:03:12,480 --> 00:03:15,360 It looks at the knowledge graphs systematically. 77 00:03:15,360 --> 00:03:19,560 Using relationships between different concepts and research findings. 78 00:03:19,560 --> 00:03:26,000 So in this case, it found research on strengthening silk with other materials. 79 00:03:26,000 --> 00:03:29,280 And connected that with research on dandelion pigments. 80 00:03:29,280 --> 00:03:31,000 So it's not random. 81 00:03:31,000 --> 00:03:32,000 It's a logical process. 82 00:03:32,000 --> 00:03:33,000 It is. 83 00:03:33,000 --> 00:03:34,000 It's data driven. 84 00:03:34,000 --> 00:03:35,600 So it's not replacing scientists then. 85 00:03:35,600 --> 00:03:36,600 No. 86 00:03:36,600 --> 00:03:37,600 It's giving them a new tool. 87 00:03:37,600 --> 00:03:38,600 Exactly. 88 00:03:38,600 --> 00:03:41,240 To explore ideas and make connections they might not have seen. 89 00:03:41,240 --> 00:03:42,240 Right. 90 00:03:42,240 --> 00:03:47,520 But it can process huge amounts of information much faster than any human team and find those 91 00:03:47,520 --> 00:03:48,760 hidden connections. 92 00:03:48,760 --> 00:03:50,120 But it's still just a tool. 93 00:03:50,120 --> 00:03:51,120 Yes. 94 00:03:51,120 --> 00:03:52,120 The ideas need to be tested. 95 00:03:52,120 --> 00:03:53,120 Of course. 96 00:03:53,120 --> 00:03:54,320 And validated in the real world. 97 00:03:54,320 --> 00:03:55,320 Absolutely. 98 00:03:55,320 --> 00:03:56,680 That's important. 99 00:03:56,680 --> 00:04:00,440 Some of the ideas might be great, but others might not be possible with the technology 100 00:04:00,440 --> 00:04:01,440 we have now. 101 00:04:01,440 --> 00:04:05,160 So it's like a blueprint for a futuristic invention. 102 00:04:05,160 --> 00:04:08,840 We might have the idea, but not the tech to build it yet. 103 00:04:08,840 --> 00:04:09,840 Exactly. 104 00:04:09,840 --> 00:04:16,080 Even if an idea is ahead of its time, it still shows us where we could go with future research. 105 00:04:16,080 --> 00:04:17,080 So it's like a guide. 106 00:04:17,080 --> 00:04:18,080 Yeah. 107 00:04:18,080 --> 00:04:19,080 More than a rule book. 108 00:04:19,080 --> 00:04:20,080 I like that. 109 00:04:20,080 --> 00:04:22,120 And it's not just about brainstorming ideas. 110 00:04:22,120 --> 00:04:25,880 The paper talks about how important critical review is. 111 00:04:25,880 --> 00:04:31,520 In both the pre-programmed and autonomous versions, there's a critic agent. 112 00:04:31,520 --> 00:04:35,320 This agent's job is to really look at the proposed research. 113 00:04:35,320 --> 00:04:37,040 So it's like quality control. 114 00:04:37,040 --> 00:04:38,040 Exactly. 115 00:04:38,040 --> 00:04:42,000 It looks for weaknesses in the idea. 116 00:04:42,000 --> 00:04:46,000 Potential problems with how they want to do the research and areas where the research 117 00:04:46,000 --> 00:04:47,000 could be better. 118 00:04:47,000 --> 00:04:49,160 So it's not just about wild ideas. 119 00:04:49,160 --> 00:04:50,880 It's about making sure they're good ideas. 120 00:04:50,880 --> 00:04:51,880 Yes. 121 00:04:51,880 --> 00:04:53,080 It's about using good science. 122 00:04:53,080 --> 00:04:54,480 That's where the real power is then. 123 00:04:54,480 --> 00:04:55,880 I think so. 124 00:04:55,880 --> 00:04:58,920 It combines creativity with a critical approach. 125 00:04:58,920 --> 00:05:00,240 Just like the best scientists. 126 00:05:00,240 --> 00:05:02,160 You keep comparing it to human scientists. 