1 00:00:00,000 --> 00:00:03,560 All right, everyone, welcome to another deep dive. 2 00:00:03,560 --> 00:00:06,360 Today, we're going to be looking at AI governance. 3 00:00:06,360 --> 00:00:08,280 Ooh, exciting. 4 00:00:08,280 --> 00:00:09,360 Yeah, it is, right? 5 00:00:09,360 --> 00:00:10,840 I feel like it's something we hear all the time, 6 00:00:10,840 --> 00:00:14,000 you know, AI governance this, AI governance that. 7 00:00:14,000 --> 00:00:16,280 But I think it can be kind of hard to wrap your head around. 8 00:00:16,280 --> 00:00:16,960 Absolutely. 9 00:00:16,960 --> 00:00:20,640 So you sent me some really cool research and articles 10 00:00:20,640 --> 00:00:23,240 about it, and I think we can unpack some of that today 11 00:00:23,240 --> 00:00:25,400 and help our listeners kind of understand, 12 00:00:25,400 --> 00:00:28,880 like, what are the key things to think about when 13 00:00:28,880 --> 00:00:30,840 we talk about AI governance? 14 00:00:30,840 --> 00:00:34,320 Like, what even is this wild west of AI 15 00:00:34,320 --> 00:00:36,080 that we hear about all the time? 16 00:00:36,080 --> 00:00:38,720 And like, why should I care about data governance 17 00:00:38,720 --> 00:00:44,440 if I'm not a data scientist or some kind of tech person? 18 00:00:44,440 --> 00:00:46,920 So maybe we can start with one of the first articles 19 00:00:46,920 --> 00:00:49,000 you sent me, which was from Gender, 20 00:00:49,000 --> 00:00:53,880 and it was about data governance, which is a term I feel 21 00:00:53,880 --> 00:00:56,600 like I hear all the time, but maybe you don't fully understand. 22 00:00:56,600 --> 00:00:59,640 Can you kind of break it down for us? 23 00:00:59,640 --> 00:01:02,760 Yeah, so what Gender really focuses on in this article 24 00:01:02,760 --> 00:01:06,920 is that data governance is really the foundation 25 00:01:06,920 --> 00:01:07,920 of AI governance. 26 00:01:07,920 --> 00:01:11,720 So AI governance begins with solid data governance, 27 00:01:11,720 --> 00:01:14,960 and that's because AI, its intelligence, 28 00:01:14,960 --> 00:01:18,520 relies on the data that fuels it. 29 00:01:18,520 --> 00:01:21,600 So if you have bad data or sensitive data 30 00:01:21,600 --> 00:01:23,680 that you're not supposed to be using, 31 00:01:23,680 --> 00:01:27,200 then it's going to lead to what we hear about AI hallucinations, 32 00:01:27,200 --> 00:01:29,920 which is where the AI just kind of makes things up 33 00:01:29,920 --> 00:01:33,200 or reveals information that could lead to a data breach 34 00:01:33,200 --> 00:01:35,880 or reputation damage to your company. 35 00:01:35,880 --> 00:01:38,000 So it's all about the quality of the data. 36 00:01:38,000 --> 00:01:39,400 It really is all about the quality of the data. 37 00:01:39,400 --> 00:01:41,080 That you're putting into these systems. 38 00:01:41,080 --> 00:01:43,680 Yeah, think of it like baking a cake. 39 00:01:43,680 --> 00:01:44,320 OK. 40 00:01:44,320 --> 00:01:45,840 You put in rotten eggs. 41 00:01:45,840 --> 00:01:46,320 Yeah. 42 00:01:46,320 --> 00:01:47,480 Not going to be a good cake. 43 00:01:47,480 --> 00:01:48,320 Makes sense. 44 00:01:48,320 --> 00:01:50,200 So it's really like garbage and garbage out. 45 00:01:50,200 --> 00:01:50,560 Yeah. 46 00:01:50,560 --> 00:01:53,280 You have to start with good data to get good results. 47 00:01:53,280 --> 00:01:55,880 So what are some real world consequences 48 00:01:55,880 --> 00:01:58,640 of using bad data in an AI? 49 00:01:58,640 --> 00:02:01,880 OK, so let's say you're a financial institution 50 00:02:01,880 --> 00:02:05,640 and you're using an AI system to assess loan applications. 51 00:02:05,640 --> 00:02:06,280 OK. 52 00:02:06,280 --> 00:02:09,600 But the data that's used to train your AI system 53 00:02:09,600 --> 00:02:13,400 has historical biases against certain demographics. 54 00:02:13,400 --> 00:02:14,640 Oh, wow. 55 00:02:14,640 --> 00:02:19,160 So that the AI could unknowingly perpetuate these biases 56 00:02:19,160 --> 00:02:21,240 and lead to unfair lending practices. 57 00:02:21,240 --> 00:02:23,000 Yeah, that's a huge deal. 58 00:02:23,000 --> 00:02:24,080 Yeah, that's a big one. 59 00:02:24,080 --> 00:02:26,880 And that's why data governance is so important in this context. 60 00:02:26,880 --> 00:02:28,760 Yeah, it's not just like a technical thing. 61 00:02:28,760 --> 00:02:31,520 It actually has real world consequences. 62 00:02:31,520 --> 00:02:33,360 Yeah, and it goes even beyond those biases, 63 00:02:33,360 --> 00:02:34,880 like you were saying. 64 00:02:34,880 --> 00:02:38,200 Inaccurate data can lead to these AI hallucinations 65 00:02:38,200 --> 00:02:40,440 where the AI generates outputs that are just 66 00:02:40,440 --> 00:02:42,960 like completely nonsensical. 67 00:02:42,960 --> 00:02:45,880 Let's say a health care provider is using an AI system 68 00:02:45,880 --> 00:02:47,560 to analyze medical images. 69 00:02:47,560 --> 00:02:48,120 OK. 70 00:02:48,120 --> 00:02:50,160 But the data that they use to train this AI 71 00:02:50,160 --> 00:02:51,560 is incomplete or corrupted. 72 00:02:51,560 --> 00:02:52,120 Right. 73 00:02:52,120 --> 00:02:54,440 The AI might misinterpret those images 74 00:02:54,440 --> 00:02:56,080 and lead to misdiagnoses. 75 00:02:56,080 --> 00:02:56,560 Oh, why not? 76 00:02:56,560 --> 00:03:00,000 Or even incorrect treatment recommendations, 77 00:03:00,000 --> 00:03:01,040 which is a scary thought. 78 00:03:01,040 --> 00:03:02,040 Yeah, that's not good. 79 00:03:02,040 --> 00:03:03,160 That's a bad situation. 80 00:03:03,160 --> 00:03:07,960 So data governance, we've established that it's important. 81 00:03:07,960 --> 00:03:11,480 But once we have this good data, what else 82 00:03:11,480 --> 00:03:13,080 do we need to think about when we're talking 83 00:03:13,080 --> 00:03:14,720 about governing these systems? 84 00:03:14,720 --> 00:03:17,600 Yeah, so then we need to think about model discovery. 85 00:03:17,600 --> 00:03:19,520 And this is once you have your data sorted. 86 00:03:19,520 --> 00:03:20,040 OK. 87 00:03:20,040 --> 00:03:22,480 Model discovery is essentially being 88 00:03:22,480 --> 00:03:27,120 aware of what specific AI models you're using 89 00:03:27,120 --> 00:03:28,160 and for what purpose. 90 00:03:28,160 --> 00:03:30,560 So we know we have good data, but we also 91 00:03:30,560 --> 00:03:33,520 need to know what specific tool we're using. 92 00:03:33,520 --> 00:03:34,280 Yes, exactly. 93 00:03:34,280 --> 00:03:34,840 OK. 94 00:03:34,840 --> 00:03:36,840 Think of it like you're doing a home repair. 95 00:03:36,840 --> 00:03:37,760 OK. 96 00:03:37,760 --> 00:03:39,080 You need to choose the right tool. 97 00:03:39,080 --> 00:03:39,680 Right. 98 00:03:39,680 --> 00:03:42,000 You wouldn't use a hammer to screw in a light bulb. 99 00:03:42,000 --> 00:03:42,720 Definitely not. 100 00:03:42,720 --> 00:03:43,040 Yeah. 101 00:03:43,040 --> 00:03:45,920 So it's all about understanding what is being used. 102 00:03:45,920 --> 00:03:46,680 That's right. 