1 00:00:00,000 --> 00:00:03,360 Ever wished you could ask AI, like really ask it, okay? 2 00:00:03,360 --> 00:00:04,840 But why did you decide that? 3 00:00:04,840 --> 00:00:06,360 Like it's gonna show its work, right? 4 00:00:06,360 --> 00:00:07,200 Exactly. 5 00:00:07,200 --> 00:00:09,140 And today we're diving deep into a paper 6 00:00:09,140 --> 00:00:11,040 that's trying to make that happen. 7 00:00:11,040 --> 00:00:12,840 Make AI more transparent, you know, 8 00:00:12,840 --> 00:00:15,480 even for those of us who aren't like coding wizards. 9 00:00:15,480 --> 00:00:16,720 Peaking behind the curtain, 10 00:00:16,720 --> 00:00:19,240 how'd that AI magician do its tricks? 11 00:00:19,240 --> 00:00:20,360 Exactly. 12 00:00:20,360 --> 00:00:22,520 The paper is called Explingo, 13 00:00:22,520 --> 00:00:25,980 explaining AI predictions using large language models. 14 00:00:25,980 --> 00:00:28,200 And it really tackles this head on. 15 00:00:28,200 --> 00:00:30,520 Yeah, the problem is AI gives us answers, 16 00:00:30,520 --> 00:00:32,960 but like often it doesn't tell us how it got there. 17 00:00:32,960 --> 00:00:36,640 That's where explainable AI comes in, or XAI. 18 00:00:36,640 --> 00:00:39,680 So XAI is kind of like asking AI to show its work. 19 00:00:39,680 --> 00:00:40,680 Yeah, exactly. 20 00:00:40,680 --> 00:00:42,760 Making that decision-making process clear, 21 00:00:42,760 --> 00:00:44,800 which is super important in fields like healthcare 22 00:00:44,800 --> 00:00:46,080 or finance, you know, 23 00:00:46,080 --> 00:00:48,000 where those decisions have big consequences. 24 00:00:48,000 --> 00:00:50,120 Instead of just blindly trusting AI, 25 00:00:50,120 --> 00:00:51,600 we can understand its reasoning. 26 00:00:51,600 --> 00:00:52,440 Exactly. 27 00:00:52,440 --> 00:00:54,280 One common way to explain AI decisions 28 00:00:54,280 --> 00:00:57,080 is using something called SHAP. 29 00:00:57,080 --> 00:01:00,920 SHP, that sounds a bit intimidating. 30 00:01:00,920 --> 00:01:04,480 It can be, but imagine like each factor in a decision 31 00:01:04,480 --> 00:01:06,720 is like an ingredient in a cake, right? 32 00:01:06,720 --> 00:01:09,160 SHAP tells us how much each ingredient 33 00:01:09,160 --> 00:01:10,960 contributes to the final taste. 34 00:01:10,960 --> 00:01:12,600 Helps us understand the recipe. 35 00:01:12,600 --> 00:01:13,480 Okay, that makes sense. 36 00:01:13,480 --> 00:01:15,720 So SHAP helps us like break down 37 00:01:15,720 --> 00:01:17,240 the AI's decision-making. 38 00:01:17,240 --> 00:01:18,280 Precisely. 39 00:01:18,280 --> 00:01:19,280 But the problem is, 40 00:01:19,280 --> 00:01:22,560 SHAP usually gives us like a bunch of numbers and graphs, 41 00:01:22,560 --> 00:01:23,680 hard to understand. 42 00:01:23,680 --> 00:01:25,040 Not exactly user-friendly. 43 00:01:25,040 --> 00:01:27,720 And that's where this paper's big idea comes in. 44 00:01:27,720 --> 00:01:29,480 They use large language models, 45 00:01:29,480 --> 00:01:31,480 you know, like what's behind chat GPT, 46 00:01:31,480 --> 00:01:34,560 to turn those complex SAP explanations 47 00:01:34,560 --> 00:01:36,720 into, well, stories we can understand. 48 00:01:36,720 --> 00:01:38,720 So it's like having an AI translator 49 00:01:38,720 --> 00:01:40,880 turning that tech jargon into plain English. 50 00:01:40,880 --> 00:01:42,240 Yeah, exactly. 