1 00:00:00,240 --> 00:00:03,800 Speaker 1: In this episode, we'll be focusing on artificial intelligence, especially 2 00:00:03,840 --> 00:00:08,039 Speaker 1: the use of natural language processing and virtual assistance powered 3 00:00:08,039 --> 00:00:11,560 Speaker 1: by Watson. To explore this topic in depth, we're going 4 00:00:11,600 --> 00:00:15,520 Speaker 1: to share two conversations we recorded with leaders at IBM. 5 00:00:15,560 --> 00:00:19,040 Speaker 1: The first is Rittka Gunner, who is Vice president for 6 00:00:19,120 --> 00:00:22,599 Speaker 1: ibm S Data and AI Expert Labs and Learning, and 7 00:00:22,640 --> 00:00:25,319 Speaker 1: the second chat will be with Jay Bellissimo, who is 8 00:00:25,320 --> 00:00:31,600 Speaker 1: IBMS General Manager for the U S Public and Federal market. Rittica, 9 00:00:31,640 --> 00:00:34,120 Speaker 1: thanks so much for joining us today. So to start off, 10 00:00:34,159 --> 00:00:36,760 Speaker 1: can you introduce yourself and talk a little bit about 11 00:00:36,800 --> 00:00:39,640 Speaker 1: your role at IBM. Yeah, thanks, Joe and Robert. This 12 00:00:39,760 --> 00:00:42,600 Speaker 1: is a pleasure to be here. My name is Ritica 13 00:00:42,600 --> 00:00:48,960 Speaker 1: Gunner and iron a organization called Data and AI Expert 14 00:00:49,280 --> 00:00:52,559 Speaker 1: Labs and Learning, and our whole mission is to be 15 00:00:52,600 --> 00:00:57,640 Speaker 1: able to help clients be really understanding of what it 16 00:00:57,680 --> 00:01:00,960 Speaker 1: means to adopt data and AI technology GS how we 17 00:01:01,040 --> 00:01:06,600 Speaker 1: help them accelerate UM the use of data and AI 18 00:01:06,720 --> 00:01:12,639 Speaker 1: across their organizations through the methodologies and through the skills 19 00:01:12,680 --> 00:01:15,040 Speaker 1: and expertise that we have working with thousands of clients, 20 00:01:15,040 --> 00:01:18,280 Speaker 1: so that is they're embarking on their AI journey they 21 00:01:18,319 --> 00:01:21,400 Speaker 1: know how to be able to do that quickly. Excellent. Now, 22 00:01:21,440 --> 00:01:23,440 Speaker 1: in this episode, we're obviously going to be chatting quite 23 00:01:23,480 --> 00:01:26,600 Speaker 1: a bit about AI. So just to ground our listeners 24 00:01:26,600 --> 00:01:29,760 Speaker 1: in the right place, can you define artificial intelligence for 25 00:01:29,840 --> 00:01:31,920 Speaker 1: us and describe the sort of a I will be 26 00:01:31,959 --> 00:01:35,920 Speaker 1: discussing here today. Yeah, so artificial intelligence, many of us 27 00:01:35,959 --> 00:01:39,880 Speaker 1: know the hype around it, but very simply, artificial intelligence 28 00:01:39,920 --> 00:01:44,640 Speaker 1: is about teaching machines to interact and think and make 29 00:01:44,680 --> 00:01:48,480 Speaker 1: decisions like humans will. So it's around us every day 30 00:01:48,560 --> 00:01:53,480 Speaker 1: when you look at how machines can actually see things 31 00:01:53,680 --> 00:01:56,720 Speaker 1: like humans do when they can speak. Many of us 32 00:01:56,760 --> 00:02:03,200 Speaker 1: have those UM consumer type applications like Alexa, Google Home. UM, 33 00:02:03,240 --> 00:02:06,240 Speaker 1: they're even UM enterprise type versions which I'm gonna talk 34 00:02:06,280 --> 00:02:09,600 Speaker 1: about today with Watson and Watson Assistant, where you can 35 00:02:09,639 --> 00:02:13,640 Speaker 1: have assistance either in the home or for commercial use. UM, 36 00:02:13,680 --> 00:02:17,360 Speaker 1: whether it's actually doing any types of areas where you 37 00:02:17,440 --> 00:02:22,960 Speaker 1: can predict or optimize or automate kind of working decisions 38 00:02:23,040 --> 00:02:28,200 Speaker 1: using artificial intelligence. Excellent. Now, you mentioned Watson AI and 39 00:02:28,280 --> 00:02:31,840 Speaker 1: Watson Assistant. UM, can you ground um those two for 40 00:02:32,000 --> 00:02:34,200 Speaker 1: us as well to tell us you know what exactly 41 00:02:34,200 --> 00:02:38,000 Speaker 1: Watson AI is and as much as is feasible at 42 00:02:38,000 --> 00:02:41,359 Speaker 1: this point in the interview, UM, what Watson Assistant is 43 00:02:41,800 --> 00:02:48,880 Speaker 1: sure UM, Watson AI is leveraging UM technologies for AI 44 00:02:49,040 --> 00:02:52,639 Speaker 1: for most enterprises and their essential workloads. When you look 45 00:02:52,639 --> 00:02:56,560 Speaker 1: at it, we've done thousands of AI projects UM, all 46 00:02:56,600 --> 00:03:01,000 Speaker 1: across the world, across almost every industy tree you can imagine, 47 00:03:01,040 --> 00:03:05,000 Speaker 1: and almost every country, and through that we've learned what 48 00:03:05,040 --> 00:03:09,239 Speaker 1: it means for clients to truly put artificial intelligence and 49 00:03:09,480 --> 00:03:12,680 Speaker 1: their most essential types of workloads. Is part of that. 50 00:03:12,760 --> 00:03:16,560 Speaker 1: We've developed Watson AI capabilities. Some of you may know 51 00:03:16,760 --> 00:03:18,799 Speaker 1: the simple kind of a p I s that are 52 00:03:18,840 --> 00:03:23,240 Speaker 1: like vision, speech, natural language processing, but it's also more 53 00:03:23,280 --> 00:03:24,880 Speaker 1: than that, and I'm going to talk about that in 54 00:03:24,960 --> 00:03:29,440 Speaker 1: terms of the applications that we see many users leveraging 55 00:03:29,480 --> 00:03:32,560 Speaker 1: AI in accelerating how they're actually doing this. So we're 56 00:03:32,560 --> 00:03:35,640 Speaker 1: going to talk about things like Watson Assistant, what we 57 00:03:35,720 --> 00:03:40,160 Speaker 1: have in terms of Watson capabilities for the financial services organizations, 58 00:03:40,280 --> 00:03:42,440 Speaker 1: or what we do to be able to apply AI 59 00:03:42,520 --> 00:03:45,080 Speaker 1: to any kind of industry. As we do that in 60 00:03:45,120 --> 00:03:48,640 Speaker 1: a way where we can actually prepackage that application. We 61 00:03:48,720 --> 00:03:52,680 Speaker 1: have Watson AI applications, Watson Assistant is one of those. 