103 00:03:46,680 --> 00:03:48,080 And for what reason. 104 00:03:48,080 --> 00:03:49,760 OK, so why is that so important? 105 00:03:49,760 --> 00:03:52,680 Like why do I need to know what model is being used? 106 00:03:52,680 --> 00:03:53,800 Like what's the big deal? 107 00:03:53,800 --> 00:03:57,040 So as AI becomes more and more popular and integrated 108 00:03:57,040 --> 00:04:01,680 into our lives, stakeholders from customers to regulators 109 00:04:01,680 --> 00:04:04,400 are going to want to know, like what's under the hood, 110 00:04:04,400 --> 00:04:06,280 how are these decisions being made? 111 00:04:06,280 --> 00:04:07,000 Right. 112 00:04:07,000 --> 00:04:10,880 And this is especially important as regulations 113 00:04:10,880 --> 00:04:12,760 like the EU AI Act emerge. 114 00:04:12,760 --> 00:04:13,040 Right. 115 00:04:13,040 --> 00:04:15,040 One of the articles you sent me mentions that. 116 00:04:15,040 --> 00:04:15,560 Yes. 117 00:04:15,560 --> 00:04:16,320 That act specifically, yeah. 118 00:04:16,320 --> 00:04:16,840 Yeah. 119 00:04:16,840 --> 00:04:20,000 The EU AI Act, it categorizes AI systems 120 00:04:20,000 --> 00:04:21,240 based on their risk levels. 121 00:04:21,240 --> 00:04:21,720 OK. 122 00:04:21,720 --> 00:04:23,800 And the higher risk your application, 123 00:04:23,800 --> 00:04:25,520 the more regulations you're going to have. 124 00:04:25,520 --> 00:04:26,000 Right. 125 00:04:26,000 --> 00:04:27,160 The stricter requirements you're going 126 00:04:27,160 --> 00:04:28,840 to have for transparency and all of that. 127 00:04:28,840 --> 00:04:31,880 So it's not even just like, oh, nice to have. 128 00:04:31,880 --> 00:04:33,640 It's like, you need to do this. 129 00:04:33,640 --> 00:04:36,240 That's going to be a legal requirement in some cases. 130 00:04:36,240 --> 00:04:36,480 OK. 131 00:04:36,480 --> 00:04:37,680 So this is getting serious. 132 00:04:37,680 --> 00:04:38,640 Yes, it is getting serious. 133 00:04:38,640 --> 00:04:38,880 OK. 134 00:04:38,880 --> 00:04:41,320 So we have this good data. 135 00:04:41,320 --> 00:04:42,600 We know what model we're using. 136 00:04:42,600 --> 00:04:45,840 But then to get into like the actual meat of what 137 00:04:45,840 --> 00:04:49,640 good AI governance looks like, in the research 138 00:04:49,640 --> 00:04:54,120 that you sent me, gender had these five pillars 139 00:04:54,120 --> 00:04:55,560 of AI governance. 140 00:04:55,560 --> 00:04:56,080 Yes. 141 00:04:56,080 --> 00:04:59,000 Can we kind of break down what each of those are? 142 00:04:59,000 --> 00:04:59,480 Yeah. 143 00:04:59,480 --> 00:05:01,360 So the first one is transparency, which we already 144 00:05:01,360 --> 00:05:02,160 kind of talked about. 145 00:05:02,160 --> 00:05:04,400 So that's knowing which models you're using, 146 00:05:04,400 --> 00:05:08,640 understanding how they work, how they reach their conclusions, 147 00:05:08,640 --> 00:05:10,040 and what data was used. 148 00:05:10,040 --> 00:05:10,360 Yeah. 149 00:05:10,360 --> 00:05:12,880 So it's not just like a black box. 150 00:05:12,880 --> 00:05:13,760 Yeah, OK. 151 00:05:13,760 --> 00:05:16,280 And this can be achieved using what's 152 00:05:16,280 --> 00:05:20,440 called explainable AI, which focuses on AI systems that 153 00:05:20,440 --> 00:05:23,440 provide clear justifications for their decisions. 154 00:05:23,440 --> 00:05:24,240 Oh, that's cool. 155 00:05:24,240 --> 00:05:24,760 Yeah. 156 00:05:24,760 --> 00:05:26,240 So it's more understandable to humans. 157 00:05:26,240 --> 00:05:26,920 I like that. 158 00:05:26,920 --> 00:05:27,440 Yeah. 159 00:05:27,440 --> 00:05:31,600 So explainable AI, pillar number one, what about number two? 160 00:05:31,600 --> 00:05:33,720 Pillar number two is ethical use. 161 00:05:33,720 --> 00:05:34,640 OK. 162 00:05:34,640 --> 00:05:38,400 And that focuses on ensuring that AI systems are used 163 00:05:38,400 --> 00:05:40,560 in a way that aligns with ethical principles. 164 00:05:40,560 --> 00:05:41,080 OK. 165 00:05:41,080 --> 00:05:43,040 So tackling things like bias. 166 00:05:43,040 --> 00:05:43,360 Right. 167 00:05:43,360 --> 00:05:45,280 And making sure that it doesn't discriminate. 168 00:05:45,280 --> 00:05:45,680 OK. 169 00:05:45,680 --> 00:05:48,960 For example, if a company is using an AI-powered recruitment 170 00:05:48,960 --> 00:05:53,000 tool, they need to make sure that this AI is not unfairly 171 00:05:53,000 --> 00:05:54,880 discriminating against certain candidates. 172 00:05:54,880 --> 00:05:55,240 Right. 173 00:05:55,240 --> 00:05:58,160 So making sure that AI is being used for good, not for evil. 174 00:05:58,160 --> 00:05:58,680 Exactly. 175 00:05:58,680 --> 00:06:01,080 And then the third pillar is accountability. 176 00:06:01,080 --> 00:06:04,400 And this is as AI takes on more and more responsibility. 177 00:06:04,400 --> 00:06:04,960 Yeah. 178 00:06:04,960 --> 00:06:08,640 It's important to establish clear lines of accountability. 179 00:06:08,640 --> 00:06:08,960 Right. 180 00:06:08,960 --> 00:06:12,240 So like if something goes wrong, we know who's responsible. 181 00:06:12,240 --> 00:06:12,760 Exactly. 182 00:06:12,760 --> 00:06:16,520 So who is liable if an AI system makes a mistake? 183 00:06:16,520 --> 00:06:16,960 OK. 184 00:06:16,960 --> 00:06:19,040 For example, let's say a self-driving car 185 00:06:19,040 --> 00:06:20,280 gets into an accident. 186 00:06:20,280 --> 00:06:20,800 Right. 187 00:06:20,800 --> 00:06:22,120 Who is accountable? 188 00:06:22,120 --> 00:06:22,620 Right. 189 00:06:22,620 --> 00:06:25,760 Is it the developer, the manufacturer, the passenger? 190 00:06:25,760 --> 00:06:26,800 That's a good question. 191 00:06:26,800 --> 00:06:26,960 Yeah. 192 00:06:26,960 --> 00:06:28,640 These are all important questions to consider. 193 00:06:28,640 --> 00:06:28,920 OK. 194 00:06:28,920 --> 00:06:30,040 So that's accountability. 195 00:06:30,040 --> 00:06:31,240 What about number four? 196 00:06:31,240 --> 00:06:32,440 Number four is compliance. 197 00:06:32,440 --> 00:06:33,240 OK. 198 00:06:33,240 --> 00:06:36,040 And this means making sure that the AI systems follow 199 00:06:36,040 --> 00:06:37,400 the rules and regulations. 200 00:06:37,400 --> 00:06:39,400 So not just doing the right thing, 201 00:06:39,400 --> 00:06:41,480 but making sure you're following all the rules too. 202 00:06:41,480 --> 00:06:41,960 Exactly. 203 00:06:41,960 --> 00:06:44,240 Like data privacy regulations, for example. 204 00:06:44,240 --> 00:06:44,520 OK. 205 00:06:44,520 --> 00:06:47,600 So compliance, and then the last one, pillar number five. 206 00:06:47,600 --> 00:06:50,080 Pillar number five is risk management. 207 00:06:50,080 --> 00:06:50,560 OK. 208 00:06:50,560 --> 00:06:53,880 And that focuses on identifying and mitigating 209 00:06:53,880 --> 00:06:55,200 any potential risks. 