1 00:00:00,500 --> 00:00:02,976 [SQUEAKING] 2 00:00:02,976 --> 00:00:04,464 [RUSTLING] 3 00:00:04,464 --> 00:00:05,952 [CLICKING] 4 00:00:10,657 --> 00:00:12,490 GARY GENSLER: I'm going to talk a little bit 5 00:00:12,490 --> 00:00:16,239 about how I see financial technology in a stack. 6 00:00:16,239 --> 00:00:19,210 We did a little bit of this in our last two classes, 7 00:00:19,210 --> 00:00:23,890 but just really thinking about the finance technology stack 8 00:00:23,890 --> 00:00:26,770 and then turn back into AI and machine learning. 9 00:00:26,770 --> 00:00:29,620 In finance, we reviewed this a bit in the last class, 10 00:00:29,620 --> 00:00:32,560 but I just want to go back to it and talk a little bit more 11 00:00:32,560 --> 00:00:37,000 about it, a little bit more granular this time around. 12 00:00:37,000 --> 00:00:42,160 And then take the bulk of the class around public policy 13 00:00:42,160 --> 00:00:45,520 frameworks and how AI fits into that, 14 00:00:45,520 --> 00:00:47,660 Artificial Intelligence, machine learning, chat 15 00:00:47,660 --> 00:00:48,800 box, and the like. 16 00:00:48,800 --> 00:00:50,740 So that's sort of the run of show. 17 00:00:50,740 --> 00:00:52,780 And Romain, you let me know when we have either 18 00:00:52,780 --> 00:00:55,930 15 or 10 minutes to go because I don't have a clock here 19 00:00:55,930 --> 00:00:59,110 on this Zoom. 20 00:00:59,110 --> 00:01:03,460 So the three readings really built upon last class's 21 00:01:03,460 --> 00:01:07,180 readings, but they had a little bit 22 00:01:07,180 --> 00:01:10,990 of a tone towards the regulatory side. 23 00:01:10,990 --> 00:01:15,700 And Oliver Wyman and the Mayer Brown-- 24 00:01:15,700 --> 00:01:17,140 Mayer Brown is a law firm. 25 00:01:17,140 --> 00:01:21,880 Oliver Wyman thinks about risk management and consultant risk. 26 00:01:21,880 --> 00:01:24,260 But each were with a little bit of a gloss 27 00:01:24,260 --> 00:01:28,040 on how to manage if you're at a board level. 28 00:01:28,040 --> 00:01:32,940 And the Oliver Wyman really went through the lines of business, 29 00:01:32,940 --> 00:01:35,490 how machine learning and artificial intelligence 30 00:01:35,490 --> 00:01:37,050 is being used. 31 00:01:37,050 --> 00:01:39,690 And the Mayer Brown went through sort of the laws. 32 00:01:39,690 --> 00:01:41,070 And we'll come back to that. 33 00:01:41,070 --> 00:01:44,730 But that's why I sort of reached out and had that. 34 00:01:44,730 --> 00:01:48,270 Now, the short 1- or 2-pager from Julie Stack from 35 00:01:48,270 --> 00:01:51,460 the Federal Reserve, I thought it was just interesting to say, 36 00:01:51,460 --> 00:01:53,220 all right, here's a senior-- 37 00:01:53,220 --> 00:01:55,950 and Julie's very well-respected in the community-- 38 00:01:55,950 --> 00:01:58,890 a senior career person at the US Federal 39 00:01:58,890 --> 00:02:00,660 Reserve writing about fintech. 40 00:02:00,660 --> 00:02:02,070 What are they thinking? 41 00:02:02,070 --> 00:02:05,880 And just like I did earlier, and we shared sometimes, 42 00:02:05,880 --> 00:02:09,600 what is the official sector thinking, as we shared 43 00:02:09,600 --> 00:02:14,160 the chair of the FDIC's speech earlier, 44 00:02:14,160 --> 00:02:16,410 and I will throughout this semester, 45 00:02:16,410 --> 00:02:18,150 I think it's helpful if you're thinking 46 00:02:18,150 --> 00:02:20,370 about this to sometimes think, all right, 47 00:02:20,370 --> 00:02:22,890 what's the official sector saying in their speeches 48 00:02:22,890 --> 00:02:25,450 and so forth? 49 00:02:25,450 --> 00:02:28,930 But that was kind of why I put those readings out there. 50 00:02:28,930 --> 00:02:32,280 I'm looking at Romain to see if there's anything, but probably 51 00:02:32,280 --> 00:02:35,050 not yet. 52 00:02:35,050 --> 00:02:36,758 And then the study questions. 53 00:02:36,758 --> 00:02:38,550 This is where it gets a little bit more fun 54 00:02:38,550 --> 00:02:40,910 and I see if we can get a little engagement. 55 00:02:40,910 --> 00:02:45,540 And again, we spoke a great deal about this already. 56 00:02:45,540 --> 00:02:48,885 But if anybody just really quickly want to give 57 00:02:48,885 --> 00:02:51,060 an articulation of this question-- 58 00:02:51,060 --> 00:02:55,920 why are these new forms of AI enabled data analytics, 59 00:02:55,920 --> 00:02:59,040 pattern recognition, speech recognition, and so forth-- 60 00:03:01,890 --> 00:03:05,160 how do they fit into the other trends, as you see it, 61 00:03:05,160 --> 00:03:08,080 as we've talked about in financial technology? 62 00:03:08,080 --> 00:03:08,970 GUEST SPEAKER: Luke? 63 00:03:08,970 --> 00:03:10,428 GARY GENSLER: This is your fun time 64 00:03:10,428 --> 00:03:13,500 to either call on people as they've volunteered 65 00:03:13,500 --> 00:03:14,130 or otherwise. 66 00:03:14,130 --> 00:03:15,400 But we've discussed this. 67 00:03:15,400 --> 00:03:17,527 I'm just trying to get it a little going here. 68 00:03:17,527 --> 00:03:19,110 GUEST SPEAKER: We can start with Luke. 69 00:03:19,110 --> 00:03:22,890 AUDIENCE: So the commonality among the industries, so 70 00:03:22,890 --> 00:03:27,990 sector agnostically, is the fact that all the companies who 71 00:03:27,990 --> 00:03:32,520 can deploy this AI to their operation system 72 00:03:32,520 --> 00:03:38,400 is to save money, save costs, so that the bottom line is better. 73 00:03:38,400 --> 00:03:42,180 GARY GENSLER: So one thing that Luke's raising is saving costs. 74 00:03:42,180 --> 00:03:45,240 Others want to sort of chime in. 75 00:03:45,240 --> 00:03:48,270 And I'm kind of also curious as others 76 00:03:48,270 --> 00:03:52,410 chime in how you see it fitting in with other emerging trends. 77 00:03:52,410 --> 00:03:55,390 We had the trends that we've already lived through, 78 00:03:55,390 --> 00:03:58,140 but we're building upon, like the internet and mobile 79 00:03:58,140 --> 00:03:59,340 and cloud. 80 00:03:59,340 --> 00:04:01,110 We have some other trends that we'll 81 00:04:01,110 --> 00:04:04,830 get to in future classes like open API. 82 00:04:04,830 --> 00:04:08,220 Just kind of curious to see if somebody 83 00:04:08,220 --> 00:04:10,630 wants to take a crack at connecting 84 00:04:10,630 --> 00:04:12,900 this piece of the technology with some 85 00:04:12,900 --> 00:04:14,390 of these other trends. 86 00:04:14,390 --> 00:04:15,902 GUEST SPEAKER: Laira? 87 00:04:15,902 --> 00:04:18,110 AUDIENCE: Yeah, I think what we discussed extensively 88 00:04:18,110 --> 00:04:22,400 in the last lecture was Erica being 89 00:04:22,400 --> 00:04:27,560 one of really good examples of how the new forms of AI 90 00:04:27,560 --> 00:04:29,570 is emerging with what's already existing 91 00:04:29,570 --> 00:04:31,970 and making it not just cheaper for the firms 92 00:04:31,970 --> 00:04:35,030 to answer mundane questions that customers have, 93 00:04:35,030 --> 00:04:38,040 but also making it more user friendly. 94 00:04:38,040 --> 00:04:39,560 So I think in terms of Erica, it's 95 00:04:39,560 --> 00:04:44,110 just a great example to show how this question kind of goes 96 00:04:44,110 --> 00:04:45,530 through. 97 00:04:45,530 --> 00:04:49,050 GARY GENSLER: So Laira's just raising-- 98 00:04:49,050 --> 00:04:52,010 and sometimes I just break these down simplistically. 99 00:04:52,010 --> 00:04:54,120 In the artificial intelligence world, 100 00:04:54,120 --> 00:04:57,260 there's the consumer-customer interface. 101 00:04:57,260 --> 00:04:59,240 Erica at Bank America is an example, 102 00:04:59,240 --> 00:05:02,330 and chat bots, and the various ways 103 00:05:02,330 --> 00:05:05,450 we communicate with customers, tying into customers, 104 00:05:05,450 --> 00:05:08,900 builds what technology? 105 00:05:08,900 --> 00:05:11,810 What is it that you use and might even 106 00:05:11,810 --> 00:05:15,430 be using right now when you're watching this course? 107 00:05:18,522 --> 00:05:20,480 GUEST SPEAKER: Eric, you also had your hand up. 108 00:05:23,440 --> 00:05:26,050 AUDIENCE: I was going to talk about something else. 109 00:05:26,050 --> 00:05:27,378 I was going to say that-- 110 00:05:27,378 --> 00:05:28,170 GARY GENSLER: Sure. 111 00:05:28,170 --> 00:05:30,140 [INAUDIBLE] and I'll tie them all together. 112 00:05:30,140 --> 00:05:30,720 Don't worry. 113 00:05:30,720 --> 00:05:33,645 I'll answer [INAUDIBLE] questions, too. 114 00:05:33,645 --> 00:05:34,270 AUDIENCE: Sure. 115 00:05:34,270 --> 00:05:39,690 I was saying that AI is being used by fintechs for better 116 00:05:39,690 --> 00:05:41,700 underwriting purposes, like using 117 00:05:41,700 --> 00:05:47,398 alternative data to better assess people's credit. 118 00:05:47,398 --> 00:05:48,440 GARY GENSLER: Absolutely. 119 00:05:48,440 --> 00:05:50,480 So it's the data analytics. 120 00:05:50,480 --> 00:05:51,920 It's the customer interface. 121 00:05:51,920 --> 00:05:55,760 That data analytics of predictive underwriting, 122 00:05:55,760 --> 00:05:57,680 whether it's in insurance, whether it's 123 00:05:57,680 --> 00:06:00,560 in lending, predictive underwriting. 124 00:06:00,560 --> 00:06:02,510 It's also on the customer side where 125 00:06:02,510 --> 00:06:04,370 we use natural language processing 126 00:06:04,370 --> 00:06:06,660 and we interface with the customers. 127 00:06:06,660 --> 00:06:08,810 Romain why don't we take one or two more? 128 00:06:08,810 --> 00:06:10,400 I'm looking for somebody who wants 129 00:06:10,400 --> 00:06:13,700 to tie it to the other technological trends 130 00:06:13,700 --> 00:06:14,520 that we see. 131 00:06:14,520 --> 00:06:18,310 GUEST SPEAKER: So let's go with Nikhil and then with Wei. 132 00:06:18,310 --> 00:06:21,040 GARY GENSLER: All right, and then we'll move on. 133 00:06:21,040 --> 00:06:23,680 AUDIENCE: I think the Oliver Wyman reading talks 134 00:06:23,680 --> 00:06:27,487 about how companies that have been using AI and machine 135 00:06:27,487 --> 00:06:28,570 learning have done better. 136 00:06:28,570 --> 00:06:30,028 I think it was asset management was 137 00:06:30,028 --> 00:06:31,630 a specific example they took. 138 00:06:31,630 --> 00:06:35,230 I think it also ties to, like another class I'm 139 00:06:35,230 --> 00:06:37,360 taking with Simon Johnson on AI talks 140 00:06:37,360 --> 00:06:40,720 about David Autor's report that says there's a superstar 141 00:06:40,720 --> 00:06:43,730 effect where firms that have access to this data 142 00:06:43,730 --> 00:06:46,780 and are using AI tend to perform better in the market. 143 00:06:46,780 --> 00:06:48,910 And I think that's a significant tie-in. 144 00:06:48,910 --> 00:06:50,620 And it's probably even more exaggerated 145 00:06:50,620 --> 00:06:52,670 in fintech specifically. 146 00:06:52,670 --> 00:06:54,790 GARY GENSLER: So let's just pause for a second. 147 00:06:54,790 --> 00:06:56,410 It's data, data. 148 00:06:56,410 --> 00:07:00,220 What we've had is this remarkable advancement 149 00:07:00,220 --> 00:07:03,740 in data analytic tools, artificial intelligence. 150 00:07:03,740 --> 00:07:06,610 But we've also had a remarkable advancement of the ability 151 00:07:06,610 --> 00:07:11,140 to store and process data through the cloud 152 00:07:11,140 --> 00:07:15,190 and just through the emergence of much faster computers 153 00:07:15,190 --> 00:07:18,550 and much more connected communications. 154 00:07:18,550 --> 00:07:22,450 So that data piece, the artificial intelligence 155 00:07:22,450 --> 00:07:24,545 and machine learning trend might not 156 00:07:24,545 --> 00:07:26,920 have been able to do as well if it weren't for everything 157 00:07:26,920 --> 00:07:29,240 that's going on, broadly speaking, 158 00:07:29,240 --> 00:07:32,050 whether it's in cloud or computing. 159 00:07:32,050 --> 00:07:35,324 And Romain the last person that was-- 160 00:07:35,324 --> 00:07:36,260 AUDIENCE: Yep. 161 00:07:36,260 --> 00:07:40,960 So I also want to maybe make a mention that in also help 162 00:07:40,960 --> 00:07:44,340 was a lot of times either collecting data analytics 163 00:07:44,340 --> 00:07:46,300 or cleaning the data analytics. 164 00:07:46,300 --> 00:07:51,100 Because a lot of time that in the old world 165 00:07:51,100 --> 00:07:54,100 there's a lot of data you potentially collect. 166 00:07:54,100 --> 00:07:56,770 First of all, I can help to better collect 167 00:07:56,770 --> 00:07:58,090 unstructured data. 168 00:07:58,090 --> 00:08:00,520 And the second found that it helps to clean 169 00:08:00,520 --> 00:08:03,340 a lot of data you collected. 170 00:08:03,340 --> 00:08:05,650 GARY GENSLER: So absolutely agreed. 