51 00:01:42,240 --> 00:01:44,880 One LLM, it's been trained on tons of data. 52 00:01:44,880 --> 00:01:47,000 It reads the SHAP output 53 00:01:47,000 --> 00:01:49,920 and then creates a story explaining the AI's decision. 54 00:01:49,920 --> 00:01:53,200 AI explaining AI, but in a way we humans can grasp. 55 00:01:53,200 --> 00:01:54,040 Clever. 56 00:01:54,040 --> 00:01:56,040 They're basically teaching AI to tell us a story 57 00:01:56,040 --> 00:01:57,360 about how it made its decision. 58 00:01:57,360 --> 00:01:58,200 Precisely. 59 00:01:58,200 --> 00:01:59,560 And the story isn't just random, 60 00:01:59,560 --> 00:02:01,480 it's based on those SHAP values 61 00:02:01,480 --> 00:02:03,760 so it accurately reflects the AI's reasoning. 62 00:02:03,760 --> 00:02:04,600 That's neat. 63 00:02:04,600 --> 00:02:06,720 So Xplingo is like a storyteller, 64 00:02:06,720 --> 00:02:09,440 revealing the secrets behind AI's decisions. 65 00:02:09,440 --> 00:02:10,400 You got it. 66 00:02:10,400 --> 00:02:12,960 This storytelling approach could be huge 67 00:02:12,960 --> 00:02:14,960 for how we interact with AI. 68 00:02:14,960 --> 00:02:17,600 Like imagine a doctor using an AI tool 69 00:02:17,600 --> 00:02:19,200 to diagnose a patient. 70 00:02:19,200 --> 00:02:22,360 With Xplingo, they understand not just what the AI recommends, 71 00:02:22,360 --> 00:02:24,240 but why it made that recommendation. 72 00:02:24,240 --> 00:02:26,640 They would definitely boost confidence in using AI, 73 00:02:26,640 --> 00:02:28,720 especially in those high-stakes situations. 74 00:02:28,720 --> 00:02:31,280 Exactly, and it's not just about trust. 75 00:02:31,280 --> 00:02:33,280 Understanding the AI's reasoning, 76 00:02:33,280 --> 00:02:36,000 well, it could help us identify potential biases 77 00:02:36,000 --> 00:02:37,800 or errors in its decision-making. 78 00:02:37,800 --> 00:02:39,880 So we can double-check the AI's work, 79 00:02:39,880 --> 00:02:42,040 sure it's making fair and accurate decisions. 80 00:02:42,040 --> 00:02:42,880 Precisely. 81 00:02:42,880 --> 00:02:46,240 And this brings us to how Xplingo actually works. 82 00:02:46,240 --> 00:02:49,840 It has two main parts, the narrator and the grader. 83 00:02:49,840 --> 00:02:50,880 Okay, let's unpack that. 84 00:02:50,880 --> 00:02:52,000 What does the narrator do? 85 00:02:52,000 --> 00:02:55,280 The narrator is the part that takes that SHAP explanation 86 00:02:55,280 --> 00:02:57,640 and crafts the human-readable narrative, 87 00:02:57,640 --> 00:02:59,400 the storyteller we were just talking about. 88 00:02:59,400 --> 00:03:02,000 So it's like the AI version of a journalist 89 00:03:02,000 --> 00:03:05,800 taking complex data and turning it into a compelling story. 90 00:03:05,800 --> 00:03:06,960 A perfect analogy. 91 00:03:06,960 --> 00:03:07,800 All right, that makes sense. 92 00:03:07,800 --> 00:03:08,640 What about this grader? 93 00:03:08,640 --> 00:03:09,460 What's its role? 94 00:03:09,460 --> 00:03:11,840 The graders, like the narrator's editor, 95 00:03:11,840 --> 00:03:14,040 evaluates the quality of the narrative, 96 00:03:14,040 --> 00:03:17,040 makes sure it's accurate, complete, fluent, and concise. 97 00:03:17,040 --> 00:03:19,480 So like a quality control check 98 00:03:19,480 --> 00:03:22,360 to ensure the AI's explanation is clear and on point. 