127 00:05:02,160 --> 00:05:03,160 Yeah. 128 00:05:03,160 --> 00:05:07,560 It seems like the researchers really want to make it clear that CyAgence is a tool. 129 00:05:07,560 --> 00:05:08,560 A replacement. 130 00:05:08,560 --> 00:05:09,560 Absolutely. 131 00:05:09,560 --> 00:05:14,280 They know that AI can't be a black box for it to be helpful in science. 132 00:05:14,280 --> 00:05:19,080 It needs to be transparent, explainable, and accountable. 133 00:05:19,080 --> 00:05:21,680 So it's about humans and AI working together. 134 00:05:21,680 --> 00:05:22,680 Yes. 135 00:05:22,680 --> 00:05:24,000 A partnership. 136 00:05:24,000 --> 00:05:25,160 Not a takeover. 137 00:05:25,160 --> 00:05:33,280 And speaking of bioinspiration, the way CyAgence works is pretty bioinspired itself. 138 00:05:33,280 --> 00:05:34,280 That's true. 139 00:05:34,280 --> 00:05:35,280 How so? 140 00:05:35,280 --> 00:05:39,800 The researchers said their multi-agent system is similar to swarm intelligence. 141 00:05:39,800 --> 00:05:40,800 Oh, yeah. 142 00:05:40,800 --> 00:05:45,840 It's like how biological systems solve problems through working together and adapting. 143 00:05:45,840 --> 00:05:49,360 So not just the results are bioinspired, but how it works is too. 144 00:05:49,360 --> 00:05:50,360 Exactly. 145 00:05:50,360 --> 00:05:52,360 The agents interact dynamically. 146 00:05:52,360 --> 00:05:55,800 Like a colony of ants or a flock of birds. 147 00:05:55,800 --> 00:06:00,320 Each agent shares what it knows and they adapt their behavior based on feedback from the 148 00:06:00,320 --> 00:06:01,320 system. 149 00:06:01,320 --> 00:06:04,560 It makes you wonder if we can create better AI. 150 00:06:04,560 --> 00:06:07,280 By studying how nature solves problems. 151 00:06:07,280 --> 00:06:08,280 Absolutely. 152 00:06:08,280 --> 00:06:11,080 The researchers see CyAgence as a first step. 153 00:06:11,080 --> 00:06:16,360 Showing what's possible when we combine AI with our understanding of nature. 154 00:06:16,360 --> 00:06:18,160 So where do we go from here? 155 00:06:18,160 --> 00:06:21,800 What are the next steps for CyAgence in AI-driven discovery? 156 00:06:21,800 --> 00:06:24,800 Well, the researchers have some exciting ideas. 157 00:06:24,800 --> 00:06:28,920 They think that using CyAgence with simulation tools would be huge. 158 00:06:28,920 --> 00:06:30,600 What do you mean? 159 00:06:30,600 --> 00:06:36,840 Even if the AI could not only come up with ideas, but also run virtual experiments. 160 00:06:36,840 --> 00:06:38,040 To test them out. 161 00:06:38,040 --> 00:06:39,040 Exactly. 162 00:06:39,040 --> 00:06:40,040 Like a virtual lab. 163 00:06:40,040 --> 00:06:41,040 That's cool. 164 00:06:41,040 --> 00:06:45,520 Where CyAgence can try out and improve its ideas before even meeting a real lab. 165 00:06:45,520 --> 00:06:49,320 And they also talk about making a universal knowledge graph. 166 00:06:49,320 --> 00:06:50,320 What would that do? 167 00:06:50,320 --> 00:06:53,800 It would connect ideas from different areas of science. 168 00:06:53,800 --> 00:06:56,840 So CyAgence could find even more unexpected connections. 169 00:06:56,840 --> 00:06:59,320 So it's not just about bioinspired materials. 170 00:06:59,320 --> 00:07:00,440 Right. 171 00:07:00,440 --> 00:07:03,720 We could be talking about new laws of physics. 172 00:07:03,720 --> 00:07:04,720 Wow. 173 00:07:04,720 --> 00:07:06,160 New medicines. 