171 00:08:05,650 --> 00:08:11,080 And often it's 80%, sometimes 90% more of a computer science 172 00:08:11,080 --> 00:08:15,630 group is in the cleaning up of data and standardizing data. 173 00:08:15,630 --> 00:08:18,100 And we'll come back to this, but a lot 174 00:08:18,100 --> 00:08:21,550 of fintech disruptors, a lot of startups 175 00:08:21,550 --> 00:08:27,580 have actually created value more around data than anything else. 176 00:08:27,580 --> 00:08:32,530 And I will say not just about data, but standardizing data. 177 00:08:32,530 --> 00:08:35,309 And later in this class, we're going to talk about Plaid 178 00:08:35,309 --> 00:08:39,070 and Credit Karma, both of which were earlier this year 179 00:08:39,070 --> 00:08:42,520 acquired, Plaid by Visa, Credit Karma 180 00:08:42,520 --> 00:08:48,730 by Intuit for $5 to $7 billion-- big, big acquisitions. 181 00:08:48,730 --> 00:08:53,340 And we're going to talk about what was the value proposition 182 00:08:53,340 --> 00:08:54,810 for Visa and Intuit? 183 00:08:54,810 --> 00:08:57,240 Why were they paying $5 or $7 billion? 184 00:08:57,240 --> 00:08:58,710 A lot of it-- 185 00:08:58,710 --> 00:09:01,680 not all of it, but a lot of it relates to data, 186 00:09:01,680 --> 00:09:06,750 but also having standardized that data, particularly 187 00:09:06,750 --> 00:09:07,740 in the case of Plaid. 188 00:09:12,530 --> 00:09:15,420 How it's affecting the competitive landscape. 189 00:09:15,420 --> 00:09:16,570 We've talked a great deal. 190 00:09:16,570 --> 00:09:18,410 Hopefully this will continue to be 191 00:09:18,410 --> 00:09:23,360 a theme throughout the semester about big incumbents, big tech, 192 00:09:23,360 --> 00:09:25,550 and fintech startups. 193 00:09:25,550 --> 00:09:29,510 I will contend in this and throughout this course 194 00:09:29,510 --> 00:09:33,590 that AI and machine learning is now moving into the technology 195 00:09:33,590 --> 00:09:35,250 stack. 196 00:09:35,250 --> 00:09:38,700 If we think of this stack as layers of technology 197 00:09:38,700 --> 00:09:43,680 that incumbents incorporate, and frankly will not 198 00:09:43,680 --> 00:09:47,030 survive if they don't incorporate, 199 00:09:47,030 --> 00:09:50,550 that AI and machine learning is being incorporated quickly 200 00:09:50,550 --> 00:09:53,910 into the financial incumbent technology stack. 201 00:09:53,910 --> 00:09:55,150 We're not fully there yet. 202 00:09:57,940 --> 00:10:00,640 And that the competitive landscape is such 203 00:10:00,640 --> 00:10:03,580 that the fintech startups and disruptors have 204 00:10:03,580 --> 00:10:07,150 been able to find cracks in the old business models. 205 00:10:07,150 --> 00:10:09,720 And using machine learning, they've been able to break in. 206 00:10:09,720 --> 00:10:11,620 And we'll talk a bit about that. 207 00:10:11,620 --> 00:10:13,240 The big tech firms, of course we've 208 00:10:13,240 --> 00:10:17,020 already talked about that they are really about networks. 209 00:10:17,020 --> 00:10:20,540 Networks that they then layer more activities upon, 210 00:10:20,540 --> 00:10:22,660 and those more activities bring them more data. 211 00:10:26,780 --> 00:10:29,360 And then we're going to talk a fair amount 212 00:10:29,360 --> 00:10:30,530 about public policy. 213 00:10:30,530 --> 00:10:34,370 But anybody who's sort of dug into the Mayer Brown reading 214 00:10:34,370 --> 00:10:38,090 want to just give two or three thoughts on the broad-- 215 00:10:38,090 --> 00:10:39,240 what are the-- 216 00:10:39,240 --> 00:10:41,840 I'll later call them the big three? 217 00:10:41,840 --> 00:10:45,930 But it's almost written right in the question for you, 218 00:10:45,930 --> 00:10:48,311 but Romain, you want to call anybody? 219 00:10:51,072 --> 00:10:52,030 GUEST SPEAKER: Michael? 220 00:10:56,530 --> 00:10:59,050 AUDIENCE: Yeah, so the reading kind of 221 00:10:59,050 --> 00:11:04,970 did touch upon bias a lot and its potential, just 222 00:11:04,970 --> 00:11:09,180 on the natural factors that a machine learning 223 00:11:09,180 --> 00:11:11,460 algorithm would trace. 224 00:11:11,460 --> 00:11:14,760 GARY GENSLER: So one of the things about machine learning 225 00:11:14,760 --> 00:11:17,160 and deep learning is that it's remarkably 226 00:11:17,160 --> 00:11:21,330 successful at extracting correlations. 227 00:11:21,330 --> 00:11:24,960 Correlations from data sometimes that we didn't see before, 228 00:11:24,960 --> 00:11:29,870 that didn't come just from a linear relationship, 229 00:11:29,870 --> 00:11:32,940 a linear relationship that we might be able to identify just 230 00:11:32,940 --> 00:11:36,210 in classical statistics. 231 00:11:36,210 --> 00:11:38,910 But in those remarkable abilities 232 00:11:38,910 --> 00:11:42,300 to extract correlations, you might see biases. 233 00:11:42,300 --> 00:11:45,930 If the data itself has a bias in it that people 234 00:11:45,930 --> 00:11:49,050 of certain gender, certain race, certain ethnic backgrounds, 235 00:11:49,050 --> 00:11:56,160 certain geographies are more likely to, in the data's mind-- 236 00:11:56,160 --> 00:11:57,560 in the data's mind-- 237 00:11:57,560 --> 00:11:59,760 are more likely to have lower income 238 00:11:59,760 --> 00:12:04,380 and in the data might have more likely to be a lower credit 239 00:12:04,380 --> 00:12:07,440 quality, then you might be embedding 240 00:12:07,440 --> 00:12:11,070 certain biases inside the data. 241 00:12:11,070 --> 00:12:14,540 And many nations around the globe, not just the US, 242 00:12:14,540 --> 00:12:17,640 have said to the credit card companies 243 00:12:17,640 --> 00:12:21,540 and the other financial firms that you shouldn't have biases 244 00:12:21,540 --> 00:12:25,130 around race, gender, ethnic background, geography, 245 00:12:25,130 --> 00:12:26,820 sometimes, and the like. 246 00:12:26,820 --> 00:12:28,980 So one is biases. 247 00:12:28,980 --> 00:12:32,340 When I consider the three big buckets 248 00:12:32,340 --> 00:12:35,460 here-- anybody want to just talk about the other two? 249 00:12:35,460 --> 00:12:36,520 Romain? 250 00:12:36,520 --> 00:12:38,218 GUEST SPEAKER: Alicia. 251 00:12:38,218 --> 00:12:38,760 AUDIENCE: Hi. 252 00:12:38,760 --> 00:12:40,710 I think we talked this last class. 253 00:12:40,710 --> 00:12:45,180 I think AI derives conclusions or correlations 254 00:12:45,180 --> 00:12:46,860 without explaining the why. 255 00:12:46,860 --> 00:12:51,180 So humans cannot understand why some guy has a better credit 256 00:12:51,180 --> 00:12:54,768 rating than another and has an issue with the law, basically. 257 00:12:54,768 --> 00:12:55,560 GARY GENSLER: Yeah. 258 00:12:55,560 --> 00:13:01,130 And why as societies have we embedded in laws-- 259 00:13:01,130 --> 00:13:02,425 and we'll talk about this. 260 00:13:02,425 --> 00:13:04,050 But if you have a point of view, why as 261 00:13:04,050 --> 00:13:06,510 societies have we embedded in laws 262 00:13:06,510 --> 00:13:09,210 that you need to be able to explain the why when 263 00:13:09,210 --> 00:13:13,590 you deny somebody credit or deny somebody a financial product? 264 00:13:13,590 --> 00:13:16,650 We did this in the United States 50 years ago 265 00:13:16,650 --> 00:13:22,560 in something called the Fair Credit Reporting Act. 266 00:13:22,560 --> 00:13:27,670 Data analytics was a big wave in the 1960s, believe it or not, 267 00:13:27,670 --> 00:13:31,410 when credit cards were invented in the 1940s and '50s. 268 00:13:31,410 --> 00:13:33,945 By the 1960s, data analytics were going, 269 00:13:33,945 --> 00:13:39,450 and the Fair Isaac Company, which became FICO, had started. 270 00:13:39,450 --> 00:13:42,570 And we embedded in law that you had to answer this question. 271 00:13:42,570 --> 00:13:44,700 Explain why you denied credit. 272 00:13:44,700 --> 00:13:47,310 But why do you think we embed that in country 273 00:13:47,310 --> 00:13:49,720 after country in our laws? 274 00:13:49,720 --> 00:13:52,580 GUEST SPEAKER: Danielle? 275 00:13:52,580 --> 00:13:56,150 AUDIENCE: So I think it's, going back to the bias question, 276 00:13:56,150 --> 00:14:00,080 to prevent bias in people who are extending credit. 277 00:14:00,080 --> 00:14:01,562 GARY GENSLER: I think you're right. 278 00:14:01,562 --> 00:14:03,020 I don't think it's the only reason, 279 00:14:03,020 --> 00:14:04,820 but I think it's a dominant reason. 280 00:14:04,820 --> 00:14:07,100 We also in the US passed something called the Equal 281 00:14:07,100 --> 00:14:10,640 Credit Opportunity Act, or it generally 282 00:14:10,640 --> 00:14:13,560 goes by the terms ECOA. 283 00:14:13,560 --> 00:14:17,180 But those two laws and another law in the US, 284 00:14:17,180 --> 00:14:19,580 Truth in Lending Act for transparency, 285 00:14:19,580 --> 00:14:23,990 were kind of this bedrock out of the 1960s data analytic credit 286 00:14:23,990 --> 00:14:25,370 card boom. 287 00:14:25,370 --> 00:14:29,490 By the early '70s, we had those three. 288 00:14:29,490 --> 00:14:34,280 Anti-bias, fairness, you might say, explainability. 289 00:14:34,280 --> 00:14:38,450 These are two bedrocks in finance in Europe 290 00:14:38,450 --> 00:14:40,580 and the US, country after country. 291 00:14:40,580 --> 00:14:43,910 What's the third challenge that comes up 292 00:14:43,910 --> 00:14:50,270 with data analytics or AI that often we find ourselves, 293 00:14:50,270 --> 00:14:52,250 and if you're starting a fintech startup 294 00:14:52,250 --> 00:14:53,630 you have to be aware of? 295 00:14:56,450 --> 00:14:59,410 Romain, any hands? 296 00:14:59,410 --> 00:15:00,490 GUEST SPEAKER: Not yet. 297 00:15:00,490 --> 00:15:02,170 We have Luke again. 298 00:15:02,170 --> 00:15:06,117 GARY GENSLER: We'll pass on Luke unless somebody else. 299 00:15:06,117 --> 00:15:07,700 GUEST SPEAKER: We have Danielle again. 300 00:15:10,410 --> 00:15:12,240 GARY GENSLER: All right, either one, 301 00:15:12,240 --> 00:15:15,890 whoever's got their mic off. 302 00:15:15,890 --> 00:15:18,280 AUDIENCE: So privacy is the last one. 303 00:15:18,280 --> 00:15:19,360 GARY GENSLER: Sure. 304 00:15:19,360 --> 00:15:22,090 AUDIENCE: For example, companies have demonstrated the ability 305 00:15:22,090 --> 00:15:25,750 to predict when consumers have certain health conditions 306 00:15:25,750 --> 00:15:29,330 or pregnancy, for example. 307 00:15:29,330 --> 00:15:32,470 There is a really famous case where a company knew 308 00:15:32,470 --> 00:15:36,010 that a consumer was pregnant based on how their shopping 309 00:15:36,010 --> 00:15:38,200 patterns changed, and there are reasons 310 00:15:38,200 --> 00:15:41,860 we've precluded employers or credit 311 00:15:41,860 --> 00:15:44,680 extenders from asking about certain parts of people's 312 00:15:44,680 --> 00:15:45,260 lives. 313 00:15:45,260 --> 00:15:48,940 But we may be unexpectedly exposed to parts of those lives 314 00:15:48,940 --> 00:15:51,050 if we're capturing data and using it. 315 00:15:51,050 --> 00:15:53,860 GARY GENSLER: So this trade-off of privacy 316 00:15:53,860 --> 00:15:58,300 versus financial services, thought 317 00:15:58,300 --> 00:16:00,810 it's not as old as sort of the fairness 318 00:16:00,810 --> 00:16:03,675 and the explainability, which in the US 319 00:16:03,675 --> 00:16:05,050 and then later in other countries 320 00:16:05,050 --> 00:16:08,130 was embedded in many laws 30 to 50 years ago, 321 00:16:08,130 --> 00:16:10,980 privacy has picked up a little bit more of a stream. 322 00:16:10,980 --> 00:16:13,240 By the late 1990s in the US, there 323 00:16:13,240 --> 00:16:16,800 was modest financial privacy protections that 324 00:16:16,800 --> 00:16:20,490 were embedded into law in 1999. 325 00:16:20,490 --> 00:16:24,150 I actually helped work on that with then-Congressman Ed 326 00:16:24,150 --> 00:16:29,060 Markey, now Senator Ed Markey of Massachusetts. 327 00:16:29,060 --> 00:16:31,710 But in Europe, they went quite a bit 328 00:16:31,710 --> 00:16:36,090 further in something called the GDPR, which we'll 329 00:16:36,090 --> 00:16:38,680 talk about a little later. 330 00:16:38,680 --> 00:16:43,560 But the General Directive-- 331 00:16:43,560 --> 00:16:47,640 P doesn't stand for privacy, but I think 332 00:16:47,640 --> 00:16:51,010 it's Protection of Regulation. 333 00:16:51,010 --> 00:16:52,800 So those three buckets-- 334 00:16:52,800 --> 00:16:54,870 those three buckets are the important ones. 335 00:16:54,870 --> 00:16:59,340 So again, AI machine learning fits into these other trends 336 00:16:59,340 --> 00:17:00,653 that we think about. 337 00:17:00,653 --> 00:17:02,070 And I'm going to walk through that 338 00:17:02,070 --> 00:17:06,780 in this class of cloud and internet and mobile and data. 339 00:17:06,780 --> 00:17:12,839 Fintech startups, big tech, and incumbents, I believe, 340 00:17:12,839 --> 00:17:16,660 are all embedding it in their technology stack. 341 00:17:16,660 --> 00:17:19,680 And you're really challenged if you don't. 342 00:17:19,680 --> 00:17:22,680 And then the big three challenges in public policy, 343 00:17:22,680 --> 00:17:26,151 explainability, bias, and privacy. 