1 00:00:00,000 --> 00:00:05,000 All right, so get ready because today we're going deep into a paper 2 00:00:05,000 --> 00:00:05,840 Yeah. 3 00:00:05,840 --> 00:00:09,840 that's making some serious waves in the AI world. 4 00:00:09,840 --> 00:00:10,440 It is. 5 00:00:10,440 --> 00:00:14,120 It's called Relational Neurosymbolic Markov Models. 6 00:00:14,120 --> 00:00:14,960 Wow. 7 00:00:14,960 --> 00:00:15,640 Yeah. 8 00:00:15,640 --> 00:00:19,280 And I know it sounds a little intimidating. 9 00:00:19,280 --> 00:00:19,780 It does. 10 00:00:19,780 --> 00:00:20,880 Something on a Star Trek. 11 00:00:20,880 --> 00:00:22,080 Yeah, a little bit, right? 12 00:00:22,080 --> 00:00:25,880 But stick with us because the ideas here 13 00:00:25,880 --> 00:00:27,280 are worth wrapping your head around. 14 00:00:27,280 --> 00:00:29,480 I think what's exciting about this paper 15 00:00:29,480 --> 00:00:32,440 is that it really gets at a fundamental challenge in AI, 16 00:00:32,440 --> 00:00:36,640 which is how do we build systems that can both learn 17 00:00:36,640 --> 00:00:39,920 from data, but also reason logically? 18 00:00:39,920 --> 00:00:43,360 Yeah, and we've talked about symbolic AI a lot on this show. 19 00:00:43,360 --> 00:00:47,960 And this idea of logic versus just brute force pattern 20 00:00:47,960 --> 00:00:48,560 recognition. 21 00:00:48,560 --> 00:00:49,760 Yeah, exactly. 22 00:00:49,760 --> 00:00:51,720 So this is kind of getting into that. 23 00:00:51,720 --> 00:00:54,320 Now, I know Markov models might sound a little scary. 24 00:00:54,320 --> 00:00:54,920 Oh, yeah. 25 00:00:54,920 --> 00:00:57,640 But you've probably encountered them 26 00:00:57,640 --> 00:00:59,000 without even realizing it. 27 00:00:59,000 --> 00:00:59,600 Yeah. 28 00:00:59,600 --> 00:01:03,360 Think about things like weather prediction apps, 29 00:01:03,360 --> 00:01:05,440 in short, or even your phone's auto correct. 30 00:01:05,440 --> 00:01:06,240 Yeah. 31 00:01:06,240 --> 00:01:11,400 They use Markov models to predict the next word based 32 00:01:11,400 --> 00:01:12,160 on the current word. 33 00:01:12,160 --> 00:01:12,760 That's right. 34 00:01:12,760 --> 00:01:17,320 And that core idea predicting the future based on the present 35 00:01:17,320 --> 00:01:19,200 is surprisingly powerful. 36 00:01:19,200 --> 00:01:21,960 It underpins so many AI applications. 37 00:01:21,960 --> 00:01:23,120 Yeah, it's really cool. 38 00:01:23,120 --> 00:01:25,280 And I think that's why a lot of people 39 00:01:25,280 --> 00:01:27,720 get a little intimidated by it because they're like, wait, 40 00:01:27,720 --> 00:01:28,920 you could predict the future. 41 00:01:28,920 --> 00:01:29,880 Yeah. 42 00:01:29,880 --> 00:01:30,960 How does that even work? 43 00:01:30,960 --> 00:01:32,840 Right, but in a limited sense, right? 44 00:01:32,840 --> 00:01:33,560 Right, exactly. 45 00:01:33,560 --> 00:01:35,280 Like we're not talking about like, you know. 46 00:01:35,280 --> 00:01:36,080 And travel. 47 00:01:36,080 --> 00:01:37,280 Yeah, exactly. 48 00:01:37,280 --> 00:01:38,480 Yeah, but it's still. 49 00:01:38,480 --> 00:01:40,080 It's still a very powerful idea. 50 00:01:40,080 --> 00:01:43,800 And it's actually pretty intuitive when you think about it. 51 00:01:43,800 --> 00:01:46,840 Yeah, and so what this paper is doing 52 00:01:46,840 --> 00:01:51,840 is taking that simple idea and like turbo charging it. 53 00:01:51,840 --> 00:01:53,280 With this neurosymbolic AI. 54 00:01:53,280 --> 00:01:54,840 Yeah, this neurosymbolic AI. 55 00:01:54,840 --> 00:01:57,120 Which is this really fascinating field 56 00:01:57,120 --> 00:01:58,680 where we're trying to bring together. 57 00:01:58,680 --> 00:01:59,180 Right. 58 00:01:59,180 --> 00:02:02,720 These two really powerful approaches to AI. 59 00:02:02,720 --> 00:02:04,800 Neural networks, which are all about learning from data. 60 00:02:04,800 --> 00:02:05,640 Right. 61 00:02:05,640 --> 00:02:10,480 And symbolic reasoning, which is about using logic and rules. 62 00:02:10,480 --> 00:02:14,440 Yeah, so instead of just mimicking patterns, 63 00:02:14,440 --> 00:02:17,040 they're actually trying to like understand 64 00:02:17,040 --> 00:02:18,920 the why behind their actions. 65 00:02:18,920 --> 00:02:21,480 Exactly, and that's where these TMMs, 66 00:02:21,480 --> 00:02:23,800 Relational Neurosymbolic Markov Models come in. 67 00:02:23,800 --> 00:02:24,560 Nesias. 68 00:02:24,560 --> 00:02:25,040 Yeah. 69 00:02:25,040 --> 00:02:25,360 I love it. 70 00:02:25,360 --> 00:02:26,080 That's a mouthful. 71 00:02:26,080 --> 00:02:26,720 It's catchy. 72 00:02:26,720 --> 00:02:28,120 It is a mouthful. 73 00:02:28,120 --> 00:02:30,600 So they're using this Nesi approach 74 00:02:30,600 --> 00:02:35,080 to make Markov models even smarter. 75 00:02:35,080 --> 00:02:37,960 But like, how does that benefit us? 76 00:02:37,960 --> 00:02:41,160 So imagine a world where you go to the doctor. 77 00:02:41,160 --> 00:02:41,400 OK. 78 00:02:41,400 --> 00:02:43,960 And the doctor's using an AI diagnostic tool. 79 00:02:43,960 --> 00:02:44,560 Right. 80 00:02:44,560 --> 00:02:47,440 And that tool not only gives you a diagnosis, 81 00:02:47,440 --> 00:02:51,160 but also explains its reasoning in clear logical steps. 82 00:02:51,160 --> 00:02:51,800 Yeah. 83 00:02:51,800 --> 00:02:54,200 That's the kind of trust and transparency 84 00:02:54,200 --> 00:02:56,400 that these Nesi MMS could bring to AI. 85 00:02:56,400 --> 00:02:57,760 And that makes a huge difference. 