62 00:03:53,000 --> 00:03:56,160 Speaker 1: Think about this not just as a simple chatbot, but 63 00:03:56,440 --> 00:04:01,200 Speaker 1: it really is about letting users have the ability to 64 00:04:01,320 --> 00:04:06,640 Speaker 1: have an assistant that understands how to respond to questions 65 00:04:06,680 --> 00:04:10,000 Speaker 1: that you've been trained on or that it has learned 66 00:04:10,040 --> 00:04:14,080 Speaker 1: through data that it's seen previously on not just really 67 00:04:14,120 --> 00:04:17,599 Speaker 1: simple questions, but really hard questions that may be sitting 68 00:04:17,600 --> 00:04:22,640 Speaker 1: in documents that are buried inside your organization, whether that 69 00:04:22,720 --> 00:04:26,360 Speaker 1: be you know, your playbooks, whether that be UM, your 70 00:04:26,480 --> 00:04:28,680 Speaker 1: run books, that you have to be able to answer questions. 71 00:04:28,800 --> 00:04:32,240 Speaker 1: It really is about answering not only the simple questions, 72 00:04:32,560 --> 00:04:35,560 Speaker 1: but training AI to answer even on the most complex 73 00:04:35,640 --> 00:04:38,359 Speaker 1: questions as well. So today we're going to be focusing 74 00:04:38,360 --> 00:04:41,200 Speaker 1: a lot on how AI is being used in the 75 00:04:41,480 --> 00:04:43,920 Speaker 1: current situation that we're facing in the world. Can can 76 00:04:43,960 --> 00:04:46,919 Speaker 1: you talk a bit to ground the problem about UM 77 00:04:46,960 --> 00:04:49,680 Speaker 1: some of the ways that the COVID nineteen pandemic is 78 00:04:49,720 --> 00:04:55,880 Speaker 1: affecting how the public interacts with businesses and institutions. Yeah, 79 00:04:55,920 --> 00:04:58,760 Speaker 1: I'll give you a little bit of a background. Three 80 00:04:58,800 --> 00:05:02,600 Speaker 1: out of four users and they have a problem, any problem, 81 00:05:02,800 --> 00:05:05,200 Speaker 1: they want to be able to answer that problem on 82 00:05:05,240 --> 00:05:07,719 Speaker 1: their own. That's why we use things like our phones, 83 00:05:08,240 --> 00:05:12,160 Speaker 1: our web internets, are our voice assistants that exist there. 84 00:05:12,160 --> 00:05:14,719 Speaker 1: Three out of four people want to be able to 85 00:05:15,120 --> 00:05:18,000 Speaker 1: find an answer without talking to someone and to be 86 00:05:18,040 --> 00:05:20,480 Speaker 1: able to do it on their own. When you look 87 00:05:20,520 --> 00:05:24,240 Speaker 1: at those users, um, two out of three of them 88 00:05:24,279 --> 00:05:27,880 Speaker 1: actually consume three or more channels. That's you know, using 89 00:05:27,920 --> 00:05:32,719 Speaker 1: your web, using your voice assistance, et cetera. So there 90 00:05:32,880 --> 00:05:35,760 Speaker 1: is a large demand for people to be able to 91 00:05:35,800 --> 00:05:38,919 Speaker 1: find answers to questions on their own and to be 92 00:05:38,960 --> 00:05:43,080 Speaker 1: able to do so in multiple mediums. Now, um, when 93 00:05:43,080 --> 00:05:47,000 Speaker 1: we look at what that means on the back end today, 94 00:05:47,040 --> 00:05:51,400 Speaker 1: we have multiple customer service agents that are answering a 95 00:05:51,400 --> 00:05:55,719 Speaker 1: lot of those questions, and over of those those customer 96 00:05:55,720 --> 00:06:00,200 Speaker 1: service agents don't have answers at hand, and so that 97 00:06:00,240 --> 00:06:04,039 Speaker 1: becomes quite frustrating for end users who really want answers 98 00:06:04,040 --> 00:06:08,520 Speaker 1: at their fingertips. So you know this, the situation that 99 00:06:08,560 --> 00:06:12,320 Speaker 1: we're embarking on is the use of more pre training 100 00:06:12,400 --> 00:06:15,480 Speaker 1: kind of AI capabilities to be able to answer those questions. 101 00:06:16,440 --> 00:06:19,840 Speaker 1: This pandemic that we have with COVID nineteen is no different. 102 00:06:20,000 --> 00:06:22,160 Speaker 1: I want you to think about the amount of uncertainty 103 00:06:22,200 --> 00:06:25,960 Speaker 1: that's there in the world because of COVID nineteen. You know, 104 00:06:25,960 --> 00:06:28,320 Speaker 1: a few months ago, at the very beginning, it was 105 00:06:28,760 --> 00:06:33,240 Speaker 1: what are the symptoms of COVID nineteen? UM, what do 106 00:06:33,480 --> 00:06:35,800 Speaker 1: how do I tell whether my own child may have 107 00:06:35,880 --> 00:06:39,240 Speaker 1: COVID nineteen? And what was happening is a lot of 108 00:06:39,240 --> 00:06:42,839 Speaker 1: the call volumes that were coming into the hospitals the 109 00:06:42,880 --> 00:06:47,719 Speaker 1: government organizations were overwhelming the system such that doctors and 110 00:06:47,880 --> 00:06:52,200 Speaker 1: nurses and public servants were spending a lot of time 111 00:06:52,240 --> 00:06:55,880 Speaker 1: answering