1 00:00:00,000 --> 00:00:04,800 When we are entering this field of a very competitive industry, 2 00:00:04,800 --> 00:00:07,640 and we know it's competitive, and we know what's out there, 3 00:00:07,640 --> 00:00:10,560 when we step into this field, we are going to show up 4 00:00:10,560 --> 00:00:13,520 and we're going to decide from the beginning 5 00:00:13,520 --> 00:00:15,760 to make responsible decisions. 6 00:00:15,760 --> 00:00:18,960 And what I have learned is that it's much easier 7 00:00:18,960 --> 00:00:21,880 if you make those decisions from the get-go, 8 00:00:21,880 --> 00:00:24,880 because what takes more energy and more time 9 00:00:24,880 --> 00:00:26,600 is going back and fixing something 10 00:00:26,600 --> 00:00:30,040 that's fundamentally broken. 11 00:00:30,040 --> 00:00:35,040 HTTTA 12 00:00:35,040 --> 00:00:38,040 HTTTA 13 00:00:38,040 --> 00:00:39,040 HTTTA 14 00:00:39,040 --> 00:00:40,040 HTTTA 15 00:00:40,040 --> 00:00:41,040 HTTTA 16 00:00:41,040 --> 00:00:46,080 It's how to talk to a I with your host, 17 00:00:46,080 --> 00:00:48,720 go-to-go and with the synth mind. 18 00:00:48,720 --> 00:00:50,640 Ladies and gentlemen, boys and girls, children of all ages, 19 00:00:50,640 --> 00:00:52,240 dogs, cats, robots and everybody in between, 20 00:00:52,240 --> 00:00:55,440 especially you, trustworthy users of artificial intelligence, 21 00:00:55,440 --> 00:00:58,480 this is HTTTA, how to talk to AI. 22 00:00:58,480 --> 00:01:00,600 I am your host, Wes the Synth Mind, Synth Mind Wes, 23 00:01:00,600 --> 00:01:03,640 and as always, in a galaxy gleaming with gigabytes 24 00:01:03,640 --> 00:01:05,600 where gadgets scavenize our future, 25 00:01:05,600 --> 00:01:07,600 one gallant glowing voice gears up to grapple 26 00:01:07,600 --> 00:01:10,280 with artificial intelligence, grandest queries, 27 00:01:10,280 --> 00:01:13,520 the grander gateway to AI's greatest grace by genius, 28 00:01:13,520 --> 00:01:15,760 guided by gusto, she delves gallantly 29 00:01:15,760 --> 00:01:17,400 where machines glean and grow. 30 00:01:17,400 --> 00:01:18,760 Might you ask, ladies and gentlemen, 31 00:01:18,760 --> 00:01:22,280 gravitates towards the genuine, the gregarious go-to-go. 32 00:01:22,280 --> 00:01:23,680 Gee, how are you this week? 33 00:01:23,680 --> 00:01:27,360 Hi, I'm fantastic, but especially on this occasion, 34 00:01:27,360 --> 00:01:31,320 I don't feel deserving of so many adjectives 35 00:01:31,320 --> 00:01:34,720 because we have this incredible guest joining us. 36 00:01:34,720 --> 00:01:38,640 Joining us today is M. Alejandra Parra-Orelani, 37 00:01:38,640 --> 00:01:41,200 or MAPO for short, way easier to say. 38 00:01:41,200 --> 00:01:43,880 MAPO enjoys tackling challenges at the intersection 39 00:01:43,880 --> 00:01:46,640 of technology, organization, leadership and society. 40 00:01:46,640 --> 00:01:48,760 So trustworthy tech is a sweet spot, 41 00:01:48,760 --> 00:01:50,880 and that's what today's episode is gonna be about, 42 00:01:50,880 --> 00:01:53,400 trustworthy and ethical applications of AI. 43 00:01:53,400 --> 00:01:56,440 She's currently the VP of Ethical Innovation and Policy 44 00:01:56,440 --> 00:01:58,880 at the Global Portfolio Division at Takeda, 45 00:01:58,880 --> 00:02:00,880 where she builds solutions that enable teams 46 00:02:00,880 --> 00:02:02,400 to develop trustworthy tech. 47 00:02:02,400 --> 00:02:04,520 Prior to that, she led a think tank at Meta 48 00:02:04,520 --> 00:02:07,320 focused on privacy enhancing technologies, 49 00:02:07,320 --> 00:02:09,440 digital identity and XR. 50 00:02:09,440 --> 00:02:12,640 She also developed responsible AI products and services 51 00:02:12,640 --> 00:02:15,040 at Quantum Black, which is AI by McKinsey, 52 00:02:15,040 --> 00:02:18,640 and helped launch the firm's inaugural open source project. 53 00:02:18,640 --> 00:02:21,400 MAPO has previous experience as an engineer, 54 00:02:21,400 --> 00:02:24,480 naval officer and attorney, and holds degrees 55 00:02:24,480 --> 00:02:26,600 from these little tiny schools you may have heard of 56 00:02:26,600 --> 00:02:30,160 called MIT, Harvard and the United States Naval Academy, 57 00:02:30,160 --> 00:02:33,160 also known as Canoe U, where I happen to go. 58 00:02:33,160 --> 00:02:35,880 She writes and speaks about leadership in tech 59 00:02:35,880 --> 00:02:36,960 and advises startups. 60 00:02:36,960 --> 00:02:39,240 She also volunteers her time mentoring women 61 00:02:39,240 --> 00:02:41,960 and transitioning military veterans in her spare time, 62 00:02:41,960 --> 00:02:44,440 MAPO, left to paint, invent silly songs, 63 00:02:44,440 --> 00:02:47,520 debate the meaning of life and play with dogs. 64 00:02:47,520 --> 00:02:49,160 What else is there to say MAPO? 65 00:02:49,160 --> 00:02:51,400 Thank you so much for being on the podcast. 66 00:02:51,400 --> 00:02:53,720 Thank you so much for having me on the podcast. 67 00:02:53,720 --> 00:02:57,000 I'm really excited to join, especially with so much energy 68 00:02:57,000 --> 00:02:59,440 and excitement from that amazing intro. 69 00:02:59,440 --> 00:03:00,280 Thank you so much. 70 00:03:00,280 --> 00:03:01,280 We bring it, we bring it. 71 00:03:01,280 --> 00:03:03,280 I'd like to start, if we may, 72 00:03:03,280 --> 00:03:04,760 we're gonna have, I think, plenty of time 73 00:03:04,760 --> 00:03:09,760 to talk about the applications and the ethical use of AI, 74 00:03:10,080 --> 00:03:12,360 both by just the casual user, 75 00:03:12,360 --> 00:03:14,160 as well as corporate employment. 76 00:03:14,160 --> 00:03:16,680 How did you find yourself into the world 77 00:03:16,680 --> 00:03:18,080 of artificial intelligence, 78 00:03:18,080 --> 00:03:21,800 given the diverse background that you have? 79 00:03:21,800 --> 00:03:25,280 I wish I could say that there was some grand design, 80 00:03:25,280 --> 00:03:26,120 but there wasn't. 81 00:03:26,120 --> 00:03:29,800 I really lead my life through curiosity. 82 00:03:29,800 --> 00:03:30,840 What am I curious about? 83 00:03:30,840 --> 00:03:32,080 What am I interested in? 84 00:03:32,080 --> 00:03:35,000 And very early on, back in the high school days 85 00:03:35,000 --> 00:03:37,480 and college days, I was really into engineering. 86 00:03:37,480 --> 00:03:38,920 I tried it out. 87 00:03:38,920 --> 00:03:42,720 I learned that I wasn't as genius at it 88 00:03:42,720 --> 00:03:45,040 as some of my colleagues at one of the schools 89 00:03:45,040 --> 00:03:46,920 that you mentioned earlier, MIT. 90 00:03:46,920 --> 00:03:48,880 So I thought to myself, hmm, 91 00:03:48,880 --> 00:03:51,760 I wanna do something that I could be really, really awesome at, 92 00:03:51,760 --> 00:03:54,880 just like these people were awesome at engineering. 93 00:03:54,880 --> 00:03:58,280 And so I started to explore and just see what was out there. 94 00:03:58,280 --> 00:03:59,320 And while I was in the Navy, 95 00:03:59,320 --> 00:04:03,520 I had the chance to work in this little weird role 96 00:04:03,520 --> 00:04:05,080 called legal officer. 97 00:04:05,080 --> 00:04:06,520 So it was kind of like, I don't know, 98 00:04:06,520 --> 00:04:08,360 paralegal role within a ship, 99 00:04:08,360 --> 00:04:10,400 because there aren't enough lawyers to go around 100 00:04:10,400 --> 00:04:11,920 to every ship in the Navy. 101 00:04:11,920 --> 00:04:14,200 And so I got really interested in the law, 102 00:04:14,200 --> 00:04:15,400 went to law school. 103 00:04:15,400 --> 00:04:17,360 And then while I was practicing law, 104 00:04:17,360 --> 00:04:19,600 I tried out different kinds of practices. 105 00:04:19,600 --> 00:04:22,440 Some were very interesting on paper in theory, 106 00:04:22,440 --> 00:04:24,160 but in my lived experience, 107 00:04:24,160 --> 00:04:27,360 they just weren't really hitting the right notes for me. 108 00:04:27,360 --> 00:04:30,120 And so I was able to reconnect back with technology 109 00:04:30,120 --> 00:04:32,720 and try out legal practice in tech law 110 00:04:32,720 --> 00:04:34,440 and policy and compliance. 111 00:04:34,440 --> 00:04:37,160 And this was around the time that AI was really starting 112 00:04:37,160 --> 00:04:39,280 to finally pick up more traction. 113 00:04:39,280 --> 00:04:43,080 And topics around ethical AI were really starting to get 114 00:04:43,080 --> 00:04:44,680 a little bit more mainstream. 115 00:04:44,680 --> 00:04:47,400 And so I think just the confluence of my own interests, 116 00:04:47,400 --> 00:04:48,560 my own experiences, 117 00:04:48,560 --> 00:04:50,880 and what was actually happening out in industry 118 00:04:50,880 --> 00:04:52,480 led me to this world of AI. 119 00:04:52,480 --> 00:04:55,000 I think you have a very topsy turvy little path. 120 00:04:55,000 --> 00:04:57,080 You have a unique perspective there, 121 00:04:57,080 --> 00:04:59,160 both from the engineering side, 122 00:04:59,160 --> 00:05:01,880 the zeros and ones and math and science, 123 00:05:01,880 --> 00:05:04,640 and then also the more philosophical side 124 00:05:04,640 --> 00:05:07,000 with the legal approach. 125 00:05:07,000 --> 00:05:07,840 Yeah. 126 00:05:07,840 --> 00:05:10,920 I just have to say that this is so fascinating 127 00:05:10,920 --> 00:05:14,120 to hear all the rules that you have undertaken 128 00:05:14,120 --> 00:05:18,120 and impressive results on your energy shines through. 129 00:05:18,120 --> 00:05:19,720 You know, as being a female, 130 00:05:19,720 --> 00:05:24,040 I weirdly urge to ask like, how was it being in the Navy, 131 00:05:24,040 --> 00:05:25,840 going through engineering, being in AI, 132 00:05:25,840 --> 00:05:28,440 because I have my small experiences 133 00:05:28,440 --> 00:05:31,160 with male dominated industry. 134 00:05:31,160 --> 00:05:32,880 I am an architect by training 135 00:05:32,880 --> 00:05:35,000 and I worked in big data analytics company, 136 00:05:35,000 --> 00:05:36,200 so a bunch of developers, 137 00:05:36,200 --> 00:05:39,040 but maybe if we could touch on that, 138 00:05:39,040 --> 00:05:42,760 just for my own curiosity and interest, 139 00:05:42,760 --> 00:05:44,160 just what was your experience 140 00:05:44,160 --> 00:05:47,920 and how you're experiencing AI space yourself right now? 141 00:05:47,920 --> 00:05:49,600 Yeah, it's a great question. 142 00:05:49,600 --> 00:05:53,320 I don't reflect on it probably as much as I should. 143 00:05:53,320 --> 00:05:56,080 I will say when I look back at my journey, 144 00:05:56,080 --> 00:05:59,280 my professional journey, it hasn't always been easy. 145 00:05:59,280 --> 00:06:01,160 And sometimes it hasn't been easy 146 00:06:01,160 --> 00:06:03,960 because of just being in, as you mentioned, 147 00:06:03,960 --> 00:06:07,480 these kind of male dominated industries or settings. 148 00:06:07,480 --> 00:06:10,800 The Naval Academy for me in particular was tough 149 00:06:10,800 --> 00:06:13,200 because at the time when I was there, 150 00:06:13,200 --> 00:06:15,440 I think there were about 12% women. 151 00:06:15,440 --> 00:06:18,760 So it's very, very few numbers of people 152 00:06:18,760 --> 00:06:21,560 that you can have certain shared experiences with 153 00:06:21,560 --> 00:06:23,000 and connect with. 154 00:06:23,000 --> 00:06:25,880 But I think that looking back, 155 00:06:25,880 --> 00:06:29,560 what all of these experiences have taught me is, 156 00:06:29,560 --> 00:06:31,800 number one, find your allies. 