86 00:02:57,760 --> 00:03:02,000 Because if you don't know why, the AI is doing what it's doing. 87 00:03:02,000 --> 00:03:03,120 It's really hard to trust it. 88 00:03:03,120 --> 00:03:04,240 Exactly. 89 00:03:04,240 --> 00:03:06,640 Or think about self-driving cars. 90 00:03:06,640 --> 00:03:07,560 Oh, yeah. 91 00:03:07,560 --> 00:03:11,120 That can not only learn from driving data, 92 00:03:11,120 --> 00:03:15,040 but also understand traffic laws at a logical level. 93 00:03:15,040 --> 00:03:18,520 That leads to safer and more reliable autonomous vehicles. 94 00:03:18,520 --> 00:03:20,080 OK, so this is already blowing my mind. 95 00:03:20,080 --> 00:03:20,880 So let's break it down. 96 00:03:20,880 --> 00:03:21,320 Right. 97 00:03:21,320 --> 00:03:24,000 How do these Nesi MMS actually work? 98 00:03:24,000 --> 00:03:26,720 So think of it as two interconnected parts. 99 00:03:26,720 --> 00:03:27,120 OK. 100 00:03:27,120 --> 00:03:29,160 First, you have these neural networks 101 00:03:29,160 --> 00:03:32,560 that process raw data, like images, 102 00:03:32,560 --> 00:03:34,960 sensor readings, things like that. 103 00:03:34,960 --> 00:03:38,240 These networks act like a translator, 104 00:03:38,240 --> 00:03:42,480 converting that complex data into a simplified symbolic 105 00:03:42,480 --> 00:03:43,840 representation. 106 00:03:43,840 --> 00:03:46,520 So they're kind of like taking this complex data 107 00:03:46,520 --> 00:03:49,360 and turning it into a language that logic can work with. 108 00:03:49,360 --> 00:03:50,120 Exactly. 109 00:03:50,120 --> 00:03:52,040 And that's where the second part comes in. 110 00:03:52,040 --> 00:03:54,640 These symbolic representations are then 111 00:03:54,640 --> 00:03:57,520 fed into this logical reasoning engine 112 00:03:57,520 --> 00:04:01,600 that can apply predefined rules and constraints. 113 00:04:01,600 --> 00:04:04,800 So it's kind of like you're giving the AI a rulebook 114 00:04:04,800 --> 00:04:10,320 and saying, OK, based on what you're seeing and the rules, 115 00:04:10,320 --> 00:04:12,600 what's the most logical action to take? 116 00:04:12,600 --> 00:04:13,720 That's a great way to put it. 117 00:04:13,720 --> 00:04:15,880 So it's learning from experience 118 00:04:15,880 --> 00:04:17,200 through those neural networks. 119 00:04:17,200 --> 00:04:20,880 But it's also applying reason and logic 120 00:04:20,880 --> 00:04:22,560 through that symbolic system. 121 00:04:22,560 --> 00:04:27,080 And the beauty of this approach is that it's probabilistic, 122 00:04:27,080 --> 00:04:29,520 meaning it can handle uncertainty, 123 00:04:29,520 --> 00:04:31,920 make predictions based on probabilities. 124 00:04:31,920 --> 00:04:33,360 Just like we do in the real world. 125 00:04:33,360 --> 00:04:34,400 Just like we do. 126 00:04:34,400 --> 00:04:34,680 All right. 127 00:04:34,680 --> 00:04:36,040 So we've got the basics down. 128 00:04:36,040 --> 00:04:37,000 Yeah. 129 00:04:37,000 --> 00:04:41,840 But to really appreciate the power of these NSEMs, 130 00:04:41,840 --> 00:04:43,680 we need to see them in action, right? 131 00:04:43,680 --> 00:04:44,560 Absolutely. 132 00:04:44,560 --> 00:04:48,760 The researchers put these NSEMs to the test 133 00:04:48,760 --> 00:04:51,800 using a video game called Minnehack, which 134 00:04:51,800 --> 00:04:55,320 is known for its complex rules and really unpredictable 135 00:04:55,320 --> 00:04:56,160 scenarios. 136 00:04:56,160 --> 00:04:57,240 So it's a good testing ground. 137 00:04:57,240 --> 00:04:57,760 Yeah. 138 00:04:57,760 --> 00:05:01,240 They wanted to see how well these models could learn, 139 00:05:01,240 --> 00:05:03,560 adapt, and generalize their knowledge. 140 00:05:03,560 --> 00:05:05,080 Video games and AI. 141 00:05:05,080 --> 00:05:06,440 Always a good combination. 142 00:05:06,440 --> 00:05:07,640 So what do they find? 143 00:05:07,640 --> 00:05:09,960 Well, they did two main experiments, 144 00:05:09,960 --> 00:05:13,440 focusing on two key capabilities of AI, 145 00:05:13,440 --> 00:05:16,120 generation and discrimination. 146 00:05:16,120 --> 00:05:16,360 OK. 147 00:05:16,360 --> 00:05:19,400 So before we dive into that, mind me what those are again. 148 00:05:19,400 --> 00:05:22,560 So generation is all about creating something new. 149 00:05:22,560 --> 00:05:22,960 OK. 150 00:05:22,960 --> 00:05:25,800 Think about composing a piece of music 151 00:05:25,800 --> 00:05:27,240 based on what it's learned. 152 00:05:27,240 --> 00:05:27,640 Yeah. 153 00:05:27,640 --> 00:05:30,280 Discrimination is more about categorizing things. 154 00:05:30,280 --> 00:05:30,780 OK. 155 00:05:30,780 --> 00:05:33,480 Identifying if a picture has a cat or a dog in it. 156 00:05:33,480 --> 00:05:33,960 Got it. 157 00:05:33,960 --> 00:05:38,400 So they were basically testing if these NSEMMs could 158 00:05:38,400 --> 00:05:40,200 create new game scenarios. 159 00:05:40,200 --> 00:05:40,920 Mm-hmm. 160 00:05:40,920 --> 00:05:41,560 Mm-hmm. 161 00:05:41,560 --> 00:05:42,240 That made sense. 162 00:05:42,240 --> 00:05:43,040 Yeah. 163 00:05:43,040 --> 00:05:46,680 And also accurately analyze existing scenarios. 164 00:05:46,680 --> 00:05:47,680 Exactly. 165 00:05:47,680 --> 00:05:48,200 OK. 166 00:05:48,200 --> 00:05:50,280 And what they found was fascinating. 167 00:05:50,280 --> 00:05:51,480 Well, don't leave me hanging. 168 00:05:51,480 --> 00:05:52,080 OK. 169 00:05:52,080 --> 00:05:52,800 What happened? 170 00:05:52,800 --> 00:05:54,960 In the generative experiment, they 171 00:05:54,960 --> 00:05:58,840 tasked the NSEMM with creating new game screens for Minnehack. 