questions versus facing the pandemic itself. And so we 112 00:06:55,960 --> 00:06:59,599 Speaker 1: saw a huge surge in request to be able to 113 00:06:59,760 --> 00:07:03,560 Speaker 1: use Who's AI powered assistants like Watson Assistant to be 114 00:07:03,640 --> 00:07:08,520 Speaker 1: able to answer those questions. And it has expanded from 115 00:07:08,560 --> 00:07:12,200 Speaker 1: just help me understand what the symptoms are, help me 116 00:07:12,280 --> 00:07:15,720 Speaker 1: understand you know where I can go get tested too? 117 00:07:16,000 --> 00:07:19,360 Speaker 1: Now other things that are downstream, like you know, what 118 00:07:19,400 --> 00:07:23,120 Speaker 1: are unemployment benefits for my state? How do I, UM, 119 00:07:23,560 --> 00:07:26,680 Speaker 1: how do I actually apply for a small business loan? 120 00:07:27,120 --> 00:07:30,160 Speaker 1: And so the demand in times that are so uncertain, 121 00:07:30,600 --> 00:07:35,240 Speaker 1: especially when you look at how every hospital, every county, 122 00:07:35,320 --> 00:07:40,640 Speaker 1: every country has their own um types of of regulations 123 00:07:40,800 --> 00:07:44,240 Speaker 1: or or or rules, a I become something that's really 124 00:07:44,320 --> 00:07:48,440 Speaker 1: powerful and that's what we've seen UM through this pandemic. 125 00:07:48,920 --> 00:07:52,880 Speaker 1: UM we have an offer with Watson Assistant that we've 126 00:07:52,920 --> 00:07:57,000 Speaker 1: put out there where we are making our technology is 127 00:07:57,040 --> 00:08:01,360 Speaker 1: available for many of these organizations are dealing with this pandemic, 128 00:08:01,920 --> 00:08:05,200 Speaker 1: and UM we are. We are trying to put these 129 00:08:05,240 --> 00:08:08,640 Speaker 1: technologies up usually in less than one day, not just 130 00:08:08,800 --> 00:08:12,800 Speaker 1: with an assistant, but a voice integrated assistant, and do 131 00:08:12,880 --> 00:08:16,160 Speaker 1: that to where our our clients can get up and 132 00:08:16,240 --> 00:08:20,680 Speaker 1: running in less than a day servicing their constituents, deflecting 133 00:08:20,920 --> 00:08:24,320 Speaker 1: up to the call volumes that usually would come into 134 00:08:24,360 --> 00:08:26,240 Speaker 1: a call center. But you have to realize that the 135 00:08:26,280 --> 00:08:30,640 Speaker 1: call centers themselves have their employees working from home and 136 00:08:30,680 --> 00:08:34,440 Speaker 1: are constrained themselves. Let me go through a few examples 137 00:08:34,679 --> 00:08:37,080 Speaker 1: that we've seen during this pandemic that I think that 138 00:08:37,559 --> 00:08:40,120 Speaker 1: will relate to not only you, but to the audience. 139 00:08:40,760 --> 00:08:44,960 Speaker 1: We have UM hundreds of assistance that are now alive 140 00:08:45,160 --> 00:08:49,000 Speaker 1: over twenty countries and a lot of them are responding 141 00:08:49,040 --> 00:08:53,640 Speaker 1: to real time UM questions around COVID nineteen. I'll give 142 00:08:53,679 --> 00:08:57,680 Speaker 1: you one example, which is my home state or my 143 00:08:57,800 --> 00:09:02,839 Speaker 1: home city of Austin, Texas. We have a Watson Assistant 144 00:09:02,920 --> 00:09:05,640 Speaker 1: on their home page for the City of Austin, providing 145 00:09:05,760 --> 00:09:10,920 Speaker 1: instant answers to their citizens about UM, the COVID nineteen 146 00:09:11,040 --> 00:09:14,400 Speaker 1: situation in Austin, where people can go get tested, and 147 00:09:14,440 --> 00:09:18,120 Speaker 1: the most up to date information there. Another example is 148 00:09:18,240 --> 00:09:21,120 Speaker 1: the Children's Healthcare of Atlanta. If you think about it, 149 00:09:21,520 --> 00:09:25,080 Speaker 1: parents are really worried about what does it mean for 150 00:09:25,120 --> 00:09:27,880 Speaker 1: their children? Do their children actually see a lot of 151 00:09:27,880 --> 00:09:31,080 Speaker 1: these symptoms UM and if they do, where should they 152 00:09:31,080 --> 00:09:33,320 Speaker 1: go to bring them in? We worked with the Children's 153 00:09:33,400 --> 00:09:35,720 Speaker 1: Healthcare of Atlanta to be able to put not just 154 00:09:35,880 --> 00:09:39,680 Speaker 1: any assistant, but a voice activated assistant on their site 155 00:09:40,040 --> 00:09:43,160 Speaker 1: within a weekend UM to be able to be up 156 00:09:43,160 --> 00:09:45,800 Speaker 1: and running. As I mentioned, we're now up in live 157 00:09:46,000 --> 00:09:49,440 Speaker 1: and over twenty countries UM. Some other examples are the 158 00:09:49,520 --> 00:09:54,000 Speaker 1: check Ministry of Health launched a virtual assistant named Annesca 159 00:09:54,080 --> 00:09:57,360 Speaker 1: to guide citizens on topics on the coronavirus according to 160 00:09:57,400 --> 00:10:01,679 Speaker 1: policies that were set by their fro MINT and you know, 161 00:10:01,800 --> 00:10:05,360 Speaker 1: in the first weekend, what they found is that only 162 00:10:05,480 --> 00:10:09,199 Speaker 1: ten percent of chats required a handover to a live agent. 