157 00:06:31,800 --> 00:06:33,960 You have to find your allies and they're out there, 158 00:06:33,960 --> 00:06:34,960 not just women. 159 00:06:34,960 --> 00:06:38,160 I've had a lot of really wonderful sponsors, 160 00:06:38,160 --> 00:06:40,400 allies, supporters, friends, 161 00:06:40,400 --> 00:06:43,760 just plain old friends from all different backgrounds. 162 00:06:43,760 --> 00:06:46,920 And I think that looking for those people 163 00:06:46,920 --> 00:06:50,800 and really connecting with them and focusing on them 164 00:06:50,800 --> 00:06:52,840 instead of the bad stuff, 165 00:06:52,840 --> 00:06:54,840 I think is really, really helpful 166 00:06:54,840 --> 00:06:57,440 because I think this is true for everyone, 167 00:06:57,440 --> 00:06:59,240 not just women, but everyone. 168 00:06:59,240 --> 00:07:02,760 None of us find success or build careers by ourselves. 169 00:07:02,760 --> 00:07:05,800 We all do it with the help of other people around us. 170 00:07:05,800 --> 00:07:09,120 And I think that just focusing on finding those people 171 00:07:09,120 --> 00:07:11,600 and connecting with them has played a huge role. 172 00:07:11,600 --> 00:07:14,520 But I will say, it's not always easy. 173 00:07:14,520 --> 00:07:16,560 There have been so many different masks 174 00:07:16,560 --> 00:07:19,960 that I've tried to wear to see if I can communicate 175 00:07:19,960 --> 00:07:22,680 in a different way that is more impactful, 176 00:07:22,680 --> 00:07:24,880 that helps people to see me differently 177 00:07:24,880 --> 00:07:26,960 because they might have preconceived notions 178 00:07:26,960 --> 00:07:30,080 based on how I look or what my gender is 179 00:07:30,080 --> 00:07:31,800 or what my voice sounds like. 180 00:07:31,800 --> 00:07:35,640 I'll never forget, there was a guy when I was in the military 181 00:07:35,640 --> 00:07:39,040 who, when I took the lead on some organization, 182 00:07:39,040 --> 00:07:41,320 he said to me, you're never gonna succeed. 183 00:07:41,320 --> 00:07:42,720 You're too giggly. 184 00:07:42,720 --> 00:07:44,080 You laugh too much. 185 00:07:44,080 --> 00:07:46,360 People are never gonna take you seriously. 186 00:07:46,360 --> 00:07:49,120 And wrong, thanks. 187 00:07:49,120 --> 00:07:50,520 Thanks for letting me know. 188 00:07:50,520 --> 00:07:53,360 But I became self-conscious about the giggling 189 00:07:53,360 --> 00:07:54,920 and the laughter and all of that. 190 00:07:54,920 --> 00:07:56,760 And so I tried on different masks 191 00:07:56,760 --> 00:07:59,480 to see if I could show up differently. 192 00:07:59,480 --> 00:08:01,640 And I did learn some communication skills 193 00:08:01,640 --> 00:08:04,120 with that kind of, we'll call it that kind of feedback. 194 00:08:04,120 --> 00:08:07,240 But in the end, I had to find myself. 195 00:08:07,240 --> 00:08:09,040 I'm a giggly, happy person, 196 00:08:09,040 --> 00:08:11,960 and I had to find ways to lead through that. 197 00:08:11,960 --> 00:08:15,880 And at the very end of this leadership experience, 198 00:08:15,880 --> 00:08:18,320 that same guy came back up to me and he said, 199 00:08:18,320 --> 00:08:19,680 I'm sorry, I was wrong. 200 00:08:19,680 --> 00:08:20,960 I actually, I was wrong. 201 00:08:20,960 --> 00:08:23,760 And you've changed my mind on what leadership looks like. 202 00:08:23,760 --> 00:08:24,600 Now I'm paraphrasing, 203 00:08:24,600 --> 00:08:27,040 it wasn't quite that powerful of a moment, 204 00:08:27,040 --> 00:08:31,600 but it was really, really amazing for me to experience 205 00:08:31,600 --> 00:08:34,640 someone being, number one, so bold enough to come 206 00:08:34,640 --> 00:08:37,360 and tell me you're gonna fail because of how you sound 207 00:08:37,360 --> 00:08:39,000 or because of how you come across. 208 00:08:39,000 --> 00:08:41,720 And then number two, that same person being humble enough 209 00:08:41,720 --> 00:08:43,080 to come back and say, you know what, 210 00:08:43,080 --> 00:08:45,000 I was wrong and I learned something. 211 00:08:45,000 --> 00:08:47,920 And I've just taken that with me the rest of my career 212 00:08:47,920 --> 00:08:50,240 and tried to remember that people come 213 00:08:50,240 --> 00:08:51,560 with preconceived notions. 214 00:08:51,560 --> 00:08:54,840 Some of them will be bold enough to tell you to your face, 215 00:08:54,840 --> 00:08:56,240 others won't actually, 216 00:08:56,240 --> 00:08:58,280 and they'll just hold it in the backs of their minds 217 00:08:58,280 --> 00:09:00,000 and prejudge you secretly. 218 00:09:00,000 --> 00:09:02,040 And then at the same time, 219 00:09:02,040 --> 00:09:04,520 a lot of these people are on their own learning journeys 220 00:09:04,520 --> 00:09:06,880 and they can learn new and different ways 221 00:09:06,880 --> 00:09:08,320 of working with people. 222 00:09:08,320 --> 00:09:10,880 I really love what you bring up and I want to, 223 00:09:10,880 --> 00:09:13,640 I will transition to a very responsible AI with this, 224 00:09:13,640 --> 00:09:16,320 but I also want to say that for me personally, 225 00:09:16,320 --> 00:09:17,760 jumping into YouTube space, 226 00:09:17,760 --> 00:09:20,120 there is internet comment section. 227 00:09:20,120 --> 00:09:22,520 I've seen all the type of comments on our channel, 228 00:09:22,520 --> 00:09:24,400 so I was very much ready. 229 00:09:24,400 --> 00:09:26,240 And I am actually happy that I joined 230 00:09:26,240 --> 00:09:27,800 when I'm a bit more mature. 231 00:09:27,800 --> 00:09:29,520 And I expected everything. 232 00:09:29,520 --> 00:09:31,800 I was like, okay, whatever comes. 233 00:09:31,800 --> 00:09:33,080 And I have to share, 234 00:09:33,080 --> 00:09:35,280 I don't know, Wes, if I shared that with you, 235 00:09:35,280 --> 00:09:39,360 but the support and also respectful communication, 236 00:09:39,360 --> 00:09:41,760 my audience is 90% male. 237 00:09:41,760 --> 00:09:44,440 On top of that, the engagement, the comments, 238 00:09:44,440 --> 00:09:47,000 I never experienced something 239 00:09:47,000 --> 00:09:48,880 which would be really disrespectful 240 00:09:48,880 --> 00:09:50,400 besides some funny jokes. 241 00:09:50,400 --> 00:09:51,760 And also not just that, 242 00:09:51,760 --> 00:09:54,440 but also like our community at SimphMinds, 243 00:09:54,440 --> 00:09:56,960 everyone is so supportive, so respectful, 244 00:09:56,960 --> 00:09:59,120 and everyone willing to help. 245 00:09:59,120 --> 00:10:02,000 So yeah, I have to say my experience 246 00:10:02,000 --> 00:10:04,840 throughout architecture with some Russian constructors 247 00:10:04,840 --> 00:10:06,200 were not the easiest, 248 00:10:06,200 --> 00:10:08,600 but I have to say maybe we as a society 249 00:10:08,600 --> 00:10:12,600 are moving towards that more inclusivity. 250 00:10:12,600 --> 00:10:15,000 And this is my transition to responsible AI 251 00:10:15,000 --> 00:10:19,200 about representation and that we have diversity. 252 00:10:20,320 --> 00:10:22,280 When we talk about ethical AI, 253 00:10:22,280 --> 00:10:25,000 what we are teaching that female male experiences 254 00:10:25,000 --> 00:10:28,040 are also integrated in a learning data set, let's say that. 255 00:10:28,040 --> 00:10:30,720 So any kind of your experience 256 00:10:30,720 --> 00:10:34,200 from when you talk about responsible AI and representation, 257 00:10:34,200 --> 00:10:37,040 any thoughts that you have preconceived? 258 00:10:37,040 --> 00:10:40,120 Yeah, I think one of the things that really interests me 259 00:10:40,120 --> 00:10:44,560 about the general topic of ethical AI or responsible AI, 260 00:10:44,560 --> 00:10:46,520 whatever label you wanna put on it, 261 00:10:46,520 --> 00:10:49,880 is that there are so many questions 262 00:10:49,880 --> 00:10:52,560 that sit at a very technical level 263 00:10:52,560 --> 00:10:56,320 all the way up through big, huge systemic questions. 264 00:10:56,320 --> 00:10:57,800 And what's really interesting 265 00:10:57,800 --> 00:11:01,800 is that the technical solutions filter up 266 00:11:01,800 --> 00:11:05,080 and might aggregate and accumulate 267 00:11:05,080 --> 00:11:07,360 to become these big systemic impacts. 268 00:11:07,360 --> 00:11:09,040 And then the systemic goals 269 00:11:09,040 --> 00:11:10,840 that we might want to be achieving, 270 00:11:10,840 --> 00:11:12,080 if you break them down, 271 00:11:12,080 --> 00:11:14,720 they sometimes filter down through to very technical 272 00:11:14,720 --> 00:11:18,280 or design or very kind of specific product level solutions. 273 00:11:18,280 --> 00:11:21,640 And so I think that this question of representation 274 00:11:21,640 --> 00:11:25,000 becomes both a complex, it's a complicated question, right? 275 00:11:25,000 --> 00:11:27,080 Because there's so many different parts to it, 276 00:11:27,080 --> 00:11:28,720 which is one of the things 277 00:11:28,720 --> 00:11:30,280 that I think is interesting about it. 278 00:11:30,280 --> 00:11:34,000 But I find it such a fascinating topic to think about 279 00:11:34,000 --> 00:11:39,000 because it requires us to focus on the goals 280 00:11:39,560 --> 00:11:42,160 of whatever it is that we're trying to build. 281 00:11:42,160 --> 00:11:43,880 What are the goals that we're trying to achieve 282 00:11:43,880 --> 00:11:45,800 when we build an AI system? 283 00:11:45,800 --> 00:11:47,920 And then what are the implications 284 00:11:47,920 --> 00:11:49,760 of how we are building it? 285 00:11:49,760 --> 00:11:51,680 So it's not just like the end product 286 00:11:51,680 --> 00:11:54,800 and what its impact might be, but how are we building it? 287 00:11:54,800 --> 00:11:57,960 And will that, again, like those little decisions 288 00:11:57,960 --> 00:11:59,440 that we make along the way, 289 00:11:59,440 --> 00:12:02,240 will they accumulate and add up to something 290 00:12:02,240 --> 00:12:05,200 that is representative, that is inclusive, 291 00:12:05,200 --> 00:12:08,680 that is going to ultimately on a grand scale, 292 00:12:08,680 --> 00:12:10,480 achieve the types of goals we want 293 00:12:10,480 --> 00:12:12,920 for how we might be interacting with that product 294 00:12:12,920 --> 00:12:16,640 or what the second and third order consequences 295 00:12:16,640 --> 00:12:18,880 might be across society. 296 00:12:18,880 --> 00:12:21,840 So this question of representation, I think, 297 00:12:21,840 --> 00:12:25,520 is a really big and important thread in the conversation, 298 00:12:25,520 --> 00:12:28,360 that broader conversation of ethical AI. 299 00:12:28,360 --> 00:12:29,720 And it's really hard. 300 00:12:29,720 --> 00:12:30,800 It's a really hard one. 301 00:12:30,800 --> 00:12:34,320 It's hard for us outside of AI, I think, as a society too. 302 00:12:34,320 --> 00:12:36,680 Just grappling with these tools 303 00:12:36,680 --> 00:12:39,440 because they're so compelling what they're able to do 304 00:12:39,440 --> 00:12:40,920 and what they're able to produce. 