344 00:17:26,151 --> 00:17:27,609 There are other challenges as well, 345 00:17:27,609 --> 00:17:30,430 but those are the big three, in a sense. 346 00:17:30,430 --> 00:17:32,280 So what do I mean by technology stack? 347 00:17:32,280 --> 00:17:34,560 Well, I think that three things are already 348 00:17:34,560 --> 00:17:38,190 embedded, the internet, mobile, and the cloud. 349 00:17:38,190 --> 00:17:41,250 And if this class were being taught at MIT in the 1980s, 350 00:17:41,250 --> 00:17:44,040 none of them would be there, and by the 1990s, 351 00:17:44,040 --> 00:17:47,220 we would have said, wow, that internet. 352 00:17:47,220 --> 00:17:51,480 The word "fintech" didn't really come about in the 1990s. 353 00:17:51,480 --> 00:17:54,300 But if we had applied it to the 1990s, 354 00:17:54,300 --> 00:17:57,840 the internet was dramatically changing. 355 00:17:57,840 --> 00:18:01,590 Mobile into the naughts in the cloud and so forth. 356 00:18:01,590 --> 00:18:06,000 I would contend you cannot really survive in the finance 357 00:18:06,000 --> 00:18:08,940 space giving customers what they need, 358 00:18:08,940 --> 00:18:12,570 whether it's in the wholesale markets of capital markets 359 00:18:12,570 --> 00:18:18,740 and payments, or in the retail markets if you haven't yet 360 00:18:18,740 --> 00:18:21,210 embedded in your technology stack. 361 00:18:21,210 --> 00:18:25,200 Now, I will note that many large financial companies 362 00:18:25,200 --> 00:18:29,110 are slow to use the cloud. 363 00:18:29,110 --> 00:18:32,110 The largest amongst them tend to want to still have 364 00:18:32,110 --> 00:18:34,180 their own data centers. 365 00:18:34,180 --> 00:18:36,430 I think you're going to see that shift dramatically 366 00:18:36,430 --> 00:18:38,440 in the 2020s. 367 00:18:38,440 --> 00:18:42,160 But I'm certainly telling you that if you start a startup, 368 00:18:42,160 --> 00:18:44,710 you cannot survive if you're trying to do your own data 369 00:18:44,710 --> 00:18:47,890 center, if you're going to already embed these 370 00:18:47,890 --> 00:18:51,610 in your what I'll call financial stack. 371 00:18:51,610 --> 00:18:57,550 The internet for connectivity, mobile in a sense for ubiquity, 372 00:18:57,550 --> 00:18:59,740 meaning that folks can be out there. 373 00:18:59,740 --> 00:19:03,730 Cloud, you're sort of renting somebody else's storage 374 00:19:03,730 --> 00:19:06,380 and often their software. 375 00:19:06,380 --> 00:19:10,490 But then the things that we're talking about in this time, 376 00:19:10,490 --> 00:19:13,520 in the 2020s that are being embedded 377 00:19:13,520 --> 00:19:19,370 into the classic standard company stack is AI, machine 378 00:19:19,370 --> 00:19:21,200 learning, and natural language processing, 379 00:19:21,200 --> 00:19:23,300 and what we'll talk about in the next class a lot 380 00:19:23,300 --> 00:19:26,310 about open API. 381 00:19:26,310 --> 00:19:27,660 Now, we're in a transition mode. 382 00:19:27,660 --> 00:19:31,080 Not every company has really embedded it in their stack. 383 00:19:31,080 --> 00:19:33,810 And these are where the opportunities really 384 00:19:33,810 --> 00:19:38,090 existed in the last dozen years in fintech. 385 00:19:38,090 --> 00:19:40,140 Fintech startups that were savvy enough 386 00:19:40,140 --> 00:19:44,220 to really build this into their product offerings 387 00:19:44,220 --> 00:19:48,810 faster than the incumbents, or, better yet, 388 00:19:48,810 --> 00:19:51,103 in a more refined, targeted way. 389 00:19:51,103 --> 00:19:52,770 And we'll talk a fair amount about that. 390 00:19:52,770 --> 00:19:55,290 Now, of course, there's other things in the stack. 391 00:19:55,290 --> 00:19:57,420 And this is not what this class is. 392 00:19:57,420 --> 00:20:00,960 Even money itself in accounting and ledgers and joint stock 393 00:20:00,960 --> 00:20:03,090 companies were all in a sense. 394 00:20:03,090 --> 00:20:05,160 We just take them completely for granted. 395 00:20:05,160 --> 00:20:09,240 By the time you're in a master's program at MIT, 396 00:20:09,240 --> 00:20:12,830 master's of finance or MBA or other graduate program, 397 00:20:12,830 --> 00:20:14,580 you're quite familiar, and you almost just 398 00:20:14,580 --> 00:20:16,080 take these for granted. 399 00:20:16,080 --> 00:20:20,480 But I can assure you at earlier decades, 400 00:20:20,480 --> 00:20:22,040 they couldn't be taken for granted. 401 00:20:22,040 --> 00:20:25,580 And some of them, like securitization and derivatives, 402 00:20:25,580 --> 00:20:28,580 will dramatically shift your ability 403 00:20:28,580 --> 00:20:31,370 if you're doing a startup to compete. 404 00:20:31,370 --> 00:20:32,870 I see some things in the chat. 405 00:20:32,870 --> 00:20:34,160 Romain, are there questions? 406 00:20:38,090 --> 00:20:39,410 GUEST SPEAKER: All good, Gary. 407 00:20:39,410 --> 00:20:40,410 GARY GENSLER: All right. 408 00:20:40,410 --> 00:20:42,240 And then the question I sort of still 409 00:20:42,240 --> 00:20:45,450 have, and I teach this quite a bit at MIT, 410 00:20:45,450 --> 00:20:47,310 is blockchain technology. 411 00:20:47,310 --> 00:20:49,020 Will that move into the stack? 412 00:20:49,020 --> 00:20:51,670 I would contend it's not really yet there. 413 00:20:51,670 --> 00:20:53,710 You can be an incumbent. 414 00:20:53,710 --> 00:20:59,950 You can be a big finance firm, a big tech, or a startup and say, 415 00:20:59,950 --> 00:21:01,800 I'm not going to compete right there. 416 00:21:01,800 --> 00:21:03,210 I'm not quite sure. 417 00:21:03,210 --> 00:21:06,400 Though, again, we look at Facebook. 418 00:21:06,400 --> 00:21:11,290 We look at Telegram, big tech companies, messaging companies, 419 00:21:11,290 --> 00:21:14,560 [INAUDIBLE] in Korea who are sort 420 00:21:14,560 --> 00:21:20,270 of pulling in some blockchain technology and looking at it. 421 00:21:20,270 --> 00:21:22,360 We see trade finance consortiums. 422 00:21:22,360 --> 00:21:24,850 And we'll talk more about this next week. 423 00:21:24,850 --> 00:21:27,040 But I would say that you will not 424 00:21:27,040 --> 00:21:29,800 survive if you're not bringing machine learning 425 00:21:29,800 --> 00:21:32,620 into your technology stack. 426 00:21:32,620 --> 00:21:34,960 You probably won't survive that long 427 00:21:34,960 --> 00:21:39,280 if you don't really have a strategy around open API 428 00:21:39,280 --> 00:21:39,790 and data. 429 00:21:42,390 --> 00:21:45,360 Romain, I pause a little bit. 430 00:21:45,360 --> 00:21:48,220 We talked last session about artificial intelligence 431 00:21:48,220 --> 00:21:49,170 and machine learning. 432 00:21:49,170 --> 00:21:50,760 We're not going to dive back in. 433 00:21:50,760 --> 00:21:52,260 I'm just going to open it if there's 434 00:21:52,260 --> 00:21:54,990 any questions about what we talked about. 435 00:21:54,990 --> 00:21:56,880 That, of course, machine learning 436 00:21:56,880 --> 00:21:59,430 is just a part of artificial intelligence. 437 00:21:59,430 --> 00:22:02,740 You narrow it down to deep learning. 438 00:22:02,740 --> 00:22:04,750 Fundamentally as a business school, 439 00:22:04,750 --> 00:22:10,030 I'm not asking each of you to be able to program with TensorFlow 440 00:22:10,030 --> 00:22:12,805 and run a TensorFlow project, even though many of you 441 00:22:12,805 --> 00:22:14,990 know how to. 442 00:22:14,990 --> 00:22:19,030 I'm sort of just saying to think about, from a business side, 443 00:22:19,030 --> 00:22:23,230 it's about extracting from data, cleaning up that data, 444 00:22:23,230 --> 00:22:27,570 standardizing that data, and often labeling it. 445 00:22:27,570 --> 00:22:30,480 Labeling it because you can learn faster. 446 00:22:30,480 --> 00:22:31,980 That's called structured learning 447 00:22:31,980 --> 00:22:35,060 rather than unstructured. 448 00:22:35,060 --> 00:22:37,550 But labeling that data and then extracting 449 00:22:37,550 --> 00:22:42,940 correlations and decision algorithms that come out of it. 450 00:22:42,940 --> 00:22:43,810 Romain, any? 451 00:22:46,610 --> 00:22:48,740 GUEST SPEAKER: Luke has raised his hand again. 452 00:22:48,740 --> 00:22:49,355 GARY GENSLER: I'm going to pause. 453 00:22:49,355 --> 00:22:50,680 AUDIENCE: Just a quick question. 454 00:22:50,680 --> 00:22:51,400 GARY GENSLER: Oh, a question. 455 00:22:51,400 --> 00:22:52,035 Yeah. 456 00:22:52,035 --> 00:22:53,160 AUDIENCE: Yeah, a question. 457 00:22:53,160 --> 00:22:57,200 So how can a country that is developing 458 00:22:57,200 --> 00:22:59,330 fintech out of not because it was 459 00:22:59,330 --> 00:23:02,150 underbanked, but rather overbanked, 460 00:23:02,150 --> 00:23:05,220 but looking for alternative investment-- 461 00:23:05,220 --> 00:23:07,190 so the likes of South Korea-- 462 00:23:07,190 --> 00:23:11,600 develop a bunch of coders or those 463 00:23:11,600 --> 00:23:14,330 with-- actually, better yet, those people who 464 00:23:14,330 --> 00:23:19,430 can draw a conclusion and extract hypotheses and build up 465 00:23:19,430 --> 00:23:23,210 better ways to build an open API, 466 00:23:23,210 --> 00:23:25,880 how can a government really step in 467 00:23:25,880 --> 00:23:28,880 to encourage that and make an ecosystem? 468 00:23:28,880 --> 00:23:30,860 Somebody's got to do something. 469 00:23:30,860 --> 00:23:33,060 And I'm not sure America have a bunch 470 00:23:33,060 --> 00:23:36,085 of great coders and great minds, and it's a melting pot. 471 00:23:36,085 --> 00:23:38,448 So [INAUDIBLE] bunch of geniuses here. 472 00:23:38,448 --> 00:23:40,740 GARY GENSLER: Yeah, I'm not sure I follow the question, 473 00:23:40,740 --> 00:23:43,830 but I'm going to take it and then move on. 474 00:23:43,830 --> 00:23:46,070 I think what you're saying is in a country 475 00:23:46,070 --> 00:23:49,820 that has a very advanced banking system, 476 00:23:49,820 --> 00:23:52,490 how can a government encourage this? 477 00:23:52,490 --> 00:23:54,680 You do it through the education system. 478 00:23:54,680 --> 00:23:57,080 You do it through, just as we do in the US, 479 00:23:57,080 --> 00:24:01,340 promoting STEM education and programs like at MIT. 480 00:24:01,340 --> 00:24:06,610 I think over time, there is a challenge 481 00:24:06,610 --> 00:24:10,300 of how you adjust laws and regulations. 482 00:24:10,300 --> 00:24:14,260 Finance is a highly regulated space in every country. 483 00:24:14,260 --> 00:24:15,410 We're dealing with trust. 484 00:24:15,410 --> 00:24:17,740 We're dealing with people's money. 485 00:24:17,740 --> 00:24:20,140 We're dealing with inherent conflicts of interests 486 00:24:20,140 --> 00:24:23,230 that you can't escape in the world of finance. 487 00:24:23,230 --> 00:24:26,710 And so trying to deal with that with regulation 488 00:24:26,710 --> 00:24:32,600 but how we adjust with these new tools that are in place. 489 00:24:32,600 --> 00:24:35,070 But that would be how I'd answer it. 490 00:24:35,070 --> 00:24:37,760 So let's go back to what we talked about and just sort of 491 00:24:37,760 --> 00:24:40,860 do a little bit more granular. 492 00:24:40,860 --> 00:24:48,060 I contend at its core that these technologies, machine learning 493 00:24:48,060 --> 00:24:51,000 and natural language processing and AI, 494 00:24:51,000 --> 00:24:53,310 need to be brought into the finance 495 00:24:53,310 --> 00:24:56,530 stack and the technology stack. 496 00:24:56,530 --> 00:24:59,080 And so every type of company, whether you're 497 00:24:59,080 --> 00:25:01,510 in payments or you're in lending, 498 00:25:01,510 --> 00:25:03,478 whether you're insurance or not, you 499 00:25:03,478 --> 00:25:05,770 want to think about how you bring it in, whether you're 500 00:25:05,770 --> 00:25:08,290 a disruptor or not. 501 00:25:08,290 --> 00:25:10,420 And so that's why I think about it 502 00:25:10,420 --> 00:25:14,140 down the line in each of these fields and not just 503 00:25:14,140 --> 00:25:16,600 about disruptors. 504 00:25:16,600 --> 00:25:19,180 And we talked about each of these in the past, 505 00:25:19,180 --> 00:25:24,010 but I pause just, again, it's a little repetitive. 506 00:25:24,010 --> 00:25:27,940 Just if there's any questions about some of these slices. 507 00:25:27,940 --> 00:25:30,340 And then remember we're going to be digging 508 00:25:30,340 --> 00:25:33,580 quite a bit into this in the next five weeks 509 00:25:33,580 --> 00:25:36,250 as well in each of these areas. 510 00:25:36,250 --> 00:25:38,131 Romain? 511 00:25:38,131 --> 00:25:40,640 GUEST SPEAKER: I don't see anyone yet. 512 00:25:40,640 --> 00:25:42,530 GARY GENSLER: All right. 513 00:25:42,530 --> 00:25:46,880 Well, what I've said is, all right, so AI is a tool. 514 00:25:46,880 --> 00:25:48,980 And this is really an interesting debate 515 00:25:48,980 --> 00:25:50,720 that people can have at MIT. 516 00:25:50,720 --> 00:25:55,130 And I've been in rooms full of five to 10 faculty 517 00:25:55,130 --> 00:25:57,710 sometimes, sometimes one on one, where we debate. 