163 00:10:09,520 --> 00:10:13,680 Speaker 1: Just think about that. That means of questions were answered 164 00:10:13,720 --> 00:10:16,760 Speaker 1: by this assistant. As I mentioned, like the first waves 165 00:10:16,800 --> 00:10:20,479 Speaker 1: we really saw we're in the public sector and in healthcare, 166 00:10:20,840 --> 00:10:23,920 Speaker 1: but now we're seeing that pervasive across all of retail, 167 00:10:24,320 --> 00:10:28,760 Speaker 1: financial services, industrial, We're seeing it across almost every industry 168 00:10:28,800 --> 00:10:32,000 Speaker 1: because there is a spike in and demand. You know, 169 00:10:32,040 --> 00:10:35,000 Speaker 1: a lot of these started out as as assistance for 170 00:10:35,240 --> 00:10:38,160 Speaker 1: citizen communities. How do you let the public know what's happening, 171 00:10:38,520 --> 00:10:40,560 Speaker 1: But think about some of the other use cases. Some 172 00:10:40,600 --> 00:10:44,560 Speaker 1: of the other ones are as I mentioned, um, you know, 173 00:10:44,800 --> 00:10:48,600 Speaker 1: actual companies and organizations that need to service in their 174 00:10:48,640 --> 00:10:53,120 Speaker 1: customers because their landscape is changing, or even companies that 175 00:10:53,160 --> 00:10:56,160 Speaker 1: need to service their internal employees. One is it okay 176 00:10:56,240 --> 00:10:58,480 Speaker 1: to come back to work? What are the regulations on 177 00:10:58,600 --> 00:11:01,040 Speaker 1: coming back to work? How are you phasing it? And 178 00:11:01,080 --> 00:11:03,400 Speaker 1: so you need to be able to have these kinds 179 00:11:03,400 --> 00:11:07,439 Speaker 1: of capabilities to respond to a lot of the uncertainty 180 00:11:07,480 --> 00:11:10,720 Speaker 1: in these times, and AI definitely helps. So I wonder 181 00:11:11,360 --> 00:11:14,560 Speaker 1: obviously a big thing when you're having, say, uh, some 182 00:11:14,640 --> 00:11:17,520 Speaker 1: kind of call routing program, an assistant that would be 183 00:11:17,520 --> 00:11:21,560 Speaker 1: dealing with calls coming in. One thing is that has 184 00:11:21,600 --> 00:11:24,880 Speaker 1: to be able to decide when somebody needs to speak 185 00:11:24,920 --> 00:11:28,480 Speaker 1: to a human operator versus uh, you know, continuing through 186 00:11:28,480 --> 00:11:32,360 Speaker 1: the call flow, whatever decision flow it naturally has what 187 00:11:32,520 --> 00:11:35,800 Speaker 1: goes into that kind of decision like and and how 188 00:11:35,840 --> 00:11:40,760 Speaker 1: important is making that decision early on? Yeah, actually, I 189 00:11:40,760 --> 00:11:44,400 Speaker 1: think that's actually a great question. Um, think about what 190 00:11:44,440 --> 00:11:46,680 Speaker 1: it means to train an assistant. You have to be 191 00:11:46,720 --> 00:11:49,800 Speaker 1: able to first when you ask a question, understand the 192 00:11:49,920 --> 00:11:53,400 Speaker 1: intent of that question. So when you're asking what are 193 00:11:53,400 --> 00:11:56,400 Speaker 1: the symptoms of COVID nineteen, you have to be able 194 00:11:56,440 --> 00:12:00,600 Speaker 1: to parse that sentence in natural language and really understand 195 00:12:00,960 --> 00:12:04,320 Speaker 1: what is that intent, because that could mean different things. 196 00:12:04,600 --> 00:12:07,600 Speaker 1: If I said, um, you know, instead of what are 197 00:12:07,600 --> 00:12:10,760 Speaker 1: the symptoms? If I said, um, where can I get 198 00:12:10,800 --> 00:12:14,400 Speaker 1: tested for COVID n team, those are two separate intents, 199 00:12:14,440 --> 00:12:16,960 Speaker 1: and so we call those things intent. We have to 200 00:12:17,080 --> 00:12:20,280 Speaker 1: train Watson assistant on the types of intent to be 201 00:12:20,320 --> 00:12:23,200 Speaker 1: able to answer, and the answers to those can change 202 00:12:23,280 --> 00:12:27,640 Speaker 1: over time. So we talked about some regulations change naturally 203 00:12:27,679 --> 00:12:31,000 Speaker 1: as these organizations actually change, and so the first thing 204 00:12:31,520 --> 00:12:34,600 Speaker 1: is having an understanding do I understand the intent? And 205 00:12:34,640 --> 00:12:38,440 Speaker 1: with what probability do I understand the intent? Majority of 206 00:12:38,480 --> 00:12:41,720 Speaker 1: the time these intents are easily understood, and so if 207 00:12:41,720 --> 00:12:44,880 Speaker 1: it's an intent that has been recognized, we can answer 208 00:12:44,920 --> 00:12:49,000 Speaker 1: that question if Watson has been trained on part of 209 00:12:49,080 --> 00:12:52,120 Speaker 1: what are those answers? And so we sometimes see that 210 00:12:52,360 --> 00:12:55,440 Speaker 1: if you can recognize the intent, then underneath there you 211 00:12:55,480 --> 00:12:58,520 Speaker 1: want to be able to answer those questions. If Watson 212 00:12:58,559 --> 00:13:01,760 Speaker 1: has a probability then range of what the user accepts 213 00:13:01,800 --> 00:13:05,800 Speaker 1: as reasonable, it will automatically respond, and if not, you 214 00:13:05,840 --> 00:13:09,040 Speaker 1: can train Watson to then kick it back immediately to 215 00:13:09,320 --> 00:13:12,960 Speaker 1: a actual, um physical person to be able to answer 216 00:13:13,000 --> 00:13:16,800 Speaker 1: those questions. Oh, that's interesting. So the uncertainty can be 217 00:13:16,880 --> 00:13:19,840 Speaker 1: the queue that we need a human to intervene here. 218 00:13:20,440 --> 00:13:23,240 Speaker 1: If you think about it, AI is all about probabilities, right, 219 00:13:23,320 --> 00:13:25,960 Speaker 1: What is the probability that the answer that you have 220 00:13:26,440 --> 00:13:30,079 Speaker 1: is most likely to answer that the user is looking for. 221 00:13:30,400 --> 00:13:33,160 Speaker 1: If that probability