174 00:07:06,160 --> 00:07:08,880 Even solutions to some of the biggest problems we face. 175 00:07:08,880 --> 00:07:09,880 It's pretty amazing. 176 00:07:09,880 --> 00:07:12,000 But of course there's still more to do. 177 00:07:12,000 --> 00:07:13,000 Like what? 178 00:07:13,000 --> 00:07:15,520 We need to keep improving these AI systems. 179 00:07:15,520 --> 00:07:16,520 I don't know. 180 00:07:16,520 --> 00:07:19,480 Make them stronger, more transparent and accountable. 181 00:07:19,480 --> 00:07:21,400 And we need to make sure they're used ethically. 182 00:07:21,400 --> 00:07:22,400 Right. 183 00:07:22,400 --> 00:07:23,400 For the good of everyone. 184 00:07:23,400 --> 00:07:24,400 Exactly. 185 00:07:24,400 --> 00:07:28,520 This research feels like a big step forward for AI-driven science. 186 00:07:28,520 --> 00:07:34,040 It shows that the future of science isn't just about what humans can do. 187 00:07:34,040 --> 00:07:37,040 It's about using AI to help us understand the universe. 188 00:07:37,040 --> 00:07:38,440 And our place in it. 189 00:07:38,440 --> 00:07:39,960 That's true. 190 00:07:39,960 --> 00:07:41,440 This research really makes you think. 191 00:07:41,440 --> 00:07:42,440 What do you mean? 192 00:07:42,440 --> 00:07:45,280 Well, it brings up some deep questions. 193 00:07:45,280 --> 00:07:46,280 Okay. 194 00:07:46,280 --> 00:07:47,280 About creativity. 195 00:07:47,280 --> 00:07:50,320 About how chance discoveries happen. 196 00:07:50,320 --> 00:07:53,640 Even about the connection between humans and AI. 197 00:07:53,640 --> 00:07:55,680 It's like this research isn't just about science. 198 00:07:55,680 --> 00:07:56,680 Right. 199 00:07:56,680 --> 00:07:58,320 It's about what it means to be human. 200 00:07:58,320 --> 00:08:01,040 In a world that's changing because of AI. 201 00:08:01,040 --> 00:08:02,040 Yeah. 202 00:08:02,040 --> 00:08:04,800 So to our listener, we'll leave you with this. 203 00:08:04,800 --> 00:08:10,080 If AI can help us understand nature, what can we learn about ourselves along the way? 204 00:08:10,080 --> 00:08:11,200 That's a good question. 205 00:08:11,200 --> 00:08:16,240 What discoveries might be possible when we combine human and artificial intelligence? 206 00:08:16,240 --> 00:08:18,120 Those are questions worth exploring. 207 00:08:18,120 --> 00:08:22,960 And with tools like sci-agents, we might find the answer sooner than we think. 208 00:08:22,960 --> 00:08:24,480 There is another interesting one. 209 00:08:24,480 --> 00:08:25,480 Okay. 210 00:08:25,480 --> 00:08:28,480 So, we have transfer performance and ramphothaca. 211 00:08:28,480 --> 00:08:29,480 Ramphothaca. 212 00:08:29,480 --> 00:08:30,480 What's that? 213 00:08:30,480 --> 00:08:32,160 That's the beak of a bird. 214 00:08:32,160 --> 00:08:33,160 Oh, okay. 215 00:08:33,160 --> 00:08:35,000 So, bird beaks and heat transfer. 216 00:08:35,000 --> 00:08:36,000 Yeah. 217 00:08:36,000 --> 00:08:37,000 Pretty interesting. 218 00:08:37,000 --> 00:08:38,000 It's a strange combination. 219 00:08:38,000 --> 00:08:39,720 Sci-agents looked at the knowledge graph. 220 00:08:39,720 --> 00:08:40,720 Okay. 221 00:08:40,720 --> 00:08:44,840 And suggested making biomimetic microfluidic chips. 222 00:08:44,840 --> 00:08:45,840 Okay. 223 00:08:45,840 --> 00:08:46,920 Based on bird beaks. 224 00:08:46,920 --> 00:08:48,160 How would that work? 225 00:08:48,160 --> 00:08:52,160 The goal was to improve how well these chips transfer heat. 226 00:08:52,160 --> 00:08:55,680 Which could be really useful for things like biomedical applications. 