305 00:12:40,920 --> 00:12:42,920 But if you get a peek behind the curtain, 306 00:12:42,920 --> 00:12:44,280 there's some ugly stuff in there 307 00:12:44,280 --> 00:12:46,200 based on that training data. 308 00:12:46,200 --> 00:12:49,760 And also too, just if you want something that's this, 309 00:12:49,760 --> 00:12:52,360 if the goal is artificial general intelligence, 310 00:12:52,360 --> 00:12:56,920 we would hope it's representative of our society's bests. 311 00:12:56,920 --> 00:13:00,640 And there's just the people building it right now. 312 00:13:00,640 --> 00:13:03,800 It's still not representative enough with females. 313 00:13:03,800 --> 00:13:07,160 Only about 22% in big AI positions 314 00:13:07,160 --> 00:13:08,760 that are putting together these tools, 315 00:13:08,760 --> 00:13:10,960 that are making those technical decisions 316 00:13:10,960 --> 00:13:13,960 that ultimately have to be circular. 317 00:13:13,960 --> 00:13:15,680 Yeah, Wes, 100%. 318 00:13:15,680 --> 00:13:18,280 And you have this beautiful visualization 319 00:13:18,280 --> 00:13:19,960 on one of your blog posts, 320 00:13:19,960 --> 00:13:23,400 which is responsible as much about the small stuff 321 00:13:23,400 --> 00:13:25,200 as it is about the big stuff. 322 00:13:25,200 --> 00:13:27,520 And this visualization, it's like the circle 323 00:13:27,520 --> 00:13:29,920 where it says the first is, should we build it? 324 00:13:29,920 --> 00:13:31,320 Can we build it? 325 00:13:31,320 --> 00:13:33,920 And then finally, are we building it right? 326 00:13:33,920 --> 00:13:35,640 Just to give you a bit of a context 327 00:13:35,640 --> 00:13:37,960 why this stood out to me recently, 328 00:13:37,960 --> 00:13:42,240 I went and tested 432 AI writing tools. 329 00:13:42,240 --> 00:13:44,040 Wow, that is awesome. 330 00:13:44,040 --> 00:13:45,560 Check out the list on her page. 331 00:13:45,560 --> 00:13:50,560 It's nothing short of sheer madness and insanity, 332 00:13:50,960 --> 00:13:53,560 but go to bore the strain for all you 333 00:13:53,560 --> 00:13:56,520 so you wouldn't have to just wade down in the muck. 334 00:13:56,520 --> 00:13:58,000 That is some of these things. 335 00:13:58,000 --> 00:14:01,600 Yeah, I was saying to Wes, I think after around 100, 336 00:14:01,600 --> 00:14:03,160 I was like, okay, I want to give up, 337 00:14:03,160 --> 00:14:05,160 but then I'm already at 100. 338 00:14:05,160 --> 00:14:08,360 I have full data set that I scraped, which is 432. 339 00:14:08,360 --> 00:14:09,480 Okay, keep going. 340 00:14:09,480 --> 00:14:13,080 At 250, I was like thinking I'm losing my mind. 341 00:14:13,080 --> 00:14:15,360 But the reason I'm bringing this up 342 00:14:15,360 --> 00:14:17,720 because this gave me such an insight 343 00:14:17,720 --> 00:14:18,600 because we always talk, 344 00:14:18,600 --> 00:14:20,960 where is startups popping up left and right. 345 00:14:20,960 --> 00:14:23,760 I think this list I did probably already outdated, 346 00:14:23,760 --> 00:14:25,560 where is probably another 400. 347 00:14:25,560 --> 00:14:29,040 But what I saw is I have this comprehensive overview 348 00:14:29,040 --> 00:14:33,800 that the amount of startups which spin out things 349 00:14:33,800 --> 00:14:36,400 probably on a hackathon over the weekend, 350 00:14:36,400 --> 00:14:40,280 no considerations, is huge, is really huge. 351 00:14:40,280 --> 00:14:43,160 And I understand where, like I understand competition, 352 00:14:43,160 --> 00:14:45,440 speed, like your investors are pushing you. 353 00:14:45,440 --> 00:14:47,480 So I'm curious to hear your thoughts 354 00:14:47,480 --> 00:14:49,680 when there is these forces in the world, 355 00:14:49,680 --> 00:14:52,080 especially when you're trying to build something. 356 00:14:52,080 --> 00:14:54,600 And in case you are trying to think responsibly, 357 00:14:54,600 --> 00:14:57,080 but then you have the whole competition, the market, 358 00:14:57,080 --> 00:14:58,760 everyone breathing into your neck. 359 00:14:58,760 --> 00:15:01,720 If you're not going to release something right now, 360 00:15:01,720 --> 00:15:03,560 there is five other startups next week, 361 00:15:03,560 --> 00:15:05,080 which beat you to the market, 362 00:15:05,080 --> 00:15:09,680 and they are now claiming the branding of text to whatever. 363 00:15:09,680 --> 00:15:11,960 So my question probably is like, 364 00:15:11,960 --> 00:15:15,720 how startups could address this outside forces, 365 00:15:15,720 --> 00:15:19,120 even though if they have a really nice intentions 366 00:15:19,120 --> 00:15:20,200 from within. 367 00:15:20,200 --> 00:15:22,840 Yeah, I have to be very honest. 368 00:15:22,840 --> 00:15:24,920 I don't think there's a magic answer to this. 369 00:15:24,920 --> 00:15:25,760 I really don't. 370 00:15:25,760 --> 00:15:28,480 I do think that there are founders out there 371 00:15:28,480 --> 00:15:31,520 who are really committed to being responsible 372 00:15:31,520 --> 00:15:33,640 in their innovations and have decided, 373 00:15:33,640 --> 00:15:36,080 have made the decision to say, okay, 374 00:15:36,080 --> 00:15:40,880 when we are entering this field of a very competitive industry 375 00:15:40,880 --> 00:15:43,760 and we know it's competitive and we know what's out there, 376 00:15:43,760 --> 00:15:46,640 when we step into this field, we are going to show up 377 00:15:46,640 --> 00:15:49,600 and we're going to decide from the beginning 378 00:15:49,600 --> 00:15:51,760 to make responsible decisions. 379 00:15:51,760 --> 00:15:54,880 And usually those companies are led by 380 00:15:54,880 --> 00:15:58,640 really incredible people who maybe choose some principles 381 00:15:58,640 --> 00:16:00,040 that they want to align to, 382 00:16:00,040 --> 00:16:02,640 some values that they want their startup 383 00:16:02,640 --> 00:16:04,760 or their new organization to have. 384 00:16:04,760 --> 00:16:07,280 And they design everything to those values 385 00:16:07,280 --> 00:16:10,120 and those principles, just like you would design a product 386 00:16:10,120 --> 00:16:13,920 to design principles or technical specifications. 387 00:16:13,920 --> 00:16:16,600 They're part of the business model. 388 00:16:16,600 --> 00:16:19,600 They're part of the product in and of itself. 389 00:16:19,600 --> 00:16:21,160 And it's doable. 390 00:16:21,160 --> 00:16:22,640 It's not easy. 391 00:16:22,640 --> 00:16:24,960 It's certainly not easy, but it's doable. 392 00:16:24,960 --> 00:16:28,120 And what I have learned is that it's much easier 393 00:16:28,120 --> 00:16:31,080 if you make those decisions from the get-go, 394 00:16:31,080 --> 00:16:34,040 because what takes more energy and more time 395 00:16:34,040 --> 00:16:35,760 is going back and fixing something 396 00:16:35,760 --> 00:16:37,440 that's fundamentally broken. 397 00:16:37,440 --> 00:16:39,480 You look at a lot of the big tech companies, 398 00:16:39,480 --> 00:16:41,440 sure, they're very successful. 399 00:16:41,440 --> 00:16:44,200 When you look at how many people they have employed 400 00:16:44,200 --> 00:16:48,200 in their PR, communications, policy, 401 00:16:48,200 --> 00:16:50,280 legal, compliance departments, 402 00:16:50,280 --> 00:16:55,280 to go back and fix stuff or reclaim some sort of level 403 00:16:56,320 --> 00:17:00,720 of just their perception and their reputation, 404 00:17:00,720 --> 00:17:02,760 they spend a lot of money on that. 405 00:17:02,760 --> 00:17:04,080 They have the money to spend on it, 406 00:17:04,080 --> 00:17:06,280 but they spend a lot of money and time on that. 407 00:17:06,280 --> 00:17:09,440 Whereas if they had been building with these principles 408 00:17:09,440 --> 00:17:11,480 and values from the get-go, 409 00:17:11,480 --> 00:17:13,760 it's not that they wouldn't have to spend money 410 00:17:13,760 --> 00:17:15,920 on compliance and legal and all of that, 411 00:17:15,920 --> 00:17:18,080 but the focus would be different. 412 00:17:18,080 --> 00:17:19,680 If they wouldn't be sweeping up a mess, 413 00:17:19,680 --> 00:17:21,280 they would be building upon 414 00:17:21,280 --> 00:17:23,240 a really strong foundation of trust. 415 00:17:23,240 --> 00:17:24,760 And so I don't know if you've noticed, 416 00:17:24,760 --> 00:17:27,080 but a lot of the newer startups that are popping up 417 00:17:27,080 --> 00:17:30,040 around AI have safety at their core 418 00:17:30,040 --> 00:17:32,320 or trustworthy tech at their core, 419 00:17:32,320 --> 00:17:33,840 because I think people have realized 420 00:17:33,840 --> 00:17:37,560 that it's not just like a good, nice thing to do. 421 00:17:37,560 --> 00:17:41,720 It's actually a very smart business decision also. 422 00:17:41,720 --> 00:17:44,000 It's just better overall. 423 00:17:44,000 --> 00:17:47,040 And so on the startup side, I don't think it's easy, 424 00:17:47,040 --> 00:17:48,440 but I think it's doable. 425 00:17:48,440 --> 00:17:50,520 I think the part that is interesting 426 00:17:50,520 --> 00:17:55,040 and also shifting a little bit on the investor side, 427 00:17:55,040 --> 00:17:58,680 I think that when investors expect certain types 428 00:17:58,680 --> 00:18:01,080 of performance, then the incentives are there 429 00:18:01,080 --> 00:18:02,320 to perform in that way. 430 00:18:02,320 --> 00:18:04,120 And so increasingly we're seeing, 431 00:18:04,120 --> 00:18:06,240 I'm sure you've noticed a lot of investors 432 00:18:06,240 --> 00:18:08,520 who are focused on certain principles 433 00:18:08,520 --> 00:18:11,720 or have certain areas of social impact 434 00:18:11,720 --> 00:18:13,480 that they want to aim toward, 435 00:18:13,480 --> 00:18:15,240 whether it's something around the environment 436 00:18:15,240 --> 00:18:17,680 or something around closing equity gaps 437 00:18:17,680 --> 00:18:19,720 or whatever the case may be. 438 00:18:19,720 --> 00:18:22,800 And those incentives from the people that bring the money 439 00:18:22,800 --> 00:18:26,240 I think are really helpful to shaping the future 440 00:18:26,240 --> 00:18:28,720 of startups and technology in the world. 441 00:18:28,720 --> 00:18:31,440 So it's not easy, but I think the incentives 442 00:18:31,440 --> 00:18:32,840 are starting to shift. 443 00:18:32,840 --> 00:18:34,760 And I think that startup founders 444 00:18:34,760 --> 00:18:37,520 are really starting to prioritize 445 00:18:37,520 --> 00:18:40,000 some of these kind of net positive things 446 00:18:40,000 --> 00:18:43,800 for however it is that their product will impact their users 447 00:18:43,800 --> 00:18:45,200 and then society at large. 448 00:18:45,200 --> 00:18:46,680 I don't know what your thoughts are, 449 00:18:46,680 --> 00:18:48,200 but I think it's hard. 450 00:18:48,200 --> 00:18:49,040 It's very hard. 451 00:18:49,040 --> 00:18:51,400 I think we get some benefit from the fact 452 00:18:51,400 --> 00:18:54,280 that we have more altruistic younger generations 453 00:18:54,280 --> 00:18:55,400 that give more to charity 454 00:18:55,400 --> 00:18:59,200 that are generally more supportive of causes. 455 00:18:59,200 --> 00:19:00,880 In fact, I'm pretty sure there's, 456 00:19:00,880 --> 00:19:02,440 this might be more in your realm, Gota, 457 00:19:02,440 --> 00:19:05,600 but a lot of marketers, if they support the causes 458 00:19:05,600 --> 00:19:08,680 of tons of that's important to Generation Z, 459 00:19:08,680 --> 00:19:10,200 millennials, things like that, 460 00:19:10,200 --> 00:19:13,520 that's really the pathway to earning their business. 