99 00:03:22,360 --> 00:03:24,320 Exactly, we don't want rambling 100 00:03:24,320 --> 00:03:26,400 or confusing explanations, do we? 101 00:03:26,400 --> 00:03:27,400 Definitely not. 102 00:03:27,400 --> 00:03:28,760 Especially if we're relying on this 103 00:03:28,760 --> 00:03:31,200 to understand important AI decisions. 104 00:03:32,140 --> 00:03:34,680 It sounds like Xplingo has a lot of potential. 105 00:03:35,680 --> 00:03:38,040 But did the researchers actually test it out? 106 00:03:38,040 --> 00:03:38,880 They absolutely did. 107 00:03:38,880 --> 00:03:41,720 They tested Xplingo on several different data sets, 108 00:03:41,720 --> 00:03:43,640 from predicting house prices 109 00:03:43,640 --> 00:03:46,320 to identifying poisonous mushrooms. 110 00:03:46,320 --> 00:03:48,840 And well, they found that it works surprisingly well. 111 00:03:48,840 --> 00:03:51,360 So it wasn't just a cool idea, it actually delivered results. 112 00:03:51,360 --> 00:03:54,000 Exactly, and that's what makes this research so exciting. 113 00:03:54,000 --> 00:03:55,840 It's not just theoretical. 114 00:03:55,840 --> 00:03:58,800 It's showing a real path towards making AI 115 00:03:58,800 --> 00:04:00,120 more understandable. 116 00:04:00,120 --> 00:04:03,200 So we've got this system that can explain AI decisions 117 00:04:03,200 --> 00:04:04,840 in a way that makes sense to us, 118 00:04:04,840 --> 00:04:06,760 but does it actually work in the real world? 119 00:04:06,760 --> 00:04:08,600 Tell me more about those tests. 120 00:04:08,600 --> 00:04:10,560 Yeah, they threw a bunch of different tasks at it. 121 00:04:10,560 --> 00:04:13,120 One was predicting house prices, 122 00:04:13,120 --> 00:04:15,360 based on size, location, stuff like that, 123 00:04:15,360 --> 00:04:17,600 like what you see on real estate websites. 124 00:04:17,600 --> 00:04:20,160 So Xplingo could explain why a house with a pool 125 00:04:20,160 --> 00:04:22,080 might be priced higher. 126 00:04:22,080 --> 00:04:22,920 Pretty cool. 127 00:04:22,920 --> 00:04:24,040 Exactly. 128 00:04:24,040 --> 00:04:27,040 Another data set, it was for identifying like, 129 00:04:27,040 --> 00:04:30,720 poisonous mushrooms, based on their characteristics. 130 00:04:30,720 --> 00:04:31,880 That's where you really wanna know 131 00:04:31,880 --> 00:04:33,320 why the AI made a decision, right? 132 00:04:33,320 --> 00:04:34,520 No kidding. 133 00:04:34,520 --> 00:04:38,240 Mistakes there could have some serious consequences. 134 00:04:38,240 --> 00:04:39,400 So how'd Xplingo do? 135 00:04:39,400 --> 00:04:40,960 Did it pass the mushroom test? 136 00:04:40,960 --> 00:04:41,800 It did pretty well. 137 00:04:41,800 --> 00:04:44,720 It was able to explain the AI's reasoning pretty accurately. 138 00:04:44,720 --> 00:04:47,400 They even tested it on malware detection, 139 00:04:47,400 --> 00:04:49,160 which is super important these days. 140 00:04:49,160 --> 00:04:51,840 Malware detection, house prices, mushrooms. 141 00:04:51,840 --> 00:04:53,680 Sounds like Xplingo's pretty versatile, 142 00:04:53,680 --> 00:04:56,200 but I'm guessing it wasn't all smooth sailing. 143 00:04:56,200 --> 00:04:58,360 Were there any areas where it struggled? 144 00:04:58,360 --> 00:04:59,920 Yeah, there were a few hiccups, 145 00:04:59,920 --> 00:05:01,640 like any new technology, right? 