227 00:08:55,680 --> 00:09:00,400 So, they're taking inspiration from how bird beaks regulate temperature. 228 00:09:00,400 --> 00:09:01,400 Exactly. 229 00:09:01,400 --> 00:09:04,240 And using those same ideas to design microchips. 230 00:09:04,240 --> 00:09:05,240 Right. 231 00:09:05,240 --> 00:09:07,280 It even suggested ways to make them... 232 00:09:07,280 --> 00:09:08,280 Like what? 233 00:09:08,280 --> 00:09:09,280 Like soft lithography. 234 00:09:09,280 --> 00:09:10,280 Okay. 235 00:09:10,280 --> 00:09:11,280 And it predicted some benefits. 236 00:09:11,280 --> 00:09:12,280 What kind of benefits? 237 00:09:12,280 --> 00:09:14,480 Like better mechanical stability. 238 00:09:14,480 --> 00:09:15,480 And biocompatibility. 239 00:09:15,480 --> 00:09:17,880 So, it's not just copying nature. 240 00:09:17,880 --> 00:09:18,880 Right. 241 00:09:18,880 --> 00:09:20,040 It's understanding how it works. 242 00:09:20,040 --> 00:09:21,040 Yes. 243 00:09:21,040 --> 00:09:23,360 And it's playing those principles in a new way. 244 00:09:23,360 --> 00:09:24,360 Exactly. 245 00:09:24,360 --> 00:09:25,560 And that's just one example. 246 00:09:25,560 --> 00:09:26,560 Oh, right. 247 00:09:26,560 --> 00:09:28,120 There are a bunch of different concept pairs. 248 00:09:28,120 --> 00:09:29,120 Yeah. 249 00:09:29,120 --> 00:09:31,480 The researchers tested Sci-agents with a bunch. 250 00:09:31,480 --> 00:09:32,680 And what did it come up with? 251 00:09:32,680 --> 00:09:34,080 Some pretty wild ideas. 252 00:09:34,080 --> 00:09:35,080 Yeah. 253 00:09:35,080 --> 00:09:36,080 Like what? 254 00:09:36,080 --> 00:09:38,160 Things like materials based on collagen. 255 00:09:38,160 --> 00:09:39,160 Uh-huh. 256 00:09:39,160 --> 00:09:40,520 For better crash-worthiness. 257 00:09:40,520 --> 00:09:41,520 Right. 258 00:09:41,520 --> 00:09:43,800 Super hydrophobic coatings. 259 00:09:43,800 --> 00:09:44,800 What are those? 260 00:09:44,800 --> 00:09:46,280 They're inspired by Nacre. 261 00:09:46,280 --> 00:09:47,280 Nacre? 262 00:09:47,280 --> 00:09:48,280 Yeah. 263 00:09:48,280 --> 00:09:49,280 You know, Mother of Pearl. 264 00:09:49,280 --> 00:09:50,280 Oh, okay. 265 00:09:50,280 --> 00:09:52,160 They're even bioelectronic devices. 266 00:09:52,160 --> 00:09:53,160 Oh. 267 00:09:53,160 --> 00:09:55,240 Using graphene and amyloid fibrils. 268 00:09:55,240 --> 00:09:57,160 Sounds like there are tons of possibilities. 269 00:09:57,160 --> 00:09:58,160 Yeah. 270 00:09:58,160 --> 00:09:59,160 Seems like it. 271 00:09:59,160 --> 00:10:02,680 But you said earlier that not all the ideas are actually possible yet. 272 00:10:02,680 --> 00:10:03,680 Yeah. 273 00:10:03,680 --> 00:10:04,680 That's true. 274 00:10:04,680 --> 00:10:05,680 How does Sci-agents deal with that? 275 00:10:05,680 --> 00:10:10,960 Well, the paper acknowledges that some of the ideas might need like really advanced 276 00:10:10,960 --> 00:10:11,960 technology. 277 00:10:11,960 --> 00:10:12,960 That doesn't exist yet. 278 00:10:12,960 --> 00:10:13,960 Right. 279 00:10:13,960 --> 00:10:15,320 Before we can actually make them. 280 00:10:15,320 --> 00:10:16,480 So it's like having a blueprint. 281 00:10:16,480 --> 00:10:17,480 Yeah. 282 00:10:17,480 --> 00:10:18,480 For something we don't know how to build. 283 00:10:18,480 --> 00:10:19,480 Exactly. 