172 00:05:58,840 --> 00:05:59,440 OK. 173 00:05:59,440 --> 00:06:02,680 That actually followed the game's very complex rule. 174 00:06:02,680 --> 00:06:05,400 So it's like giving the AI a paint brush and saying, 175 00:06:05,400 --> 00:06:08,640 all right, paint me a plausible Minnehack scene. 176 00:06:08,640 --> 00:06:09,480 Exactly. 177 00:06:09,480 --> 00:06:12,720 And what surprised them was that the NSEMM wasn't just 178 00:06:12,720 --> 00:06:14,800 mimicking what it had seen. 179 00:06:14,800 --> 00:06:16,880 It was generating new game states 180 00:06:16,880 --> 00:06:19,000 that were logically consistent. 181 00:06:19,000 --> 00:06:20,600 That's a big deal. 182 00:06:20,600 --> 00:06:21,280 Yeah, it is. 183 00:06:21,280 --> 00:06:22,760 Because it's one thing to copy. 184 00:06:22,760 --> 00:06:23,200 Right. 185 00:06:23,200 --> 00:06:25,240 It's another to truly understand. 186 00:06:25,240 --> 00:06:26,000 Yeah. 187 00:06:26,000 --> 00:06:27,520 And create something new. 188 00:06:27,520 --> 00:06:28,240 Exactly. 189 00:06:28,240 --> 00:06:32,000 It suggests that the NSEMM was able to internalize 190 00:06:32,000 --> 00:06:35,080 the logic of the game at a much deeper level. 191 00:06:35,080 --> 00:06:35,680 Wow. 192 00:06:35,680 --> 00:06:37,080 Now, the second experiment. 193 00:06:37,080 --> 00:06:37,560 OK. 194 00:06:37,560 --> 00:06:41,960 The discriminative task got even more interesting. 195 00:06:41,960 --> 00:06:42,240 All right. 196 00:06:42,240 --> 00:06:43,280 I'm on the edge of my seat. 197 00:06:43,280 --> 00:06:43,720 OK. 198 00:06:43,720 --> 00:06:44,520 Tell me more. 199 00:06:44,520 --> 00:06:46,240 This time they wanted to see. 200 00:06:46,240 --> 00:06:46,440 Yeah. 201 00:06:46,440 --> 00:06:50,800 If the NSEMM could predict whether a player would die. 202 00:06:50,800 --> 00:06:51,280 OK. 203 00:06:51,280 --> 00:06:53,040 In a given Minnehack sequence. 204 00:06:53,040 --> 00:06:54,880 Ooh, a life or death scenario. 205 00:06:54,880 --> 00:06:55,360 Yeah. 206 00:06:55,360 --> 00:06:59,160 So they just show the AI a bunch of gameplay. 207 00:06:59,160 --> 00:07:02,320 Well, not exactly to make it more challenging. 208 00:07:02,320 --> 00:07:06,520 They only gave the AI the symbolic representation 209 00:07:06,520 --> 00:07:07,640 of the game state. 210 00:07:07,640 --> 00:07:07,960 Hold on. 211 00:07:07,960 --> 00:07:11,120 So they were asking the AI to predict the outcome? 212 00:07:11,120 --> 00:07:11,400 Yeah. 213 00:07:11,400 --> 00:07:12,680 Without even seeing the game. 214 00:07:12,680 --> 00:07:13,280 Right. 215 00:07:13,280 --> 00:07:15,880 That's like trying to figure out who's winning a chess match. 216 00:07:15,880 --> 00:07:16,160 Yeah. 217 00:07:16,160 --> 00:07:17,720 Just by looking at the notation. 218 00:07:17,720 --> 00:07:18,080 Right. 219 00:07:18,080 --> 00:07:19,400 And like the actual board. 220 00:07:19,400 --> 00:07:19,760 OK. 221 00:07:19,760 --> 00:07:20,400 That's tough. 222 00:07:20,400 --> 00:07:20,720 Yeah. 223 00:07:20,720 --> 00:07:22,560 And to make it even tougher. 224 00:07:22,560 --> 00:07:23,160 Yeah. 225 00:07:23,160 --> 00:07:25,840 They intentionally left out some information. 226 00:07:25,840 --> 00:07:26,360 Oh, no. 227 00:07:26,360 --> 00:07:28,720 About how the enemies in the game behaved. 228 00:07:28,720 --> 00:07:30,360 So they're really putting it through the ringer. 229 00:07:30,360 --> 00:07:31,040 Yeah. 230 00:07:31,040 --> 00:07:32,520 Like an incomplete rule book. 231 00:07:32,520 --> 00:07:33,000 Wow. 232 00:07:33,000 --> 00:07:33,440 OK. 233 00:07:33,440 --> 00:07:34,160 So how to do? 234 00:07:34,160 --> 00:07:35,600 Remarkably well. 235 00:07:35,600 --> 00:07:36,200 Really? 236 00:07:36,200 --> 00:07:38,320 Even with this missing information. 237 00:07:38,320 --> 00:07:38,640 OK. 238 00:07:38,640 --> 00:07:42,080 The NSEMMs were able to combine the rules they did know. 239 00:07:42,080 --> 00:07:43,000 Mm-hmm. 240 00:07:43,000 --> 00:07:44,800 With the patterns they'd learned. 241 00:07:44,800 --> 00:07:45,200 Right. 242 00:07:45,200 --> 00:07:46,840 To pretty accurately predict. 243 00:07:46,840 --> 00:07:47,280 Wow. 244 00:07:47,280 --> 00:07:49,080 Whether the player would survive. 245 00:07:49,080 --> 00:07:50,480 So it's like they were able to fill in. 246 00:07:50,480 --> 00:07:50,880 Yeah. 247 00:07:50,880 --> 00:07:52,080 The gaps in their knowledge. 248 00:07:52,080 --> 00:07:52,760 Mm-hmm. 249 00:07:52,760 --> 00:07:54,040 By thinking logically. 250 00:07:54,040 --> 00:07:54,760 Exactly. 251 00:07:54,760 --> 00:07:56,880 Not just relying on what they'd seen before. 252 00:07:56,880 --> 00:07:57,720 That's right. 253 00:07:57,720 --> 00:07:59,280 And to push it even further. 254 00:07:59,280 --> 00:07:59,960 OK. 255 00:07:59,960 --> 00:08:04,280 They tested these models on increasingly complex versions 256 00:08:04,280 --> 00:08:04,960 of the game. 257 00:08:04,960 --> 00:08:05,600 Right. 258 00:08:05,600 --> 00:08:09,080 Changing the map size, the enemy count, the sequence length. 259 00:08:09,080 --> 00:08:10,200 So they really wanted to see. 260 00:08:10,200 --> 00:08:10,600 Yeah. 261 00:08:10,600 --> 00:08:12,720 If it could handle anything they could throw at it. 262 00:08:12,720 --> 00:08:13,920 Exactly. 263 00:08:13,920 --> 00:08:14,960 So what was the verdict? 264 00:08:14,960 --> 00:08:19,400 The NSEMMs consistently outperformed the other models. 