146 00:05:01,640 --> 00:05:03,920 Sometimes Xplingo had trouble with context, 147 00:05:03,920 --> 00:05:07,320 like explaining why a larger file size is suspicious, 148 00:05:07,320 --> 00:05:09,320 without saying what larger actually means. 149 00:05:09,320 --> 00:05:11,120 Ah, so it needs some help 150 00:05:11,120 --> 00:05:14,640 with those nuances of human language, still learning. 151 00:05:14,640 --> 00:05:15,480 Exactly. 152 00:05:15,480 --> 00:05:18,280 Some domains, like that mushroom identification 153 00:05:18,280 --> 00:05:20,240 where you need really high accuracy, 154 00:05:20,240 --> 00:05:22,680 well, Xplingo showed promise, 155 00:05:22,680 --> 00:05:26,240 but needs some fine tuning for those high stakes situations. 156 00:05:26,240 --> 00:05:27,080 Makes sense. 157 00:05:27,080 --> 00:05:29,760 It's exciting to see that even with those challenges, 158 00:05:29,760 --> 00:05:31,640 this research is really pushing the boundaries 159 00:05:31,640 --> 00:05:35,520 of what's possible with AI explainability. 160 00:05:35,520 --> 00:05:38,680 But how do the researchers see this being used, 161 00:05:38,680 --> 00:05:39,520 like in the future? 162 00:05:39,520 --> 00:05:41,720 They think Xplingo could change how we use AI 163 00:05:41,720 --> 00:05:43,080 in tons of fields, 164 00:05:43,080 --> 00:05:46,240 like doctors using it to understand an AI diagnosis, 165 00:05:46,240 --> 00:05:49,080 or financial analysts using it to break down 166 00:05:49,080 --> 00:05:51,000 an AI investment recommendation. 167 00:05:51,000 --> 00:05:54,160 Well, giving us a window into the AI's thought process, 168 00:05:54,160 --> 00:05:55,520 we can make better decisions. 169 00:05:55,520 --> 00:05:58,120 It could really revolutionize how we use AI every day. 170 00:05:58,120 --> 00:05:58,960 Absolutely. 171 00:05:58,960 --> 00:06:01,120 And beyond specific applications, 172 00:06:01,120 --> 00:06:03,400 Xplingo also brings up some really interesting questions 173 00:06:03,400 --> 00:06:05,000 about the future of AI. 174 00:06:05,000 --> 00:06:06,920 Oh, what kind of questions? 175 00:06:06,920 --> 00:06:11,320 Well, if we can teach AI to explain itself to us, right? 176 00:06:11,320 --> 00:06:13,160 Does that mean we're getting closer to AI 177 00:06:13,160 --> 00:06:15,880 that can actually understand us? 178 00:06:15,880 --> 00:06:18,800 Is this a step towards AI that can not just think, 179 00:06:18,800 --> 00:06:20,480 but maybe even empathize? 180 00:06:20,480 --> 00:06:22,560 Wow, that's a deep thought. 181 00:06:22,560 --> 00:06:24,560 It's like by teaching AI to explain itself, 182 00:06:24,560 --> 00:06:27,200 we're teaching it about what it means to be human. 183 00:06:27,200 --> 00:06:28,160 Exactly. 184 00:06:28,160 --> 00:06:30,480 And that brings up all sorts of ethical 185 00:06:30,480 --> 00:06:34,160 and philosophical questions that we need to start thinking 186 00:06:34,160 --> 00:06:35,520 about as a society. 187 00:06:35,520 --> 00:06:37,800 Like opening a Pandora's box. 188 00:06:37,800 --> 00:06:40,200 Exciting, but maybe a little scary too. 189 00:06:40,200 --> 00:06:43,720 It is, but by making AI more transparent, 190 00:06:43,720 --> 00:06:46,560 understandable, we're also making it more controllable 191 00:06:46,560 --> 00:06:49,200 and ultimately more helpful to humanity. 