210 00:06:55,200 --> 00:06:57,160 So being prepared? 211 00:06:57,160 --> 00:06:57,800 Exactly. 212 00:06:57,800 --> 00:06:58,800 Like thinking ahead? 213 00:06:58,800 --> 00:06:59,520 Yes. 214 00:06:59,520 --> 00:07:02,880 Thinking ahead, taking a proactive approach 215 00:07:02,880 --> 00:07:05,480 to AI safety and security. 216 00:07:05,480 --> 00:07:09,600 So considering data breaches, algorithmic bias, 217 00:07:09,600 --> 00:07:11,560 and any unintended consequences. 218 00:07:11,560 --> 00:07:11,840 Right. 219 00:07:11,840 --> 00:07:14,200 Because sometimes you don't know what those are. 220 00:07:14,200 --> 00:07:14,240 Right. 221 00:07:14,240 --> 00:07:14,760 Exactly. 222 00:07:14,760 --> 00:07:18,480 And so risk management is just having a plan in place 223 00:07:18,480 --> 00:07:19,720 to address these issues. 224 00:07:19,720 --> 00:07:22,560 So it's really about thinking about all these different facets 225 00:07:22,560 --> 00:07:24,200 of what could go wrong. 226 00:07:24,200 --> 00:07:24,880 Yeah. 227 00:07:24,880 --> 00:07:25,080 OK. 228 00:07:25,080 --> 00:07:28,000 So those five pillars, I really like that framework. 229 00:07:28,000 --> 00:07:30,560 I think it's helpful to kind of break down 230 00:07:30,560 --> 00:07:32,520 what responsible AI development looks like. 231 00:07:32,520 --> 00:07:33,000 Exactly. 232 00:07:33,000 --> 00:07:34,360 It's a really good framework. 233 00:07:34,360 --> 00:07:36,560 So we've talked about data governance. 234 00:07:36,560 --> 00:07:38,520 But in one of the other pieces you sent me, 235 00:07:38,520 --> 00:07:42,840 Matter talks about this idea of the wild west of AI. 236 00:07:42,840 --> 00:07:43,480 Yes. 237 00:07:43,480 --> 00:07:44,400 That's a good one. 238 00:07:44,400 --> 00:07:45,600 What does that even mean? 239 00:07:45,600 --> 00:07:47,880 Like, help me understand that analogy. 240 00:07:47,880 --> 00:07:48,080 Yeah. 241 00:07:48,080 --> 00:07:49,480 So what Matter's really talking about 242 00:07:49,480 --> 00:07:52,240 is that in a lot of companies, there's 243 00:07:52,240 --> 00:07:55,920 a lack of control and visibility when it comes to AI. 244 00:07:55,920 --> 00:07:59,080 So people are experimenting with AI in different departments 245 00:07:59,080 --> 00:08:01,720 using different tools, different data sets. 246 00:08:01,720 --> 00:08:04,240 And oftentimes there aren't clear guidelines in place. 247 00:08:04,240 --> 00:08:06,240 So it's like everyone's just kind of doing their own thing. 248 00:08:06,240 --> 00:08:06,520 Yeah. 249 00:08:06,520 --> 00:08:06,880 Exactly. 250 00:08:06,880 --> 00:08:07,920 And there's no one to kind of. 251 00:08:07,920 --> 00:08:09,280 Yeah, there's no sheriff in town. 252 00:08:09,280 --> 00:08:10,200 Wrangle it all in. 253 00:08:10,200 --> 00:08:10,880 Exactly. 254 00:08:10,880 --> 00:08:11,400 OK. 255 00:08:11,400 --> 00:08:13,040 And this can be really risky. 256 00:08:13,040 --> 00:08:14,800 What kind of risks are we talking about here? 257 00:08:14,800 --> 00:08:19,240 So if you don't know where AI is being used in your company, 258 00:08:19,240 --> 00:08:21,480 you can't effectively govern it. 259 00:08:21,480 --> 00:08:21,960 OK. 260 00:08:21,960 --> 00:08:25,040 You might have different teams using different models. 261 00:08:25,040 --> 00:08:25,480 Yes. 262 00:08:25,480 --> 00:08:27,280 That have conflicting assumptions. 263 00:08:27,280 --> 00:08:27,720 Right. 264 00:08:27,720 --> 00:08:29,400 Or even contradictory goals. 265 00:08:29,400 --> 00:08:29,640 Yeah. 266 00:08:29,640 --> 00:08:32,080 So it's like having a bunch of chefs in a kitchen 267 00:08:32,080 --> 00:08:33,520 all making different dishes. 268 00:08:33,520 --> 00:08:34,200 Yes. 269 00:08:34,200 --> 00:08:36,000 And you don't know what you're going to get. 270 00:08:36,000 --> 00:08:36,480 Yeah. 271 00:08:36,480 --> 00:08:37,120 It could be a mess. 272 00:08:37,120 --> 00:08:38,520 And you don't know what you're going to get. 273 00:08:38,520 --> 00:08:39,920 It could be a disaster. 274 00:08:39,920 --> 00:08:43,120 So then how do we address that? 275 00:08:43,120 --> 00:08:44,960 How do we wrangle this Wild West? 276 00:08:44,960 --> 00:08:45,440 Yeah. 277 00:08:45,440 --> 00:08:50,320 So what you can do is establish a centralized AI governance 278 00:08:50,320 --> 00:08:50,840 function. 279 00:08:50,840 --> 00:08:51,720 OK. 280 00:08:51,720 --> 00:08:55,480 So it could be a dedicated team or individual 281 00:08:55,480 --> 00:08:58,240 that's responsible for implementing these policies 282 00:08:58,240 --> 00:08:59,040 across the board. 283 00:08:59,040 --> 00:09:00,040 So like a sheriff? 284 00:09:00,040 --> 00:09:00,600 Yeah, exactly. 285 00:09:00,600 --> 00:09:01,480 The sheriff comes in. 286 00:09:01,480 --> 00:09:02,840 To kind of keep everyone in line. 287 00:09:02,840 --> 00:09:03,360 Yeah. 288 00:09:03,360 --> 00:09:07,080 But it sounds like that could be a big lift for smaller 289 00:09:07,080 --> 00:09:10,160 organizations that maybe don't have a ton of resources. 290 00:09:10,160 --> 00:09:10,480 Right. 291 00:09:10,480 --> 00:09:13,080 And the good news is that AI governance doesn't 292 00:09:13,080 --> 00:09:15,320 have to be an all or nothing proposition. 293 00:09:15,320 --> 00:09:15,720 OK. 294 00:09:15,720 --> 00:09:17,640 You can implement it in stages. 295 00:09:17,640 --> 00:09:18,040 OK. 296 00:09:18,040 --> 00:09:20,000 Starting with the most critical areas. 297 00:09:20,000 --> 00:09:20,400 Right. 298 00:09:20,400 --> 00:09:22,000 And then gradually expanding it. 299 00:09:22,000 --> 00:09:23,200 So it just starts somewhere. 300 00:09:23,200 --> 00:09:23,840 Start somewhere. 301 00:09:23,840 --> 00:09:24,920 Take those first steps. 302 00:09:24,920 --> 00:09:25,200 OK. 303 00:09:25,200 --> 00:09:26,880 So we've talked about good data. 304 00:09:26,880 --> 00:09:27,400 Right. 305 00:09:27,400 --> 00:09:29,800 And the importance of like model discovery. 306 00:09:29,800 --> 00:09:31,840 But in another one of the articles you sent, 307 00:09:31,840 --> 00:09:33,800 Roy had a really interesting point. 308 00:09:33,800 --> 00:09:34,280 Yes. 309 00:09:34,280 --> 00:09:38,400 About the need for a strong engineering perspective 310 00:09:38,400 --> 00:09:40,560 when we're thinking about AI governance. 311 00:09:40,560 --> 00:09:40,840 Yeah. 312 00:09:40,840 --> 00:09:43,760 He really emphasized the importance 313 00:09:43,760 --> 00:09:47,000 of defining the problem you're trying to solve with AI. 314 00:09:47,000 --> 00:09:47,680 OK. 315 00:09:47,680 --> 00:09:51,640 Before you even think about data or algorithms. 316 00:09:51,640 --> 00:09:53,000 So before you get all technical. 317 00:09:53,000 --> 00:09:53,280 Yeah. 318 00:09:53,280 --> 00:09:54,080 Before you get into all that. 319 00:09:54,080 --> 00:09:57,640 You need to think about like the big picture of what 320 00:09:57,640 --> 00:09:59,160 we're even trying to do. 321 00:09:59,160 --> 00:09:59,720 Exactly. 