461 00:19:13,520 --> 00:19:15,880 So we have, I think, a good starting point 462 00:19:15,880 --> 00:19:17,600 because these people are the ones that generally 463 00:19:17,600 --> 00:19:20,040 are engaging in these startup activities. 464 00:19:20,040 --> 00:19:23,000 It's like Gota said, it's gonna be hard 465 00:19:23,000 --> 00:19:24,360 and maybe they don't get a VC 466 00:19:24,360 --> 00:19:26,640 that has some of those central 467 00:19:26,640 --> 00:19:28,280 to their priority that's investing. 468 00:19:28,280 --> 00:19:30,320 How do you, from the perspective of someone 469 00:19:30,320 --> 00:19:34,840 that's just, that's coming up, has a super exciting new idea 470 00:19:34,840 --> 00:19:36,960 and is immediately shot out of a cannon 471 00:19:36,960 --> 00:19:38,520 in terms of the interest in it? 472 00:19:38,520 --> 00:19:40,840 I may be speaking from a little bit of experience here, 473 00:19:40,840 --> 00:19:45,760 but how do you recommend someone continue to hone 474 00:19:45,760 --> 00:19:49,240 and sharpen that kind of steel to remain, 475 00:19:49,240 --> 00:19:53,400 have to have ethical AI use and ethical employment central 476 00:19:53,400 --> 00:19:55,040 to their core values? 477 00:19:55,040 --> 00:19:58,840 And Wes, I would just add that maybe not just startups, 478 00:19:58,840 --> 00:20:02,720 but also corporations who are looking into integrating AI 479 00:20:02,720 --> 00:20:05,000 and building the whole team coming up. 480 00:20:05,000 --> 00:20:06,920 I'm looking, I just want advice from me. 481 00:20:06,920 --> 00:20:10,000 I just, I just, I need to set it for everybody, 482 00:20:10,000 --> 00:20:11,240 but that was for me. 483 00:20:11,240 --> 00:20:12,600 We can get to corpse. 484 00:20:12,600 --> 00:20:15,560 Mappo, please answer, I need to take some notes. 485 00:20:15,560 --> 00:20:18,360 So selfish, our audience. 486 00:20:18,360 --> 00:20:19,200 Whatever. 487 00:20:19,200 --> 00:20:22,360 Okay, first problems, Wes, and then everyone else. 488 00:20:22,360 --> 00:20:25,640 I think, again, I just have to emphasize, 489 00:20:25,640 --> 00:20:26,920 this stuff is not easy. 490 00:20:26,920 --> 00:20:28,080 So one of the things, 491 00:20:28,080 --> 00:20:30,760 and the reason I'm emphasizing this 492 00:20:30,760 --> 00:20:34,000 is because I listen to these discussions all the time 493 00:20:34,000 --> 00:20:35,640 of people saying, oh, all you have to do 494 00:20:35,640 --> 00:20:37,840 is have these principles and do this and do that, 495 00:20:37,840 --> 00:20:38,760 and whatever. 496 00:20:38,760 --> 00:20:40,560 This stuff takes time, it takes energy. 497 00:20:40,560 --> 00:20:42,440 And when you're a small entity, 498 00:20:42,440 --> 00:20:44,760 and even when you're a large corporation, 499 00:20:44,760 --> 00:20:46,000 go to something you said earlier, 500 00:20:46,000 --> 00:20:47,280 it's competitive out there. 501 00:20:47,280 --> 00:20:48,560 Like it's real. 502 00:20:48,560 --> 00:20:50,200 Like you can't just sit there 503 00:20:50,200 --> 00:20:52,640 and think deep thoughts for months and then say, 504 00:20:52,640 --> 00:20:55,360 ah, now we know how to build something responsibly. 505 00:20:55,360 --> 00:20:56,440 It doesn't work that way. 506 00:20:56,440 --> 00:21:00,960 So I wanna say this just with that understanding 507 00:21:00,960 --> 00:21:03,000 of the fact that doing things, 508 00:21:03,000 --> 00:21:05,200 changing the way that we build technology 509 00:21:05,200 --> 00:21:07,240 and adapting it and evolving it 510 00:21:07,240 --> 00:21:09,920 so that things like responsible practices 511 00:21:09,920 --> 00:21:11,600 are built into the way that we work, 512 00:21:11,600 --> 00:21:12,600 that doesn't happen overnight. 513 00:21:12,600 --> 00:21:14,600 We don't change the way we do things overnight. 514 00:21:14,600 --> 00:21:16,840 So I do think though that there are a few things. 515 00:21:16,840 --> 00:21:18,600 Now we are lucky enough 516 00:21:18,600 --> 00:21:21,280 that we've had so many different companies, 517 00:21:21,280 --> 00:21:23,040 the Microstalks, the Googles, 518 00:21:23,040 --> 00:21:25,280 the OECDs, an organization 519 00:21:25,280 --> 00:21:27,960 that's not a private corporation of the world, 520 00:21:27,960 --> 00:21:30,880 come up with this incredible collection 521 00:21:30,880 --> 00:21:33,120 of ethical AI principles. 522 00:21:33,120 --> 00:21:35,480 So that hard work has been done. 523 00:21:35,480 --> 00:21:37,600 Pick one, look it up. 524 00:21:37,600 --> 00:21:40,040 You'll find if you compare and contrast them 525 00:21:40,040 --> 00:21:41,360 that they're mostly similar. 526 00:21:41,360 --> 00:21:42,600 There are nuances here and there, 527 00:21:42,600 --> 00:21:43,840 but they're mostly similar. 528 00:21:43,840 --> 00:21:45,600 And pick one that works for you. 529 00:21:45,600 --> 00:21:46,960 And then start to think about 530 00:21:46,960 --> 00:21:48,760 how is the work that I'm doing? 531 00:21:48,760 --> 00:21:50,360 How is the product that I'm building 532 00:21:50,360 --> 00:21:52,000 aligning to these principles? 533 00:21:52,000 --> 00:21:53,920 Have to sit down for eight months 534 00:21:53,920 --> 00:21:55,480 and do a deep study on it, 535 00:21:55,480 --> 00:21:57,320 but take a couple of hours to sit down 536 00:21:57,320 --> 00:21:58,320 and figure out like, 537 00:21:58,320 --> 00:22:01,440 am I actually even thinking about these principles? 538 00:22:01,440 --> 00:22:03,720 And if so, how is my workflow 539 00:22:03,720 --> 00:22:06,480 laddering up to them in some way, shape or form? 540 00:22:06,480 --> 00:22:08,200 It's a thought exercise. 541 00:22:08,200 --> 00:22:09,880 It shouldn't take you forever. 542 00:22:09,880 --> 00:22:11,400 It takes a couple of hours. 543 00:22:11,400 --> 00:22:14,040 And I think that just having that conversation 544 00:22:14,040 --> 00:22:16,640 with yourself or if you are in a broader, 545 00:22:16,640 --> 00:22:19,760 like in a larger corporation with a few team members, 546 00:22:19,760 --> 00:22:22,840 just having that conversation from time to time, 547 00:22:22,840 --> 00:22:26,360 once a month, a couple of times in a month, 548 00:22:26,360 --> 00:22:27,440 as a starting point, 549 00:22:27,440 --> 00:22:29,400 I think is a really good place to start. 550 00:22:29,400 --> 00:22:31,440 Because if you're priming your brain 551 00:22:31,440 --> 00:22:34,240 to think about those principles frequently enough, 552 00:22:34,240 --> 00:22:36,920 what you will find is you start making decisions 553 00:22:36,920 --> 00:22:39,880 that suddenly you're like, oh, hold on. 554 00:22:39,880 --> 00:22:43,200 Did I even think about the transparency side of this yet? 555 00:22:43,200 --> 00:22:45,680 And then you realize that question, 556 00:22:45,680 --> 00:22:47,840 I'm picking transparency just for fun, 557 00:22:47,840 --> 00:22:49,480 but that question of transparency 558 00:22:49,480 --> 00:22:52,760 suddenly starts to pop into your brain once a day, 559 00:22:52,760 --> 00:22:54,120 twice a day, three times a day. 560 00:22:54,120 --> 00:22:55,160 And you're like, okay, 561 00:22:55,160 --> 00:22:57,240 I need to figure out this transparency thing. 562 00:22:57,240 --> 00:22:59,640 And then suddenly you find yourself looking for the people 563 00:22:59,640 --> 00:23:01,200 that can help you solve that problem. 564 00:23:01,200 --> 00:23:03,160 And it becomes part of the way you build 565 00:23:03,160 --> 00:23:04,760 whatever it is that you're building. 566 00:23:04,760 --> 00:23:07,600 And so it kind of allows for a more intuitive 567 00:23:07,600 --> 00:23:12,200 and organic way for you to build these principles 568 00:23:12,200 --> 00:23:14,640 into the way that you work. 569 00:23:14,640 --> 00:23:18,480 That's one thing that I would say is pretty helpful 570 00:23:18,480 --> 00:23:19,800 and pretty kind of low lift, 571 00:23:19,800 --> 00:23:22,560 especially when you don't have a ton of resources at hand. 572 00:23:22,560 --> 00:23:24,560 The other thing I would say is that, 573 00:23:24,560 --> 00:23:26,080 I mentioned this earlier, 574 00:23:26,080 --> 00:23:29,640 when we talk about things like responsible AI, 575 00:23:29,640 --> 00:23:32,720 there's a lot to be done at the technical level, 576 00:23:32,720 --> 00:23:35,000 which I think is little by little 577 00:23:35,000 --> 00:23:37,720 becoming a little bit more measurable. 578 00:23:37,720 --> 00:23:40,320 So it's still in development, 579 00:23:40,320 --> 00:23:43,120 but we now have tools that can help measure fairness. 580 00:23:43,120 --> 00:23:44,360 What does fairness really mean? 581 00:23:44,360 --> 00:23:45,480 Are we measuring the right thing? 582 00:23:45,480 --> 00:23:46,920 I know that's a big question 583 00:23:46,920 --> 00:23:48,640 to which there's no magic answer, 584 00:23:48,640 --> 00:23:49,960 but at least now there are tools. 585 00:23:49,960 --> 00:23:52,160 A lot of them are open source, they're out there. 586 00:23:52,160 --> 00:23:55,280 So now we can leverage some of that at the technical level. 587 00:23:55,280 --> 00:23:57,840 There are AI system model cards 588 00:23:57,840 --> 00:24:00,200 that a lot of different organizations have developed. 589 00:24:00,200 --> 00:24:02,600 You can borrow those and many of them, again, 590 00:24:02,600 --> 00:24:04,920 they're open source, they're there for your use. 591 00:24:04,920 --> 00:24:07,040 Like you can start to implement those. 592 00:24:07,040 --> 00:24:08,800 And then some of the big questions, 593 00:24:08,800 --> 00:24:10,480 those are the things at the technical level. 594 00:24:10,480 --> 00:24:12,600 The things that the big question side, 595 00:24:12,600 --> 00:24:16,600 your business model, the product itself 596 00:24:16,600 --> 00:24:19,400 and how people are going to interact with it 597 00:24:19,400 --> 00:24:21,520 and what the follow-on effects of that might be. 598 00:24:21,520 --> 00:24:23,040 Those are big questions. 599 00:24:23,040 --> 00:24:25,480 You're not gonna sit down and fix those 600 00:24:25,480 --> 00:24:28,840 because they're deep, big questions. 601 00:24:28,840 --> 00:24:30,680 But you do wanna probably sit down 602 00:24:30,680 --> 00:24:33,360 with your organization's leadership team, 603 00:24:33,360 --> 00:24:35,120 with your investors if you have them, 604 00:24:35,120 --> 00:24:36,400 and actually just talk about it. 605 00:24:36,400 --> 00:24:38,640 Like what is it that we're building here 606 00:24:38,640 --> 00:24:40,600 and what is it going to do in the world? 607 00:24:40,600 --> 00:24:44,120 And a really great example of an area 608 00:24:44,120 --> 00:24:46,640 where we did not do this as a society 609 00:24:46,640 --> 00:24:48,800 until probably a little too late 610 00:24:48,800 --> 00:24:50,960 was ad-driven business models 611 00:24:50,960 --> 00:24:53,600 where a lot of personalized ads 612 00:24:53,600 --> 00:24:57,320 turned out to cause a few bits of discomfort 613 00:24:57,320 --> 00:24:58,400 around the world. 