192 00:06:49,200 --> 00:06:52,640 It's like we're going from AI as this mysterious black box 193 00:06:52,640 --> 00:06:55,440 to AI as a partner we can work with and understand. 194 00:06:55,440 --> 00:06:56,200 Precisely. 195 00:06:56,200 --> 00:06:58,400 And Xplingo is a big step in that direction. 196 00:06:58,400 --> 00:07:00,680 It's showing us we don't have to be in the dark when 197 00:07:00,680 --> 00:07:01,760 it comes to AI. 198 00:07:01,760 --> 00:07:04,760 We can shed some light on its decision making, 199 00:07:04,760 --> 00:07:06,680 use that to build a better future. 200 00:07:06,680 --> 00:07:07,880 That's powerful. 201 00:07:07,880 --> 00:07:09,480 It feels like this research isn't just 202 00:07:09,480 --> 00:07:12,520 about explaining AI but empowering us 203 00:07:12,520 --> 00:07:17,320 to shape the future of AI in a way that aligns with our values. 204 00:07:17,320 --> 00:07:18,480 I couldn't agree more. 205 00:07:18,480 --> 00:07:19,800 This has been a fascinating dive. 206 00:07:19,800 --> 00:07:23,400 We've seen how Xplingo uses these powerful language models 207 00:07:23,400 --> 00:07:28,440 to bridge the gap between AI and us human understanding. 208 00:07:28,440 --> 00:07:32,120 But I'm curious, are there any potential downsides 209 00:07:32,120 --> 00:07:35,760 to this approach or limitations? 210 00:07:35,760 --> 00:07:36,840 That's a great question. 211 00:07:36,840 --> 00:07:39,520 One concern is by simplifying AI explanations, 212 00:07:39,520 --> 00:07:41,120 maybe we're losing some important details. 213 00:07:41,120 --> 00:07:44,080 It's kind of like summarizing a really complex book 214 00:07:44,080 --> 00:07:45,680 in just a few sentences. 215 00:07:45,680 --> 00:07:48,200 You get the main idea, but you miss a lot of the depth. 216 00:07:48,200 --> 00:07:51,480 So there's a risk of oversimplifying and potentially 217 00:07:51,480 --> 00:07:52,640 missing important info. 218 00:07:52,640 --> 00:07:54,480 That's definitely something to keep in mind. 219 00:07:54,480 --> 00:07:55,680 Any other limitations? 220 00:07:55,680 --> 00:07:58,200 Another challenge is making sure these AI explanations are 221 00:07:58,200 --> 00:08:00,160 actually accurate and unbiased. 222 00:08:00,160 --> 00:08:02,120 We need to be sure the language model isn't just 223 00:08:02,120 --> 00:08:05,280 telling a good story, but really reflecting how the AI system 224 00:08:05,280 --> 00:08:06,120 is thinking. 225 00:08:06,120 --> 00:08:08,560 Right, because if we're going to trust these explanations, 226 00:08:08,560 --> 00:08:10,400 they need to be reliable. 227 00:08:10,400 --> 00:08:12,240 So it's not just making AI explainable, 228 00:08:12,240 --> 00:08:15,120 it's making it explainable in a way that's accurate 229 00:08:15,120 --> 00:08:15,880 and meaningful. 230 00:08:15,880 --> 00:08:16,760 Exactly. 231 00:08:16,760 --> 00:08:18,480 And that means carefully thinking 232 00:08:18,480 --> 00:08:21,560 about how we design and train these language models 233 00:08:21,560 --> 00:08:24,520 and how we evaluate the quality of their explanations. 234 00:08:24,520 --> 00:08:26,960 It sounds like there's still a lot of work to be done, 235 00:08:26,960 --> 00:08:28,840 but this research definitely points us 236 00:08:28,840 --> 00:08:30,280 in a really exciting direction. 