322 00:09:59,720 --> 00:10:01,360 And like why are we doing this? 323 00:10:01,360 --> 00:10:01,920 Yeah. 324 00:10:01,920 --> 00:10:03,320 What problem are we trying to solve? 325 00:10:03,320 --> 00:10:04,320 Right. 326 00:10:04,320 --> 00:10:07,000 Because you can end up creating a whole new set of problems. 327 00:10:07,000 --> 00:10:08,040 You can create more problems. 328 00:10:08,040 --> 00:10:09,000 You're not careful. 329 00:10:09,000 --> 00:10:11,360 And it's not just identifying the problem. 330 00:10:11,360 --> 00:10:14,840 It's also understanding the context in which the AI will 331 00:10:14,840 --> 00:10:16,680 be operating. 332 00:10:16,680 --> 00:10:19,120 What do you mean by context? 333 00:10:19,120 --> 00:10:22,520 So let's say a company is developing an AI system 334 00:10:22,520 --> 00:10:24,920 to personalize product recommendations. 335 00:10:24,920 --> 00:10:25,360 OK. 336 00:10:25,360 --> 00:10:26,520 The context matters. 337 00:10:26,520 --> 00:10:26,720 Right. 338 00:10:26,720 --> 00:10:30,400 So a recommendation engine for like a clothing store 339 00:10:30,400 --> 00:10:32,040 is going to be very different than one 340 00:10:32,040 --> 00:10:33,120 for a health care website. 341 00:10:33,120 --> 00:10:33,600 Exactly. 342 00:10:33,600 --> 00:10:36,320 You have to think about things like user privacy, data 343 00:10:36,320 --> 00:10:40,000 security, and potential ethical implications. 344 00:10:40,000 --> 00:10:40,320 Right. 345 00:10:40,320 --> 00:10:43,800 Because like what might be an OK recommendation in one 346 00:10:43,800 --> 00:10:46,560 context could be really bad in another. 347 00:10:46,560 --> 00:10:47,240 Exactly. 348 00:10:47,240 --> 00:10:50,240 So you need to think about all that stuff up front 349 00:10:50,240 --> 00:10:52,560 before you even get into the technical bits. 350 00:10:52,560 --> 00:10:53,920 Before you write any code. 351 00:10:53,920 --> 00:10:54,440 OK. 352 00:10:54,440 --> 00:10:56,760 And I think Roy's point about diverse perspectives 353 00:10:56,760 --> 00:10:58,280 is really key here. 354 00:10:58,280 --> 00:11:00,880 It's not just the engineers who should be in the room 355 00:11:00,880 --> 00:11:02,320 when we're making these decisions, right? 356 00:11:02,320 --> 00:11:02,880 Absolutely not. 357 00:11:02,880 --> 00:11:04,640 Like who else should we be talking to? 358 00:11:04,640 --> 00:11:07,920 We should be talking to data scientists, product managers, 359 00:11:07,920 --> 00:11:10,760 legal experts, even end users. 360 00:11:10,760 --> 00:11:11,760 Yeah. 361 00:11:11,760 --> 00:11:12,360 That makes sense. 362 00:11:12,360 --> 00:11:14,000 It has to be a collaborative effort. 363 00:11:14,000 --> 00:11:15,600 Because if you just have a bunch of engineers, 364 00:11:15,600 --> 00:11:17,520 they might build something super cool that 365 00:11:17,520 --> 00:11:19,120 is totally unusable. 366 00:11:19,120 --> 00:11:19,480 Exactly. 367 00:11:19,480 --> 00:11:20,400 Or like, you know. 368 00:11:20,400 --> 00:11:21,200 Unethical. 369 00:11:21,200 --> 00:11:21,800 Unethical. 370 00:11:21,800 --> 00:11:22,200 Yeah. 371 00:11:22,200 --> 00:11:23,840 So it's important to get everyone's perspective. 372 00:11:23,840 --> 00:11:24,160 OK. 373 00:11:24,160 --> 00:11:26,960 So we've talked about a lot of internal things 374 00:11:26,960 --> 00:11:28,320 that companies need to do. 375 00:11:28,320 --> 00:11:28,760 Yeah. 376 00:11:28,760 --> 00:11:31,440 But one of the articles you sent me also mentioned third party 377 00:11:31,440 --> 00:11:33,480 risks, which I thought was really interesting. 378 00:11:33,480 --> 00:11:33,680 Yeah. 379 00:11:33,680 --> 00:11:34,760 That's a big one. 380 00:11:34,760 --> 00:11:36,160 So can you talk a little bit about that? 381 00:11:36,160 --> 00:11:37,480 Like what is that? 382 00:11:37,480 --> 00:11:41,200 So third party risks are essentially 383 00:11:41,200 --> 00:11:45,760 even if your organization has strong AI governance 384 00:11:45,760 --> 00:11:50,520 practices, you could still be exposed to risks 385 00:11:50,520 --> 00:11:52,640 if the vendors that you're working with 386 00:11:52,640 --> 00:11:54,120 don't have those same standards. 387 00:11:54,120 --> 00:11:57,360 So it's like you're only as strong as your weakest link. 388 00:11:57,360 --> 00:11:59,480 You're only as strong as your weakest link, exactly. 389 00:11:59,480 --> 00:11:59,880 OK. 390 00:11:59,880 --> 00:12:01,880 So let's say you're a bank and you're 391 00:12:01,880 --> 00:12:05,400 using a third party vendor for customer service chatbots. 392 00:12:05,400 --> 00:12:06,240 OK. 393 00:12:06,240 --> 00:12:10,320 And that vendor might be using an AI model that's 394 00:12:10,320 --> 00:12:12,800 trained on data that has bias. 395 00:12:12,800 --> 00:12:13,760 Right. 396 00:12:13,760 --> 00:12:17,560 Even if your AI system internally is perfectly fine, 397 00:12:17,560 --> 00:12:18,760 you're still going to be at risk. 398 00:12:18,760 --> 00:12:19,080 Right. 399 00:12:19,080 --> 00:12:20,720 Because you're using this other system. 400 00:12:20,720 --> 00:12:21,280 Right. 401 00:12:21,280 --> 00:12:22,240 Yeah, that's tricky. 402 00:12:22,240 --> 00:12:23,240 It's very tricky, yeah. 403 00:12:23,240 --> 00:12:27,760 So how do you even know to ask these questions of your vendors? 404 00:12:27,760 --> 00:12:32,560 Yeah, well, I think including AI governance requirements 405 00:12:32,560 --> 00:12:34,560 in your contracts with your vendors 406 00:12:34,560 --> 00:12:36,080 is a good place to start. 407 00:12:36,080 --> 00:12:39,840 So making it really clear upfront what your expectations are. 408 00:12:39,840 --> 00:12:41,680 Yeah, like what AI models are you using? 409 00:12:41,680 --> 00:12:43,640 What data are you using to train them? 410 00:12:43,640 --> 00:12:43,960 Right. 411 00:12:43,960 --> 00:12:45,520 How are you mitigating bias? 412 00:12:45,520 --> 00:12:48,240 So it's really about having those conversations early on. 413 00:12:48,240 --> 00:12:48,800 Early on. 414 00:12:48,800 --> 00:12:50,560 And then, of course, ongoing monitoring. 415 00:12:50,560 --> 00:12:51,200 OK. 416 00:12:51,200 --> 00:12:53,080 You have to make sure that your vendors are actually 417 00:12:53,080 --> 00:12:54,520 adhering to those standards. 418 00:12:54,520 --> 00:12:56,840 So trusting but verify. 419 00:12:56,840 --> 00:12:57,440 Exactly. 420 00:12:57,440 --> 00:12:58,640 Trusting but verifying. 421 00:12:58,640 --> 00:12:59,360 Got it. 422 00:12:59,360 --> 00:13:02,800 And this can involve things like document review, audits, 423 00:13:02,800 --> 00:13:06,920 or even running tests on the AI system itself. 424 00:13:06,920 --> 00:13:09,120 OK, so this is getting very complicated. 425 00:13:09,120 --> 00:13:10,560 It can be very complicated. 426 00:13:10,560 --> 00:13:11,560 It's a lot to think about. 427 00:13:11,560 --> 00:13:12,720 It is a lot to think about. 