265 00:08:19,400 --> 00:08:19,680 OK. 266 00:08:19,680 --> 00:08:23,680 Especially when things got really complex and unpredictable. 267 00:08:23,680 --> 00:08:24,240 Yeah. 268 00:08:24,240 --> 00:08:26,400 They showed this impressive ability 269 00:08:26,400 --> 00:08:28,960 to generalize their knowledge. 270 00:08:28,960 --> 00:08:29,480 Mm-hmm. 271 00:08:29,480 --> 00:08:31,480 And adapt to new situations. 272 00:08:31,480 --> 00:08:33,360 Which is really crucial. 273 00:08:33,360 --> 00:08:33,640 Right. 274 00:08:33,640 --> 00:08:36,440 For any AI system that's going to operate in the real world. 275 00:08:36,440 --> 00:08:36,680 Yeah. 276 00:08:36,680 --> 00:08:38,760 Because the real world is messy. 277 00:08:38,760 --> 00:08:39,520 It is. 278 00:08:39,520 --> 00:08:40,440 It's unpredictable. 279 00:08:40,440 --> 00:08:40,760 It is. 280 00:08:40,760 --> 00:08:41,720 Things change all the time. 281 00:08:41,720 --> 00:08:42,480 That's right. 282 00:08:42,480 --> 00:08:45,440 So you need an AI that can roll with the punches. 283 00:08:45,440 --> 00:08:46,080 Exactly. 284 00:08:46,080 --> 00:08:49,160 And I'm starting to see why this research is generating 285 00:08:49,160 --> 00:08:50,240 so much buzz. 286 00:08:50,240 --> 00:08:51,760 It's really exciting stuff. 287 00:08:51,760 --> 00:08:52,320 It is. 288 00:08:52,320 --> 00:08:55,840 But before we get too carried away with all the potential here. 289 00:08:55,840 --> 00:08:56,640 Mm-hmm. 290 00:08:56,640 --> 00:08:59,440 Let's kind of step back and consider 291 00:08:59,440 --> 00:09:03,160 both the strengths and the limitations of this approach. 292 00:09:03,160 --> 00:09:03,960 Of course. 293 00:09:03,960 --> 00:09:06,760 Because no technology is perfect. 294 00:09:06,760 --> 00:09:08,360 No technology is perfect. 295 00:09:08,360 --> 00:09:08,920 Right. 296 00:09:08,920 --> 00:09:09,360 You're right. 297 00:09:09,360 --> 00:09:10,560 No technology is perfect. 298 00:09:10,560 --> 00:09:11,120 Yeah. 299 00:09:11,120 --> 00:09:11,880 But uh. 300 00:09:11,880 --> 00:09:13,280 But this one's pretty cool. 301 00:09:13,280 --> 00:09:14,160 It is pretty cool. 302 00:09:14,160 --> 00:09:15,400 On the positive side. 303 00:09:15,400 --> 00:09:16,000 Yeah. 304 00:09:16,000 --> 00:09:18,720 These NECMs they've shown. 305 00:09:18,720 --> 00:09:21,760 These remarkable generalization abilities. 306 00:09:21,760 --> 00:09:22,760 Right. 307 00:09:22,760 --> 00:09:25,320 They can handle incomplete information. 308 00:09:25,320 --> 00:09:26,000 Yeah. 309 00:09:26,000 --> 00:09:27,440 Really gracefully. 310 00:09:27,440 --> 00:09:28,240 Which is important. 311 00:09:28,240 --> 00:09:29,280 Which is huge. 312 00:09:29,280 --> 00:09:30,480 In the real world. 313 00:09:30,480 --> 00:09:31,560 Yeah. 314 00:09:31,560 --> 00:09:33,840 And the use of these logical rules. 315 00:09:33,840 --> 00:09:34,680 Mm-hmm. 316 00:09:34,680 --> 00:09:37,240 Make some way more transparent. 317 00:09:37,240 --> 00:09:37,760 Yeah. 318 00:09:37,760 --> 00:09:40,480 Than your typical black box AI. 319 00:09:40,480 --> 00:09:42,040 And that transparency is huge. 320 00:09:42,040 --> 00:09:42,400 Yeah. 321 00:09:42,400 --> 00:09:45,520 Especially if we're talking about AI making decisions. 322 00:09:45,520 --> 00:09:45,720 Mm-hmm. 323 00:09:45,720 --> 00:09:47,040 That affect our lives. 324 00:09:47,040 --> 00:09:47,680 Right. 325 00:09:47,680 --> 00:09:48,120 You know. 326 00:09:48,120 --> 00:09:51,080 It's reassuring to know that we can understand 327 00:09:51,080 --> 00:09:53,440 why the AI is doing what it's doing. 328 00:09:53,440 --> 00:09:54,200 Exactly. 329 00:09:54,200 --> 00:09:56,200 It's not just a magic black box. 330 00:09:56,200 --> 00:09:56,480 Yeah. 331 00:09:56,480 --> 00:09:57,280 No magic. 332 00:09:57,280 --> 00:09:57,480 Yeah. 333 00:09:57,480 --> 00:09:58,280 However. 334 00:09:58,280 --> 00:09:58,960 But. 335 00:09:58,960 --> 00:10:00,480 There are hurdles to overcome. 336 00:10:00,480 --> 00:10:01,160 OK. 337 00:10:01,160 --> 00:10:03,400 The paper itself acknowledges. 338 00:10:03,400 --> 00:10:03,880 Mm-hmm. 339 00:10:03,880 --> 00:10:06,200 This issue of scalability. 340 00:10:06,200 --> 00:10:06,640 Right. 341 00:10:06,640 --> 00:10:07,760 These NECMMs. 342 00:10:07,760 --> 00:10:08,280 Yeah. 343 00:10:08,280 --> 00:10:10,240 At least in their current form. 344 00:10:10,240 --> 00:10:10,640 Mm-hmm. 345 00:10:10,640 --> 00:10:13,600 Can be computationally expensive to train. 346 00:10:13,600 --> 00:10:14,560 So it's kind of like. 347 00:10:14,560 --> 00:10:14,920 Yeah. 348 00:10:14,920 --> 00:10:16,680 Having a brilliant chef. 349 00:10:16,680 --> 00:10:17,240 Oh, right. 350 00:10:17,240 --> 00:10:20,680 Who can create these incredible dishes. 351 00:10:20,680 --> 00:10:21,120 Mm-hmm. 352 00:10:21,120 --> 00:10:23,240 But only in a tiny kitchen. 353 00:10:23,240 --> 00:10:23,640 Yeah. 354 00:10:23,640 --> 00:10:25,160 With limited equipment. 355 00:10:25,160 --> 00:10:26,080 That's a great analogy. 356 00:10:26,080 --> 00:10:27,720 They need a bigger space. 357 00:10:27,720 --> 00:10:28,080 Mm-hmm. 358 00:10:28,080 --> 00:10:29,280 To really shine. 359 00:10:29,280 --> 00:10:30,280 Exactly. 360 00:10:30,280 --> 00:10:33,440 And beyond just the computational challenges. 361 00:10:33,440 --> 00:10:33,800 Yeah. 362 00:10:33,800 --> 00:10:35,960 There's the ongoing task of finding. 363 00:10:35,960 --> 00:10:36,520 Oh. 364 00:10:36,520 --> 00:10:37,840 That sweet spot. 365 00:10:37,840 --> 00:10:38,200 Right. 