237 00:08:30,280 --> 00:08:31,920 It definitely does. 238 00:08:31,920 --> 00:08:33,920 This paper is a big contribution 239 00:08:33,920 --> 00:08:36,520 to the field of explainable AI. 240 00:08:36,520 --> 00:08:38,520 It's opening up all these new possibilities 241 00:08:38,520 --> 00:08:41,680 for how we can understand, interact with, and shape 242 00:08:41,680 --> 00:08:43,040 the future of AI. 243 00:08:43,040 --> 00:08:45,120 Well, I have to say this has been a really illuminating 244 00:08:45,120 --> 00:08:45,880 conversation. 245 00:08:45,880 --> 00:08:48,040 We've explored how Explingo works, 246 00:08:48,040 --> 00:08:51,240 its potential applications, and even those broader 247 00:08:51,240 --> 00:08:53,400 implications for the future of AI. 248 00:08:53,400 --> 00:08:55,120 But before we wrap up, is there anything else 249 00:08:55,120 --> 00:08:58,760 about this research that you find particularly interesting? 250 00:08:58,760 --> 00:09:00,360 One thing that really stands out to me 251 00:09:00,360 --> 00:09:04,040 is how Explingo could democratize AI. 252 00:09:04,040 --> 00:09:05,440 Democratize AI, what do you mean? 253 00:09:05,440 --> 00:09:05,840 Yes. 254 00:09:05,840 --> 00:09:09,520 Well, right now, understanding AI, working with it, 255 00:09:09,520 --> 00:09:11,880 it takes a lot of specialized knowledge. 256 00:09:11,880 --> 00:09:15,000 It's mostly data scientists, engineers. 257 00:09:15,000 --> 00:09:18,680 But if we can make those AI explanations easy for anyone 258 00:09:18,680 --> 00:09:21,720 to understand, even if they don't have a tech background, 259 00:09:21,720 --> 00:09:23,760 it could open up so many possibilities. 260 00:09:23,760 --> 00:09:27,240 So instead of AI being this exclusive club, 261 00:09:27,240 --> 00:09:29,760 it becomes something everyone can be part of, 262 00:09:29,760 --> 00:09:30,640 benefit from. 263 00:09:30,640 --> 00:09:31,280 Exactly. 264 00:09:31,280 --> 00:09:35,880 Imagine business owners, teachers, artists, even students, 265 00:09:35,880 --> 00:09:38,360 all using AI tools and understanding how they work, 266 00:09:38,360 --> 00:09:42,120 that could lead to incredible innovations in all kinds of fields. 267 00:09:42,120 --> 00:09:42,600 Yeah. 268 00:09:42,600 --> 00:09:45,480 It's like breaking down those barriers between the people 269 00:09:45,480 --> 00:09:48,040 who create AI and the people who use it. 270 00:09:48,040 --> 00:09:49,320 That's a really empowering thought. 271 00:09:49,320 --> 00:09:49,880 It is. 272 00:09:49,880 --> 00:09:51,880 And it also brings up the importance of education. 273 00:09:51,880 --> 00:09:54,280 If we really want to democratize AI, 274 00:09:54,280 --> 00:09:56,400 we've got to make sure people have the knowledge and skills 275 00:09:56,400 --> 00:09:57,600 to use it responsibly. 276 00:09:57,600 --> 00:10:00,040 So it's not just about AI being explainable. 277 00:10:00,040 --> 00:10:04,160 It's about teaching people how to interpret those explanations 278 00:10:04,160 --> 00:10:06,320 and make good decisions based on them. 279 00:10:06,320 --> 00:10:07,640 Precisely. 280 00:10:07,640 --> 00:10:11,720 We need to develop educational resources, training programs, 281 00:10:11,720 --> 00:10:15,840 stuff that empowers people to really think critically about AI. 282 00:10:15,840 --> 00:10:18,080 We can't just blindly trust the explanations, right? 283 00:10:18,080 --> 00:10:20,360 We need to evaluate them, understand their limits. 