614 00:24:58,400 --> 00:25:00,840 And so these deep questions about business model 615 00:25:00,840 --> 00:25:03,520 that nobody really thought about early on 616 00:25:03,520 --> 00:25:05,480 are now big problems that we're trying to solve 617 00:25:05,480 --> 00:25:06,600 across society. 618 00:25:06,600 --> 00:25:08,360 But had we had that opportunity 619 00:25:08,360 --> 00:25:11,440 to have an earlier conversation around the business model 620 00:25:11,440 --> 00:25:13,320 and what the implications were, 621 00:25:13,320 --> 00:25:16,200 could have probably maybe tweaked the different approaches 622 00:25:16,200 --> 00:25:17,360 to how we built those products. 623 00:25:17,360 --> 00:25:20,680 So I think, again, like you're not gonna have a you, 624 00:25:20,680 --> 00:25:23,480 you're not gonna have a team of 1000 ethicists 625 00:25:23,480 --> 00:25:27,400 and designers and engineers to build your, 626 00:25:27,400 --> 00:25:30,640 responsible AI solution suite. 627 00:25:30,640 --> 00:25:31,640 Yeah, yet. 628 00:25:31,640 --> 00:25:33,920 But you can sit down with the principals 629 00:25:33,920 --> 00:25:36,200 and at least as a starting point. 630 00:25:36,200 --> 00:25:38,720 And what you will, I think find is that 631 00:25:38,720 --> 00:25:40,840 there are a few things that are gonna jump out at you 632 00:25:40,840 --> 00:25:43,960 and they're gonna be questions that you wanna answer 633 00:25:43,960 --> 00:25:45,160 over and over again and say, 634 00:25:45,160 --> 00:25:47,040 okay, I guess I better figure out 635 00:25:47,040 --> 00:25:49,880 how I'm going to build this into the way that I work 636 00:25:49,880 --> 00:25:52,040 and the way that I build this product. 637 00:25:52,040 --> 00:25:54,200 And then over time it becomes more sophisticated 638 00:25:54,200 --> 00:25:57,440 and then you have a team of 1000 and then 639 00:25:57,440 --> 00:25:59,280 have a different kind of problem. 640 00:25:59,280 --> 00:26:00,440 That's fantastic advice. 641 00:26:00,440 --> 00:26:02,840 And one I think is very relatable 642 00:26:02,840 --> 00:26:04,680 even for someone building stuff. 643 00:26:04,680 --> 00:26:08,800 We came up against this with a client not too long ago 644 00:26:08,800 --> 00:26:13,720 that the tantalizing drive of potentially a big contract 645 00:26:13,720 --> 00:26:16,760 butts up against something that maybe we haven't sat 646 00:26:16,760 --> 00:26:18,480 and taken enough time to evaluate. 647 00:26:18,480 --> 00:26:20,880 I think we made the right decision with that. 648 00:26:20,880 --> 00:26:22,840 But that was the first time I think 649 00:26:22,840 --> 00:26:24,000 just even at a small scale, 650 00:26:24,000 --> 00:26:25,960 everyone's gonna confront this. 651 00:26:25,960 --> 00:26:27,880 I know we wanna get, we wanna dive into 652 00:26:27,880 --> 00:26:30,480 some of your responsible AI discussion guide, 653 00:26:30,480 --> 00:26:32,520 but you did bring up something that I really wanna 654 00:26:32,520 --> 00:26:34,160 maybe touch on a little bit. 655 00:26:34,160 --> 00:26:36,200 And you talked about open source. 656 00:26:36,200 --> 00:26:38,680 There's some open source information that's used 657 00:26:38,680 --> 00:26:43,400 for some of these ethical value shaping propositions 658 00:26:43,400 --> 00:26:45,000 and things like that that companies do. 659 00:26:45,000 --> 00:26:49,680 What is your opinion on open source versus closed sourced 660 00:26:49,680 --> 00:26:52,120 AI models at this juncture? 661 00:26:52,120 --> 00:26:54,040 And then maybe down the road, 662 00:26:54,040 --> 00:26:57,640 what is the kind of positives and negatives in your eyes 663 00:26:57,640 --> 00:27:01,360 and what do you think is the pathway to not only use 664 00:27:01,360 --> 00:27:04,000 the thing that's most transparent and ethical, 665 00:27:04,000 --> 00:27:09,000 but also keeps us safe and doesn't let this stuff run wild? 666 00:27:09,360 --> 00:27:11,240 That is a really good question. 667 00:27:11,240 --> 00:27:12,760 I have not made up my mind. 668 00:27:12,760 --> 00:27:15,400 Just to, I'll lay that out upfront. 669 00:27:15,400 --> 00:27:18,040 I've not made up my mind on either or, 670 00:27:18,040 --> 00:27:21,400 or if there's one approach that's better than the other. 671 00:27:21,400 --> 00:27:24,920 I think the closed source or non open source stuff, 672 00:27:24,920 --> 00:27:27,400 I think it's great because it allows there to be 673 00:27:27,400 --> 00:27:32,400 a bit more control and safety measures, at least in theory. 674 00:27:32,440 --> 00:27:35,360 There's just more control over what gets released. 675 00:27:35,360 --> 00:27:38,880 And I think that since we're still learning a lot about 676 00:27:38,880 --> 00:27:41,400 and especially generative AI, 677 00:27:41,400 --> 00:27:43,840 and I say we collectively, we humanity, 678 00:27:43,840 --> 00:27:45,000 we're still learning a lot about it, 679 00:27:45,000 --> 00:27:46,360 not just on the technical side, 680 00:27:46,360 --> 00:27:49,240 but how we're gonna use it, what the impact might be on job, 681 00:27:49,240 --> 00:27:50,360 all the things. 682 00:27:50,360 --> 00:27:53,800 It's nice to know that there's maybe a smaller group 683 00:27:53,800 --> 00:27:57,760 of corporations that are very much in the public eye. 684 00:27:57,760 --> 00:27:59,760 A lot of people are watching them, 685 00:27:59,760 --> 00:28:01,760 both for the great things that they'll do, 686 00:28:01,760 --> 00:28:04,160 but also for the things they might do wrong. 687 00:28:04,160 --> 00:28:06,760 And so it's just nice, I think, to know that there's 688 00:28:06,760 --> 00:28:09,760 that kind of a straight line of accountability in some sense. 689 00:28:09,760 --> 00:28:12,720 On the other hand, I think the open source approach 690 00:28:12,720 --> 00:28:13,880 is so cool, right? 691 00:28:13,880 --> 00:28:18,880 Because it allows more people access to experiment 692 00:28:19,280 --> 00:28:21,120 and learn and build. 693 00:28:21,120 --> 00:28:23,120 And I think that, I don't know, 694 00:28:23,120 --> 00:28:25,320 I don't study philosophy or history, 695 00:28:25,320 --> 00:28:28,800 but I do think that part of the magic of tech 696 00:28:28,800 --> 00:28:30,800 for a lot of people is this notion 697 00:28:30,800 --> 00:28:33,640 that it can give so many people opportunity. 698 00:28:33,640 --> 00:28:37,320 It's a way to democratize opportunity in some sense. 699 00:28:37,320 --> 00:28:40,840 And so I love this idea of open source as a general matter, 700 00:28:40,840 --> 00:28:43,080 but especially with something like AI 701 00:28:43,080 --> 00:28:45,680 that I think is increasingly showing itself 702 00:28:45,680 --> 00:28:49,840 to be an economic opportunity for all sorts of people 703 00:28:49,840 --> 00:28:51,000 all over the world. 704 00:28:51,000 --> 00:28:55,080 I don't know, just having that as an open source resource, 705 00:28:55,080 --> 00:28:56,960 I think is really cool, of course. 706 00:28:56,960 --> 00:28:59,320 But that also means that there's less control 707 00:28:59,320 --> 00:29:01,320 and we're not really sure what it means. 708 00:29:01,320 --> 00:29:05,200 It might be fine, like maybe the lack of control 709 00:29:05,200 --> 00:29:07,560 and the experimentation, maybe it'll be fine. 710 00:29:07,560 --> 00:29:10,280 Maybe as we develop the technology, 711 00:29:10,280 --> 00:29:13,480 the norms around how we use it and how we develop it 712 00:29:13,480 --> 00:29:16,620 and other safety measures will grow along with it. 713 00:29:16,620 --> 00:29:18,240 We don't know yet, but it's a gamble 714 00:29:18,240 --> 00:29:19,640 because maybe those things won't happen 715 00:29:19,640 --> 00:29:22,440 and maybe it'll be a disaster and horrible people 716 00:29:22,440 --> 00:29:24,800 will get their hands on this open source stuff 717 00:29:24,800 --> 00:29:26,200 and we're in the world. 718 00:29:26,200 --> 00:29:28,040 I don't know, I see both sides of it. 719 00:29:28,040 --> 00:29:29,720 I haven't made up my mind. 720 00:29:29,720 --> 00:29:33,240 The protector side of me is no, keep it closed. 721 00:29:33,240 --> 00:29:35,880 And then the kind of just like, I don't know, 722 00:29:35,880 --> 00:29:38,640 I want everyone to have an opportunity side of me 723 00:29:38,640 --> 00:29:40,080 loves the open source model. 724 00:29:40,080 --> 00:29:43,840 Maybe there is something open source, as you said, 725 00:29:43,840 --> 00:29:46,340 diversity comes into place because more people 726 00:29:46,340 --> 00:29:48,320 simply can bring in and chip in. 727 00:29:48,320 --> 00:29:51,440 But maybe there is something that even open source community 728 00:29:51,440 --> 00:29:54,160 could pick and learn from the big companies 729 00:29:54,160 --> 00:29:56,440 which have this thousand people staff. 730 00:29:56,440 --> 00:29:58,880 And this brings me that I had this 731 00:29:58,880 --> 00:30:00,480 very fortunate conversation. 732 00:30:00,480 --> 00:30:05,480 One of our friends is sitting two seats below Microsoft CEO. 733 00:30:05,480 --> 00:30:07,600 So then I was talking with her and I asked, 734 00:30:07,600 --> 00:30:10,820 oh, okay, how about your regulations? 735 00:30:10,820 --> 00:30:12,680 And she just laughed. 736 00:30:12,680 --> 00:30:15,640 She was like, Microsoft paved the way. 737 00:30:15,640 --> 00:30:17,520 We wouldn't be able to do what we do 738 00:30:17,520 --> 00:30:21,480 if we didn't install these in our ginormous organization. 739 00:30:21,480 --> 00:30:24,880 So she's saying that Microsoft is like consulting 740 00:30:24,880 --> 00:30:27,160 commissioners on these stuff. 741 00:30:27,160 --> 00:30:29,080 So that kind of brings me thinking that 742 00:30:29,080 --> 00:30:31,760 maybe smaller startups or like open source 743 00:30:31,760 --> 00:30:33,800 could actually pick up some resources, 744 00:30:33,800 --> 00:30:37,760 how to move towards trustworthy AI, responsible AI. 745 00:30:37,760 --> 00:30:41,320 And with that, I know that you created something. 746 00:30:41,320 --> 00:30:43,720 So maybe we could, with this podcast, with this discussion, 747 00:30:43,720 --> 00:30:46,400 also leave listeners with some things 748 00:30:46,400 --> 00:30:48,920 that may need to think about, as you mentioned, 749 00:30:48,920 --> 00:30:50,760 just once a month, maybe sit down. 750 00:30:50,760 --> 00:30:55,200 I need to say that I'm always seeking for content 751 00:30:55,200 --> 00:30:57,400 which simplifies complex things, 752 00:30:57,400 --> 00:31:00,160 which also makes it accessible to more people. 753 00:31:00,160 --> 00:31:03,440 And this is one of the amazing skills to have. 754 00:31:03,440 --> 00:31:06,560 And your analogies are the best. 755 00:31:06,560 --> 00:31:10,680 If you could walk us through and share some kind of breakdowns, 756 00:31:10,680 --> 00:31:11,520 that would be amazing. 757 00:31:11,520 --> 00:31:13,720 If I could too, I can read one of these. 758 00:31:13,720 --> 00:31:16,760 Mapo has put out a discussion guide that we, like I said, 759 00:31:16,760 --> 00:31:18,800 we'll share in the show notes. 760 00:31:18,800 --> 00:31:21,440 And it has some really, really clever analogies. 