366 00:10:38,200 --> 00:10:41,280 Between the neural network learning. 367 00:10:41,280 --> 00:10:41,880 Yeah. 368 00:10:41,880 --> 00:10:43,320 And the symbolic reasoning. 369 00:10:43,320 --> 00:10:44,360 It's a balancing act. 370 00:10:44,360 --> 00:10:45,480 It is a balancing act. 371 00:10:45,480 --> 00:10:47,920 It's like finding the perfect recipe for AI. 372 00:10:47,920 --> 00:10:48,240 Right. 373 00:10:48,240 --> 00:10:48,560 Exactly. 374 00:10:48,560 --> 00:10:49,640 But all in all. 375 00:10:49,640 --> 00:10:52,200 It seems like these NECMs. 376 00:10:52,200 --> 00:10:53,200 Mm-hmm. 377 00:10:53,200 --> 00:10:55,600 Are a promising step forward. 378 00:10:55,600 --> 00:10:55,800 Yeah. 379 00:10:55,800 --> 00:10:56,840 In AI research. 380 00:10:56,840 --> 00:10:57,160 I think. 381 00:10:57,160 --> 00:10:59,400 It's not just about making AI smarter. 382 00:10:59,400 --> 00:10:59,640 Right. 383 00:10:59,640 --> 00:11:00,920 It's about making it. 384 00:11:00,920 --> 00:11:02,920 It's about making it more reliable. 385 00:11:02,920 --> 00:11:03,160 Yeah. 386 00:11:03,160 --> 00:11:03,920 More understandable. 387 00:11:03,920 --> 00:11:04,920 We're beaptable. 388 00:11:04,920 --> 00:11:05,160 Yeah. 389 00:11:05,160 --> 00:11:06,400 All the things that we want. 390 00:11:06,400 --> 00:11:07,160 Exactly. 391 00:11:07,160 --> 00:11:10,120 Yeah. That shift from pure pattern recognition. 392 00:11:10,120 --> 00:11:12,400 To AI that can actually reason. 393 00:11:12,400 --> 00:11:12,600 Right. 394 00:11:12,600 --> 00:11:15,560 That's what excites me the most about this research. 395 00:11:15,560 --> 00:11:15,680 No. 396 00:11:15,680 --> 00:11:16,880 I can't help but wonder. 397 00:11:16,880 --> 00:11:17,640 Yeah. 398 00:11:17,640 --> 00:11:20,400 Like what are some real world applications. 399 00:11:20,400 --> 00:11:20,880 Sure. 400 00:11:20,880 --> 00:11:22,120 Beyond video games. 401 00:11:22,120 --> 00:11:22,720 Mm-hmm. 402 00:11:22,720 --> 00:11:25,120 Where these NECMMs could make a difference. 403 00:11:25,120 --> 00:11:26,880 The potential is vast. 404 00:11:26,880 --> 00:11:27,480 Yeah. 405 00:11:27,480 --> 00:11:30,880 One area that's ripe for disruption is robotics. 406 00:11:30,880 --> 00:11:31,600 Oh yeah. 407 00:11:31,600 --> 00:11:33,600 Imagine robots in manufacturing. 408 00:11:33,600 --> 00:11:34,160 Mm-hmm. 409 00:11:34,160 --> 00:11:35,280 Or healthcare. 410 00:11:35,280 --> 00:11:35,760 OK. 411 00:11:35,760 --> 00:11:40,680 That can not only perform these complex tasks. 412 00:11:40,680 --> 00:11:41,040 Right. 413 00:11:41,040 --> 00:11:46,280 But also adapt to unexpected changes in their environment. 414 00:11:46,280 --> 00:11:46,640 Right. 415 00:11:46,640 --> 00:11:47,800 Because that's the thing. 416 00:11:47,800 --> 00:11:48,000 Yeah. 417 00:11:48,000 --> 00:11:50,800 The real world is not a controlled laboratory 418 00:11:50,800 --> 00:11:51,800 where all things happen. 419 00:11:51,800 --> 00:11:52,360 All the time. 420 00:11:52,360 --> 00:11:53,480 You've got to be able to roll with it. 421 00:11:53,480 --> 00:11:54,080 That's right. 422 00:11:54,080 --> 00:11:55,680 And all while adhering. 423 00:11:55,680 --> 00:11:56,160 Yeah. 424 00:11:56,160 --> 00:11:59,440 To a set of like logical safety protocols. 425 00:11:59,440 --> 00:12:00,160 Exactly. 426 00:12:00,160 --> 00:12:03,400 So we have these robots that are not only more versatile. 427 00:12:03,400 --> 00:12:03,840 Right. 428 00:12:03,840 --> 00:12:06,280 More reliable but also safer. 429 00:12:06,280 --> 00:12:06,760 Yeah. 430 00:12:06,760 --> 00:12:08,720 We don't want robots going rogue. 431 00:12:08,720 --> 00:12:09,520 No we don't. 432 00:12:09,520 --> 00:12:13,240 And it's a far cry from those kind of rigid. 433 00:12:13,240 --> 00:12:13,600 Right. 434 00:12:13,600 --> 00:12:16,280 Pre-programmed robots we see today. 435 00:12:16,280 --> 00:12:19,040 It's like going from Arumba to Rosie the Robot. 436 00:12:19,040 --> 00:12:19,880 Exactly. 437 00:12:19,880 --> 00:12:21,480 And speaking of collaboration. 438 00:12:21,480 --> 00:12:22,160 Oh yeah. 439 00:12:22,160 --> 00:12:23,560 What about education. 440 00:12:23,560 --> 00:12:24,720 Oh that's a good one. 441 00:12:24,720 --> 00:12:28,120 Could these NECMMs change how we learn. 442 00:12:28,120 --> 00:12:28,680 I think so. 443 00:12:28,680 --> 00:12:29,360 Absolutely. 444 00:12:29,360 --> 00:12:32,400 We already have AI powered learning platforms. 445 00:12:32,400 --> 00:12:33,040 Right. 446 00:12:33,040 --> 00:12:36,480 But imagine a personalized AI tutor. 447 00:12:36,480 --> 00:12:36,840 OK. 448 00:12:36,840 --> 00:12:39,400 That can understand your unique learning style. 449 00:12:39,400 --> 00:12:39,960 Yeah. 450 00:12:39,960 --> 00:12:42,600 Taylor it's approach to your needs. 451 00:12:42,600 --> 00:12:46,200 Explain complex concepts in a way that makes sense to you. 452 00:12:46,200 --> 00:12:48,080 So it's like having a personal tutor. 453 00:12:48,080 --> 00:12:48,400 Yeah. 454 00:12:48,400 --> 00:12:50,200 Who's available 24 seven. 455 00:12:50,200 --> 00:12:50,800 Exactly. 456 00:12:50,800 --> 00:12:52,480 And never gets tired of explaining things. 457 00:12:52,480 --> 00:12:54,720 And because these AI tutors. 458 00:12:54,720 --> 00:12:55,480 Right. 459 00:12:55,480 --> 00:12:59,200 Would be grounded in logical reasoning. 460 00:12:59,200 --> 00:13:00,960 We could have more confidence that they're 461 00:13:00,960 --> 00:13:02,760 teaching accurate information. 