428 00:13:12,720 --> 00:13:14,720 And it sounds like these third-party risks 429 00:13:14,720 --> 00:13:17,120 are something we really need to be paying attention to. 430 00:13:17,120 --> 00:13:20,120 You really do, because AI is constantly evolving. 431 00:13:20,120 --> 00:13:20,560 Yeah. 432 00:13:20,560 --> 00:13:23,440 And new risks emerge all the time, so you have to stay informed. 433 00:13:23,440 --> 00:13:24,200 Wow, OK. 434 00:13:24,200 --> 00:13:27,120 So it's like a constant process of learning. 435 00:13:27,120 --> 00:13:27,620 Yes. 436 00:13:27,620 --> 00:13:29,040 OK, so we've covered so much today. 437 00:13:29,040 --> 00:13:31,520 I feel like I have a much better understanding 438 00:13:31,520 --> 00:13:33,280 of what AI governance is. 439 00:13:33,280 --> 00:13:33,780 Yeah. 440 00:13:33,780 --> 00:13:35,480 Maybe we can just take a second to kind of summarize 441 00:13:35,480 --> 00:13:36,360 what we've talked about. 442 00:13:36,360 --> 00:13:37,200 Yeah, good idea. 443 00:13:37,200 --> 00:13:41,720 So number one, AI governance is a lot more than just data 444 00:13:41,720 --> 00:13:42,440 governance. 445 00:13:42,440 --> 00:13:43,000 Right. 446 00:13:43,000 --> 00:13:45,160 It's really about all these different things, 447 00:13:45,160 --> 00:13:47,440 like transparency, ethics. 448 00:13:47,440 --> 00:13:51,240 Accountability, compliance, risk management. 449 00:13:51,240 --> 00:13:52,640 Yeah, it's a lot. 450 00:13:52,640 --> 00:13:53,160 It is. 451 00:13:53,160 --> 00:13:55,680 And then, you know, we talked about how data governance is 452 00:13:55,680 --> 00:13:59,760 so important, because the quality of the data 453 00:13:59,760 --> 00:14:03,560 that you're putting in really affects the results. 454 00:14:03,560 --> 00:14:04,320 The results. 455 00:14:04,320 --> 00:14:06,680 And then we also talked about model discovery 456 00:14:06,680 --> 00:14:10,880 and why it matters to know what model you're using. 457 00:14:10,880 --> 00:14:12,440 Exactly for what purpose. 458 00:14:12,440 --> 00:14:12,940 Yeah. 459 00:14:12,940 --> 00:14:13,840 Transparency is key. 460 00:14:13,840 --> 00:14:15,640 And then we got into those five pillars, 461 00:14:15,640 --> 00:14:17,480 which I thought was really helpful. 462 00:14:17,480 --> 00:14:18,200 Really good frame rate. 463 00:14:18,200 --> 00:14:20,120 And kind of break down all the things 464 00:14:20,120 --> 00:14:21,680 that we need to be thinking about. 465 00:14:21,680 --> 00:14:22,180 Yes. 466 00:14:22,180 --> 00:14:25,120 And then, of course, this idea of the Wild West. 467 00:14:25,120 --> 00:14:26,440 The Wild West, yes. 468 00:14:26,440 --> 00:14:27,800 Which is a little bit scary. 469 00:14:27,800 --> 00:14:28,640 It is a little scary. 470 00:14:28,640 --> 00:14:32,080 But it sounds like there are things that organizations can do 471 00:14:32,080 --> 00:14:34,200 to kind of wrangle that. 472 00:14:34,200 --> 00:14:37,360 And then, of course, these third party risks, which I think 473 00:14:37,360 --> 00:14:38,840 is something that we often forget about. 474 00:14:38,840 --> 00:14:40,320 We do, yes. 475 00:14:40,320 --> 00:14:41,120 So it's important. 476 00:14:41,120 --> 00:14:41,800 Yeah, it's a lot. 477 00:14:41,800 --> 00:14:44,640 But I feel like we have a really good foundation. 478 00:14:44,640 --> 00:14:45,720 I think so, too. 479 00:14:45,720 --> 00:14:48,440 For understanding AI governance. 480 00:14:48,440 --> 00:14:49,160 Yes. 481 00:14:49,160 --> 00:14:49,520 Now. 482 00:14:49,520 --> 00:14:50,600 Good place to start. 483 00:14:50,600 --> 00:14:51,440 Yeah. 484 00:14:51,440 --> 00:14:53,200 It really is a lot to consider, isn't it? 485 00:14:53,200 --> 00:14:53,520 Yeah. 486 00:14:53,520 --> 00:14:55,080 You know, one thing that I think is really interesting 487 00:14:55,080 --> 00:14:59,440 that Matter brings up is this idea of really understanding 488 00:14:59,440 --> 00:15:03,160 where the AI is being used within a company. 489 00:15:03,160 --> 00:15:03,660 Right. 490 00:15:03,660 --> 00:15:06,600 It's one thing to have these high level policies in place. 491 00:15:06,600 --> 00:15:11,120 But you need to actually know the specifics of, 492 00:15:11,120 --> 00:15:12,520 is it being used in marketing? 493 00:15:12,520 --> 00:15:14,320 Is it being used in finance? 494 00:15:14,320 --> 00:15:18,200 So it's not just a company having an AI ethics statement up 495 00:15:18,200 --> 00:15:18,760 on their website. 496 00:15:18,760 --> 00:15:21,920 It's actually knowing where is this stuff happening. 497 00:15:21,920 --> 00:15:22,420 Yeah. 498 00:15:22,420 --> 00:15:26,740 You have to know which teams are using, which tools, what 499 00:15:26,740 --> 00:15:27,560 data are they using? 500 00:15:27,560 --> 00:15:29,040 Like what are they actually using it for? 501 00:15:29,040 --> 00:15:29,440 Yeah. 502 00:15:29,440 --> 00:15:32,320 Because if you don't have that, it's kind of like you're 503 00:15:32,320 --> 00:15:34,880 just trying to steer a ship, but you don't know where you are 504 00:15:34,880 --> 00:15:35,480 on the map. 505 00:15:35,480 --> 00:15:38,120 You don't know what direction you're going in. 506 00:15:38,120 --> 00:15:42,080 So it's like, even if you have this great AI governance team 507 00:15:42,080 --> 00:15:43,680 and you have these wonderful policies, 508 00:15:43,680 --> 00:15:47,960 if you don't know where the rubber meets the road, 509 00:15:47,960 --> 00:15:49,920 then it's not going to be very effective. 510 00:15:49,920 --> 00:15:50,800 That's not going to work. 511 00:15:50,800 --> 00:15:51,080 Yeah. 512 00:15:51,080 --> 00:15:53,480 OK, can you give me a specific example 513 00:15:53,480 --> 00:15:58,320 of what could go wrong if you don't have this granular level 514 00:15:58,320 --> 00:15:59,160 of understanding? 515 00:15:59,160 --> 00:15:59,280 Yeah. 516 00:15:59,280 --> 00:16:02,160 Let's say your marketing team is using an AI tool 517 00:16:02,160 --> 00:16:04,160 to personalize email campaigns. 518 00:16:04,160 --> 00:16:05,040 OK. 519 00:16:05,040 --> 00:16:08,960 But they don't realize that the data that is feeding into that 520 00:16:08,960 --> 00:16:13,360 tool contains sensitive customer information that should not 521 00:16:13,360 --> 00:16:15,320 be used for marketing purposes. 522 00:16:15,320 --> 00:16:15,960 No. 523 00:16:15,960 --> 00:16:16,480 Right. 524 00:16:16,480 --> 00:16:18,360 So then you have a potential privates for violation. 525 00:16:18,360 --> 00:16:19,080 That's a problem. 526 00:16:19,080 --> 00:16:19,360 Yeah. 527 00:16:19,360 --> 00:16:20,320 And reputational damage. 528 00:16:20,320 --> 00:16:21,960 Yeah, that's a big one. 529 00:16:21,960 --> 00:16:24,560 And if you don't know where the AI is being used 530 00:16:24,560 --> 00:16:26,280 and what data it's accessing, it's 531 00:16:26,280 --> 00:16:28,360 really easy for those oversights to occur. 