284 00:10:19,480 --> 00:10:21,280 We have the idea. 285 00:10:21,280 --> 00:10:22,680 But not the tools yet. 286 00:10:22,680 --> 00:10:24,480 But those ideas could still be useful. 287 00:10:24,480 --> 00:10:25,480 Definitely. 288 00:10:25,480 --> 00:10:28,480 Even if they're ahead of their time, they give scientists a direction. 289 00:10:28,480 --> 00:10:29,480 Okay. 290 00:10:29,480 --> 00:10:30,480 A place to start their research. 291 00:10:30,480 --> 00:10:31,480 So it's more of a guide. 292 00:10:31,480 --> 00:10:32,480 Yes. 293 00:10:32,480 --> 00:10:33,480 Than a rule book. 294 00:10:33,480 --> 00:10:34,480 Right. 295 00:10:34,480 --> 00:10:35,480 And it's not just about coming up with ideas. 296 00:10:35,480 --> 00:10:36,480 Okay. 297 00:10:36,480 --> 00:10:40,200 The paper also talks about how important it is to have a critical review. 298 00:10:40,200 --> 00:10:44,680 So with both versions of Sci-agents, there's a critic agent. 299 00:10:44,680 --> 00:10:45,680 Interesting. 300 00:10:45,680 --> 00:10:46,680 What does that do? 301 00:10:46,680 --> 00:10:50,480 The first job is to look closely at the research that's being proposed. 302 00:10:50,480 --> 00:10:51,480 Oh. 303 00:10:51,480 --> 00:10:52,880 So it's like quality control. 304 00:10:52,880 --> 00:10:53,880 Yeah. 305 00:10:53,880 --> 00:10:54,880 Exactly. 306 00:10:54,880 --> 00:10:55,880 The critic looks for weak spots. 307 00:10:55,880 --> 00:10:56,880 Okay. 308 00:10:56,880 --> 00:11:01,280 Problems with the methods and areas where the research could be improved. 309 00:11:01,280 --> 00:11:03,680 So it's not just about having any idea. 310 00:11:03,680 --> 00:11:04,680 Right. 311 00:11:04,680 --> 00:11:06,080 It's about making sure it's a good idea. 312 00:11:06,080 --> 00:11:07,080 Exactly. 313 00:11:07,080 --> 00:11:09,480 It's about having a strong scientific foundation. 314 00:11:09,480 --> 00:11:11,800 That's where the real strength of Sci-agents comes in then. 315 00:11:11,800 --> 00:11:12,800 I think so. 316 00:11:12,800 --> 00:11:13,800 It's the balance. 317 00:11:13,800 --> 00:11:14,800 Between creativity. 318 00:11:14,800 --> 00:11:15,800 Yeah. 319 00:11:15,800 --> 00:11:17,200 And being critical. 320 00:11:17,200 --> 00:11:18,200 Like a good scientist. 321 00:11:18,200 --> 00:11:19,200 Exactly. 322 00:11:19,200 --> 00:11:22,560 It's interesting that the paper keeps coming back to that comparison. 323 00:11:22,560 --> 00:11:25,680 Between Sci-agents and human scientists. 324 00:11:25,680 --> 00:11:32,240 It seems like the researchers are really trying to emphasize that Sci-agents is a tool. 325 00:11:32,240 --> 00:11:33,240 Not a replacement. 326 00:11:33,240 --> 00:11:34,240 Absolutely. 327 00:11:34,240 --> 00:11:36,600 They understand that AI can't be a mystery. 328 00:11:36,600 --> 00:11:38,280 It can't be a black box. 329 00:11:38,280 --> 00:11:39,280 Right. 330 00:11:39,280 --> 00:11:43,400 For it to be truly helpful in science, it needs to be transparent. 331 00:11:43,400 --> 00:11:45,560 We need to be able to understand it. 332 00:11:45,560 --> 00:11:47,280 And it needs to be accountable. 333 00:11:47,280 --> 00:11:49,960 So it's about humans and AI working together. 334 00:11:49,960 --> 00:11:50,960 Yes. 335 00:11:50,960 --> 00:11:51,960 It's a collaboration. 336 00:11:51,960 --> 00:11:52,960 Yeah. 337 00:11:52,960 --> 00:11:53,960 Not a replacement. 338 00:11:53,960 --> 00:11:54,960 And speaking of bio-inspiration. 339 00:11:54,960 --> 00:11:55,960 Yeah. 340 00:11:55,960 --> 00:12:00,040 The way Sci-agents works is actually pretty bio-inspired itself. 