462 00:13:02,760 --> 00:13:03,080 Yeah. 463 00:13:03,080 --> 00:13:03,840 That's a big deal. 464 00:13:03,840 --> 00:13:05,120 Consistent information. 465 00:13:05,120 --> 00:13:06,000 Especially these days. 466 00:13:06,000 --> 00:13:06,320 It is. 467 00:13:06,320 --> 00:13:09,120 It could revolutionize education as we know it. 468 00:13:09,120 --> 00:13:10,880 Personalized robots. 469 00:13:10,880 --> 00:13:11,080 Yeah. 470 00:13:11,080 --> 00:13:12,640 AI tutors. 471 00:13:12,640 --> 00:13:14,640 It seems like the possibilities are endless. 472 00:13:14,640 --> 00:13:16,040 They really are. 473 00:13:16,040 --> 00:13:18,240 But there's one area we haven't touched on yet. 474 00:13:18,240 --> 00:13:18,960 OK. 475 00:13:18,960 --> 00:13:20,800 Transportation. 476 00:13:20,800 --> 00:13:21,320 Yes. 477 00:13:21,320 --> 00:13:23,600 Given all this talk about self-driving cars. 478 00:13:23,600 --> 00:13:24,240 Right. 479 00:13:24,240 --> 00:13:26,520 I'm curious how these NECMMs. 480 00:13:26,520 --> 00:13:26,920 Yeah. 481 00:13:26,920 --> 00:13:28,720 Could fit into that picture. 482 00:13:28,720 --> 00:13:29,680 That's a great point. 483 00:13:29,680 --> 00:13:31,320 Because that's a really big one. 484 00:13:31,320 --> 00:13:31,840 It is. 485 00:13:31,840 --> 00:13:32,920 With AI, right? 486 00:13:32,920 --> 00:13:33,200 Yeah. 487 00:13:33,200 --> 00:13:38,120 Self-driving cars are like the poster child for AI. 488 00:13:38,120 --> 00:13:38,720 Right. 489 00:13:38,720 --> 00:13:39,680 For good reason. 490 00:13:39,680 --> 00:13:40,080 Yeah. 491 00:13:40,080 --> 00:13:41,080 The stakes are high. 492 00:13:41,080 --> 00:13:41,480 Yeah. 493 00:13:41,480 --> 00:13:42,680 People's lives are at stake. 494 00:13:42,680 --> 00:13:42,880 Yeah. 495 00:13:42,880 --> 00:13:43,680 Exactly. 496 00:13:43,680 --> 00:13:46,840 So by combining this data driven learning. 497 00:13:46,840 --> 00:13:50,160 With logical reasoning about traffic rules. 498 00:13:50,160 --> 00:13:50,640 OK. 499 00:13:50,640 --> 00:13:51,840 Safety protocols. 500 00:13:51,840 --> 00:13:52,240 Right. 501 00:13:52,240 --> 00:13:54,920 We could create self-driving systems that 502 00:13:54,920 --> 00:13:57,240 are not only more reliable. 503 00:13:57,240 --> 00:13:57,520 Yeah. 504 00:13:57,520 --> 00:13:59,160 But also more predictable. 505 00:13:59,160 --> 00:13:59,880 And cautious. 506 00:13:59,880 --> 00:14:00,600 And cautious. 507 00:14:00,600 --> 00:14:02,560 Which I think is what a lot of people are worried about. 508 00:14:02,560 --> 00:14:03,520 They are. 509 00:14:03,520 --> 00:14:05,840 Is that the car is going to be too cautious. 510 00:14:05,840 --> 00:14:06,040 Right. 511 00:14:06,040 --> 00:14:08,520 And it's going to cause more problems than it's solved. 512 00:14:08,520 --> 00:14:09,760 But the thing is. 513 00:14:09,760 --> 00:14:10,200 Yeah. 514 00:14:10,200 --> 00:14:15,240 If you can bake in that logical reasoning about safety. 515 00:14:15,240 --> 00:14:15,600 Right. 516 00:14:15,600 --> 00:14:17,880 Then you can actually create a system. 517 00:14:17,880 --> 00:14:18,360 Yeah. 518 00:14:18,360 --> 00:14:21,640 That understands why it needs to be cautious 519 00:14:21,640 --> 00:14:22,800 in certain situations. 520 00:14:22,800 --> 00:14:26,640 So it's not just reacting to what's happening on the road. 521 00:14:26,640 --> 00:14:28,560 It's actually understanding. 522 00:14:28,560 --> 00:14:29,800 Understanding the why. 523 00:14:29,800 --> 00:14:30,160 Yeah. 524 00:14:30,160 --> 00:14:31,920 Behind certain driving decisions. 525 00:14:31,920 --> 00:14:32,400 Yeah. 526 00:14:32,400 --> 00:14:35,680 So it can anticipate potential hazards. 527 00:14:35,680 --> 00:14:38,040 Make informed decisions at intersections. 528 00:14:38,040 --> 00:14:38,760 Yeah. 529 00:14:38,760 --> 00:14:41,240 Even explain its actions to the passenger. 530 00:14:41,240 --> 00:14:42,000 Exactly. 531 00:14:42,000 --> 00:14:44,120 Which builds trust and confidence. 532 00:14:44,120 --> 00:14:44,760 It does. 533 00:14:44,760 --> 00:14:45,400 In the system. 534 00:14:45,400 --> 00:14:45,680 Right. 535 00:14:45,680 --> 00:14:48,600 And that trust is really what's going to make or break 536 00:14:48,600 --> 00:14:50,320 the adoption of these technologies. 537 00:14:50,320 --> 00:14:52,280 It sounds like NECI MMS. 538 00:14:52,280 --> 00:14:52,800 Yeah. 539 00:14:52,800 --> 00:14:57,480 Could be the key to unlocking a future of truly intelligent 540 00:14:57,480 --> 00:14:58,640 transportation systems. 541 00:14:58,640 --> 00:14:59,200 I think so. 542 00:14:59,200 --> 00:15:00,920 And that's just scratching the surface. 543 00:15:00,920 --> 00:15:02,800 We're talking about a technology that 544 00:15:02,800 --> 00:15:06,600 could potentially revolutionize any failed. 545 00:15:06,600 --> 00:15:09,640 Where AI needs to reason. 546 00:15:09,640 --> 00:15:10,760 Adapt. 547 00:15:10,760 --> 00:15:11,240 Adapt. 548 00:15:11,240 --> 00:15:12,360 Interact with the world. 549 00:15:12,360 --> 00:15:12,760 Yeah. 550 00:15:12,760 --> 00:15:15,160 In a way that aligns with their expectations. 551 00:15:15,160 --> 00:15:16,040 It's exciting. 552 00:15:16,040 --> 00:15:16,760 It is exciting. 553 00:15:16,760 --> 00:15:21,640 So the takeaway here is that while NECI MMS. 554 00:15:21,640 --> 00:15:22,280 Yeah. 555 00:15:22,280 --> 00:15:22,920 Are still. 556 00:15:22,920 --> 00:15:24,200 And they're early stages. 557 00:15:24,200 --> 00:15:25,160 Yeah, they're early stages. 558 00:15:25,160 --> 00:15:27,720 The potential impact is enormous. 559 00:15:27,720 --> 00:15:28,160 Right. 