is high enough and in the tolerance 222 00:13:33,280 --> 00:13:35,840 Speaker 1: range that you have, then you can actually give that 223 00:13:35,880 --> 00:13:39,240 Speaker 1: as an answer. Yeah, that makes sense. So to to 224 00:13:39,360 --> 00:13:42,560 Speaker 1: unpack the technology a little bit more, UM, I think 225 00:13:42,800 --> 00:13:44,959 Speaker 1: from the caller's point of view, I think we've all 226 00:13:44,960 --> 00:13:48,320 Speaker 1: had these experiences where we know that in some cases 227 00:13:48,360 --> 00:13:51,880 Speaker 1: automated menus can be very frustrating, right, Like, uh, I'm 228 00:13:51,920 --> 00:13:54,920 Speaker 1: imagining calling customer service at a credit card company or 229 00:13:54,960 --> 00:13:59,480 Speaker 1: an Internet service provider. Obviously, like we realize we're talking 230 00:13:59,520 --> 00:14:02,360 Speaker 1: to a machine been and without a human operator, we 231 00:14:02,440 --> 00:14:07,000 Speaker 1: might worry that some important information we have is being ignored, 232 00:14:07,160 --> 00:14:10,320 Speaker 1: or some nuance of our case that that we're trying 233 00:14:10,360 --> 00:14:15,120 Speaker 1: to deal with doesn't fit within the automated routing stream. UM, So, 234 00:14:15,200 --> 00:14:18,880 Speaker 1: what are the ways that properly designed AI assistants can 235 00:14:18,960 --> 00:14:23,320 Speaker 1: help make automated call routing both more practically helpful to 236 00:14:23,480 --> 00:14:27,160 Speaker 1: callers but also more emotionally reassuring. There are many things 237 00:14:27,240 --> 00:14:29,120 Speaker 1: that are happening in this space that I think are 238 00:14:29,160 --> 00:14:32,560 Speaker 1: really exciting. Look most of the time when you call 239 00:14:32,600 --> 00:14:35,640 Speaker 1: into a company for customer service, when you get kind 240 00:14:35,640 --> 00:14:38,560 Speaker 1: of the automated machines that are like press Wan press too, 241 00:14:38,920 --> 00:14:43,040 Speaker 1: those are very deterministic UM systems. Those are not using 242 00:14:43,120 --> 00:14:46,920 Speaker 1: artificial intelligence. If you look at the real value of 243 00:14:46,960 --> 00:14:51,680 Speaker 1: having assistant like Watson Assistant, it's about being able to 244 00:14:51,720 --> 00:14:56,760 Speaker 1: answer your questions when you want them as you want them, 245 00:14:56,800 --> 00:15:00,400 Speaker 1: not in a pre integrated workflow that you have with 246 00:15:00,600 --> 00:15:04,880 Speaker 1: other deterministic systems. And that's has a completely different experience 247 00:15:04,960 --> 00:15:08,560 Speaker 1: because you can start in one part of a conversation 248 00:15:08,920 --> 00:15:12,200 Speaker 1: you can go multiple levels deep. You can then go 249 00:15:12,360 --> 00:15:16,320 Speaker 1: start another thread. And because the way the AI components 250 00:15:16,360 --> 00:15:20,560 Speaker 1: as well as the logic works, you are speaking naturally, 251 00:15:20,800 --> 00:15:23,640 Speaker 1: you are getting answers in a natural way, and you 252 00:15:23,680 --> 00:15:28,520 Speaker 1: are actually going between threads multiple levels deep. That is 253 00:15:28,680 --> 00:15:32,720 Speaker 1: engaging as you would as a human And what you 254 00:15:32,800 --> 00:15:34,960 Speaker 1: can do with wads and assistant and what makes it 255 00:15:35,080 --> 00:15:40,320 Speaker 1: such a beautiful experience comparative to what customer service experiences 256 00:15:40,360 --> 00:15:44,600 Speaker 1: like when you have those deterministic voice capabilities. Now, one 257 00:15:44,600 --> 00:15:47,960 Speaker 1: of the um aspects of calling and engaging with the 258 00:15:48,320 --> 00:15:50,520 Speaker 1: human operator on the side of the line is that 259 00:15:50,800 --> 00:15:54,000 Speaker 1: this generally there is an expectation that they might be 260 00:15:54,040 --> 00:15:58,760 Speaker 1: empathetic and give appropriate responses for say, delicate situations like 261 00:15:58,880 --> 00:16:02,400 Speaker 1: unemployment claims or health fears. Uh. You know, And again 262 00:16:02,440 --> 00:16:05,640 Speaker 1: that may be the expectation, if not the actual reality. 263 00:16:05,680 --> 00:16:09,240 Speaker 1: But then how do you tackle that from an AI standpoint? 264 00:16:09,240 --> 00:16:11,840 Speaker 1: How do you make sure that it is at least 265 00:16:11,840 --> 00:16:15,080 Speaker 1: seeming uh to speak with with empathy and give those 266 00:16:15,120 --> 00:16:18,080 Speaker 1: appropriate responses to someone who is in need. You know, 267 00:16:18,200 --> 00:16:20,920 Speaker 1: It's part of understanding the intent that we have not 268 00:16:21,000 --> 00:16:24,640 Speaker 1: only with Watson, but other parts of our portfolio. We 269 00:16:24,680 --> 00:16:28,160 Speaker 1: can understand the tone and the sentiment of a user, 270 00:16:28,600 --> 00:16:31,320 Speaker 1: and so we can understand if that tone is positive. 271 00:16:31,440 --> 00:16:34,000 Speaker 1: We can understand if the tone is that the end 272 00:16:34,080 --> 00:16:38,240 Speaker 1: user is irritated or if they're extremely angry. A lot 273 00:16:38,280 --> 00:16:40,840 Speaker 1: of that can be understood not only by the words 274 00:16:41,000 --> 00:16:43,880 Speaker 1: that the user may be typing UM into the screen, 275 00:16:43,920 --> 00:16:47,320 Speaker 1: but by the way their tone is when they even speak. 