518 00:25:57,710 --> 00:26:01,610 Is AI a service, or is AI a tool? 519 00:26:01,610 --> 00:26:06,050 And I would say that it's an interesting debate. 520 00:26:06,050 --> 00:26:08,420 Most of the time we land on that it's more 521 00:26:08,420 --> 00:26:09,770 a tool than a service. 522 00:26:09,770 --> 00:26:12,785 But every new technology, every new technology 523 00:26:12,785 --> 00:26:16,310 that comes along, whether it was the telephone, whether it 524 00:26:16,310 --> 00:26:20,480 was the railroads, whether it was airplanes, 525 00:26:20,480 --> 00:26:22,190 every new technology that comes along 526 00:26:22,190 --> 00:26:26,390 has some attributes of being a new industry, a new tool, 527 00:26:26,390 --> 00:26:29,670 the railroad industry, for instance. 528 00:26:29,670 --> 00:26:34,410 And yet many businesses use the railroad to do their shipping. 529 00:26:34,410 --> 00:26:40,160 AI and machine learning is more like a tool 530 00:26:40,160 --> 00:26:41,810 than it is a service. 531 00:26:41,810 --> 00:26:45,700 But it doesn't mean it's always just a tool. 532 00:26:45,700 --> 00:26:48,200 I did some research over the last few days, just a list 533 00:26:48,200 --> 00:26:50,480 where we could go through any one of these. 534 00:26:50,480 --> 00:26:51,620 AI as a service. 535 00:26:51,620 --> 00:26:54,560 Here I've listed 10 or 12 companies 536 00:26:54,560 --> 00:27:00,080 in finance that are actually doing AI sort of as a service. 537 00:27:00,080 --> 00:27:05,890 AlphaSense was started in 2011 before the whole rage of AI, 538 00:27:05,890 --> 00:27:09,470 but they were data analytics as a search engine for finance 539 00:27:09,470 --> 00:27:13,580 firms to search key terms and key words 540 00:27:13,580 --> 00:27:16,820 in registration statements and other statements that 541 00:27:16,820 --> 00:27:19,310 are filed with these various securities regulators 542 00:27:19,310 --> 00:27:20,510 around the globe. 543 00:27:20,510 --> 00:27:24,860 Sort of think of it as the Google for financial documents. 544 00:27:24,860 --> 00:27:29,120 Well, Google certainly has moved dramatically into AI space. 545 00:27:29,120 --> 00:27:31,850 AlphaSense did as well. 546 00:27:31,850 --> 00:27:34,760 There's a number of these in the insurance sector 547 00:27:34,760 --> 00:27:39,050 who really are around taking photographs of automobiles 548 00:27:39,050 --> 00:27:42,830 at an accident scene, and then based upon those automobile 549 00:27:42,830 --> 00:27:49,460 photographs or accident data, to use machine learning. 550 00:27:49,460 --> 00:27:54,890 And so Cape Analytics, Tractable are both firms 551 00:27:54,890 --> 00:27:57,410 that are in essence providing services 552 00:27:57,410 --> 00:27:59,300 to insurance companies. 553 00:27:59,300 --> 00:28:02,000 They have not yet, as to the best of my knowledge, Cape 554 00:28:02,000 --> 00:28:05,660 Analytics or Tractable, decided to have direct consumer 555 00:28:05,660 --> 00:28:06,350 interface. 556 00:28:06,350 --> 00:28:08,360 They're not selling insurance. 557 00:28:08,360 --> 00:28:12,070 They're selling a software analytics tool 558 00:28:12,070 --> 00:28:13,930 to insurance companies. 559 00:28:13,930 --> 00:28:16,330 And similarly, like ComplyAdvantage 560 00:28:16,330 --> 00:28:21,250 in the money laundering space or Featurespace in anti-fraud. 561 00:28:21,250 --> 00:28:25,450 They're saying, we can build something for fraud detection. 562 00:28:25,450 --> 00:28:28,870 We can build something for this world 563 00:28:28,870 --> 00:28:30,970 of anti-money-laundering compliance. 564 00:28:30,970 --> 00:28:35,260 We can build the software, and we'll put our product out 565 00:28:35,260 --> 00:28:41,710 there for the banking sector to basically rent us rather 566 00:28:41,710 --> 00:28:43,720 than building their own system. 567 00:28:43,720 --> 00:28:47,230 And you see others, document processing and the like. 568 00:28:47,230 --> 00:28:49,720 And even Zest AI-- 569 00:28:49,720 --> 00:28:55,480 Zest AI, founded in 2009, before this conceptual framework 570 00:28:55,480 --> 00:28:59,890 and the big movement, but Zest AI in credit underwriting 571 00:28:59,890 --> 00:29:02,950 software, basically providing-- 572 00:29:02,950 --> 00:29:07,600 broadly speaking, I'm calling it AI in finance as a service, 573 00:29:07,600 --> 00:29:11,430 rather than building it right into the stack. 574 00:29:11,430 --> 00:29:13,900 Romain I'm going to pause for a bit. 575 00:29:15,992 --> 00:29:17,700 GUEST SPEAKER: If you have any questions, 576 00:29:17,700 --> 00:29:19,730 please rate the blue little hand, 577 00:29:19,730 --> 00:29:21,140 as you probably know by now. 578 00:29:25,280 --> 00:29:26,767 I don't see anything, Gary. 579 00:29:26,767 --> 00:29:28,850 GARY GENSLER: And I'd say this, that each of these 580 00:29:28,850 --> 00:29:32,180 go back to some of the sectors back here. 581 00:29:32,180 --> 00:29:36,290 So asset management, we have sentiment analysis. 582 00:29:36,290 --> 00:29:41,800 We have-- I'm not going to pronounce the company's name 583 00:29:41,800 --> 00:29:44,590 right, but Dataminr. 584 00:29:44,590 --> 00:29:48,740 Dataminr, which can actually do market sentiment analysis. 585 00:29:48,740 --> 00:29:50,770 And if you're a hedge fund, you might 586 00:29:50,770 --> 00:29:57,000 sort of rent into Dataminr and get that sentiment analysis. 587 00:29:57,000 --> 00:30:00,190 You have several services that are 588 00:30:00,190 --> 00:30:02,360 doing fraud detection and regulatory, 589 00:30:02,360 --> 00:30:05,090 the anti-money-laundering slices. 590 00:30:05,090 --> 00:30:06,620 Credit and insurance-- 591 00:30:06,620 --> 00:30:11,410 I picked three that are doing insurance underwriting. 592 00:30:11,410 --> 00:30:13,690 But if you're Bank of America or J.P, 593 00:30:13,690 --> 00:30:17,770 Morgan or Cap One in the credit card business, 594 00:30:17,770 --> 00:30:23,130 you're going to be embedding this right into your business, 595 00:30:23,130 --> 00:30:24,000 by and large. 596 00:30:24,000 --> 00:30:26,350 Not always, not always, but by and large, 597 00:30:26,350 --> 00:30:28,870 you're embedding it right into your business. 598 00:30:28,870 --> 00:30:30,850 GUEST SPEAKER: We have a question from Geetha. 599 00:30:30,850 --> 00:30:32,736 GARY GENSLER: Please. 600 00:30:32,736 --> 00:30:33,850 AUDIENCE: Hey, Gary. 601 00:30:33,850 --> 00:30:34,800 Geetha here. 602 00:30:34,800 --> 00:30:36,530 I work for our Capital One. 603 00:30:36,530 --> 00:30:40,350 I'm in the credit lending space. 604 00:30:40,350 --> 00:30:40,850 One-- 605 00:30:40,850 --> 00:30:42,392 GARY GENSLER: You're going to correct 606 00:30:42,392 --> 00:30:43,740 anything I say about Cap One. 607 00:30:43,740 --> 00:30:45,610 Please. 608 00:30:45,610 --> 00:30:48,370 AUDIENCE: No. 609 00:30:48,370 --> 00:30:51,310 One thing that I find really surprising with the regulations 610 00:30:51,310 --> 00:30:58,280 is that if we develop our own AI models, regulations-- 611 00:30:58,280 --> 00:31:00,760 anybody, like when auditing happens, 612 00:31:00,760 --> 00:31:03,900 they are very specific about explainability, 613 00:31:03,900 --> 00:31:05,500 interpretability. 614 00:31:05,500 --> 00:31:10,490 But if you were to use a vendor, Shape Security, Akamai, 615 00:31:10,490 --> 00:31:13,982 they don't care too much about explainability. 616 00:31:13,982 --> 00:31:16,550 That I always found surprising. 617 00:31:16,550 --> 00:31:19,540 Why is it that within the realms of bank, 618 00:31:19,540 --> 00:31:21,550 they're so specific about regulations, 619 00:31:21,550 --> 00:31:26,040 but when we use a vendor, the extent to which they 620 00:31:26,040 --> 00:31:27,670 care as how you use Akamai. 621 00:31:27,670 --> 00:31:29,575 You use Shape, and that's it. 622 00:31:29,575 --> 00:31:31,910 GARY GENSLER: Gita, I'm so glad that you raised this. 623 00:31:31,910 --> 00:31:40,390 I think I earlier had said in our introductory class 624 00:31:40,390 --> 00:31:49,090 that one of the competitive advantages of disruptors 625 00:31:49,090 --> 00:31:56,070 is that they have certain asymmetries. 626 00:31:56,070 --> 00:31:59,130 Incumbents, like Cap One-- and if I 627 00:31:59,130 --> 00:32:01,050 might say you working for Cap One, 628 00:32:01,050 --> 00:32:04,910 which is one of the big seven firms in credit cards, 629 00:32:04,910 --> 00:32:07,200 is truly one of the incumbents. 630 00:32:07,200 --> 00:32:10,590 Incumbents tend to need to protect their business models. 631 00:32:10,590 --> 00:32:12,270 And part of what they're protecting 632 00:32:12,270 --> 00:32:17,590 is also the reputational and regulatory risks. 633 00:32:17,590 --> 00:32:25,640 But the disruptors have a little bit of a different perspective. 634 00:32:25,640 --> 00:32:30,470 They're not generally protecting any in inherent or incumbent 635 00:32:30,470 --> 00:32:31,850 business model. 636 00:32:31,850 --> 00:32:34,310 And yes, they're also willing to take 637 00:32:34,310 --> 00:32:36,500 more risk with the regulators. 638 00:32:36,500 --> 00:32:38,570 I'm not saying whether they should or shouldn't. 639 00:32:38,570 --> 00:32:41,640 I'm just saying this is kind of the facts on the ground 640 00:32:41,640 --> 00:32:44,420 that disruptors are a little bit more 641 00:32:44,420 --> 00:32:50,680 towards the end of basically begging for forgiveness 642 00:32:50,680 --> 00:32:52,720 rather than asking for permission, 643 00:32:52,720 --> 00:32:54,790 if you sort of remember how you might 644 00:32:54,790 --> 00:32:57,880 have been with your parents or if any of you have children. 645 00:32:57,880 --> 00:33:01,300 And incumbents are more into asking 646 00:33:01,300 --> 00:33:04,750 for permission of your at least internal GC, your General 647 00:33:04,750 --> 00:33:05,260 Counsel. 648 00:33:05,260 --> 00:33:06,100 Can we do this? 649 00:33:06,100 --> 00:33:07,610 Can we do that? 650 00:33:07,610 --> 00:33:09,010 And so I think the vendors-- 651 00:33:09,010 --> 00:33:11,560 in that case, the vendors are a little bit more 652 00:33:11,560 --> 00:33:14,325 willing to take risks when explainability 653 00:33:14,325 --> 00:33:17,740 and the explainability that's inherent in US law 654 00:33:17,740 --> 00:33:20,950 and the Fair Credit Reporting Act and the like. 655 00:33:20,950 --> 00:33:25,260 And it doesn't mean that it's more legal for a disruptor 656 00:33:25,260 --> 00:33:28,290 to do it than for Cap One to do it. 657 00:33:28,290 --> 00:33:32,070 It's just their business model tends 658 00:33:32,070 --> 00:33:35,430 to be a little bit more accepting 659 00:33:35,430 --> 00:33:38,010 of that regulatory compliance risk. 660 00:33:38,010 --> 00:33:42,270 Secondly, and I think this is probably 661 00:33:42,270 --> 00:33:45,810 a bit of a misreading of the risks, 662 00:33:45,810 --> 00:33:53,760 but sometimes the thought is if the vendor does something, 663 00:33:53,760 --> 00:33:57,680 it sort of insulates the big firm. 664 00:33:57,680 --> 00:33:59,640 Now, my understanding-- again, I'm not 665 00:33:59,640 --> 00:34:01,170 a lawyer-- but my understanding, it 666 00:34:01,170 --> 00:34:04,020 doesn't really insulate Cap One, or it doesn't really 667 00:34:04,020 --> 00:34:07,350 insulate Bank of America if their vendor does something 668 00:34:07,350 --> 00:34:10,440 that's blatantly a violation. 669 00:34:10,440 --> 00:34:11,760 I don't think it does. 670 00:34:11,760 --> 00:34:14,940 But sometimes there is a bit of that mindset as well. 671 00:34:14,940 --> 00:34:16,835 Does that help? 672 00:34:16,835 --> 00:34:17,460 AUDIENCE: Yeah. 673 00:34:17,460 --> 00:34:18,000 Yeah. 674 00:34:18,000 --> 00:34:20,760 And the last thing is-- not taking too much time, just 675 00:34:20,760 --> 00:34:23,310 a comment. 676 00:34:23,310 --> 00:34:25,620 One other thing I find very intriguing with vendors 677 00:34:25,620 --> 00:34:29,040 is that they often get the data of incumbents, 678 00:34:29,040 --> 00:34:31,880 like maybe Bank of America, Capital One, 679 00:34:31,880 --> 00:34:34,510 and they charge us back for that data. 680 00:34:34,510 --> 00:34:39,429 That's the other thing I find [INAUDIBLE].. 681 00:34:39,429 --> 00:34:41,460 GARY GENSLER: So here, this is about how 682 00:34:41,460 --> 00:34:44,070 companies can capture data. 683 00:34:44,070 --> 00:34:47,010 We're going to talk a lot about this one, open API. 684 00:34:47,010 --> 00:34:51,150 Just to the intersection, this is one of the key things 685 00:34:51,150 --> 00:34:56,440 to take from these series of classes. 