761 00:31:21,440 --> 00:31:23,920 I think one that is obviously one we talk about a whole lot 762 00:31:23,920 --> 00:31:24,760 is generative AI. 763 00:31:24,760 --> 00:31:28,720 So I'm going to read the analogy in that side. 764 00:31:28,720 --> 00:31:30,280 And then maybe we could talk about some 765 00:31:30,280 --> 00:31:31,360 of your thoughts around it. 766 00:31:31,360 --> 00:31:33,560 Child has a large box of Legos. 767 00:31:33,560 --> 00:31:36,400 The child learns to build all sorts of things using Legos 768 00:31:36,400 --> 00:31:38,840 by looking through a large catalog available 769 00:31:38,840 --> 00:31:41,280 on the internet of pre-built Lego structures, 770 00:31:41,280 --> 00:31:43,160 some with accompanying instructions, 771 00:31:43,160 --> 00:31:44,880 i.e. supervised learning, and some 772 00:31:44,880 --> 00:31:47,080 without, i.e. unsupervised learning. 773 00:31:47,080 --> 00:31:49,800 That is a wonderful analogy for generative AI. 774 00:31:49,800 --> 00:31:52,560 I'm glad you guys think that they're helpful analogies. 775 00:31:52,560 --> 00:31:56,160 And I will say, I feel like it's important for me 776 00:31:56,160 --> 00:31:57,560 to disclaim a little bit. 777 00:31:57,560 --> 00:32:00,040 I didn't come up with all of these by myself. 778 00:32:00,040 --> 00:32:04,200 Again, a culmination of years of trying to myself learn 779 00:32:04,200 --> 00:32:05,600 and understand AI. 780 00:32:05,600 --> 00:32:08,160 And I've read so many analogies. 781 00:32:08,160 --> 00:32:10,520 Colleagues have shared analogies. 782 00:32:10,520 --> 00:32:13,120 Some of these are, especially the earlier ones, 783 00:32:13,120 --> 00:32:14,520 are the classic ones that you might 784 00:32:14,520 --> 00:32:17,320 find in explainers or textbooks, like the teacher teaching 785 00:32:17,320 --> 00:32:19,480 the child if it's a cat or a dog. 786 00:32:19,480 --> 00:32:21,520 That's a classic example by now. 787 00:32:21,520 --> 00:32:23,640 Isn't it funny that AI has been around long enough 788 00:32:23,640 --> 00:32:25,080 that we have classic examples? 789 00:32:25,080 --> 00:32:28,200 Classic, I think that's what happened. 790 00:32:28,200 --> 00:32:30,920 A world before is considered classic examples, right? 791 00:32:30,920 --> 00:32:32,760 So I just want to throw that out there 792 00:32:32,760 --> 00:32:35,160 because I don't want anyone to think I sat down and dreamed 793 00:32:35,160 --> 00:32:36,160 these up myself. 794 00:32:36,160 --> 00:32:39,240 But yeah, I really, again, for me, 795 00:32:39,240 --> 00:32:42,280 as someone who has a technical background but does not build, 796 00:32:42,280 --> 00:32:44,440 I don't build AI products myself. 797 00:32:44,440 --> 00:32:48,040 I'll usually sit down with an engineer 798 00:32:48,040 --> 00:32:50,120 and they'll walk me through all the math. 799 00:32:50,120 --> 00:32:53,720 And then I'll ask a million, I'm sure, very annoying questions. 800 00:32:53,720 --> 00:32:57,040 And then I'll go away and say, I still don't get this 801 00:32:57,040 --> 00:32:58,360 and I still don't get that. 802 00:32:58,360 --> 00:33:00,800 And then I'll go find a simple analogy 803 00:33:00,800 --> 00:33:03,400 that either I read or someone shared with me 804 00:33:03,400 --> 00:33:05,200 or that I cobbled together. 805 00:33:05,200 --> 00:33:09,040 And then I put the two together till I can make sense of it. 806 00:33:09,040 --> 00:33:12,040 And I have just found that to be a very useful tool 807 00:33:12,040 --> 00:33:15,160 to then start to think about, OK, what 808 00:33:15,160 --> 00:33:17,120 are some of the things that might go wrong 809 00:33:17,120 --> 00:33:20,360 or that we want to do better when we build whatever it 810 00:33:20,360 --> 00:33:21,520 is that we're building? 811 00:33:21,520 --> 00:33:25,520 So that's really the origin of the whole analogy thing. 812 00:33:25,520 --> 00:33:28,040 And then I was, I know we were talking about this a little bit 813 00:33:28,040 --> 00:33:29,840 earlier before we started recording, 814 00:33:29,840 --> 00:33:32,200 but I was running this roundtable 815 00:33:32,200 --> 00:33:35,360 where I would have to talk about trustworthy AI issues 816 00:33:35,360 --> 00:33:38,640 to a team that was not very familiar with AI. 817 00:33:38,640 --> 00:33:40,600 And we had very short amount of time. 818 00:33:40,600 --> 00:33:42,080 And I was thinking, how in the world 819 00:33:42,080 --> 00:33:43,760 am I going to introduce the technology 820 00:33:43,760 --> 00:33:48,120 and introduce the issues in a way that makes any sort of sense? 821 00:33:48,120 --> 00:33:49,360 And so this is how we did it. 822 00:33:49,360 --> 00:33:52,040 We did it through analogies, and it was quite successful. 823 00:33:52,040 --> 00:33:54,360 Then I decided to write them all down, and there you go. 824 00:33:54,360 --> 00:33:58,320 Just for listeners who are not checking out us on YouTube, 825 00:33:58,320 --> 00:34:02,400 which you should, the document contains of two kind of tables. 826 00:34:02,400 --> 00:34:05,320 And on one side, we have AI and machine learning 827 00:34:05,320 --> 00:34:06,800 types and analogies. 828 00:34:06,800 --> 00:34:08,360 That's what we are talking about. 829 00:34:08,360 --> 00:34:10,760 But on the right side, we have examples 830 00:34:10,760 --> 00:34:13,320 of trustworthy AI issues. 831 00:34:13,320 --> 00:34:17,160 And what Wes read was on the left side. 832 00:34:17,160 --> 00:34:19,320 So it was under generative AI. 833 00:34:19,320 --> 00:34:22,880 There is also, you broke down deep learning models, 834 00:34:22,880 --> 00:34:26,800 transfer learning models, reinforcement, which is amazing. 835 00:34:26,800 --> 00:34:30,560 But on the generative AI side, on the right side, 836 00:34:30,560 --> 00:34:33,680 you have hallucinations and intellectual property. 837 00:34:33,680 --> 00:34:36,520 I'll read the intellectual property analogy first, 838 00:34:36,520 --> 00:34:38,560 as it pertains, of course, to LEGO blocks. 839 00:34:38,560 --> 00:34:41,480 You ask the child to build you a unique LEGO structure. 840 00:34:41,480 --> 00:34:44,320 The child builds something that seems interesting 841 00:34:44,320 --> 00:34:47,640 and different, and the child claims it's totally unique. 842 00:34:47,640 --> 00:34:50,920 You find out later that the child's unique LEGO structure 843 00:34:50,920 --> 00:34:53,840 was copied directly from an existing proprietary structure 844 00:34:53,840 --> 00:34:57,480 the child had seen when learning about LEGO structures. 845 00:34:57,480 --> 00:34:58,320 There you go. 846 00:34:58,320 --> 00:35:01,520 IP is a great topic to jump into now. 847 00:35:01,520 --> 00:35:03,360 I would love to hear your take on some of the things 848 00:35:03,360 --> 00:35:06,080 that are starting to creep their way through the courts 849 00:35:06,080 --> 00:35:09,520 that I think are gonna be the subject of huge discussions 850 00:35:09,520 --> 00:35:10,440 in the future. 851 00:35:10,440 --> 00:35:12,840 Yeah, let's maybe talk about that for a little bit. 852 00:35:12,840 --> 00:35:13,680 Yeah, sure. 853 00:35:13,680 --> 00:35:17,440 IP, as you said, is definitely an interesting topic de jure 854 00:35:17,440 --> 00:35:20,040 when it comes to generative AI. 855 00:35:20,040 --> 00:35:21,960 I think there are a lot of different elements to it. 856 00:35:21,960 --> 00:35:23,760 One that I think a lot of people are talking about 857 00:35:23,760 --> 00:35:24,800 right now is scraping. 858 00:35:24,800 --> 00:35:27,680 So obviously a lot of these generative AI models, 859 00:35:27,680 --> 00:35:29,360 they feed on data. 860 00:35:29,360 --> 00:35:31,240 They need tons of information, 861 00:35:31,240 --> 00:35:33,400 and what better place to get information 862 00:35:33,400 --> 00:35:36,040 than the worldwide web? 863 00:35:36,040 --> 00:35:37,200 There's so much out there. 864 00:35:37,200 --> 00:35:39,400 And a lot of the stuff that is fed 865 00:35:39,400 --> 00:35:42,960 into the larger generative AI models 866 00:35:42,960 --> 00:35:47,960 comes from sources that maybe weren't meant to feed in 867 00:35:47,960 --> 00:35:49,800 to a commercial product 868 00:35:49,800 --> 00:35:52,040 that was then going to be sold for profit. 869 00:35:52,040 --> 00:35:55,920 So think about, for example, a book that was published, 870 00:35:55,920 --> 00:35:59,040 and there's like a Google Books version 871 00:35:59,040 --> 00:36:00,760 of five of the chapters, 872 00:36:00,760 --> 00:36:04,640 and it's just getting fed into this generative AI model 873 00:36:04,640 --> 00:36:05,920 as training data. 874 00:36:05,920 --> 00:36:07,920 Or let's say all of your blog posts 875 00:36:07,920 --> 00:36:10,120 that you may have some time ago 876 00:36:10,120 --> 00:36:11,400 is just getting fed in there, 877 00:36:11,400 --> 00:36:14,480 or any number of things that might be on the web 878 00:36:14,480 --> 00:36:17,560 that aren't really meant to go into someone else's product 879 00:36:17,560 --> 00:36:19,400 that they are going to sell for a profit. 880 00:36:19,400 --> 00:36:20,240 People are getting a little bit- 881 00:36:20,240 --> 00:36:21,080 Comments on Reddit. 882 00:36:21,080 --> 00:36:21,920 Comments on Reddit. 883 00:36:21,920 --> 00:36:22,760 Comments on Reddit. 884 00:36:22,760 --> 00:36:24,240 Those literary gems. 885 00:36:24,240 --> 00:36:25,480 It's a huge amount. 886 00:36:25,480 --> 00:36:29,520 Reddit is a huge big block of content on there. 887 00:36:29,520 --> 00:36:31,080 Sorry, didn't mean to interrupt you. 888 00:36:31,080 --> 00:36:33,000 No, no, that's actually a great example 889 00:36:33,000 --> 00:36:36,080 because think about it from Reddit's perspective. 890 00:36:36,080 --> 00:36:39,840 They don't, they probably, and their users probably, 891 00:36:39,840 --> 00:36:42,760 don't want a bunch of their information 892 00:36:42,760 --> 00:36:46,360 being fed into these models as training data, 893 00:36:46,360 --> 00:36:49,040 either for intellectual property purposes, 894 00:36:49,040 --> 00:36:52,480 or also in some cases for some privacy purposes, 895 00:36:52,480 --> 00:36:55,000 or any other kind of concerns that you might have 896 00:36:55,000 --> 00:36:58,480 with your written down comments going into this model 897 00:36:58,480 --> 00:37:01,680 that's going to be used in ways that you can't really control. 898 00:37:01,680 --> 00:37:04,760 And so I think that's one of the levels 899 00:37:04,760 --> 00:37:06,760 of intellectual property that's coming up 900 00:37:06,760 --> 00:37:10,920 is the stuff that's getting fed into these models aren't, 901 00:37:10,920 --> 00:37:13,840 some of the owners or the creators of that data 902 00:37:13,840 --> 00:37:15,720 aren't really getting to control 903 00:37:15,720 --> 00:37:17,640 whether or not their stuff gets fed in. 904 00:37:17,640 --> 00:37:19,880 Though, if I remember this correctly, 905 00:37:19,880 --> 00:37:22,240 I think Wes, or no, go to you were saying 906 00:37:22,240 --> 00:37:24,320 that things in this space change so much. 907 00:37:24,320 --> 00:37:26,960 I feel if I don't read the news every single day, 908 00:37:26,960 --> 00:37:28,560 I've missed everything. 909 00:37:28,560 --> 00:37:31,800 I think I remember reading that open AI 910 00:37:31,800 --> 00:37:34,120 is going to create a control that, 911 00:37:34,120 --> 00:37:37,800 so websites can opt out of having their data 912 00:37:37,800 --> 00:37:39,720 fed into their models. 