276 00:16:47,800 --> 00:16:51,000 Speaker 1: And because of that, we can actually respond in similar 277 00:16:51,040 --> 00:16:54,720 Speaker 1: ways where we have the right level of empathy to 278 00:16:55,040 --> 00:16:59,720 Speaker 1: respond back with. In some cases where we find that, 279 00:17:00,080 --> 00:17:03,720 Speaker 1: you know, suppose a user is extremely angry, we can 280 00:17:03,800 --> 00:17:07,960 Speaker 1: pass them off immediately to a human and put that 281 00:17:08,040 --> 00:17:10,159 Speaker 1: human in the loop. And that's why, you know, when 282 00:17:10,200 --> 00:17:12,719 Speaker 1: you think about artificial intelligence, I always think about it 283 00:17:12,760 --> 00:17:16,280 Speaker 1: as an ingredient in the broader picture, and it actually 284 00:17:16,480 --> 00:17:20,240 Speaker 1: is something that helps your overall application or your overall 285 00:17:20,240 --> 00:17:25,080 Speaker 1: customer service experience and is human assisted. Like the human 286 00:17:25,400 --> 00:17:27,720 Speaker 1: needs to be part of that loop at some point 287 00:17:27,800 --> 00:17:32,879 Speaker 1: if if there is an escalation UM. So, understanding tone 288 00:17:32,920 --> 00:17:36,240 Speaker 1: and sentiment is an extremely important thing, especially when you're 289 00:17:36,280 --> 00:17:40,240 Speaker 1: dealing with customer service. UM. You know, we see that 290 00:17:40,600 --> 00:17:43,840 Speaker 1: quite often. You know, you want to understand, for example, 291 00:17:44,000 --> 00:17:47,040 Speaker 1: if you have been bill double and you are extremely 292 00:17:47,080 --> 00:17:49,359 Speaker 1: angry about it. The assistant itself can take care of 293 00:17:49,400 --> 00:17:52,320 Speaker 1: that with empathy, but perhaps it's something that you want 294 00:17:52,359 --> 00:17:56,600 Speaker 1: to pass to a a a human agent to be 295 00:17:56,640 --> 00:17:58,639 Speaker 1: able to have a little bit more handholding in that 296 00:17:58,640 --> 00:18:02,680 Speaker 1: particular situation. What we've actually found is, you know when 297 00:18:02,840 --> 00:18:06,919 Speaker 1: huge enterprises, when when large enterprises get started with AI capabilities, 298 00:18:07,119 --> 00:18:09,800 Speaker 1: they'll actually start in a way where a I can 299 00:18:09,880 --> 00:18:13,240 Speaker 1: also be used for their assistance. If you think about it, 300 00:18:13,720 --> 00:18:19,240 Speaker 1: you want every every assistant in your organization, every human assistant, 301 00:18:19,480 --> 00:18:22,479 Speaker 1: to be able to give the same answer in the 302 00:18:22,560 --> 00:18:26,080 Speaker 1: same way, so that there's no discrepancies on what the 303 00:18:26,160 --> 00:18:28,320 Speaker 1: right answers are. So we see assistance not only being 304 00:18:28,440 --> 00:18:32,360 Speaker 1: used directly for the customer, but also for agents within 305 00:18:32,520 --> 00:18:36,159 Speaker 1: organizations to have the same response to all types of 306 00:18:36,240 --> 00:18:40,240 Speaker 1: questions coming in. So when we when we think about bias, 307 00:18:40,280 --> 00:18:43,960 Speaker 1: we tend to think about human activities and human institutions, 308 00:18:43,960 --> 00:18:46,439 Speaker 1: but of course AI as a human creation is susceptible 309 00:18:46,480 --> 00:18:50,600 Speaker 1: to bias as well. Can you explain how bias creeps 310 00:18:50,600 --> 00:18:54,359 Speaker 1: in UH? And then how do we prevent AI systems 311 00:18:54,400 --> 00:18:58,000 Speaker 1: from succumbing to these same errors? If you think about it, 312 00:18:58,240 --> 00:19:02,040 Speaker 1: bias exists every where in the world in the natural 313 00:19:02,040 --> 00:19:04,439 Speaker 1: world today. You know, if you take a look at 314 00:19:04,480 --> 00:19:08,680 Speaker 1: AI systems, AI systems are naturally trained by data data 315 00:19:08,720 --> 00:19:13,240 Speaker 1: that has existed there in the world for decades centuries. 316 00:19:13,800 --> 00:19:16,720 Speaker 1: I'll give an example today. If I would train an 317 00:19:16,720 --> 00:19:23,399 Speaker 1: AI system on how to approve alone for a particular person, 318 00:19:23,560 --> 00:19:26,800 Speaker 1: and I use data from the past fifty years, more 319 00:19:26,840 --> 00:19:29,760 Speaker 1: than likely you will see bias in the data for 320 00:19:29,800 --> 00:19:33,960 Speaker 1: the last fifty years. Were given everything equal that men 321 00:19:34,080 --> 00:19:36,480 Speaker 1: were more apt to be able to get alan with 322 00:19:36,600 --> 00:19:39,879 Speaker 1: all the same attributes than a female would. That is 323 00:19:39,960 --> 00:19:43,600 Speaker 1: biased that exists in data that real people have approved 324 00:19:43,720 --> 00:19:47,639 Speaker 1: loans over the past five decades in that particular case. 325 00:19:48,080 --> 00:19:52,360 Speaker 1: You know, if we train AI on that same data today, 326 00:19:52,760 --> 00:19:55,000 Speaker 1: we would want to make sure we take that data 327 00:19:55,080 --> 00:19:59,880 Speaker 1: and we remove that bias, because that is a fact 328 00:20:00,359 --> 00:20:02,560 Speaker 1: that we would want to say, Okay, there's bias that 329 00:20:02,640 --> 00:20:04,600 Speaker 1: exists in the data that we have. We can see 330 00:20:04,600 --> 00:20:07,240 Speaker 1: the bias, and that's a bias that I don't want 331 00:20:07,240 --> 00:20:12,120 Speaker 1: to be able to have. Another