560 00:15:28,160 --> 00:15:29,240 We're not saying. 561 00:15:29,240 --> 00:15:29,720 Right. 562 00:15:29,720 --> 00:15:31,200 This is the solution to everything. 563 00:15:31,200 --> 00:15:32,160 The VL end all. 564 00:15:32,160 --> 00:15:32,640 Right. 565 00:15:32,640 --> 00:15:33,360 But it's definitely. 566 00:15:33,360 --> 00:15:34,480 It's a step in the right direction. 567 00:15:34,480 --> 00:15:35,880 It is a step in the right direction. 568 00:15:35,880 --> 00:15:39,280 And the journey is only going to get more exciting from here. 569 00:15:39,280 --> 00:15:40,080 I think so. 570 00:15:40,080 --> 00:15:43,040 As researchers continue to refine these models. 571 00:15:43,040 --> 00:15:44,280 And address those. 572 00:15:44,280 --> 00:15:44,560 Right. 573 00:15:44,560 --> 00:15:45,840 Scalability challenges. 574 00:15:45,840 --> 00:15:47,120 Scalability challenges. 575 00:15:47,120 --> 00:15:48,200 Who knows what. 576 00:15:48,200 --> 00:15:48,400 Right. 577 00:15:48,400 --> 00:15:51,360 Round breaking applications. 578 00:15:51,360 --> 00:15:51,920 We'll see. 579 00:15:51,920 --> 00:15:53,360 We'll see in the years to come. 580 00:15:53,360 --> 00:15:55,760 It's like we've opened a door. 581 00:15:55,760 --> 00:15:56,240 Yeah. 582 00:15:56,240 --> 00:15:57,000 To this. 583 00:15:57,000 --> 00:16:00,080 To this whole new era of AI. 584 00:16:00,080 --> 00:16:01,680 Whole new era of AI. 585 00:16:01,680 --> 00:16:04,800 Where machines can not only learn from data. 586 00:16:04,800 --> 00:16:07,200 But they can understand the world around them. 587 00:16:07,200 --> 00:16:07,560 Yeah. 588 00:16:07,560 --> 00:16:09,200 In a way that's closer. 589 00:16:09,200 --> 00:16:09,560 Yeah. 590 00:16:09,560 --> 00:16:11,480 To how we humans think. 591 00:16:11,480 --> 00:16:13,120 And with that new understanding. 592 00:16:13,120 --> 00:16:13,640 Mm-hmm. 593 00:16:13,640 --> 00:16:15,440 Comes a greater potential. 594 00:16:15,440 --> 00:16:15,960 Right. 595 00:16:15,960 --> 00:16:17,280 For collaboration. 596 00:16:17,280 --> 00:16:18,160 That's problem solving. 597 00:16:18,160 --> 00:16:19,120 For problem solving. 598 00:16:19,120 --> 00:16:20,000 And perhaps even. 599 00:16:20,000 --> 00:16:20,400 Yeah. 600 00:16:20,400 --> 00:16:22,240 A deeper understanding of ourselves. 601 00:16:22,240 --> 00:16:23,120 That's a great point. 602 00:16:23,120 --> 00:16:23,720 Mm. 603 00:16:23,720 --> 00:16:24,040 Yeah. 604 00:16:24,040 --> 00:16:25,920 It really is mind blowing to think about like. 605 00:16:25,920 --> 00:16:26,360 It is. 606 00:16:26,360 --> 00:16:27,520 All the ripple effects. 607 00:16:27,520 --> 00:16:28,040 Yeah. 608 00:16:28,040 --> 00:16:29,120 This research could have. 609 00:16:29,120 --> 00:16:29,920 Mm-hmm. 610 00:16:29,920 --> 00:16:32,840 We've talked about robots, education, transportation. 611 00:16:32,840 --> 00:16:33,160 Right. 612 00:16:33,160 --> 00:16:35,800 Like it feels like Nessiems could be like. 613 00:16:35,800 --> 00:16:36,640 Mm-hmm. 614 00:16:36,640 --> 00:16:37,840 The missing link. 615 00:16:37,840 --> 00:16:38,200 Yeah. 616 00:16:38,200 --> 00:16:42,000 To bridging that gap between AI as we know it. 617 00:16:42,000 --> 00:16:42,480 Yeah. 618 00:16:42,480 --> 00:16:44,840 And AI as we like imagine it. 619 00:16:44,840 --> 00:16:45,240 Mm-hmm. 620 00:16:45,240 --> 00:16:46,520 In science fiction. 621 00:16:46,520 --> 00:16:49,000 It's certainly a powerful concept. 622 00:16:49,000 --> 00:16:49,400 Yeah. 623 00:16:49,400 --> 00:16:51,880 But it's important to remember that we're still. 624 00:16:51,880 --> 00:16:53,480 Early and early stages. 625 00:16:53,480 --> 00:16:56,000 We're not going to solve all the world's problems. 626 00:16:56,000 --> 00:16:56,600 Not yet. 627 00:16:56,600 --> 00:16:58,800 With Nessiems' tomorrow. 628 00:16:58,800 --> 00:16:59,760 Not tomorrow. 629 00:16:59,760 --> 00:17:00,160 But. 630 00:17:00,160 --> 00:17:02,040 But it's hard not to get excited. 631 00:17:02,040 --> 00:17:03,080 It is exciting. 632 00:17:03,080 --> 00:17:04,080 About the possibilities. 633 00:17:04,080 --> 00:17:08,400 Imagine a world where AI can not only diagnose diseases. 634 00:17:08,400 --> 00:17:10,080 But explain its reasoning. 635 00:17:10,080 --> 00:17:10,440 Yeah. 636 00:17:10,440 --> 00:17:11,320 To doctors. 637 00:17:11,320 --> 00:17:13,240 In a way that helps them make better decisions. 638 00:17:13,240 --> 00:17:14,240 That would be huge. 639 00:17:14,240 --> 00:17:16,200 Or imagine AI systems. 640 00:17:16,200 --> 00:17:16,560 Yeah. 641 00:17:16,560 --> 00:17:18,120 That can help scientists. 642 00:17:18,120 --> 00:17:18,480 Mm-hmm. 643 00:17:18,480 --> 00:17:21,240 Model complex systems like climate change. 644 00:17:21,240 --> 00:17:21,360 Right. 645 00:17:21,360 --> 00:17:24,160 With greater accuracy and predictability. 646 00:17:24,160 --> 00:17:25,800 So we can make more informed decisions. 647 00:17:25,800 --> 00:17:26,480 Working. 648 00:17:26,480 --> 00:17:27,840 About our planet's future. 649 00:17:27,840 --> 00:17:29,120 And let's not forget. 650 00:17:29,120 --> 00:17:29,480 Yeah. 651 00:17:29,480 --> 00:17:33,560 The potential for AI-powered personal assistants. 652 00:17:33,560 --> 00:17:33,840 OK. 653 00:17:33,840 --> 00:17:36,200 That can actually understand our needs. 654 00:17:36,200 --> 00:17:36,640 Right. 655 00:17:36,640 --> 00:17:38,520 Anticipate our requests. 656 00:17:38,520 --> 00:17:39,120 Ooh. 657 00:17:39,120 --> 00:17:40,560 Like a mind reader. 658 00:17:40,560 --> 00:17:41,200 Almost. 