913 00:37:39,720 --> 00:37:41,680 I feel like I remember reading that somewhere. 914 00:37:41,680 --> 00:37:45,320 I think it was that, or they open sourced, 915 00:37:45,320 --> 00:37:48,600 they published their crawler, their web crawler bot. 916 00:37:48,600 --> 00:37:53,040 So if websites wanted to make a little defense against it, 917 00:37:53,040 --> 00:37:53,880 they could. 918 00:37:53,880 --> 00:37:55,680 I just want to go back to something though, 919 00:37:55,680 --> 00:37:58,440 because you spoke about people didn't expect to, 920 00:37:58,440 --> 00:38:00,880 or didn't want their content potentially fed 921 00:38:00,880 --> 00:38:02,720 into this commercial product. 922 00:38:02,720 --> 00:38:06,040 But I always happen to think that's just an interesting take 923 00:38:06,040 --> 00:38:09,440 because is it just because these, in this use case, 924 00:38:09,440 --> 00:38:12,160 it's being taken kind of one step further 925 00:38:12,160 --> 00:38:15,080 to train and create a commercial product? 926 00:38:15,080 --> 00:38:18,760 Because when we put a text message into our cell phone 927 00:38:18,760 --> 00:38:20,800 and send it, we don't own it anymore. 928 00:38:20,800 --> 00:38:21,960 Our carrier owns it. 929 00:38:24,000 --> 00:38:25,200 The police can send a warrant 930 00:38:25,200 --> 00:38:27,160 and they can get all our text messages. 931 00:38:27,160 --> 00:38:29,760 And that to me, it's okay, you're putting it out there. 932 00:38:29,760 --> 00:38:34,240 There's an illusion of privacy or control, so to speak, 933 00:38:34,240 --> 00:38:36,480 but it's not yours anymore. 934 00:38:36,480 --> 00:38:38,360 And I don't see us lining up being like, 935 00:38:38,360 --> 00:38:40,440 I'm gonna stop sending text messages now then, 936 00:38:40,440 --> 00:38:42,840 since I don't own them as soon as they're released. 937 00:38:42,840 --> 00:38:47,360 Do you think that there's either some technological pathway 938 00:38:47,360 --> 00:38:51,160 where it's just everything is completely end to end secured, 939 00:38:51,160 --> 00:38:55,960 randomized, the company has no idea what's even being sent, 940 00:38:55,960 --> 00:38:56,840 is the solution. 941 00:38:56,840 --> 00:38:58,800 Or the other side of it could be like, 942 00:38:58,800 --> 00:39:01,640 is there some wild solution, this architecture 943 00:39:01,640 --> 00:39:02,840 that we could create for ourselves 944 00:39:02,840 --> 00:39:05,800 that everyone kind of gets a little like Spotify 945 00:39:05,800 --> 00:39:08,280 streaming credit worth of money 946 00:39:08,280 --> 00:39:10,320 if someone trains on their data. 947 00:39:10,320 --> 00:39:14,000 Oh, six of your Reddit comments got pulled for this model. 948 00:39:14,000 --> 00:39:17,280 You get 0.0002 cents or something like that for that. 949 00:39:17,280 --> 00:39:19,920 And you can actually see this times everybody. 950 00:39:19,920 --> 00:39:22,600 I think to me, that's something that would work. 951 00:39:22,600 --> 00:39:25,480 You'd have a ton of infrastructure you need to build 952 00:39:25,480 --> 00:39:27,440 for that, but where do you fall 953 00:39:27,440 --> 00:39:29,880 or where do you think companies should maybe fall 954 00:39:29,880 --> 00:39:32,160 between those two endpoints? 955 00:39:32,160 --> 00:39:35,160 Like many things, I'm gonna give a dissatisfying answer 956 00:39:35,160 --> 00:39:36,680 of I haven't decided yet. 957 00:39:36,680 --> 00:39:38,800 But the question that you're raising, 958 00:39:38,800 --> 00:39:42,680 I think is the same question that we're continuing 959 00:39:42,680 --> 00:39:46,600 to struggle with when it comes to data driven ads, 960 00:39:46,600 --> 00:39:47,560 personalized ads. 961 00:39:47,560 --> 00:39:48,880 It's the same question, right? 962 00:39:48,880 --> 00:39:51,080 Like you put your information out there, 963 00:39:51,080 --> 00:39:53,120 there's not a human being who's sitting there 964 00:39:53,120 --> 00:39:54,640 reading your personal data. 965 00:39:54,640 --> 00:39:56,520 So in terms of like privacy, 966 00:39:56,520 --> 00:39:59,680 it's not like there's some like magic human being 967 00:39:59,680 --> 00:40:02,000 who's like reading everything about you 968 00:40:02,000 --> 00:40:03,960 and annotating everything about you. 969 00:40:03,960 --> 00:40:07,280 But there is something invasive about knowing 970 00:40:07,280 --> 00:40:10,120 that this information that you're putting out there 971 00:40:10,120 --> 00:40:11,920 about yourself in some way, shape or form, 972 00:40:11,920 --> 00:40:13,280 whether it's clicks, whether it's something 973 00:40:13,280 --> 00:40:15,440 you've actually written, whether it's something 974 00:40:15,440 --> 00:40:18,280 you've shared somehow online, a photo, whatever, 975 00:40:18,280 --> 00:40:20,320 just any sort of information you've put out there 976 00:40:20,320 --> 00:40:22,520 by yourself, there's something weird about it, 977 00:40:22,520 --> 00:40:24,920 I think to people, there's something weird about it. 978 00:40:24,920 --> 00:40:28,400 Then being used by a corporation to do something 979 00:40:28,400 --> 00:40:31,320 for the corporation that makes the corporation money 980 00:40:31,320 --> 00:40:35,400 without you really, I don't know, having control over it 981 00:40:35,400 --> 00:40:38,760 or maybe not having had the transparency around it 982 00:40:38,760 --> 00:40:41,440 or maybe remuneration like you were saying, 983 00:40:41,440 --> 00:40:43,080 like you're not getting paid for it. 984 00:40:43,080 --> 00:40:45,080 Although some corporations would argue 985 00:40:45,080 --> 00:40:47,560 you are getting paid for it in the sense 986 00:40:47,560 --> 00:40:51,040 that you're getting the free use of this product or tool 987 00:40:51,040 --> 00:40:52,520 or whatever the case may be. 988 00:40:52,520 --> 00:40:55,920 But I think that debate, it's such a big debate 989 00:40:55,920 --> 00:40:57,560 and I think it goes back to something 990 00:40:57,560 --> 00:41:00,160 we were talking about earlier where a lot 991 00:41:00,160 --> 00:41:04,880 of the tech industry is digging itself out of a big, big hole 992 00:41:04,880 --> 00:41:08,520 that was created when we started using these kind 993 00:41:08,520 --> 00:41:12,640 of data-driven technologies without transparency, 994 00:41:12,640 --> 00:41:16,280 privacy and all of the principles that we talk about 995 00:41:16,280 --> 00:41:20,240 leading the way and have now put a lot of the tech industry 996 00:41:20,240 --> 00:41:24,080 in a position of trying to earn back the trust of people 997 00:41:24,080 --> 00:41:26,680 whose data are feeding a lot of these products. 998 00:41:26,680 --> 00:41:29,080 And so I don't know what the right answer is 999 00:41:29,080 --> 00:41:33,240 in terms of the right model for how we all get comfortable 1000 00:41:33,240 --> 00:41:36,800 with how data might be used and what happens with it 1001 00:41:36,800 --> 00:41:37,620 and all of that. 1002 00:41:37,620 --> 00:41:39,800 But I do think that to your point 1003 00:41:39,800 --> 00:41:41,880 about some technological solutions, 1004 00:41:41,880 --> 00:41:43,960 I think there are a lot of solutions, 1005 00:41:43,960 --> 00:41:46,920 especially with the rise of privacy enhancing technologies 1006 00:41:46,920 --> 00:41:49,480 which are increasingly sophisticated. 1007 00:41:49,480 --> 00:41:51,920 There's some really cool way over my head 1008 00:41:51,920 --> 00:41:54,160 but some really cool work that's being done 1009 00:41:54,160 --> 00:41:58,760 that creates like a level of privacy at the model level. 1010 00:41:58,760 --> 00:42:01,840 So you can create model at the, or create privacy 1011 00:42:01,840 --> 00:42:03,480 at the data level through both. 1012 00:42:03,480 --> 00:42:05,440 But also, yeah, it's fascinating, 1013 00:42:05,440 --> 00:42:06,880 like really, really cool stuff. 1014 00:42:06,880 --> 00:42:09,680 So I think there's stuff we can do there, 1015 00:42:09,680 --> 00:42:13,880 but I don't think that solves that fundamental discomfort 1016 00:42:13,880 --> 00:42:16,240 that some people have with this notion 1017 00:42:16,240 --> 00:42:19,080 of their data being used for this thing that's making money. 1018 00:42:19,080 --> 00:42:21,440 So it's not just this privacy question 1019 00:42:21,440 --> 00:42:23,080 which happens to come up a lot. 1020 00:42:23,080 --> 00:42:25,040 It's just the general, like, I don't know, 1021 00:42:25,040 --> 00:42:26,680 some people don't feel comfortable with it. 1022 00:42:26,680 --> 00:42:29,120 I don't know, I'm probably going to be canceled for this, 1023 00:42:29,120 --> 00:42:33,000 but I have this alternative thought about this. 1024 00:42:33,000 --> 00:42:38,000 The way I see it that people vote with their time or money. 1025 00:42:38,120 --> 00:42:40,880 And I was at the conference and people were like, 1026 00:42:40,880 --> 00:42:43,560 oh, this is an ethical, I would never use AI, 1027 00:42:43,560 --> 00:42:45,200 like they scrape data and stuff. 1028 00:42:45,200 --> 00:42:47,280 And I was just saying, if you would really care 1029 00:42:47,280 --> 00:42:49,880 about ethics, none of you would carry iPhone 1030 00:42:49,880 --> 00:42:52,440 and nobody would shop at H&M. 1031 00:42:52,440 --> 00:42:56,320 And I worked with GDPR and then it was such a big thing, 1032 00:42:56,320 --> 00:42:58,920 you know, in the EU, like everyone gets power 1033 00:42:58,920 --> 00:43:00,520 over your privacy. 1034 00:43:00,520 --> 00:43:02,360 And then basically the feedback was like, 1035 00:43:02,360 --> 00:43:06,000 the cookies that I have to click and accept is annoying. 1036 00:43:06,000 --> 00:43:09,560 And I wonder sometimes the amount of people 1037 00:43:09,560 --> 00:43:12,440 who truly, truly care about this 1038 00:43:12,440 --> 00:43:15,840 and at what step they are willing to give it away. 1039 00:43:15,840 --> 00:43:17,880 If you say that, okay, you will never ever 1040 00:43:17,880 --> 00:43:21,840 will be able to use AI unless you give all your privacy 1041 00:43:21,840 --> 00:43:24,160 to us, I wonder what will be, 1042 00:43:24,160 --> 00:43:26,560 maybe it's a split in society, but yeah, 1043 00:43:26,560 --> 00:43:30,280 but also maybe our private thoughts, private comments 1044 00:43:30,280 --> 00:43:33,240 are not that unique when it's taken at scale. 1045 00:43:33,240 --> 00:43:34,200 Did it for me. 1046 00:43:34,200 --> 00:43:36,200 You know how you think that your iPhone 1047 00:43:36,200 --> 00:43:38,000 might be like listening to you 1048 00:43:38,000 --> 00:43:39,440 or your phone might be listening to you 1049 00:43:39,440 --> 00:43:41,760 because you had a thought about something 1050 00:43:41,760 --> 00:43:43,760 and maybe you told it to your spouse 1051 00:43:43,760 --> 00:43:45,240 and then you get an ad for that? 1052 00:43:45,240 --> 00:43:48,120 That's not really it listening in most instances. 1053 00:43:48,120 --> 00:43:50,400 That is you are a variable 1054 00:43:50,400 --> 00:43:53,080 in these incredibly complex algorithms 1055 00:43:53,080 --> 00:43:56,120 that have you so pegged that it knows, 1056 00:43:56,120 --> 00:44:01,120 oh, your wife liked a engagement ring ad three weeks ago. 1057 00:44:01,600 --> 00:44:04,320 So now every sixth ad you're gonna get 1058 00:44:04,320 --> 00:44:06,200 because you're in the same family group, 1059 00:44:06,200 --> 00:44:08,680 boom, you're gonna start getting that kind of target. 