example maybe about claims 332 00:20:12,160 --> 00:20:16,560 Speaker 1: and claims approvals for auto insurance based on age. We know, 333 00:20:16,920 --> 00:20:19,720 Speaker 1: as an example, if I take the last fifty years 334 00:20:19,720 --> 00:20:24,480 Speaker 1: of data that claims that come in from younger generations 335 00:20:24,560 --> 00:20:29,040 Speaker 1: are probably um more prone to some sort of fraud 336 00:20:29,359 --> 00:20:33,240 Speaker 1: than older generations, and that particular case, you might have 337 00:20:33,440 --> 00:20:36,960 Speaker 1: a higher fraud rate at you know, the at ages 338 00:20:37,040 --> 00:20:39,480 Speaker 1: eighteen to twenty four as an example. That may be 339 00:20:39,640 --> 00:20:41,160 Speaker 1: a bias that you want to be able to keep 340 00:20:41,160 --> 00:20:44,960 Speaker 1: in your system of approving or not approving claims, because 341 00:20:44,960 --> 00:20:47,199 Speaker 1: that is a reasonable bias to be able to have 342 00:20:47,320 --> 00:20:50,320 Speaker 1: to say, I want to double check before we approve 343 00:20:50,480 --> 00:20:54,040 Speaker 1: or not approve those type of claims. So bias exists 344 00:20:54,119 --> 00:20:56,760 Speaker 1: naturally in the data that we have. And given that 345 00:20:56,920 --> 00:21:00,320 Speaker 1: AI is only as good as the data that you 346 00:21:00,359 --> 00:21:03,119 Speaker 1: train it on, you need to be able to take 347 00:21:03,400 --> 00:21:05,760 Speaker 1: the algorithms that you create from AI, you need to 348 00:21:05,800 --> 00:21:08,840 Speaker 1: be able to detect where that bias exists. And then 349 00:21:09,160 --> 00:21:12,560 Speaker 1: in cases like I mentioned where you're talking about age 350 00:21:12,600 --> 00:21:15,400 Speaker 1: and loans, you want to be able to remove that bias. 351 00:21:15,480 --> 00:21:19,040 Speaker 1: And that is one of the most critical factors to 352 00:21:19,160 --> 00:21:22,560 Speaker 1: making AI mainstream. Like I'm a firm believer that this 353 00:21:22,640 --> 00:21:25,520 Speaker 1: is a decade for AI to go mainstream, and for 354 00:21:25,600 --> 00:21:29,680 Speaker 1: it to be able to go mainstream means that organizations 355 00:21:29,720 --> 00:21:32,800 Speaker 1: need to be able to have trust in AI and 356 00:21:32,840 --> 00:21:36,400 Speaker 1: how that AI is working. And that's why being able 357 00:21:36,440 --> 00:21:41,040 Speaker 1: to take any model and understand where the biases in 358 00:21:41,080 --> 00:21:44,200 Speaker 1: that particular model, and then to be able to understand 359 00:21:44,280 --> 00:21:48,879 Speaker 1: things like, um, you know where, how can I explain 360 00:21:49,160 --> 00:21:52,000 Speaker 1: how that particular model made that decision? What are the 361 00:21:52,080 --> 00:21:54,439 Speaker 1: factors that went in too? For AI to make a 362 00:21:54,440 --> 00:21:59,360 Speaker 1: decision becomes extremely critical. Um So the ability for organizations 363 00:21:59,400 --> 00:22:02,520 Speaker 1: to make that AI mainstream is can I detect bias? 364 00:22:02,600 --> 00:22:05,160 Speaker 1: Can I remove it? Can I explain what AI is doing? 365 00:22:05,400 --> 00:22:07,760 Speaker 1: And that's a lot of what our teams within IBM 366 00:22:07,840 --> 00:22:10,800 Speaker 1: are working really hard at. A lot of the research 367 00:22:10,880 --> 00:22:14,840 Speaker 1: technologies that we've had are now in our products, and 368 00:22:15,040 --> 00:22:18,920 Speaker 1: we're helping many of our many organizations large and small, 369 00:22:19,440 --> 00:22:22,560 Speaker 1: be able to take the AI capabilities they have and 370 00:22:22,640 --> 00:22:27,640 Speaker 1: to put explainability and biased detection, UM and fairness recommendations 371 00:22:27,840 --> 00:22:30,120 Speaker 1: into their AI components because that's the only way we're 372 00:22:30,119 --> 00:22:32,720 Speaker 1: going to get AI to scale. Yeah, I guess, I 373 00:22:32,720 --> 00:22:35,159 Speaker 1: guess what I was wondering there is if this is 374 00:22:35,200 --> 00:22:38,240 Speaker 1: a case where like a diversity within the tech world 375 00:22:38,280 --> 00:22:41,960 Speaker 1: actually has measurable effects on whether these types of bias 376 00:22:42,119 --> 00:22:45,399 Speaker 1: uh end up making it through, uh, you know, or 377 00:22:45,440 --> 00:22:48,080 Speaker 1: go unnoticed in the design stage. I think that's a 378 00:22:48,119 --> 00:22:52,639 Speaker 1: good point. Look. Diversity in technology and especially in artificial 379 00:22:52,680 --> 00:22:58,440 Speaker 1: intelligence is critical UM, and that's diversity in all kinds 380 00:22:58,440 --> 00:23:02,960 Speaker 1: of backgrounds. I would say from UM, having diverse perspectives 381 00:23:02,960 --> 00:23:06,520 Speaker 1: and diverse point of views help you create a that 382 00:23:06,720 --> 00:23:09,720 Speaker 1: is more beneficial for society and for the community. Let 383 00:23:09,720 --> 00:23:14,000 Speaker 1: me give you a couple of interesting statistics UM. You know, 384 00:23:14,280 --> 00:23:18,520 Speaker 1: we we have recently put a lot of focus on 385 00:23:18,840 --> 00:23:23,440 Speaker 1: women and artificial intelligence today. It's estimated that less than 386 00:23:23,560 --> 00:23:28,360 Speaker 1: twenty of A professionals in the marketplace today are female. 