686 00:34:56,440 --> 00:35:00,520 Machine learning is nowhere if it doesn't have data. 687 00:35:00,520 --> 00:35:06,280 Data is facilitated by a lot of startups getting access 688 00:35:06,280 --> 00:35:09,280 to that which the incumbents already have. 689 00:35:09,280 --> 00:35:13,210 So around the globe, in Europe, in the UK, in Brazil, 690 00:35:13,210 --> 00:35:17,470 in Canada, US, Australia, there are significant efforts 691 00:35:17,470 --> 00:35:20,260 to promote what's called open API-- 692 00:35:20,260 --> 00:35:24,100 Application Program Interface. 693 00:35:24,100 --> 00:35:29,290 In essence, that is you or I permissioning a company 694 00:35:29,290 --> 00:35:34,960 to get my or your data from an incumbent financial firm. 695 00:35:34,960 --> 00:35:39,100 And so we permission somebody to get data, in Gita's example, 696 00:35:39,100 --> 00:35:41,580 from Cap One. 697 00:35:41,580 --> 00:35:43,950 Then they use that data, the startup. 698 00:35:43,950 --> 00:35:46,110 And then Gita's saying that Cap One then 699 00:35:46,110 --> 00:35:50,050 has to pay a fee to some startup, 700 00:35:50,050 --> 00:35:54,060 even though the data had initially come from Cap One. 701 00:35:54,060 --> 00:35:56,460 And that's a perfect set up to two examples 702 00:35:56,460 --> 00:35:58,410 I just want to talk about. 703 00:35:58,410 --> 00:36:00,690 I want to talk about two large mergers that 704 00:36:00,690 --> 00:36:03,280 were announced in 2020. 705 00:36:03,280 --> 00:36:04,920 The first one is Credit Karma. 706 00:36:04,920 --> 00:36:08,710 Now, I don't know if we could do by a show of hands, 707 00:36:08,710 --> 00:36:14,430 but how many-- just raising the blue hands in the windows-- 708 00:36:14,430 --> 00:36:17,940 how many people, if you can go into participant buttons, 709 00:36:17,940 --> 00:36:21,060 have actually used Credit Karma, that you would consider 710 00:36:21,060 --> 00:36:23,190 yourself one of these members? 711 00:36:23,190 --> 00:36:26,010 And Romain you'll tell me what it looks like. 712 00:36:32,730 --> 00:36:35,502 GUEST SPEAKER: We have at least 10 students so far. 713 00:36:35,502 --> 00:36:37,210 GARY GENSLER: No, but if you scroll down. 714 00:36:37,210 --> 00:36:40,260 So all right, so it's not as big a percent as I thought. 715 00:36:43,990 --> 00:36:45,770 Let me go back. 716 00:36:45,770 --> 00:36:51,620 So Credit Karma started in 2007. 717 00:36:51,620 --> 00:36:55,250 The entrepreneur who started it couldn't get a free credit 718 00:36:55,250 --> 00:36:55,910 report. 719 00:36:55,910 --> 00:37:01,790 So they say, why don't I start a credit report platform? 720 00:37:01,790 --> 00:37:05,180 13 years later, they were able to sell for $7 billion 721 00:37:05,180 --> 00:37:07,070 to Intuit. 722 00:37:07,070 --> 00:37:09,550 Now, you might not be familiar with Intuit. 723 00:37:09,550 --> 00:37:14,870 Their main products at Intuit are tax software, TurboTax. 724 00:37:14,870 --> 00:37:18,080 They also have something called Quicken Books. 725 00:37:18,080 --> 00:37:20,670 And I believe it's possible they have a third product. 726 00:37:20,670 --> 00:37:21,860 They might even have Mint. 727 00:37:25,030 --> 00:37:28,240 But Intuit saw they wanted to buy Credit Karma that had never 728 00:37:28,240 --> 00:37:29,620 gone public. 729 00:37:29,620 --> 00:37:31,975 Credit Karma apparently had nearly a billion dollars 730 00:37:31,975 --> 00:37:35,830 in revenue last year, and yet Credit Karma 731 00:37:35,830 --> 00:37:37,918 is still a free app. 732 00:37:37,918 --> 00:37:39,460 How is it that something that doesn't 733 00:37:39,460 --> 00:37:42,660 charge anything can have a billion dollars in revenue? 734 00:37:42,660 --> 00:37:46,140 It's that they're commercializing data. 735 00:37:46,140 --> 00:37:51,740 And remarkably, 106 million members-- 736 00:37:51,740 --> 00:37:53,135 106 million members. 737 00:37:57,120 --> 00:38:03,230 8 billion daily decisions, credit decisions 738 00:38:03,230 --> 00:38:07,670 or other analytic decisions that they have. 739 00:38:07,670 --> 00:38:11,960 And so Intuit is saying, why are they buying Credit Karma? 740 00:38:11,960 --> 00:38:15,170 Even at seven times revenue, that's a healthy price. 741 00:38:15,170 --> 00:38:18,710 They're buying it largely around data and data analytics. 742 00:38:18,710 --> 00:38:23,450 And credit card has figured out how to basically commercialize 743 00:38:23,450 --> 00:38:28,420 that flow of data on over 100 million accounts. 744 00:38:28,420 --> 00:38:29,540 And how do they do that? 745 00:38:29,540 --> 00:38:32,080 They do it by cross marketing. 746 00:38:32,080 --> 00:38:35,030 So they're marketing not just to us. 747 00:38:35,030 --> 00:38:37,610 But then they're also, with many financial firms, 748 00:38:37,610 --> 00:38:40,450 they're going back and say, this account here, 749 00:38:40,450 --> 00:38:42,820 this is a worthy thing. 750 00:38:42,820 --> 00:38:44,680 So they make arrangements. 751 00:38:44,680 --> 00:38:46,630 They enter into contractual arrangements 752 00:38:46,630 --> 00:38:50,200 with financial institutions and then market to us 753 00:38:50,200 --> 00:38:52,340 to take a mortgage, to take an auto loan, 754 00:38:52,340 --> 00:38:53,895 to take a personal loan. 755 00:38:57,780 --> 00:38:58,850 Plaid. 756 00:38:58,850 --> 00:39:03,100 Plaid's a company that we'll talk a lot about when 757 00:39:03,100 --> 00:39:05,680 we talk about open API. 758 00:39:05,680 --> 00:39:09,420 This was software that started just seven years ago. 759 00:39:09,420 --> 00:39:14,565 Two developers, straight kind of hard-core computer scientists 760 00:39:14,565 --> 00:39:17,388 who had went to work for Bain. 761 00:39:17,388 --> 00:39:19,930 And for anybody who's thinking about working for a consulting 762 00:39:19,930 --> 00:39:23,770 firm, this is not a vote against Bain or BCG or others. 763 00:39:23,770 --> 00:39:26,920 But they basically decided after a year at Bain 764 00:39:26,920 --> 00:39:29,080 to go out and do their own startup. 765 00:39:29,080 --> 00:39:35,760 And it was a startup to facilitate financial disruptors 766 00:39:35,760 --> 00:39:39,810 or fintech companies accessing data at banks. 767 00:39:39,810 --> 00:39:41,580 And there was not a standard. 768 00:39:41,580 --> 00:39:44,010 There was not a standard for this open API. 769 00:39:44,010 --> 00:39:48,030 So they created at a hackathon-- 770 00:39:48,030 --> 00:39:51,630 at a hackathon that they actually won back, 771 00:39:51,630 --> 00:39:55,020 I think, in 2012 or 2013-- 772 00:39:55,020 --> 00:39:57,410 they were in their late 20s at the time, by the way, 773 00:39:57,410 --> 00:39:59,280 if you're just trying to figure out. 774 00:39:59,280 --> 00:40:01,770 Seven years later, in their mid to late 30s, 775 00:40:01,770 --> 00:40:04,890 they sell their business for $5.3 billion. 776 00:40:04,890 --> 00:40:09,285 But it all starts at Bain, computer scientists creating 777 00:40:09,285 --> 00:40:11,990 open API software. 778 00:40:11,990 --> 00:40:14,880 Well, what happened over those seven years, 779 00:40:14,880 --> 00:40:18,240 11,000 financial companies signed up 780 00:40:18,240 --> 00:40:23,360 to use that standard protocol to do open API. 781 00:40:23,360 --> 00:40:27,180 And all the other side, 2,600 fintech developers. 782 00:40:27,180 --> 00:40:30,390 And if anyone here has taken Michael Cusumano's class 783 00:40:30,390 --> 00:40:35,090 on platforms, this is the classic sweet spot 784 00:40:35,090 --> 00:40:37,640 of creating a platform company, when 785 00:40:37,640 --> 00:40:41,570 you have this two-sided many-to-many market. 786 00:40:41,570 --> 00:40:45,830 Many fintech developers want to access financial data 787 00:40:45,830 --> 00:40:46,940 at financial firms. 788 00:40:46,940 --> 00:40:49,820 Many financial firms don't want to deal with thousands 789 00:40:49,820 --> 00:40:51,800 of fintech developers. 790 00:40:51,800 --> 00:40:55,490 And so inside of this many-to-many market, 791 00:40:55,490 --> 00:40:58,800 Plaid creates a software, a standard software 792 00:40:58,800 --> 00:41:00,940 for that to happen. 793 00:41:00,940 --> 00:41:03,420 But what did they build on top of that? 794 00:41:03,420 --> 00:41:06,380 They built data aggregation. 795 00:41:06,380 --> 00:41:09,002 They announced a $5.3 billion merger to Visa. 796 00:41:09,002 --> 00:41:10,460 There's a lot of people that debate 797 00:41:10,460 --> 00:41:14,570 whether it was a good idea because the estimate by Forbes 798 00:41:14,570 --> 00:41:17,910 is there was only about $110 million of revenue. 799 00:41:17,910 --> 00:41:22,510 I mean, now we're talking 40 or 50 times revenue. 800 00:41:22,510 --> 00:41:27,980 But 200 million accounts are linked. 801 00:41:27,980 --> 00:41:31,070 We'll chat about this more because those 11,000 802 00:41:31,070 --> 00:41:34,130 financial firms could all stop using Plaid 803 00:41:34,130 --> 00:41:35,810 and go to one of Plaid's competitors 804 00:41:35,810 --> 00:41:38,330 now that Plaid's bought by Visa. 805 00:41:38,330 --> 00:41:42,020 But this gives you the sense of the power and the value, 806 00:41:42,020 --> 00:41:47,930 the economic value of data, machine learning, and the like. 807 00:41:47,930 --> 00:41:49,010 Romain questions? 808 00:41:52,685 --> 00:41:53,810 GUEST SPEAKER: None so far. 809 00:41:53,810 --> 00:41:56,560 It seems like the class is quiet today. 810 00:41:56,560 --> 00:41:58,600 GARY GENSLER: All right. 811 00:41:58,600 --> 00:42:02,340 So I'm going to talk a little bit about financial policy. 812 00:42:02,340 --> 00:42:05,980 How does this all fit in, in the next half hour. 813 00:42:05,980 --> 00:42:09,925 Broadly, first, is just a sense of-- 814 00:42:09,925 --> 00:42:13,600 I'm trying to get rid of it this participant window here 815 00:42:13,600 --> 00:42:16,090 for a minute. 816 00:42:16,090 --> 00:42:19,450 So broad public policy frameworks 817 00:42:19,450 --> 00:42:22,990 have been around for thousands of years, since the Hammurabi 818 00:42:22,990 --> 00:42:26,770 code, since Roman and Greek times, sometimes 819 00:42:26,770 --> 00:42:30,090 embedded even in religious law. 820 00:42:30,090 --> 00:42:32,680 That's the nature of money. 821 00:42:32,680 --> 00:42:35,040 But four slip streams, and all four 822 00:42:35,040 --> 00:42:38,310 will be relevant for fintech as we talk through 823 00:42:38,310 --> 00:42:41,250 not just AI, but all sectors. 824 00:42:41,250 --> 00:42:43,140 One is money and lending. 825 00:42:43,140 --> 00:42:47,160 We've, over centuries, often get official sector 826 00:42:47,160 --> 00:42:49,230 as a point of view, sometimes even 827 00:42:49,230 --> 00:42:52,620 limiting interest rates and the like. 828 00:42:52,620 --> 00:42:54,960 Two is financial stability. 829 00:42:54,960 --> 00:42:56,490 We think about a crisis. 830 00:42:56,490 --> 00:42:59,010 We're living through this corona crisis right now. 831 00:42:59,010 --> 00:43:02,040 Central banks around the globe are, 832 00:43:02,040 --> 00:43:06,120 with an eye towards promoting the economy, 833 00:43:06,120 --> 00:43:10,440 also thinking about how to ensure for financial stability. 834 00:43:10,440 --> 00:43:12,240 The reverse of financial stability 835 00:43:12,240 --> 00:43:17,190 was happening in 2008 crisis, that that crisis where banks 836 00:43:17,190 --> 00:43:19,530 were faltering and closing up. 837 00:43:19,530 --> 00:43:23,070 And then that led to millions of people losing their jobs, 838 00:43:23,070 --> 00:43:27,460 millions of people losing their homes, and the like. 839 00:43:27,460 --> 00:43:30,150 So financial stability, I grab a couple pictures 840 00:43:30,150 --> 00:43:35,340 here out of the Great Depression, 841 00:43:35,340 --> 00:43:38,170 an earlier period of crisis. 842 00:43:38,170 --> 00:43:42,010 But what we'll talk a lot about is the third and fourth bucket. 843 00:43:42,010 --> 00:43:45,550 The third bucket of protecting consumers and investors. 844 00:43:45,550 --> 00:43:48,460 Consumer protection we can think of even just in terms 845 00:43:48,460 --> 00:43:53,440 of ensuring that if we buy a crib for our children 846 00:43:53,440 --> 00:43:54,860 that it's safe. 847 00:43:54,860 --> 00:43:59,350 If we buy a car that it actually is safe on the road. 848 00:43:59,350 --> 00:44:04,090 So consumer protection refers to things much broader 849 00:44:04,090 --> 00:44:04,810 than finance. 850 00:44:04,810 --> 00:44:07,930 Investor protection is the concept that, yes, 851 00:44:07,930 --> 00:44:10,120 we can take risk in markets. 852 00:44:10,120 --> 00:44:13,030 We're all allowed to take risk in markets. 853 00:44:13,030 --> 00:44:16,600 But that the markets themselves and the issuers, the people 854 00:44:16,600 --> 00:44:21,960 raising money, should explain to us at least the material 855 00:44:21,960 --> 00:44:25,920 pieces of information upon which we would take those risks. 856 00:44:25,920 --> 00:44:29,910 And that the markets themselves have a certain transparency 857 00:44:29,910 --> 00:44:34,220 and we protect against fraud and manipulation and the like. 858 00:44:34,220 --> 00:44:36,520 And then guarding against illicit activity. 