341 00:12:00,040 --> 00:12:01,040 It is. 342 00:12:01,040 --> 00:12:02,040 How so? 343 00:12:02,040 --> 00:12:07,800 Well, the researchers said that their multi-agent system is similar to swarm intelligence. 344 00:12:07,800 --> 00:12:08,800 Interesting. 345 00:12:08,800 --> 00:12:13,760 It's like how biological systems solve complex problems by working together and adapting. 346 00:12:13,760 --> 00:12:16,120 So it's not just the results that are bio-inspired. 347 00:12:16,120 --> 00:12:17,120 Right. 348 00:12:17,120 --> 00:12:19,120 But the way Sci-agents works is too. 349 00:12:19,120 --> 00:12:20,120 Exactly. 350 00:12:20,120 --> 00:12:21,120 It's like a swarm. 351 00:12:21,120 --> 00:12:22,120 Okay. 352 00:12:22,120 --> 00:12:23,120 The agents are all interacting. 353 00:12:23,120 --> 00:12:24,720 A colony of ants. 354 00:12:24,720 --> 00:12:26,440 Or a flock of birds. 355 00:12:26,440 --> 00:12:28,360 Each agent has its own expertise. 356 00:12:28,360 --> 00:12:29,360 Okay. 357 00:12:29,360 --> 00:12:33,120 And they change their behavior based on feedback from the system. 358 00:12:33,120 --> 00:12:34,320 So it's adapting all the time. 359 00:12:34,320 --> 00:12:35,320 Yeah. 360 00:12:35,320 --> 00:12:36,680 It's constantly learning and changing. 361 00:12:36,680 --> 00:12:40,120 It makes you wonder if we can create even better AI systems. 362 00:12:40,120 --> 00:12:41,120 I think so. 363 00:12:41,120 --> 00:12:43,000 By studying how nature does things. 364 00:12:43,000 --> 00:12:44,000 Yeah. 365 00:12:44,000 --> 00:12:45,120 That's what the researchers think too. 366 00:12:45,120 --> 00:12:47,240 They see Sci-agents as a first step. 367 00:12:47,240 --> 00:12:48,440 A proof of concept. 368 00:12:48,440 --> 00:12:49,440 Right. 369 00:12:49,440 --> 00:12:54,480 It shows what's possible when we combine AI with a deep understanding of nature's design 370 00:12:54,480 --> 00:12:55,480 principles. 371 00:12:55,480 --> 00:12:56,480 So where do we go from here? 372 00:12:56,480 --> 00:12:58,320 That's the big question. 373 00:12:58,320 --> 00:13:01,320 What are the next steps for Sci-agents? 374 00:13:01,320 --> 00:13:05,720 And for AI-driven discovery in general, the researchers have some interesting ideas for 375 00:13:05,720 --> 00:13:06,880 the future. 376 00:13:06,880 --> 00:13:07,880 Like what? 377 00:13:07,880 --> 00:13:11,560 They think that combining Sci-agents with simulation tools would be a big step. 378 00:13:11,560 --> 00:13:13,800 So the AI could actually run experiments. 379 00:13:13,800 --> 00:13:14,800 Exactly. 380 00:13:14,800 --> 00:13:16,240 It could test out its ideas virtually. 381 00:13:16,240 --> 00:13:18,000 Kind of like a virtual laboratory. 382 00:13:18,000 --> 00:13:19,000 Yeah. 383 00:13:19,000 --> 00:13:20,000 Before even needing a real lab. 384 00:13:20,000 --> 00:13:21,840 That would be amazing. 385 00:13:21,840 --> 00:13:25,120 And they also talk about creating a universal knowledge graph. 386 00:13:25,120 --> 00:13:26,120 Okay. 387 00:13:26,120 --> 00:13:27,120 What would that be? 388 00:13:27,120 --> 00:13:30,720 Well, it would connect information from all different areas of science. 389 00:13:30,720 --> 00:13:32,880 So Sci-agents could make even more connections. 390 00:13:32,880 --> 00:13:33,880 Right. 391 00:13:33,880 --> 00:13:34,960 Connections that we might never think of. 392 00:13:34,960 --> 00:13:37,760 So it's not just about bio-inspired materials anymore? 