532 00:16:28,360 --> 00:16:31,720 So it's really all about having the systems and processes 533 00:16:31,720 --> 00:16:33,840 in place to keep track of this stuff. 534 00:16:33,840 --> 00:16:35,360 Yeah, it's not just about the intentions. 535 00:16:35,360 --> 00:16:37,520 It's like, how are you actually going to implement this? 536 00:16:37,520 --> 00:16:37,920 Yeah. 537 00:16:37,920 --> 00:16:39,560 How are you going to monitor this stuff? 538 00:16:39,560 --> 00:16:42,640 So this kind of connects back to the model discovery 539 00:16:42,640 --> 00:16:43,880 that we were talking about. 540 00:16:43,880 --> 00:16:47,480 It's like, you not only need to know what model you have, 541 00:16:47,480 --> 00:16:49,960 but you need to know where it's being used. 542 00:16:49,960 --> 00:16:53,680 Yeah, where it's being used and for what purpose. 543 00:16:53,680 --> 00:16:54,160 Yeah. 544 00:16:54,160 --> 00:16:55,720 It's all connected. 545 00:16:55,720 --> 00:16:57,640 And I imagine this becomes even more important 546 00:16:57,640 --> 00:17:01,120 when you start talking about third party risks and vendors. 547 00:17:01,120 --> 00:17:01,720 Absolutely. 548 00:17:01,720 --> 00:17:03,840 Because it's one thing if it's your own company, 549 00:17:03,840 --> 00:17:06,480 but if you're sharing data with other people 550 00:17:06,480 --> 00:17:08,600 and you don't even know what they're doing with it, 551 00:17:08,600 --> 00:17:10,200 that can be really scary. 552 00:17:10,200 --> 00:17:15,040 Right, because even if you have great data security practices 553 00:17:15,040 --> 00:17:18,800 in place, if you're sharing that data with a company that 554 00:17:18,800 --> 00:17:19,920 doesn't, you're evolving. 555 00:17:19,920 --> 00:17:20,520 It's vulnerable. 556 00:17:20,520 --> 00:17:22,640 It's like having a really secure house, 557 00:17:22,640 --> 00:17:26,440 but then you just give your key to a random person. 558 00:17:26,440 --> 00:17:27,200 Right, exactly. 559 00:17:27,200 --> 00:17:28,320 Like, that's not very helpful. 560 00:17:28,320 --> 00:17:29,320 It defeats the purpose. 561 00:17:29,320 --> 00:17:31,160 OK, so we're working with all these vendors. 562 00:17:31,160 --> 00:17:32,640 We're sharing all this data. 563 00:17:32,640 --> 00:17:35,640 How do we even know what questions to ask them? 564 00:17:35,640 --> 00:17:37,520 Right, so you want to ask them things like, 565 00:17:37,520 --> 00:17:39,320 what AI models are you using? 566 00:17:39,320 --> 00:17:41,000 Where did you get the data to train them? 567 00:17:41,000 --> 00:17:43,640 What are you doing to make sure that it's ethical? 568 00:17:43,640 --> 00:17:45,480 How are you mitigating bias? 569 00:17:45,480 --> 00:17:48,360 Yeah, so it's really like having those open and honest 570 00:17:48,360 --> 00:17:49,520 conversations. 571 00:17:49,520 --> 00:17:50,040 Exactly. 572 00:17:50,040 --> 00:17:51,640 With those third parties to make sure 573 00:17:51,640 --> 00:17:52,960 that everyone's on the same page. 574 00:17:52,960 --> 00:17:54,560 Yeah, it's not just about checking a box. 575 00:17:54,560 --> 00:17:56,160 It's like actually understanding what's going on. 576 00:17:56,160 --> 00:18:00,360 Right, because they might have a great AI ethics statement too. 577 00:18:00,360 --> 00:18:02,360 But if they're not actually doing the work, 578 00:18:02,360 --> 00:18:03,640 then it doesn't really matter. 579 00:18:03,640 --> 00:18:04,320 Exactly. 580 00:18:04,320 --> 00:18:08,080 OK, so this is really about a deeper level of. 581 00:18:08,080 --> 00:18:09,320 Yeah, it's about the trust. 582 00:18:09,320 --> 00:18:10,400 Trust and understanding. 583 00:18:10,400 --> 00:18:11,000 And communication. 584 00:18:11,000 --> 00:18:11,800 Yeah. 585 00:18:11,800 --> 00:18:14,240 OK, and this kind of brings us back to Roy's point 586 00:18:14,240 --> 00:18:16,480 about diverse perspectives. 587 00:18:16,480 --> 00:18:19,040 It's not just the engineers who need 588 00:18:19,040 --> 00:18:19,800 to be in the room. 589 00:18:19,800 --> 00:18:25,960 We need the ethicists, the legal experts, the data 590 00:18:25,960 --> 00:18:28,520 scientists, all these different people. 591 00:18:28,520 --> 00:18:29,400 And the end users. 592 00:18:29,400 --> 00:18:30,360 Right, the end users. 593 00:18:30,360 --> 00:18:32,520 Because if you just have a bunch of engineers, 594 00:18:32,520 --> 00:18:34,920 you might end up with something that's really 595 00:18:34,920 --> 00:18:38,920 technically brilliant, but completely misses the mark 596 00:18:38,920 --> 00:18:41,400 in terms of how it's actually going to be used. 597 00:18:41,400 --> 00:18:46,720 It's like designing a car that's super fast and efficient, 598 00:18:46,720 --> 00:18:48,160 but it has no brakes. 599 00:18:48,160 --> 00:18:49,040 Yeah, exactly. 600 00:18:49,040 --> 00:18:50,040 What's the point? 601 00:18:50,040 --> 00:18:52,720 It's cool, but it's not going to work. 602 00:18:52,720 --> 00:18:55,520 So you need all these different people to come together. 603 00:18:55,520 --> 00:18:55,920 Absolutely. 604 00:18:55,920 --> 00:18:58,160 To kind of check each other's blind spots. 605 00:18:58,160 --> 00:18:58,680 Yeah. 606 00:18:58,680 --> 00:19:00,720 I like that it's like a team effort. 607 00:19:00,720 --> 00:19:02,080 It is a team effort, for sure. 608 00:19:02,080 --> 00:19:06,120 And this isn't just about building AI responsibly 609 00:19:06,120 --> 00:19:07,120 within a company. 610 00:19:07,120 --> 00:19:11,600 It's also about building trust with the public. 611 00:19:11,600 --> 00:19:12,280 Absolutely. 612 00:19:12,280 --> 00:19:15,840 Right, because we hear all these stories about AI gone wrong. 613 00:19:15,840 --> 00:19:18,360 And people are understandably, I think, a little bit 614 00:19:18,360 --> 00:19:21,160 hesitant about what this all means. 615 00:19:21,160 --> 00:19:22,360 It's scary for people. 616 00:19:22,360 --> 00:19:25,440 So if companies are being really transparent and showing 617 00:19:25,440 --> 00:19:27,640 that they're thinking about these things, 618 00:19:27,640 --> 00:19:30,040 maybe we can start to feel a little bit more comfortable. 619 00:19:30,040 --> 00:19:32,320 Exactly, and it's not just transparency. 620 00:19:32,320 --> 00:19:34,240 It's accountability, too. 621 00:19:34,240 --> 00:19:36,640 You have to be willing to take responsibility 622 00:19:36,640 --> 00:19:37,760 if things go wrong. 623 00:19:37,760 --> 00:19:39,760 Right, because they probably will at some point. 624 00:19:39,760 --> 00:19:40,520 Oh, they will. 625 00:19:40,520 --> 00:19:41,400 It's inevitable. 626 00:19:41,400 --> 00:19:42,240 It's technology. 627 00:19:42,240 --> 00:19:43,000 Things break. 628 00:19:43,000 --> 00:19:43,720 Things break. 629 00:19:43,720 --> 00:19:46,800 So it's more about how do we deal with that? 630 00:19:46,800 --> 00:19:47,760 When it happens. 631 00:19:47,760 --> 00:19:49,040 Yeah, how do we respond? 