859 00:44:36,520 --> 00:44:38,260 This is one that we've really layered 860 00:44:38,260 --> 00:44:42,710 over the financial sector in the last 40-odd years. 861 00:44:42,710 --> 00:44:46,730 In an earlier era, 19th century, earlier 20th century, 862 00:44:46,730 --> 00:44:49,220 we didn't have as much about this, even though, of course, 863 00:44:49,220 --> 00:44:51,370 we did guard against bank robbers. 864 00:44:51,370 --> 00:44:53,830 But I'm talking about illicit activity outside 865 00:44:53,830 --> 00:44:55,420 of the financial sector-- 866 00:44:55,420 --> 00:44:59,320 money laundering, terrorism, and even sanctions. 867 00:44:59,320 --> 00:45:02,645 So these four slip [streams in a sense, are there. 868 00:45:05,530 --> 00:45:10,930 So how does it fit back to AI and policy in finance? 869 00:45:10,930 --> 00:45:13,450 So I've talked about what I have come 870 00:45:13,450 --> 00:45:17,740 to call the big three, biases, fairness, and inclusion; 871 00:45:17,740 --> 00:45:20,640 explainability; and privacy. 872 00:45:20,640 --> 00:45:24,730 And what we mean by that is if you take a whole data 873 00:45:24,730 --> 00:45:28,620 set, millions or tens of millions of pieces of data, 874 00:45:28,620 --> 00:45:32,920 and extract correlations, and you find patterns, 875 00:45:32,920 --> 00:45:35,260 some of those patterns might have biases. 876 00:45:35,260 --> 00:45:38,500 And those biases can exist because we 877 00:45:38,500 --> 00:45:42,860 as a society are not perfect. 878 00:45:42,860 --> 00:45:45,100 We have biases even in what we've already done. 879 00:45:48,010 --> 00:45:50,150 And so now you're extracting and you might be 880 00:45:50,150 --> 00:45:53,000 embedding some of those biases. 881 00:45:53,000 --> 00:45:55,880 Secondly, sometimes it will happen 882 00:45:55,880 --> 00:45:58,970 just out of how you build your protocols, 883 00:45:58,970 --> 00:46:01,400 how you build your actual questions 884 00:46:01,400 --> 00:46:04,220 and query on the data. 885 00:46:04,220 --> 00:46:11,650 But I assure you that most data sets have some biases in them. 886 00:46:11,650 --> 00:46:14,320 You just might not be aware of them. 887 00:46:14,320 --> 00:46:16,710 And even if you have a perfect data set, 888 00:46:16,710 --> 00:46:20,340 the protocols themselves might sort of build some biases 889 00:46:20,340 --> 00:46:22,350 on top of them. 890 00:46:22,350 --> 00:46:25,690 And we're finding this in AI policy not just in finance. 891 00:46:25,690 --> 00:46:27,420 It's true in the criminal justice system. 892 00:46:27,420 --> 00:46:31,970 It's true in hiring, that using machine learning, 893 00:46:31,970 --> 00:46:36,930 you have to sort of say, wait, is there bias? 894 00:46:36,930 --> 00:46:43,860 And the laws here in the US that are most relevant in finance 895 00:46:43,860 --> 00:46:46,330 started with something called the Equal Credit Opportunity 896 00:46:46,330 --> 00:46:47,420 Act. 897 00:46:47,420 --> 00:46:51,560 And we'll talk a little bit more about that. 898 00:46:51,560 --> 00:46:56,390 Explainability and transparency, as we talked about earlier, 899 00:46:56,390 --> 00:47:00,200 is sort of like a cousin or a sister to the bias issue. 900 00:47:00,200 --> 00:47:02,330 And in the US, it was 50 years ago 901 00:47:02,330 --> 00:47:06,050 that we passed these twin laws within four years. 902 00:47:06,050 --> 00:47:08,790 The second law was the Fair Credit Reporting Act. 903 00:47:08,790 --> 00:47:12,590 And this was the concept about holding that data, 904 00:47:12,590 --> 00:47:15,560 but also being able to explain it. 905 00:47:15,560 --> 00:47:17,810 Romain I see the chat button. 906 00:47:17,810 --> 00:47:19,980 GUEST SPEAKER: We have no question from Jorge. 907 00:47:19,980 --> 00:47:21,890 GARY GENSLER: Please. 908 00:47:21,890 --> 00:47:24,950 AUDIENCE: Yes, professor, thank you so much. 909 00:47:24,950 --> 00:47:27,920 I just want to have a little bit more color 910 00:47:27,920 --> 00:47:31,110 on financial inclusion, and specifically 911 00:47:31,110 --> 00:47:34,820 on what type of data, what models are used? 912 00:47:34,820 --> 00:47:38,240 What's the forefront of data modeling 913 00:47:38,240 --> 00:47:43,760 for using AI and to help financial inclusion? 914 00:47:43,760 --> 00:47:45,525 Thank you. 915 00:47:45,525 --> 00:47:47,900 GARY GENSLER: I'm not sure, Jorje, I follow the question. 916 00:47:47,900 --> 00:47:53,270 Let me see if I do it, but please keep your audio on 917 00:47:53,270 --> 00:47:56,180 so we can engage here. 918 00:47:56,180 --> 00:47:59,730 Biases are sort of the reverse of inclusion. 919 00:47:59,730 --> 00:48:01,940 So financial inclusion is a concept 920 00:48:01,940 --> 00:48:06,140 that everyone in society has fair access 921 00:48:06,140 --> 00:48:12,140 and open, equal access in some way to the extension of credit, 922 00:48:12,140 --> 00:48:14,870 to insurance, to financial advice, 923 00:48:14,870 --> 00:48:19,110 to investment products and savings products 924 00:48:19,110 --> 00:48:23,030 that they wish to, or payment products as well. 925 00:48:23,030 --> 00:48:25,280 And the reverse of inclusion is sometimes 926 00:48:25,280 --> 00:48:27,110 that somebody is excluded. 927 00:48:27,110 --> 00:48:35,320 And excluding someone could be excluding them 928 00:48:35,320 --> 00:48:37,360 on something that is allowed. 929 00:48:37,360 --> 00:48:40,630 Like I might exclude somebody only earning $50,000 930 00:48:40,630 --> 00:48:43,180 a year from an investment product 931 00:48:43,180 --> 00:48:47,260 which is for high-risk investors, 932 00:48:47,260 --> 00:48:51,160 depending upon how the country is arranged. 933 00:48:51,160 --> 00:48:54,490 But in the US, we have a little bit of this concept 934 00:48:54,490 --> 00:48:59,230 that sophisticated investors can be investing 935 00:48:59,230 --> 00:49:02,260 in products of higher risk. 936 00:49:02,260 --> 00:49:05,080 Or at least they get less disclosure. 937 00:49:05,080 --> 00:49:09,140 But that's how inclusion and bias are kind of the-- 938 00:49:09,140 --> 00:49:10,500 they complement each other. 939 00:49:10,500 --> 00:49:12,610 The greater inclusion you have-- 940 00:49:12,610 --> 00:49:16,630 you can get to greater inclusion if you have fewer biases, 941 00:49:16,630 --> 00:49:18,700 in a [sense fairness. 942 00:49:18,700 --> 00:49:19,760 But [INAUDIBLE]. 943 00:49:19,760 --> 00:49:21,370 AUDIENCE: No, I was just-- 944 00:49:21,370 --> 00:49:23,100 I totally get that. 945 00:49:23,100 --> 00:49:27,070 I was just trying to understand what type of models, 946 00:49:27,070 --> 00:49:31,890 what type of data, what is the forefront of AI currently? 947 00:49:31,890 --> 00:49:34,180 Because I totally get it's gathering data 948 00:49:34,180 --> 00:49:35,480 and finding patterns. 949 00:49:35,480 --> 00:49:40,820 But digging a little bit more on that, what type of-- 950 00:49:40,820 --> 00:49:44,110 GARY GENSLER: So let's say that one pattern 951 00:49:44,110 --> 00:49:46,060 that we know about already-- 952 00:49:46,060 --> 00:49:51,040 this is a classic pattern in credit extension. 953 00:49:51,040 --> 00:49:55,870 I don't know how many of you know what retreading a tire is. 954 00:49:55,870 --> 00:49:59,110 A retread means that you're putting rubber on your tire. 955 00:49:59,110 --> 00:50:02,230 Instead of replacing your tire on your automobile, 956 00:50:02,230 --> 00:50:06,020 you're actually paying to put new rubber on the tire-- 957 00:50:06,020 --> 00:50:07,750 tire retreading. 958 00:50:07,750 --> 00:50:11,710 It's been known for decades that people who retread their tires 959 00:50:11,710 --> 00:50:15,080 are a little lower income, generally speaking. 960 00:50:15,080 --> 00:50:20,860 And actually there's research-- 961 00:50:20,860 --> 00:50:22,600 I don't mean academic research, but there 962 00:50:22,600 --> 00:50:27,580 is research in the credit business 963 00:50:27,580 --> 00:50:31,180 that retreading tires means that you're probably 964 00:50:31,180 --> 00:50:34,080 a little higher credit risk. 965 00:50:34,080 --> 00:50:36,300 Now bring it forward to now. 966 00:50:36,300 --> 00:50:39,270 Bring it forward to the 2020 environment. 967 00:50:39,270 --> 00:50:41,250 And let's say that you can follow 968 00:50:41,250 --> 00:50:44,370 that those people who bought tire retreading, 969 00:50:44,370 --> 00:50:49,380 or even if you went to a website on your laptop 970 00:50:49,380 --> 00:50:52,470 about tire retreading, let's say that's built 971 00:50:52,470 --> 00:50:55,290 in to an algorithm that's going to give you 972 00:50:55,290 --> 00:50:59,450 lower extension of credit. 973 00:50:59,450 --> 00:51:04,100 That might be allowed, or it might embed a different bias. 974 00:51:04,100 --> 00:51:08,930 It might be that tire retreading shops are perfectly acceptable 975 00:51:08,930 --> 00:51:13,730 in certain communities, either ethnic communities, 976 00:51:13,730 --> 00:51:16,520 or gender-based, or racial communities, 977 00:51:16,520 --> 00:51:20,090 that it's just perfectly-- 978 00:51:20,090 --> 00:51:21,990 it's not about creditworthiness. 979 00:51:21,990 --> 00:51:27,110 So it's how you extract certain patterns that 980 00:51:27,110 --> 00:51:34,460 are about credit extension but not about race, ethnicity, 981 00:51:34,460 --> 00:51:38,290 cultural backgrounds, and the like. 982 00:51:38,290 --> 00:51:42,070 And if you hold just for a second, 983 00:51:42,070 --> 00:51:44,070 we're going to talk a little bit more about this 984 00:51:44,070 --> 00:51:47,100 because I'm going to talk about the Equal Credit Opportunity 985 00:51:47,100 --> 00:51:48,517 Act. 986 00:51:48,517 --> 00:51:49,350 AUDIENCE: Thank you. 987 00:51:52,070 --> 00:51:54,280 GARY GENSLER: Romain we good? 988 00:51:54,280 --> 00:51:55,530 GUEST SPEAKER: All good, Gary. 989 00:51:55,530 --> 00:52:00,820 GARY GENSLER: So beyond what I'm sort of calling the big three, 990 00:52:00,820 --> 00:52:02,470 I list four other things. 991 00:52:02,470 --> 00:52:04,620 But they're all really relevant. 992 00:52:04,620 --> 00:52:09,480 They're relevant to, broadly speaking, the official sector. 993 00:52:09,480 --> 00:52:11,100 But they're also relevant as you think 994 00:52:11,100 --> 00:52:13,260 about going into these businesses 995 00:52:13,260 --> 00:52:15,150 the use of alternative data. 996 00:52:15,150 --> 00:52:16,710 And we'll come back to that. 997 00:52:16,710 --> 00:52:20,190 Basically, we've had data analytics in consumer finance 998 00:52:20,190 --> 00:52:22,110 since the 1960s. 999 00:52:22,110 --> 00:52:27,260 We have in 30-plus countries used these FICO scores. 1000 00:52:27,260 --> 00:52:32,670 But beyond what is built into the classic data 1001 00:52:32,670 --> 00:52:36,110 set, what about new data? 1002 00:52:36,110 --> 00:52:39,560 We have issues about whether the algorithms themselves will 1003 00:52:39,560 --> 00:52:42,440 be correlated or even collude. 1004 00:52:42,440 --> 00:52:45,680 And this is absolutely the case that one machine learning 1005 00:52:45,680 --> 00:52:48,800 algorithm and another machine learning algorithm 1006 00:52:48,800 --> 00:52:52,010 can actually train against each other. 1007 00:52:52,010 --> 00:52:55,280 We've already seen this in high-frequency trading. 1008 00:52:55,280 --> 00:52:57,590 Even if the humans aren't talking, 1009 00:52:57,590 --> 00:53:00,140 the machines will start to actually 1010 00:53:00,140 --> 00:53:04,340 have a sense of cooperation. 1011 00:53:04,340 --> 00:53:06,460 And when is cooperation collusion? 1012 00:53:06,460 --> 00:53:08,870 When is it that they're spoofing each other 1013 00:53:08,870 --> 00:53:10,670 or doing something against each other 1014 00:53:10,670 --> 00:53:13,600 in a high-frequency world? 1015 00:53:13,600 --> 00:53:15,640 I deeply-- and this is one of the areas 1016 00:53:15,640 --> 00:53:18,580 I want to research more with colleagues. 1017 00:53:18,580 --> 00:53:23,050 I deeply am concerned that a future crisis-- 1018 00:53:23,050 --> 00:53:24,192 it's remarkable. 1019 00:53:24,192 --> 00:53:25,900 We're in the middle of the corona crisis, 1020 00:53:25,900 --> 00:53:30,740 but a future crisis we'll find algorithmic correlation. 1021 00:53:30,740 --> 00:53:33,310 And this is certainly the case in smaller developing 1022 00:53:33,310 --> 00:53:37,690 countries, where a Baidu from China or a Google from the US 1023 00:53:37,690 --> 00:53:41,110 might come in, and they might come in with their approach 1024 00:53:41,110 --> 00:53:42,940 to artificial intelligence. 1025 00:53:42,940 --> 00:53:44,810 Or a large financial firm-- 1026 00:53:44,810 --> 00:53:48,340 it could be a European, Asian, or US financial firm comes 1027 00:53:48,340 --> 00:53:50,650 into that smaller country, and they kind of 1028 00:53:50,650 --> 00:53:54,340 dominate the thinking about how to do underwriting, 1029 00:53:54,340 --> 00:53:57,460 and they're the big network effect. 