393 00:13:37,760 --> 00:13:38,760 No. 394 00:13:38,760 --> 00:13:39,760 Not at all. 395 00:13:39,760 --> 00:13:42,120 We're talking about discovering new laws of physics. 396 00:13:42,120 --> 00:13:43,120 Wow. 397 00:13:43,120 --> 00:13:44,120 Or designing new medicines. 398 00:13:44,120 --> 00:13:45,120 Mm-hmm. 399 00:13:45,120 --> 00:13:48,440 Maybe even finding solutions to some of the world's biggest problems. 400 00:13:48,440 --> 00:13:50,280 That's pretty incredible. 401 00:13:50,280 --> 00:13:53,640 But like you said before, there's still work to be done. 402 00:13:53,640 --> 00:13:54,800 Of course. 403 00:13:54,800 --> 00:13:55,800 There always is. 404 00:13:55,800 --> 00:13:58,480 We need to keep making these AI systems better. 405 00:13:58,480 --> 00:13:59,480 Okay. 406 00:13:59,480 --> 00:14:00,480 So making them more reliable? 407 00:14:00,480 --> 00:14:01,480 Yes. 408 00:14:01,480 --> 00:14:02,480 Yes. 409 00:14:02,480 --> 00:14:03,480 More transparent too. 410 00:14:03,480 --> 00:14:04,480 Yeah. 411 00:14:04,480 --> 00:14:05,480 So we can understand how they work. 412 00:14:05,480 --> 00:14:06,480 And accountable. 413 00:14:06,480 --> 00:14:07,480 Right. 414 00:14:07,480 --> 00:14:08,480 We need to make sure they're used for good. 415 00:14:08,480 --> 00:14:09,480 For everyone's benefit. 416 00:14:09,480 --> 00:14:10,480 Exactly. 417 00:14:10,480 --> 00:14:12,680 This research feels like a really important step forward. 418 00:14:12,680 --> 00:14:13,680 I agree. 419 00:14:13,680 --> 00:14:15,120 It shows the potential of AI. 420 00:14:15,120 --> 00:14:17,000 For helping us understand the world around us. 421 00:14:17,000 --> 00:14:20,000 And maybe even ourselves. 422 00:14:20,000 --> 00:14:22,000 This research brings up some deep questions. 423 00:14:22,000 --> 00:14:24,000 What do you mean? 424 00:14:24,000 --> 00:14:27,880 Well, it makes you think about what creativity really is. 425 00:14:27,880 --> 00:14:28,880 Mm-hmm. 426 00:14:28,880 --> 00:14:31,960 And the role of chance in making discoveries. 427 00:14:31,960 --> 00:14:36,280 And it even makes you think about the relationship between humans and AI. 428 00:14:36,280 --> 00:14:37,840 It's about more than just science then. 429 00:14:37,840 --> 00:14:38,840 I think so. 430 00:14:38,840 --> 00:14:39,840 Yeah. 431 00:14:39,840 --> 00:14:42,000 It's about what it means to be human in a world with AI. 432 00:14:42,000 --> 00:14:44,160 That's a good point. 433 00:14:44,160 --> 00:14:47,640 So to our listener, we'll leave you with this thought. 434 00:14:47,640 --> 00:14:48,640 Okay. 435 00:14:48,640 --> 00:14:50,760 If AI can help us unlock the secrets of nature. 436 00:14:50,760 --> 00:14:51,760 Yeah. 437 00:14:51,760 --> 00:14:54,120 What might we learn about ourselves along the way? 438 00:14:54,120 --> 00:14:56,640 That's a fascinating question to consider. 439 00:14:56,640 --> 00:15:02,400 What incredible discoveries might be waiting for us when we combine human intelligence 440 00:15:02,400 --> 00:15:04,320 with artificial intelligence? 441 00:15:04,320 --> 00:15:05,960 It's exciting to think about. 442 00:15:05,960 --> 00:15:07,880 And with tools like Psi Agents. 443 00:15:07,880 --> 00:15:08,880 Yeah. 444 00:15:08,880 --> 00:15:11,600 We might be on the verge of something truly amazing. 445 00:15:11,600 --> 00:15:14,160 Thanks for joining us on this deep dive into Psi Agents. 446 00:15:14,160 --> 00:15:41,720 It's been a pleasure.