659 00:17:41,200 --> 00:17:41,520 Yeah. 660 00:17:41,520 --> 00:17:43,760 And even help us solve creative problems. 661 00:17:43,760 --> 00:17:43,960 OK. 662 00:17:43,960 --> 00:17:44,960 So we're talking about AI. 663 00:17:44,960 --> 00:17:45,400 Yeah. 664 00:17:45,400 --> 00:17:48,840 That's not just like doing things for us. 665 00:17:48,840 --> 00:17:49,200 Right. 666 00:17:49,200 --> 00:17:50,520 But it's like helping us. 667 00:17:50,520 --> 00:17:51,840 Helping us think better. 668 00:17:51,840 --> 00:17:53,440 Think better be more creative. 669 00:17:53,440 --> 00:17:54,240 Exactly. 670 00:17:54,240 --> 00:17:54,800 Wow. 671 00:17:54,800 --> 00:17:56,720 These are all exciting possibilities. 672 00:17:56,720 --> 00:17:57,480 They are. 673 00:17:57,480 --> 00:18:02,440 And I believe NECI-MMs represent a significant step. 674 00:18:02,440 --> 00:18:02,920 OK. 675 00:18:02,920 --> 00:18:05,800 Towards realizing that vision. 676 00:18:05,800 --> 00:18:07,400 It's really cool to see. 677 00:18:07,400 --> 00:18:08,080 It is. 678 00:18:08,080 --> 00:18:09,760 How this field is progressing. 679 00:18:09,760 --> 00:18:10,120 Yeah. 680 00:18:10,120 --> 00:18:14,120 Because for a long time it felt like AI was kind of stuck. 681 00:18:14,120 --> 00:18:14,520 Right. 682 00:18:14,520 --> 00:18:15,040 In this. 683 00:18:15,040 --> 00:18:16,280 In a rut. 684 00:18:16,280 --> 00:18:16,480 Yeah. 685 00:18:16,480 --> 00:18:18,760 Just like pattern recognition. 686 00:18:18,760 --> 00:18:19,640 Brute force. 687 00:18:19,640 --> 00:18:20,240 Brute force. 688 00:18:20,240 --> 00:18:20,640 Yeah. 689 00:18:20,640 --> 00:18:25,760 And now it's like we're starting to see AI that can actually 690 00:18:25,760 --> 00:18:26,760 reason. 691 00:18:26,760 --> 00:18:27,640 Right. 692 00:18:27,640 --> 00:18:28,360 And understand. 693 00:18:28,360 --> 00:18:30,560 And that's the game changer. 694 00:18:30,560 --> 00:18:31,320 That is the game changer. 695 00:18:31,320 --> 00:18:34,320 By combining the strengths of data driven learning. 696 00:18:34,320 --> 00:18:34,680 Right. 697 00:18:34,680 --> 00:18:36,920 With the power of logical reasoning. 698 00:18:36,920 --> 00:18:39,200 We're opening up this whole new frontier. 699 00:18:39,200 --> 00:18:40,440 And with any new frontier. 700 00:18:40,440 --> 00:18:41,000 Right. 701 00:18:41,000 --> 00:18:42,320 There are going to be challenges. 702 00:18:42,320 --> 00:18:43,040 Oh absolutely. 703 00:18:43,040 --> 00:18:44,240 Unexpected discoveries. 704 00:18:44,240 --> 00:18:45,120 Yeah. 705 00:18:45,120 --> 00:18:46,120 Setbacks. 706 00:18:46,120 --> 00:18:46,880 Probably. 707 00:18:46,880 --> 00:18:47,680 Probably. 708 00:18:47,680 --> 00:18:49,320 But that's what makes it so exciting. 709 00:18:49,320 --> 00:18:49,800 Exactly. 710 00:18:49,800 --> 00:18:51,720 It's a constant process. 711 00:18:51,720 --> 00:18:52,040 Yeah. 712 00:18:52,040 --> 00:18:53,160 Of exploration. 713 00:18:53,160 --> 00:18:53,960 Experimentation. 714 00:18:53,960 --> 00:18:54,680 Experimentation. 715 00:18:54,680 --> 00:18:55,680 Pushing the boundaries. 716 00:18:55,680 --> 00:18:56,920 Pushing the boundaries. 717 00:18:56,920 --> 00:18:57,240 Yeah. 718 00:18:57,240 --> 00:18:59,280 Of what's possible. 719 00:18:59,280 --> 00:19:00,840 I think that's a great point to end on. 720 00:19:00,840 --> 00:19:01,800 I think so too. 721 00:19:01,800 --> 00:19:06,880 The future of AI is not just about creating machines. 722 00:19:06,880 --> 00:19:07,440 Right. 723 00:19:07,440 --> 00:19:08,640 That are smarter than us. 724 00:19:08,640 --> 00:19:09,120 Right. 725 00:19:09,120 --> 00:19:10,800 It's not about just outsmarting us. 726 00:19:10,800 --> 00:19:11,680 It's about. 727 00:19:11,680 --> 00:19:15,120 It's about creating machines that can help us understand. 728 00:19:15,120 --> 00:19:15,560 Yeah. 729 00:19:15,560 --> 00:19:17,440 Our selves and the world around us. 730 00:19:17,440 --> 00:19:18,840 That's right. 731 00:19:18,840 --> 00:19:20,280 In new and profound ways. 732 00:19:20,280 --> 00:19:22,760 And I think that's a really beautiful vision. 733 00:19:22,760 --> 00:19:23,680 It is a beautiful vision. 734 00:19:23,680 --> 00:19:25,120 For the future of AI. 735 00:19:25,120 --> 00:19:26,600 So to everyone listening. 736 00:19:26,600 --> 00:19:26,920 Yeah. 737 00:19:26,920 --> 00:19:31,000 Who wants to like dive deeper into this fascinating world 738 00:19:31,000 --> 00:19:33,040 of neuro-symbolic AI. 739 00:19:33,040 --> 00:19:34,360 We'll be sure to include a link. 740 00:19:34,360 --> 00:19:34,560 Yeah. 741 00:19:34,560 --> 00:19:36,520 We'll include links in the show notes. 742 00:19:36,520 --> 00:19:38,000 To the full paper. 743 00:19:38,000 --> 00:19:38,920 To all the resources. 744 00:19:38,920 --> 00:19:39,840 All the resources. 745 00:19:39,840 --> 00:19:41,520 So you can check it out for yourself. 746 00:19:41,520 --> 00:19:42,320 And as always. 747 00:19:42,320 --> 00:19:43,160 Yeah. 748 00:19:43,160 --> 00:19:44,760 Keep those minds curious. 749 00:19:44,760 --> 00:19:45,680 Stay curious. 750 00:19:45,680 --> 00:19:48,440 And stay tuned for more exciting developments. 751 00:19:48,440 --> 00:19:51,160 Because there are going to be more in the world of AI. 752 00:19:51,160 --> 00:19:52,160 Thanks for joining us. 753 00:19:52,160 --> 00:19:52,520 Yeah. 754 00:19:52,520 --> 00:19:54,560 Thanks for going on this deep dive with us, everyone. 755 00:19:54,560 --> 00:19:55,440 Until next time. 756 00:19:55,440 --> 00:20:19,440 Until next time.