1030 00:53:57,460 --> 00:53:59,500 And all of a sudden, the crisis of 2037 1031 00:53:59,500 --> 00:54:02,710 might be that everybody is extending 1032 00:54:02,710 --> 00:54:08,890 credit kind of consistently in the same way. 1033 00:54:08,890 --> 00:54:12,640 So it's basically less resilient. 1034 00:54:12,640 --> 00:54:14,460 We're living through a moment of crisis 1035 00:54:14,460 --> 00:54:17,730 right now where we're testing the resiliency of humankind 1036 00:54:17,730 --> 00:54:18,870 through the corona crisis. 1037 00:54:18,870 --> 00:54:23,960 But I'm talking one in the financial side. 1038 00:54:23,960 --> 00:54:26,120 And then the question is, how does machine learning 1039 00:54:26,120 --> 00:54:29,640 fit into current regulatory frameworks? 1040 00:54:29,640 --> 00:54:31,780 Around the globe, a lot's been written, 1041 00:54:31,780 --> 00:54:35,910 but it's all at a very high level, and it's non-binding. 1042 00:54:35,910 --> 00:54:38,340 But it's these top three, as I've mentioned. 1043 00:54:38,340 --> 00:54:41,130 So the alternatives-- this was a question earlier-- 1044 00:54:41,130 --> 00:54:44,920 is the official sector can stay neutral and say, listen, 1045 00:54:44,920 --> 00:54:46,650 this is just a tool. 1046 00:54:46,650 --> 00:54:48,240 We're still going to regulate lending. 1047 00:54:48,240 --> 00:54:49,907 We're going to regulate capital markets. 1048 00:54:49,907 --> 00:54:52,470 We're going to regulate everything the way we did. 1049 00:54:52,470 --> 00:54:55,320 And new activities will just come into those frameworks. 1050 00:54:55,320 --> 00:54:58,890 Maybe we'll clarify a little bit around the fringes. 1051 00:54:58,890 --> 00:55:00,930 Secondly, you can adjust. 1052 00:55:00,930 --> 00:55:02,070 Turn the dial. 1053 00:55:02,070 --> 00:55:05,070 When the internet came along in the 1990s, 1054 00:55:05,070 --> 00:55:07,650 at first it was like technology neutral. 1055 00:55:07,650 --> 00:55:10,350 And then pretty much every regulator around the globe 1056 00:55:10,350 --> 00:55:13,620 had to adjust. 1057 00:55:13,620 --> 00:55:16,150 What did it mean if there was an online bulletin 1058 00:55:16,150 --> 00:55:19,860 board that was trading stocks, where buyers and sellers can 1059 00:55:19,860 --> 00:55:21,100 meet? 1060 00:55:21,100 --> 00:55:23,470 Was that what's called an exchange? 1061 00:55:23,470 --> 00:55:25,570 Should it be regulated like the New York Stock 1062 00:55:25,570 --> 00:55:27,010 Exchange or London Stock Exchange, 1063 00:55:27,010 --> 00:55:29,860 or regulated maybe a little differently? 1064 00:55:29,860 --> 00:55:32,920 Where our securities regulator, where the Europeans ended up 1065 00:55:32,920 --> 00:55:37,560 in the 1990s was to regulate these online platforms 1066 00:55:37,560 --> 00:55:40,470 like exchanges but not identical. 1067 00:55:40,470 --> 00:55:45,400 So then we had a regime of fully regulated exchanges 1068 00:55:45,400 --> 00:55:48,840 and these online electronic trading platforms. 1069 00:55:48,840 --> 00:55:52,350 And that was then later adopted in Asia as well, 1070 00:55:52,350 --> 00:55:55,260 with some variations. 1071 00:55:55,260 --> 00:55:57,450 The other thing is that the official sector often 1072 00:55:57,450 --> 00:55:59,940 tries to promote this-- promote the innovations, 1073 00:55:59,940 --> 00:56:03,920 promote the technologies, or promote open banking, 1074 00:56:03,920 --> 00:56:05,810 as I've said. 1075 00:56:05,810 --> 00:56:10,010 But an interesting piece of this all is activities. 1076 00:56:10,010 --> 00:56:13,550 Should we think about machine learning as a tool, 1077 00:56:13,550 --> 00:56:15,320 like a hammer that everybody is going 1078 00:56:15,320 --> 00:56:18,770 to be using, like electricity, like the telephone 1079 00:56:18,770 --> 00:56:20,250 that everybody is using? 1080 00:56:20,250 --> 00:56:23,410 Or should we think about it, as I said earlier, some companies 1081 00:56:23,410 --> 00:56:26,650 are providing AI as a service? 1082 00:56:26,650 --> 00:56:28,960 Activities, a technology, a tool. 1083 00:56:28,960 --> 00:56:32,950 And the official sector grapples with this sometimes. 1084 00:56:32,950 --> 00:56:35,800 To date, mostly they've stayed technology 1085 00:56:35,800 --> 00:56:37,600 neutral with a little bit of promoting 1086 00:56:37,600 --> 00:56:42,010 early stage activity and the promoting of the open banking. 1087 00:56:42,010 --> 00:56:43,540 Romain questions? 1088 00:56:43,540 --> 00:56:45,500 GUEST SPEAKER: We have 15 minutes left. 1089 00:56:45,500 --> 00:56:47,620 GARY GENSLER: OK. 1090 00:56:47,620 --> 00:56:50,190 Alternative data. 1091 00:56:50,190 --> 00:56:52,550 This is data that you can extract, 1092 00:56:52,550 --> 00:56:55,190 whether it's banking and checking information, 1093 00:56:55,190 --> 00:56:59,270 or as Alibaba does in China, taking a whole cash flow 1094 00:56:59,270 --> 00:56:59,840 approach. 1095 00:56:59,840 --> 00:57:02,540 Saying I can see everything about your business. 1096 00:57:02,540 --> 00:57:04,790 It's called cash flow underwriting. 1097 00:57:04,790 --> 00:57:07,700 Here in the US, a payment company, Toast, 1098 00:57:07,700 --> 00:57:10,760 was able to do-- until restaurants closed down-- 1099 00:57:10,760 --> 00:57:14,360 able to cash flow underwriting around their restaurants 1100 00:57:14,360 --> 00:57:17,490 because they had that payment data. 1101 00:57:17,490 --> 00:57:21,600 All the way down to your app usage and browsing history. 1102 00:57:21,600 --> 00:57:23,640 I believe this is a trend that we've 1103 00:57:23,640 --> 00:57:28,020 been on that will be accelerated by the corona crisis. 1104 00:57:28,020 --> 00:57:32,120 That this crisis, we're finding whether it's 1105 00:57:32,120 --> 00:57:36,140 the large firms like Google or even smaller ones, 1106 00:57:36,140 --> 00:57:41,630 want to contribute to trying to thwart the virus by following 1107 00:57:41,630 --> 00:57:45,610 us or our location devices. 1108 00:57:45,610 --> 00:57:48,320 Our location devices, of course, are called cell phones 1109 00:57:48,320 --> 00:57:50,350 and smartphones. 1110 00:57:50,350 --> 00:57:52,720 But such vast parts of the population 1111 00:57:52,720 --> 00:57:55,840 have them that with location tracking, 1112 00:57:55,840 --> 00:58:04,510 we can possibly thwart or even contain this virus by watching 1113 00:58:04,510 --> 00:58:05,950 how we track ourselves. 1114 00:58:05,950 --> 00:58:08,410 I think that we will shift a little bit further 1115 00:58:08,410 --> 00:58:11,800 into data sharing, and that will look back 1116 00:58:11,800 --> 00:58:16,450 and 2020 will maybe be, and 2021, a pivot point. 1117 00:58:16,450 --> 00:58:18,550 But what it does mean, even in finance, 1118 00:58:18,550 --> 00:58:22,180 that a lot of this data is going to be available somewhere, 1119 00:58:22,180 --> 00:58:26,150 even more, maybe, than currently is available. 1120 00:58:26,150 --> 00:58:28,870 So there are actually alternative data fintech 1121 00:58:28,870 --> 00:58:29,440 companies. 1122 00:58:29,440 --> 00:58:33,640 CB Insights, which is a leader in tracking fintech, 1123 00:58:33,640 --> 00:58:35,080 puts together this chart. 1124 00:58:35,080 --> 00:58:37,630 And we don't have time to go through these companies, 1125 00:58:37,630 --> 00:58:41,650 but these are companies that are sort of marketing themselves 1126 00:58:41,650 --> 00:58:44,290 in this alternative data set, almost 1127 00:58:44,290 --> 00:58:49,450 like capturing the data, and then data and AI as a service. 1128 00:58:51,970 --> 00:58:54,510 I want to just talk about Apple Credit Card for a second 1129 00:58:54,510 --> 00:58:57,390 because it's where you can also stub your toe. 1130 00:58:57,390 --> 00:59:00,510 Apple Credit Card with a big rollout in conjunction with 1131 00:59:00,510 --> 00:59:03,720 Goldman Sachs' Marcus and MasterCard-- 1132 00:59:03,720 --> 00:59:09,230 so it's a really interesting combination of big tech, 1133 00:59:09,230 --> 00:59:11,270 big finance together-- 1134 00:59:11,270 --> 00:59:13,130 rolls out a credit card product. 1135 00:59:13,130 --> 00:59:15,240 I think it was in November-- 1136 00:59:15,240 --> 00:59:18,620 yeah, November of this past year. 1137 00:59:18,620 --> 00:59:23,480 And in rolling it out in a very proud rollout, 1138 00:59:23,480 --> 00:59:29,480 an entrepreneur, here going by the Twitter account DHH, which 1139 00:59:29,480 --> 00:59:33,710 you might know of this, goes on and finds 1140 00:59:33,710 --> 00:59:36,740 that he is provided greater credit than his wife, 1141 00:59:36,740 --> 00:59:39,750 and he and his wife are both billionaires, 1142 00:59:39,750 --> 00:59:43,110 like literally worth a lot of money. 1143 00:59:43,110 --> 00:59:46,560 Maybe it was only they were centimillionaires, but worth 1144 00:59:46,560 --> 00:59:47,820 a lot of money. 1145 00:59:47,820 --> 00:59:50,850 Joint tax account, joint assets, and he 1146 00:59:50,850 --> 00:59:55,650 was being provided 10 to 20 times more credit. 1147 00:59:55,650 --> 00:59:58,960 So he took to the Twittersphere and sort of made this. 1148 00:59:58,960 --> 01:00:02,850 And he says his wife had spoke to two Apple reps, 1149 01:00:02,850 --> 01:00:04,990 both very nice. 1150 01:00:04,990 --> 01:00:08,020 But basically saying, I don't know why. 1151 01:00:08,020 --> 01:00:09,220 It's just the algorithm. 1152 01:00:09,220 --> 01:00:11,820 It's just the algorithm. 1153 01:00:11,820 --> 01:00:15,490 What really hurt Apple even more than that was within days, 1154 01:00:15,490 --> 01:00:18,970 the next day, Steve Wozniak, one of the co-founders of Apple, 1155 01:00:18,970 --> 01:00:20,440 put this Twitter out. 1156 01:00:20,440 --> 01:00:23,080 "I'm a current Apple employee and the founder, 1157 01:00:23,080 --> 01:00:25,340 and the same thing happened to us, 10 times." 1158 01:00:25,340 --> 01:00:29,590 Meaning Steve got 10 times the credit as his wife 1159 01:00:29,590 --> 01:00:31,480 in the same algorithm. 1160 01:00:31,480 --> 01:00:35,120 "Some say to blame Goldman Sachs," et cetera, et cetera. 1161 01:00:35,120 --> 01:00:36,860 But Apple shares the responsibility. 1162 01:00:36,860 --> 01:00:39,500 Not a good rollout. 1163 01:00:39,500 --> 01:00:42,170 So for Apple Credit Card, they'll survive. 1164 01:00:42,170 --> 01:00:43,760 Apple is a big company. 1165 01:00:43,760 --> 01:00:47,240 They'll probably fix these biases, but not 1166 01:00:47,240 --> 01:00:50,660 a particularly good rollout, as you can well imagine, 1167 01:00:50,660 --> 01:00:51,830 in their models. 1168 01:00:51,830 --> 01:00:52,928 Romain. 1169 01:00:52,928 --> 01:00:54,470 GUEST SPEAKER: Alida has her hand up. 1170 01:00:54,470 --> 01:00:55,637 GARY GENSLER: Please, Alida. 1171 01:00:58,282 --> 01:01:02,140 AUDIENCE: Yes, you mentioned cash flow lending earlier. 1172 01:01:02,140 --> 01:01:06,100 And that really falls under the merchant cash advance business. 1173 01:01:06,100 --> 01:01:08,363 There's been a lot of debate about that being-- 1174 01:01:08,363 --> 01:01:09,780 it's not very regulated right now, 1175 01:01:09,780 --> 01:01:11,940 but that becoming more regulated. 1176 01:01:11,940 --> 01:01:17,780 Is an entry point like Toast and other companies that have just 1177 01:01:17,780 --> 01:01:19,090 launched these new products-- 1178 01:01:19,090 --> 01:01:21,770 [INAUDIBLE] speed up the regulations 1179 01:01:21,770 --> 01:01:23,340 around that business? 1180 01:01:23,340 --> 01:01:28,710 GARY GENSLER: So Romain or Alida I missed a word. 1181 01:01:28,710 --> 01:01:31,746 Which company did you say in there? 1182 01:01:31,746 --> 01:01:32,500 AUDIENCE: Toast. 1183 01:01:32,500 --> 01:01:33,750 GARY GENSLER: Oh, Toast, OK. 1184 01:01:33,750 --> 01:01:36,740 So Toast-- for people to be familiar, 1185 01:01:36,740 --> 01:01:41,050 Toast started in the restaurant payment space. 1186 01:01:41,050 --> 01:01:45,740 And it was basically trying to provide hardware, tablets. 1187 01:01:45,740 --> 01:01:48,060 They thought that the point of sale 1188 01:01:48,060 --> 01:01:53,215 would be facilitated if servers had a tablet. 1189 01:01:53,215 --> 01:01:55,590 And so they were sort of in the hardware, software space. 1190 01:01:55,590 --> 01:01:57,757 They found themselves getting into the payment space 1191 01:01:57,757 --> 01:02:00,270 very quickly, and then, of course, data. 1192 01:02:00,270 --> 01:02:03,630 And with that data, they could do a whole cash flow 1193 01:02:03,630 --> 01:02:04,230 underwriting. 1194 01:02:04,230 --> 01:02:06,030 And then they started Toast Capital, 1195 01:02:06,030 --> 01:02:08,430 where they would make loans, small business loans, 1196 01:02:08,430 --> 01:02:10,810 to these restaurants. 