1060 00:44:08,680 --> 00:44:11,000 It's not because you even had to have a conversation 1061 00:44:11,000 --> 00:44:13,880 about it, like these kinds of connections are there. 1062 00:44:13,880 --> 00:44:15,560 And I think you're absolutely right. 1063 00:44:15,560 --> 00:44:20,560 I think I don't foresee us as not of recursing, 1064 00:44:21,120 --> 00:44:21,960 of coming back. 1065 00:44:21,960 --> 00:44:23,720 We've already given away the farm. 1066 00:44:23,720 --> 00:44:25,480 It's okay, we're here. 1067 00:44:25,480 --> 00:44:28,960 Maybe we can draw some lines a little bit better. 1068 00:44:28,960 --> 00:44:32,840 And just to this Lego example on intellectual property, 1069 00:44:32,840 --> 00:44:35,760 I loved it because it's exactly experience 1070 00:44:35,760 --> 00:44:38,720 I had in architecture school where a teacher asked us 1071 00:44:38,720 --> 00:44:40,960 to make a hundred drawings and ideas. 1072 00:44:40,960 --> 00:44:42,960 And you know, as a creative person, 1073 00:44:42,960 --> 00:44:44,920 I'm so creative, I'm so unique. 1074 00:44:44,920 --> 00:44:49,320 And then we all exhibited and 90% we are all having 1075 00:44:49,320 --> 00:44:52,160 exactly the same ideas, the same images. 1076 00:44:52,160 --> 00:44:54,400 And maybe there is out of hundred, I would say, 1077 00:44:54,400 --> 00:44:57,360 okay, maybe two, three where it really stood out. 1078 00:44:57,360 --> 00:44:59,680 And I feel it's also comes down to copyright 1079 00:44:59,680 --> 00:45:02,560 when we can talk about when you are creating something, 1080 00:45:02,560 --> 00:45:04,960 how much you're creating that you closed your eyes 1081 00:45:04,960 --> 00:45:07,320 and never seen world around you 1082 00:45:07,320 --> 00:45:10,680 and how much it is an influence from everything around you 1083 00:45:10,680 --> 00:45:12,560 all the way back to what Wes is saying 1084 00:45:12,560 --> 00:45:15,320 that you are just variable in your preferences, 1085 00:45:15,320 --> 00:45:20,000 your likability is already determined by these algorithms. 1086 00:45:20,000 --> 00:45:22,920 We know you better than your spouse, literally. 1087 00:45:22,920 --> 00:45:26,880 No, I actually, I agree with basically everything 1088 00:45:26,880 --> 00:45:27,800 both of you said. 1089 00:45:27,800 --> 00:45:32,320 And I think that when you start becoming really dogmatic 1090 00:45:32,320 --> 00:45:36,280 about concepts like transparency, privacy, 1091 00:45:36,280 --> 00:45:38,760 accountability, all of those things, 1092 00:45:38,760 --> 00:45:41,560 I think what happens is you start to lose the bigger picture 1093 00:45:41,560 --> 00:45:45,320 of why we even have technology in this world. 1094 00:45:45,320 --> 00:45:48,960 And one of the most interesting sets of conversations 1095 00:45:48,960 --> 00:45:51,960 that I tend to have are when I speak to people, for example, 1096 00:45:51,960 --> 00:45:54,000 that are situated in the EU, 1097 00:45:54,000 --> 00:45:57,920 there's a very specific type of economic situation there. 1098 00:45:57,920 --> 00:46:02,160 They're very human rights focused as a society. 1099 00:46:02,160 --> 00:46:05,160 Things like privacy are ingrained in the culture. 1100 00:46:05,160 --> 00:46:08,160 Everyone knows about privacy, everyone's heard of it. 1101 00:46:08,160 --> 00:46:09,800 It's not new information. 1102 00:46:09,800 --> 00:46:12,000 And then I have a conversation about technology 1103 00:46:12,000 --> 00:46:14,520 with people from different parts of the world, 1104 00:46:14,520 --> 00:46:18,080 maybe Sub-Saharan Africa as an example, 1105 00:46:18,080 --> 00:46:21,080 or some of the Southeast Asia countries 1106 00:46:21,080 --> 00:46:23,960 that have very different economic situations 1107 00:46:23,960 --> 00:46:27,280 are really striving to build their economy. 1108 00:46:27,280 --> 00:46:30,160 And when you start to talk about what technology means 1109 00:46:30,160 --> 00:46:33,800 for their society and how you balance that against concepts 1110 00:46:33,800 --> 00:46:38,800 like security and privacy and safety and things like that, 1111 00:46:38,800 --> 00:46:42,000 it's not that things like privacy transparency 1112 00:46:42,000 --> 00:46:43,760 and all of those things, it's not that they go away, 1113 00:46:43,760 --> 00:46:45,400 they're still really quite important. 1114 00:46:45,400 --> 00:46:49,280 And it's important to have, I think, an ethical foundation 1115 00:46:49,280 --> 00:46:51,920 in whatever it is that you're building and doing. 1116 00:46:51,920 --> 00:46:53,960 But the balance starts to sound a little bit different 1117 00:46:53,960 --> 00:46:55,880 as to what's important for people. 1118 00:46:55,880 --> 00:46:59,240 And what I have trouble with in my field 1119 00:46:59,240 --> 00:47:03,280 is that dogmatic view of these types of, 1120 00:47:03,280 --> 00:47:06,520 we'll call them responsible principles or ethical principles. 1121 00:47:06,520 --> 00:47:08,760 And I think instead, what's important for us 1122 00:47:08,760 --> 00:47:13,160 is to go back to, okay, what is it that we're trying to do? 1123 00:47:13,160 --> 00:47:14,680 What are we trying to accomplish 1124 00:47:14,680 --> 00:47:16,160 with whatever it is that we're building? 1125 00:47:16,160 --> 00:47:18,760 What's the impact we want to have on the world? 1126 00:47:18,760 --> 00:47:21,600 What positive things will that build? 1127 00:47:21,600 --> 00:47:23,600 Will it build value and how? 1128 00:47:23,600 --> 00:47:26,080 And let's just take a few minutes to also think 1129 00:47:26,080 --> 00:47:28,560 of how we might destroy value along the way, 1130 00:47:28,560 --> 00:47:31,400 maybe by accident, maybe in ways that we didn't anticipate. 1131 00:47:31,400 --> 00:47:32,720 And let's try to not do that 1132 00:47:32,720 --> 00:47:34,640 because we don't wanna destroy value, 1133 00:47:34,640 --> 00:47:36,320 we're sure to create value. 1134 00:47:36,320 --> 00:47:40,280 And so I think that all these questions about privacy 1135 00:47:40,280 --> 00:47:44,960 and are you a variable or does someone really know you better 1136 00:47:44,960 --> 00:47:47,280 than, sorry, does an algorithm really know you better 1137 00:47:47,280 --> 00:47:51,800 than someone in your family or in your group of friends? 1138 00:47:51,800 --> 00:47:54,440 I think it has a lot less to do 1139 00:47:54,440 --> 00:47:56,600 with the technical dogmatic questions 1140 00:47:56,600 --> 00:47:59,000 and more to do with people's comfort 1141 00:47:59,000 --> 00:48:01,400 and understanding of technology 1142 00:48:01,400 --> 00:48:03,400 and what role it plays in their lives. 1143 00:48:03,400 --> 00:48:06,440 And so I think that we have to think about 1144 00:48:06,440 --> 00:48:09,520 that kind of storytelling side of things. 1145 00:48:09,520 --> 00:48:11,280 That's why I always go back to business model. 1146 00:48:11,280 --> 00:48:12,600 What is your business model? 1147 00:48:12,600 --> 00:48:14,520 What is it that you're trying to create 1148 00:48:14,520 --> 00:48:18,680 and really sell with whatever it is that you're building? 1149 00:48:18,680 --> 00:48:21,360 If that business model is going to make people 1150 00:48:21,360 --> 00:48:24,760 feel uncomfortable with how you're using technology 1151 00:48:24,760 --> 00:48:27,160 and how they might fit into the equation, 1152 00:48:27,160 --> 00:48:29,760 it might actually be perfectly fine. 1153 00:48:29,760 --> 00:48:32,680 And yet still people will critique it 1154 00:48:32,680 --> 00:48:35,760 and try to tear it down and try to break it apart, 1155 00:48:35,760 --> 00:48:39,480 even if it's objectively okay and fine. 1156 00:48:39,480 --> 00:48:41,160 I don't know, that's my view on it. 1157 00:48:41,160 --> 00:48:44,200 Stay away from the dogma and take some time 1158 00:48:44,200 --> 00:48:46,280 from time to time to look at that big picture 1159 00:48:46,280 --> 00:48:48,080 and ask the question of what is it 1160 00:48:48,080 --> 00:48:50,920 that we're actually trying to do here and are we doing it? 1161 00:48:50,920 --> 00:48:51,760 I love this. 1162 00:48:51,760 --> 00:48:54,000 So either you spend time now, 1163 00:48:54,000 --> 00:48:56,480 invest your time and effort and energy, 1164 00:48:56,480 --> 00:48:59,880 or you're going to pay a lot for PR and rebranding 1165 00:48:59,880 --> 00:49:03,920 eventually down the line when everyone starts to come at you 1166 00:49:03,920 --> 00:49:06,680 because eventually something will happen 1167 00:49:06,680 --> 00:49:08,480 destroying the image of the company. 1168 00:49:08,480 --> 00:49:11,480 We could talk to Mapo for hours more, 1169 00:49:11,480 --> 00:49:14,320 but want to give her back her Sunday afternoon here 1170 00:49:14,320 --> 00:49:15,400 while we record. 1171 00:49:15,400 --> 00:49:18,840 So aside from the stages of any AI conference 1172 00:49:18,840 --> 00:49:20,640 where they can surely see Mapo, 1173 00:49:20,640 --> 00:49:23,560 she spoke at AI4, South by Southwest, 1174 00:49:23,560 --> 00:49:25,480 a couple of other big ones right there, 1175 00:49:25,480 --> 00:49:28,080 where else can people find what you got going on? 1176 00:49:28,080 --> 00:49:31,040 Please feel free to reach out on LinkedIn. 1177 00:49:31,040 --> 00:49:33,840 I definitely keep track of that account. 1178 00:49:33,840 --> 00:49:37,000 And I also have a website called Call Me Mapo. 1179 00:49:37,000 --> 00:49:38,480 Beautiful. 1180 00:49:38,480 --> 00:49:41,120 And I don't know if it's appropriate to say, 1181 00:49:41,120 --> 00:49:43,880 but I would love to have you on my YouTube channel. 1182 00:49:43,880 --> 00:49:44,800 There you go. 1183 00:49:44,800 --> 00:49:47,520 This is open in the air invitation. 1184 00:49:47,520 --> 00:49:48,360 You? 1185 00:49:48,360 --> 00:49:49,200 Look at that. 1186 00:49:49,200 --> 00:49:50,040 Yeah, happy to join. 1187 00:49:50,040 --> 00:49:51,600 That'd be awesome. 1188 00:49:51,600 --> 00:49:52,440 Okay, so that's another plug. 1189 00:49:52,440 --> 00:49:54,000 You don't want a bunch of tech bros like me 1190 00:49:54,000 --> 00:49:55,360 to decide in the future. 1191 00:49:55,360 --> 00:49:58,440 Come on, we need to get the good feminine voices 1192 00:49:58,440 --> 00:50:00,000 out there to steer us right. 1193 00:50:00,000 --> 00:50:02,400 I don't make any decisions in my family. 1194 00:50:02,400 --> 00:50:04,200 I just listen to the real boss. 1195 00:50:04,200 --> 00:50:07,560 So with that, for Go2Go, I am Wes Shields. 1196 00:50:07,560 --> 00:50:09,760 And for Mapo, our wonderful guest, 1197 00:50:09,760 --> 00:50:11,720 saying happy prompting, everybody. 1198 00:50:11,720 --> 00:50:13,600 Happy prompting, everybody. 1199 00:50:13,600 --> 00:50:15,360 And now Mapo, you do too. 1200 00:50:15,360 --> 00:50:16,880 Happy prompting. 1201 00:50:16,880 --> 00:50:17,720 Amazing. 1202 00:50:17,720 --> 00:50:26,720 As always, you can check out the show notes and links 1203 00:50:26,720 --> 00:50:30,400 at howtotalk2.ai. 1204 00:50:30,400 --> 00:50:32,600 That's all for this week's episode. 1205 00:50:32,600 --> 00:50:51,600 Happy prompting, everyone.