632 00:19:49,040 --> 00:19:50,560 And I think as we're talking about this, 633 00:19:50,560 --> 00:19:54,840 it strikes me that it's not just about big companies, right? 634 00:19:54,840 --> 00:19:56,560 This is stuff that we as individuals 635 00:19:56,560 --> 00:19:57,920 need to be thinking about, too. 636 00:19:57,920 --> 00:19:58,420 Absolutely. 637 00:19:58,420 --> 00:20:00,440 We're all impacted by AI. 638 00:20:00,440 --> 00:20:03,160 Yeah, think about the algorithms that 639 00:20:03,160 --> 00:20:06,880 are shaping our social media feeds or recommending 640 00:20:06,880 --> 00:20:10,800 products to us or even facial recognition technology. 641 00:20:10,800 --> 00:20:12,240 Like this stuff is everywhere. 642 00:20:12,240 --> 00:20:16,000 And I think it's easy to just kind of let it wash over us. 643 00:20:16,000 --> 00:20:17,280 Yeah, just accept it. 644 00:20:17,280 --> 00:20:20,360 But we should be thinking critically about, 645 00:20:20,360 --> 00:20:22,000 how is this affecting me? 646 00:20:22,000 --> 00:20:24,040 Right, how is it using my data? 647 00:20:24,040 --> 00:20:24,540 Yeah. 648 00:20:24,540 --> 00:20:26,040 What data is it collecting? 649 00:20:26,040 --> 00:20:28,160 And one of the articles that you sent me 650 00:20:28,160 --> 00:20:29,840 mentioned data literacy. 651 00:20:29,840 --> 00:20:30,320 Oh, yeah. 652 00:20:30,320 --> 00:20:31,440 And I thought that was really interesting. 653 00:20:31,440 --> 00:20:33,160 Can you talk a little bit more about that? 654 00:20:33,160 --> 00:20:34,680 What does that even mean? 655 00:20:34,680 --> 00:20:37,160 So data literacy is basically being 656 00:20:37,160 --> 00:20:40,160 able to understand and work with data. 657 00:20:40,160 --> 00:20:40,660 OK. 658 00:20:40,660 --> 00:20:44,080 So it's being able to read it, interpret it, analyze it, 659 00:20:44,080 --> 00:20:45,800 and communicate about data. 660 00:20:45,800 --> 00:20:48,600 So kind of like being able to speak the language of data. 661 00:20:48,600 --> 00:20:49,720 Exactly, yeah. 662 00:20:49,720 --> 00:20:52,760 So if you understand how data works, how it's collected, 663 00:20:52,760 --> 00:20:56,440 analyzed, and used or misused, then you 664 00:20:56,440 --> 00:20:58,320 can understand how AI is making decisions. 665 00:20:58,320 --> 00:21:02,040 OK, so this is about us being more informed citizens 666 00:21:02,040 --> 00:21:05,360 in a world that's increasingly shaped by data. 667 00:21:05,360 --> 00:21:06,520 Absolutely, yeah. 668 00:21:06,520 --> 00:21:09,000 OK, can you give me a specific example 669 00:21:09,000 --> 00:21:12,320 of how data literacy might be helpful? 670 00:21:12,320 --> 00:21:15,640 Yeah, let's say you're using a health app that's powered by AI. 671 00:21:15,640 --> 00:21:16,280 OK. 672 00:21:16,280 --> 00:21:20,000 Data literacy would help you interpret that information 673 00:21:20,000 --> 00:21:22,980 that that app is giving you and make better decisions 674 00:21:22,980 --> 00:21:23,720 about your health. 675 00:21:23,720 --> 00:21:24,360 That makes sense. 676 00:21:24,360 --> 00:21:25,920 Or if you're reading a news article 677 00:21:25,920 --> 00:21:28,520 that's citing a study that was done with AI. 678 00:21:28,520 --> 00:21:29,040 Right. 679 00:21:29,040 --> 00:21:30,680 Data literacy would help you understand, 680 00:21:30,680 --> 00:21:31,840 is that a valid study? 681 00:21:31,840 --> 00:21:34,800 OK, so it's about not just taking things at face value, 682 00:21:34,800 --> 00:21:37,640 but really understanding where did this come from? 683 00:21:37,640 --> 00:21:39,800 Yeah, what are they basing these conclusions on? 684 00:21:39,800 --> 00:21:43,760 Yeah, OK, so how do we become more data literate? 685 00:21:43,760 --> 00:21:47,280 Is this something that we all need to go back to school for? 686 00:21:47,280 --> 00:21:48,720 Not necessarily. 687 00:21:48,720 --> 00:21:49,220 OK. 688 00:21:49,220 --> 00:21:51,800 There's a lot of resources available online, courses, 689 00:21:51,800 --> 00:21:54,200 tutorials, books, articles. 690 00:21:54,200 --> 00:21:57,680 Start by being curious about the data you encounter every day. 691 00:21:57,680 --> 00:21:58,840 What do you mean by that? 692 00:21:58,840 --> 00:22:02,440 So next time you see a chart or a graph in a news article, 693 00:22:02,440 --> 00:22:03,800 don't just scroll past it. 694 00:22:03,800 --> 00:22:04,440 Yeah. 695 00:22:04,440 --> 00:22:05,600 Take a minute to look at it. 696 00:22:05,600 --> 00:22:06,880 What is it showing you? 697 00:22:06,880 --> 00:22:07,360 OK. 698 00:22:07,360 --> 00:22:08,880 What are the units of measurement? 699 00:22:08,880 --> 00:22:11,840 Are there any trends or patterns you can see? 700 00:22:11,840 --> 00:22:16,080 So it's like being more mindful and engaged 701 00:22:16,080 --> 00:22:18,160 with the data that we're seeing all the time. 702 00:22:18,160 --> 00:22:20,160 Yeah, because we see data everywhere. 703 00:22:20,160 --> 00:22:20,800 Yeah, that's true. 704 00:22:20,800 --> 00:22:23,360 And I think, like you said, we just let it wash over us. 705 00:22:23,360 --> 00:22:25,440 Yeah, it's like background noise almost. 706 00:22:25,440 --> 00:22:26,120 Right. 707 00:22:26,120 --> 00:22:26,840 But it's not. 708 00:22:26,840 --> 00:22:27,840 It's important. 709 00:22:27,840 --> 00:22:28,040 Yeah. 710 00:22:28,040 --> 00:22:29,080 It's very important. 711 00:22:29,080 --> 00:22:32,360 And I think, as we're talking about all this, it's like AI 712 00:22:32,360 --> 00:22:32,960 governance. 713 00:22:32,960 --> 00:22:36,040 It's not just this abstract concept. 714 00:22:36,040 --> 00:22:38,360 It's something that affects all of us 715 00:22:38,360 --> 00:22:40,160 in very real ways. 716 00:22:40,160 --> 00:22:41,600 It's not just for organization. 717 00:22:41,600 --> 00:22:42,920 It's for individuals, too. 718 00:22:42,920 --> 00:22:45,200 OK, well, this has been a really great conversation. 719 00:22:45,200 --> 00:22:47,280 I feel like we've covered so much. 720 00:22:47,280 --> 00:22:47,880 We have. 721 00:22:47,880 --> 00:22:50,720 Maybe we can just take a second to recap 722 00:22:50,720 --> 00:22:52,760 the key takeaways from this part. 723 00:22:52,760 --> 00:22:53,760 Yeah, good idea. 724 00:22:53,760 --> 00:22:56,360 So we started by talking about understanding 725 00:22:56,360 --> 00:23:00,920 where AI is being used within organizations. 726 00:23:00,920 --> 00:23:02,200 Yeah, and having a map. 727 00:23:02,200 --> 00:23:03,320 Yeah, having that map. 728 00:23:03,320 --> 00:23:05,640 Of the AI landscape, which sounds really cool, actually. 729 00:23:05,640 --> 00:23:06,480 It is cool. 730 00:23:06,480 --> 00:23:08,840 And then we talked about how this understanding is 731 00:23:08,840 --> 00:23:10,600 so important for like. 732 00:23:10,600 --> 00:23:11,760 For data privacy. 733 00:23:11,760 --> 00:23:13,680 Data privacy, security. 734 00:23:13,680 --> 00:23:16,160 Yeah, especially when you're working with third parties. 