1197 01:02:10,810 --> 01:02:15,150 And I think they have 25,000 or 30,000 restaurants that 1198 01:02:15,150 --> 01:02:16,400 are in their client list. 1199 01:02:16,400 --> 01:02:19,530 And they did a round C funding earlier this year 1200 01:02:19,530 --> 01:02:20,920 at $4.9 billion. 1201 01:02:20,920 --> 01:02:25,010 So this is working, of course until the crisis. 1202 01:02:25,010 --> 01:02:27,480 And so then the question is about regulation 1203 01:02:27,480 --> 01:02:29,880 about cash flow underwriting. 1204 01:02:29,880 --> 01:02:34,290 I'm not familiar enough with how Toast 1205 01:02:34,290 --> 01:02:37,780 feels, even though I've met the founders and things like that. 1206 01:02:37,780 --> 01:02:39,720 They're a Boston company. 1207 01:02:39,720 --> 01:02:43,620 But I think they're dealing with the same set of regulations 1208 01:02:43,620 --> 01:02:46,350 that everyone is, which I'm going to turn to right now 1209 01:02:46,350 --> 01:02:48,750 in the last eight minutes. 1210 01:02:48,750 --> 01:02:50,600 But cash flow underwriting, Alida, 1211 01:02:50,600 --> 01:02:53,100 are you worried about something specifically 1212 01:02:53,100 --> 01:02:54,570 about cash flow underwriting? 1213 01:02:54,570 --> 01:02:59,523 Because maybe I'll learn from your concern. 1214 01:02:59,523 --> 01:03:01,190 AUDIENCE: I [INAUDIBLE] if you look at-- 1215 01:03:01,190 --> 01:03:05,080 I would consider that to be like a merchant cash advance 1216 01:03:05,080 --> 01:03:06,720 product, and those are not actually 1217 01:03:06,720 --> 01:03:10,890 considered loans in the regulations. 1218 01:03:10,890 --> 01:03:14,370 Now, there's a lot of movement toward having 1219 01:03:14,370 --> 01:03:16,920 those products being considered loans and then 1220 01:03:16,920 --> 01:03:22,470 fall under different regulatory standards that [INAUDIBLE].. 1221 01:03:22,470 --> 01:03:25,110 GARY GENSLER: So what Alida is raising-- 1222 01:03:25,110 --> 01:03:25,770 I'm sorry. 1223 01:03:25,770 --> 01:03:31,890 What Alida raising is, again, in every political and regulatory 1224 01:03:31,890 --> 01:03:33,960 process, there is some definition 1225 01:03:33,960 --> 01:03:37,050 of what falls within a regulation, what falls out. 1226 01:03:37,050 --> 01:03:39,600 You might think, where are the borders and boundaries 1227 01:03:39,600 --> 01:03:41,760 of a regulatory environment? 1228 01:03:41,760 --> 01:03:44,790 What's defined, really, as a security 1229 01:03:44,790 --> 01:03:46,410 in the cryptocurrency space? 1230 01:03:46,410 --> 01:03:49,500 What's defined as an exchange, an exchange regulation? 1231 01:03:49,500 --> 01:03:53,110 And hear Alida is saying, what's defined as a loan? 1232 01:03:53,110 --> 01:03:56,250 And in Toast's case, doing cash flow underwriting, 1233 01:03:56,250 --> 01:04:00,930 it might be considered a cash advance rather than a loan. 1234 01:04:00,930 --> 01:04:03,540 So let me do a little research on that, 1235 01:04:03,540 --> 01:04:06,030 and we'll come back to that maybe in a future class 1236 01:04:06,030 --> 01:04:09,180 when we talk about payments. 1237 01:04:09,180 --> 01:04:12,330 But in terms of the consumer credit law environment, 1238 01:04:12,330 --> 01:04:15,120 we talked about the Equal Credit Opportunity Act. 1239 01:04:15,120 --> 01:04:18,240 The key thing is not only whether you 1240 01:04:18,240 --> 01:04:19,950 have disparate treatment, but whether you 1241 01:04:19,950 --> 01:04:22,300 have disparate impact. 1242 01:04:22,300 --> 01:04:26,250 So back to my retreading analysis, if for some reason 1243 01:04:26,250 --> 01:04:29,430 you've been reviewing and your machine algorithms say, 1244 01:04:29,430 --> 01:04:33,810 aha, all these folks that are getting retreads 1245 01:04:33,810 --> 01:04:37,710 should have lower credit, that might be OK 1246 01:04:37,710 --> 01:04:42,430 unless you find that you're treating 1247 01:04:42,430 --> 01:04:46,030 different protected classes differently. 1248 01:04:46,030 --> 01:04:48,558 Are you treating people of different backgrounds 1249 01:04:48,558 --> 01:04:50,350 differently, different genders differently, 1250 01:04:50,350 --> 01:04:55,390 as Apple certainly was in Steve Wozniak and his wife? 1251 01:04:55,390 --> 01:04:58,270 Fair Housing Act, Fair Credit Reporting Act-- the Fair 1252 01:04:58,270 --> 01:05:01,210 Housing Act has a lot of these same protected class 1253 01:05:01,210 --> 01:05:02,260 perspectives. 1254 01:05:02,260 --> 01:05:05,050 Fair Credit Reporting-- and you read this in the Mayer Brown 1255 01:05:05,050 --> 01:05:06,010 piece-- 1256 01:05:06,010 --> 01:05:08,140 the Fair Credit Reporting Act, you 1257 01:05:08,140 --> 01:05:11,770 can find yourself as a fintech company or a data aggregator. 1258 01:05:11,770 --> 01:05:14,200 The Plaids and the other data aggregators 1259 01:05:14,200 --> 01:05:17,110 could find that they were, in fact, coming under the Fair 1260 01:05:17,110 --> 01:05:18,730 Credit Reporting Act. 1261 01:05:18,730 --> 01:05:22,840 They were those vendors that Cap One might be using. 1262 01:05:22,840 --> 01:05:25,840 And they, the vendor, might become Fair Credit Reporting 1263 01:05:25,840 --> 01:05:27,110 Act companies. 1264 01:05:27,110 --> 01:05:28,960 And there is usually a boundary there. 1265 01:05:28,960 --> 01:05:32,410 Again, this is not a law class, but these are to highlight. 1266 01:05:32,410 --> 01:05:35,690 States also have Unfair and Deceptive Acts and Practices 1267 01:05:35,690 --> 01:05:36,190 Act. 1268 01:05:36,190 --> 01:05:38,860 When I was chairman of the Maryland Consumer Financial 1269 01:05:38,860 --> 01:05:41,980 Protection Commission, we went to the state legislature 1270 01:05:41,980 --> 01:05:42,520 in Maryland. 1271 01:05:42,520 --> 01:05:44,350 This is a year and a half ago. 1272 01:05:44,350 --> 01:05:48,970 And said that Maryland's Unfair and Deceptive Practices 1273 01:05:48,970 --> 01:05:53,030 Act, UDAP, should be updated to include abusive. 1274 01:05:53,030 --> 01:05:57,420 So we sort of broadened it a little bit. 1275 01:05:57,420 --> 01:06:01,450 And then privacy laws. 1276 01:06:01,450 --> 01:06:03,850 This should say general direct-- 1277 01:06:03,850 --> 01:06:11,110 general-- I'll correct the GDPR, but Protection Regulation, 1278 01:06:11,110 --> 01:06:12,700 and then in the US. 1279 01:06:12,700 --> 01:06:15,710 And those are the buckets that really matter. 1280 01:06:15,710 --> 01:06:21,022 Sort of trying to close out, AI, finance, and geopolitics. 1281 01:06:21,022 --> 01:06:24,210 We've got nearly 200 countries. 1282 01:06:24,210 --> 01:06:28,170 And those 200 countries, we're interconnected. 1283 01:06:28,170 --> 01:06:31,540 We're very interconnected globally. 1284 01:06:31,540 --> 01:06:33,600 And we've got a lot of standards setters, 1285 01:06:33,600 --> 01:06:35,490 but those standards setters do not 1286 01:06:35,490 --> 01:06:38,440 have the authority of lawmakers. 1287 01:06:38,440 --> 01:06:42,360 So whether it's the Organization of Economic-- 1288 01:06:42,360 --> 01:06:47,400 OECD, or the guidelines of things like the securities 1289 01:06:47,400 --> 01:06:51,690 regulators and the anti-crime regulators or the banking 1290 01:06:51,690 --> 01:06:55,500 regulators, these are not enforceable standards. 1291 01:06:55,500 --> 01:06:58,980 So what we have is competing models for AI, finance, 1292 01:06:58,980 --> 01:07:01,610 and policy. 1293 01:07:01,610 --> 01:07:03,770 And I note that because if you're 1294 01:07:03,770 --> 01:07:05,330 thinking about starting a company 1295 01:07:05,330 --> 01:07:08,593 and you operate globally, sometimes the global law 1296 01:07:08,593 --> 01:07:09,260 will affect you. 1297 01:07:09,260 --> 01:07:12,410 GDPR from Europe has already affected how 1298 01:07:12,410 --> 01:07:14,480 we deal with privacy in the US. 1299 01:07:14,480 --> 01:07:18,230 California institutes the California Consumer Protection 1300 01:07:18,230 --> 01:07:21,830 Act, it influences the whole country. 1301 01:07:21,830 --> 01:07:23,910 So sometimes that works that way. 1302 01:07:23,910 --> 01:07:24,873 Romain. 1303 01:07:24,873 --> 01:07:26,790 GUEST SPEAKER: We have a question from Akshay. 1304 01:07:26,790 --> 01:07:28,060 GARY GENSLER: Please, Akshay. 1305 01:07:28,060 --> 01:07:29,060 AUDIENCE: Hi, professor. 1306 01:07:29,060 --> 01:07:33,130 So the gender bias algorithm that you mentioned 1307 01:07:33,130 --> 01:07:38,750 about the Apple Credit Card, so the only thing that we 1308 01:07:38,750 --> 01:07:41,300 can control here is the data that we're using. 1309 01:07:41,300 --> 01:07:44,570 If we are not using any gender data, 1310 01:07:44,570 --> 01:07:48,800 and if algorithm turns out and is creating biases 1311 01:07:48,800 --> 01:07:54,350 without even using a particular data which is considered 1312 01:07:54,350 --> 01:07:57,410 racist or sexist, so would that be 1313 01:07:57,410 --> 01:08:02,970 counted as breaking the laws? 1314 01:08:02,970 --> 01:08:05,550 GARY GENSLER: So here-- it's a great question, Akshay. 1315 01:08:05,550 --> 01:08:08,290 And again, I caution that this is not a legal class. 1316 01:08:08,290 --> 01:08:13,380 But embedded in the US law and often in other laws 1317 01:08:13,380 --> 01:08:17,920 is this concept about disparate treatment and disparate impact. 1318 01:08:17,920 --> 01:08:23,130 And what you're asking is, what if you didn't mean to do it? 1319 01:08:23,130 --> 01:08:26,670 What if there was no intent, and it's just-- wow, 1320 01:08:26,670 --> 01:08:29,819 that you've extracted this correlation and all of a sudden 1321 01:08:29,819 --> 01:08:31,979 there's disparate impact? 1322 01:08:31,979 --> 01:08:34,580 You could have a problem in a court. 1323 01:08:34,580 --> 01:08:39,300 And it's a very established 50-year-old-- 1324 01:08:39,300 --> 01:08:41,819 there's a lot of case law around, 1325 01:08:41,819 --> 01:08:44,189 when would a disparate impact cause you 1326 01:08:44,189 --> 01:08:46,710 those headaches and anxiety? 1327 01:08:46,710 --> 01:08:49,470 And it relates a lot to explainability. 1328 01:08:49,470 --> 01:08:52,010 If you can come back to explainability 1329 01:08:52,010 --> 01:08:56,100 and you can truly lay out, this is why, 1330 01:08:56,100 --> 01:08:57,689 and it has nothing to do with gender, 1331 01:08:57,689 --> 01:09:01,529 nothing to do with race, sexual orientation, and backgrounds, 1332 01:09:01,529 --> 01:09:04,667 and so forth, a protected class, you're 1333 01:09:04,667 --> 01:09:06,250 going to be better in that court case. 1334 01:09:06,250 --> 01:09:12,899 But it would be far better to have no disparate impact. 1335 01:09:12,899 --> 01:09:17,800 Then you're in a much broadly more safe area. 1336 01:09:17,800 --> 01:09:18,590 AUDIENCE: Got it. 1337 01:09:18,590 --> 01:09:20,600 Thank you. 1338 01:09:20,600 --> 01:09:22,370 GARY GENSLER: Romain other questions? 1339 01:09:22,370 --> 01:09:25,180 GUEST SPEAKER: Perhaps one last question from Luke. 1340 01:09:25,180 --> 01:09:28,146 GARY GENSLER: Oh my God, Luke, you're always in there. 1341 01:09:28,146 --> 01:09:29,479 AUDIENCE: I just had a question. 1342 01:09:29,479 --> 01:09:31,300 GARY GENSLER: Before Luke goes, is there anybody else? 1343 01:09:31,300 --> 01:09:32,604 Just, I want to make sure. 1344 01:09:32,604 --> 01:09:33,990 OK, Luke, you can go. 1345 01:09:33,990 --> 01:09:35,323 AUDIENCE: It was not a question. 1346 01:09:35,323 --> 01:09:37,899 It was a comment to Akshay's question. 1347 01:09:37,899 --> 01:09:39,910 I'm sure he doesn't support this. 1348 01:09:39,910 --> 01:09:42,069 But what he asked is, isn't it the same thing 1349 01:09:42,069 --> 01:09:46,180 if a person gives a racist or a misogynist comment or a hate 1350 01:09:46,180 --> 01:09:47,500 crime comment? 1351 01:09:47,500 --> 01:09:51,399 And if they didn't know about it, is he liable for it? 1352 01:09:51,399 --> 01:09:53,770 Should a corporate be held responsible 1353 01:09:53,770 --> 01:09:56,348 in the same way a human would be? 1354 01:09:56,348 --> 01:09:58,390 GARY GENSLER: Well, I don't know if that's really 1355 01:09:58,390 --> 01:10:03,430 where Akshay was going, but I see your point is that 1356 01:10:03,430 --> 01:10:06,208 basically-- and it depends on the country, Akshay and Luke. 1357 01:10:06,208 --> 01:10:07,750 It really does depend on the country. 1358 01:10:07,750 --> 01:10:10,240 But here in the US, we'd have this conceptual framework 1359 01:10:10,240 --> 01:10:14,550 [? of ?] disparate treatment, disparate impact. 1360 01:10:14,550 --> 01:10:17,010 And then explainability is from another law. 1361 01:10:17,010 --> 01:10:22,910 But it really should be based on fairness and inclusion. 1362 01:10:22,910 --> 01:10:27,710 Everybody's got the same fair shot, regardless 1363 01:10:27,710 --> 01:10:32,970 of where the data comes from at all. 1364 01:10:32,970 --> 01:10:37,900 So I think that sort of-- we're almost out of time.