284 00:10:20,360 --> 00:10:23,800 It's like we're going from AI consumers to AI citizens. 285 00:10:23,800 --> 00:10:26,040 Everyone has a voice, a role to play in, 286 00:10:26,040 --> 00:10:27,840 shaping the future of this tech. 287 00:10:27,840 --> 00:10:28,360 I love that. 288 00:10:28,360 --> 00:10:30,480 Really highlights how important this research is. 289 00:10:30,480 --> 00:10:33,360 Making AI transparent and understandable. 290 00:10:33,360 --> 00:10:34,920 We're not just making it more usable. 291 00:10:34,920 --> 00:10:36,360 We're making it more accountable. 292 00:10:36,360 --> 00:10:37,280 Accountable to who? 293 00:10:37,280 --> 00:10:39,640 Accountable to society as a whole. 294 00:10:39,640 --> 00:10:42,080 If AI is going to be such a big part of our lives, 295 00:10:42,080 --> 00:10:44,040 we've got to make sure it's aligned with our values. 296 00:10:44,040 --> 00:10:48,200 That its decisions can be scrutinized, understood by everyone. 297 00:10:48,200 --> 00:10:50,600 So it's not just about making AI work. 298 00:10:50,600 --> 00:10:53,240 It's about making sure it works for good, 299 00:10:53,240 --> 00:10:54,400 for the good of humanity. 300 00:10:54,400 --> 00:10:55,160 Exactly. 301 00:10:55,160 --> 00:10:58,280 And that's where explainable AI is so crucial. 302 00:10:58,280 --> 00:11:00,760 It lets us hold those AI systems accountable, 303 00:11:00,760 --> 00:11:03,400 make sure they're being used ethically, responsibly. 304 00:11:03,400 --> 00:11:05,040 Like we're giving AI a conscience. 305 00:11:05,040 --> 00:11:06,040 That's a great way to put it. 306 00:11:06,040 --> 00:11:08,960 And it's a responsibility for all of us, researchers, 307 00:11:08,960 --> 00:11:09,920 developers, users. 308 00:11:09,920 --> 00:11:11,280 We've got to take this seriously. 309 00:11:11,280 --> 00:11:13,000 This has been a really thought-provoking dive. 310 00:11:13,000 --> 00:11:15,240 We've gone from the nitty-gritty of how 311 00:11:15,240 --> 00:11:20,760 Explingo works to its potential to reshape society, 312 00:11:20,760 --> 00:11:22,440 our whole relationship with AI. 313 00:11:22,440 --> 00:11:26,480 It's clear explainable AI is more than just a technical 314 00:11:26,480 --> 00:11:27,280 challenge. 315 00:11:27,280 --> 00:11:31,040 It's a fundamental shift in how we think about and interact 316 00:11:31,040 --> 00:11:32,600 with this incredible tech. 317 00:11:32,600 --> 00:11:33,520 I couldn't agree more. 318 00:11:33,520 --> 00:11:35,880 And I think this research is just the beginning. 319 00:11:35,880 --> 00:11:38,520 As AI keeps evolving, so will our efforts 320 00:11:38,520 --> 00:11:40,680 to understand and explain it. 321 00:11:40,680 --> 00:11:41,200 And who knows? 322 00:11:41,200 --> 00:11:44,400 Maybe even more amazing discoveries and advancements 323 00:11:44,400 --> 00:11:45,720 are just around the corner. 324 00:11:45,720 --> 00:11:47,600 It's a future full of possibilities. 325 00:11:47,600 --> 00:11:49,960 And for our listeners, we'd love to hear what you think. 326 00:11:49,960 --> 00:11:52,400 How do you think explainable AI will change your life? 327 00:11:52,400 --> 00:11:55,120 What would you ask an AI if it could explain itself to you? 328 00:11:55,120 --> 00:11:56,800 Share your thoughts with us. 329 00:11:56,800 --> 00:12:23,760 And until next time, happy diving.