735 00:23:16,160 --> 00:23:17,800 Right, because you don't have control over that. 736 00:23:17,800 --> 00:23:19,000 Yeah, you don't know what they're doing. 737 00:23:19,000 --> 00:23:21,960 And then we talked about diverse perspectives and how. 738 00:23:21,960 --> 00:23:23,720 Yeah, we need everyone in the room. 739 00:23:23,720 --> 00:23:25,240 It's not just a one-person job. 740 00:23:25,240 --> 00:23:26,080 It's a team effort. 741 00:23:26,080 --> 00:23:27,040 It's a team effort. 742 00:23:27,040 --> 00:23:27,520 Yeah. 743 00:23:27,520 --> 00:23:30,520 And then we talked about this idea of data literacy, 744 00:23:30,520 --> 00:23:32,640 which I think is so important. 745 00:23:32,640 --> 00:23:33,440 Absolutely. 746 00:23:33,440 --> 00:23:36,440 All of us as individuals, to start understanding. 747 00:23:36,440 --> 00:23:38,040 Yeah, the more we understand data, 748 00:23:38,040 --> 00:23:39,560 the more we can understand AI. 749 00:23:39,560 --> 00:23:39,880 Yeah. 750 00:23:39,880 --> 00:23:41,480 And how it's impacting our lives. 751 00:23:41,480 --> 00:23:44,840 This is really about empowering individuals 752 00:23:44,840 --> 00:23:49,120 to be more informed and take control of their data 753 00:23:49,120 --> 00:23:50,760 and their interactions with AI. 754 00:23:50,760 --> 00:23:52,720 We need to be active participants. 755 00:23:52,720 --> 00:23:54,400 Yeah, not just passive consumers. 756 00:23:54,400 --> 00:23:55,720 Not passive, yes. 757 00:23:55,720 --> 00:23:58,360 This has been a really great conversation. 758 00:23:58,360 --> 00:23:58,800 It has. 759 00:23:58,800 --> 00:24:00,560 But we do have more ground to cover. 760 00:24:00,560 --> 00:24:01,400 Yes, we do. 761 00:24:01,400 --> 00:24:05,360 Including some specific steps that individuals and organizations 762 00:24:05,360 --> 00:24:05,960 can take. 763 00:24:05,960 --> 00:24:06,480 Yes. 764 00:24:06,480 --> 00:24:09,160 So stay tuned for part three. 765 00:24:09,160 --> 00:24:12,680 I think it's easy to get caught up in all the scary stuff 766 00:24:12,680 --> 00:24:13,480 about AI. 767 00:24:13,480 --> 00:24:13,960 Yeah. 768 00:24:13,960 --> 00:24:17,720 But it's important to remember, it can also be used for good. 769 00:24:17,720 --> 00:24:18,640 Oh, absolutely. 770 00:24:18,640 --> 00:24:21,600 AI has the potential to revolutionize 771 00:24:21,600 --> 00:24:23,440 so many aspects of our lives. 772 00:24:23,440 --> 00:24:25,760 Right, like health care, education. 773 00:24:25,760 --> 00:24:28,800 Sustainability, social justice, you name it. 774 00:24:28,800 --> 00:24:30,560 Yeah, it could really help us solve 775 00:24:30,560 --> 00:24:32,680 some of these big problems that we're facing. 776 00:24:32,680 --> 00:24:33,360 Huge problems. 777 00:24:33,360 --> 00:24:36,440 Like, can you imagine AI helping us develop 778 00:24:36,440 --> 00:24:37,760 new medical treatments? 779 00:24:37,760 --> 00:24:39,400 Yeah, like, that would be amazing. 780 00:24:39,400 --> 00:24:42,800 Or like, personalized learning for every student. 781 00:24:42,800 --> 00:24:44,200 I would be so cool. 782 00:24:44,200 --> 00:24:44,720 Yeah. 783 00:24:44,720 --> 00:24:48,840 So it's really about how do we harness that power? 784 00:24:48,840 --> 00:24:50,320 Yeah, harness that power for good 785 00:24:50,320 --> 00:24:53,320 and make sure that we're not letting it run away from us. 786 00:24:53,320 --> 00:24:53,840 Exactly. 787 00:24:53,840 --> 00:24:55,720 We need to create a framework that 788 00:24:55,720 --> 00:25:00,120 allows us to unlock that potential, but also 789 00:25:00,120 --> 00:25:01,800 safeguard against the harms. 790 00:25:01,800 --> 00:25:04,600 Right, so it's like innovation, yes, 791 00:25:04,600 --> 00:25:08,400 but not at the expense of our humanity. 792 00:25:08,400 --> 00:25:10,040 Exactly, it's all about balance. 793 00:25:10,040 --> 00:25:10,360 Yeah. 794 00:25:10,360 --> 00:25:11,480 And it's a collective effort. 795 00:25:11,480 --> 00:25:14,200 Like, it's not just on governments or tech companies. 796 00:25:14,200 --> 00:25:14,760 Right. 797 00:25:14,760 --> 00:25:15,720 It's on all of us. 798 00:25:15,720 --> 00:25:17,360 Yeah, we all have a role to play. 799 00:25:17,360 --> 00:25:17,760 Absolutely. 800 00:25:17,760 --> 00:25:20,160 In making sure that AI is developed ethically. 801 00:25:20,160 --> 00:25:22,800 Yes, we need to be asking questions. 802 00:25:22,800 --> 00:25:23,280 Yeah. 803 00:25:23,280 --> 00:25:24,600 Demanding transparency. 804 00:25:24,600 --> 00:25:26,440 Yeah, holding these companies accountable. 805 00:25:26,440 --> 00:25:26,920 Yes. 806 00:25:26,920 --> 00:25:29,920 So it's not about being afraid of AI. 807 00:25:29,920 --> 00:25:35,440 It's about shaping it to be what we want it to be. 808 00:25:35,440 --> 00:25:42,000 OK, so as we're wrapping up this deep dive into AI governance, 809 00:25:42,000 --> 00:25:45,000 I want to leave our listeners with a question. 810 00:25:45,000 --> 00:25:49,320 What role do you want to play in shaping the future of AI? 811 00:25:49,320 --> 00:25:50,280 It's an important question. 812 00:25:50,280 --> 00:25:52,000 Yeah, like what questions are you 813 00:25:52,000 --> 00:25:53,600 going to ask in your workplace? 814 00:25:53,600 --> 00:25:53,800 Right. 815 00:25:53,800 --> 00:25:55,800 How are you going to engage with this topic 816 00:25:55,800 --> 00:25:56,880 in your communities? 817 00:25:56,880 --> 00:26:00,360 Exactly, because this isn't just some far off thing. 818 00:26:00,360 --> 00:26:02,360 This is happening now. 819 00:26:02,360 --> 00:26:03,960 AI is already here. 820 00:26:03,960 --> 00:26:05,760 It's affecting our lives. 821 00:26:05,760 --> 00:26:06,840 In so many ways. 822 00:26:06,840 --> 00:26:08,800 And we need to make sure that we're 823 00:26:08,800 --> 00:26:10,200 being thoughtful about it. 824 00:26:10,200 --> 00:26:10,720 Absolutely. 825 00:26:10,720 --> 00:26:11,680 We have to be intentional. 826 00:26:11,680 --> 00:26:14,880 Yeah, and that we're using it for good. 827 00:26:14,880 --> 00:26:17,360 And remember, we talked about data literacy. 828 00:26:17,360 --> 00:26:19,160 Yes, data literacy is key. 829 00:26:19,160 --> 00:26:21,200 Like the more you understand data, 830 00:26:21,200 --> 00:26:22,840 the more you're going to understand AI. 831 00:26:22,840 --> 00:26:23,360 Exactly. 832 00:26:23,360 --> 00:26:24,280 They're interconnected. 833 00:26:24,280 --> 00:26:25,880 So keep learning. 834 00:26:25,880 --> 00:26:26,240 Yes. 835 00:26:26,240 --> 00:26:27,840 Keep asking those questions. 836 00:26:27,840 --> 00:26:29,640 Keep pushing for a better future. 837 00:26:29,640 --> 00:26:32,720 Yeah, a future where AI is a force for good. 838 00:26:32,720 --> 00:26:34,040 Absolutely. 839 00:26:34,040 --> 00:26:36,160 Thanks for joining us for this deep dive. 840 00:26:36,160 --> 00:26:37,600 Yes, thanks for listening. 841 00:26:37,600 --> 00:26:57,760 And we'll see you next time.