387 00:23:29,359 --> 00:23:31,720 Speaker 1: That's not where we want to be able to see 388 00:23:32,240 --> 00:23:36,720 Speaker 1: UM the representation of A on females because as you 389 00:23:36,800 --> 00:23:40,879 Speaker 1: have more diverse perspectives and points of view, you can 390 00:23:40,920 --> 00:23:46,080 Speaker 1: create better outcomes for users and better AI algorithms. And 391 00:23:46,119 --> 00:23:48,720 Speaker 1: so it's one of the reasons why IBM has put 392 00:23:48,800 --> 00:23:53,080 Speaker 1: such a huge focus on women and AI. This year 393 00:23:53,240 --> 00:23:56,760 Speaker 1: is the second year that we announced a Woman in 394 00:23:57,040 --> 00:24:04,040 Speaker 1: AI program where we have UM celebrated over thirty females 395 00:24:04,640 --> 00:24:07,680 Speaker 1: in AI professions and the journey that they have taken 396 00:24:08,160 --> 00:24:12,199 Speaker 1: through their career. We have a few goals in and 397 00:24:12,280 --> 00:24:14,919 Speaker 1: being able to do that. UM, let me let me 398 00:24:14,960 --> 00:24:17,560 Speaker 1: first take a step back and tell you, like our 399 00:24:17,640 --> 00:24:21,600 Speaker 1: our effort in doing this was to really promote gender 400 00:24:21,640 --> 00:24:26,000 Speaker 1: equality and AI and showcasing these over thirty leaders across 401 00:24:26,040 --> 00:24:28,359 Speaker 1: a variety of industries I think they're in like twelve 402 00:24:28,400 --> 00:24:32,200 Speaker 1: countries was really important to us to demonstrate not only 403 00:24:32,240 --> 00:24:35,520 Speaker 1: the power of AI, but the power of diversity in 404 00:24:35,600 --> 00:24:39,920 Speaker 1: AI and the kinds of accomplishments that these organizations are 405 00:24:39,960 --> 00:24:43,960 Speaker 1: doing in the technologies that they're implementing with Watson and 406 00:24:44,359 --> 00:24:48,359 Speaker 1: other capabilities. So you know, what we what we have 407 00:24:48,480 --> 00:24:53,760 Speaker 1: found is that you know, by having and highlighting a 408 00:24:53,760 --> 00:24:57,879 Speaker 1: lot of these females, we can interest other younger generation 409 00:24:58,560 --> 00:25:02,119 Speaker 1: of women to be able to embark in this AI career. 410 00:25:03,080 --> 00:25:05,520 Speaker 1: This is especially touching for me and I think UM 411 00:25:05,560 --> 00:25:07,879 Speaker 1: a lot more active in the women in AI field 412 00:25:08,160 --> 00:25:11,760 Speaker 1: being a computer scientists working on data science and data 413 00:25:11,800 --> 00:25:14,800 Speaker 1: science products for a while as I have young children 414 00:25:14,960 --> 00:25:20,120 Speaker 1: and my my daughter who is nine years old, came 415 00:25:20,160 --> 00:25:23,359 Speaker 1: to me UM one day after I sent her to 416 00:25:23,400 --> 00:25:26,280 Speaker 1: a Python programming class and said I don't like it. 417 00:25:26,480 --> 00:25:28,840 Speaker 1: And when I sat down with her and really understood 418 00:25:28,880 --> 00:25:31,280 Speaker 1: the threat of why don't you like the class that 419 00:25:31,320 --> 00:25:33,680 Speaker 1: you're in. You know, it's a programming class where we're 420 00:25:33,720 --> 00:25:36,280 Speaker 1: engineers at heart, and you know it's one of the 421 00:25:36,280 --> 00:25:38,879 Speaker 1: things that you have to go do. She said, I 422 00:25:38,960 --> 00:25:42,600 Speaker 1: was the only girl. Everyone was coding Minecraft mobs, and 423 00:25:43,080 --> 00:25:45,760 Speaker 1: I really wasn't interested in that. I was sitting by myself, 424 00:25:46,280 --> 00:25:48,480 Speaker 1: and it was in that moment that I thought about, 425 00:25:48,520 --> 00:25:50,920 Speaker 1: you know, people need to be able to see themselves 426 00:25:51,119 --> 00:25:54,439 Speaker 1: in role models, and that's why I think programs like 427 00:25:54,480 --> 00:25:57,000 Speaker 1: Women in AI are so important, because as we want 428 00:25:57,000 --> 00:26:00,359 Speaker 1: more diversity, people need to be able to see themselves 429 00:26:00,480 --> 00:26:02,879 Speaker 1: very clearly in that. I'm pretty proud of some of 430 00:26:02,920 --> 00:26:05,080 Speaker 1: the things that IBM has done to be able to 431 00:26:05,119 --> 00:26:08,359 Speaker 1: put more women in AI, not only within our organization, 432 00:26:08,520 --> 00:26:11,560 Speaker 1: but to be able to promote that with the thousands 433 00:26:11,560 --> 00:26:14,200 Speaker 1: of clients that we work through in this second inaugural 434 00:26:14,240 --> 00:26:17,879 Speaker 1: program that we have. Now, when the COVID nineteen pandemic 435 00:26:18,040 --> 00:26:22,080 Speaker 1: ultimately subsides, what lasting impact do you hope to have 436 00:26:22,240 --> 00:26:24,800 Speaker 1: with the work that you're doing right now? You know, 437 00:26:24,920 --> 00:26:28,199 Speaker 1: I'm so proud of our teams that have risen to 438 00:26:28,240 --> 00:26:32,920 Speaker 1: the occasion of helping these hundreds of organizations implement AI 439 00:26:33,040 --> 00:26:35,960 Speaker 1: capabilities during this time of crisis. If you look at it, 440 00:26:35,960 --> 00:26:39,400 Speaker 1: it is helping answer some of the most pervasive questions 441 00:26:39,480 --> 00:26:42,919 Speaker 1: across all of these organizations in an extremely timely manner. 442 00:26:44,080 --> 00:26:46,080 Speaker 1: I think the change is that a lot of these 443 00:26:46,160 --> 00:26:50,159 Speaker 1: organizations are embarking on are here to stay, and not 444 00:26:50,240 --> 00:26:52,679 Speaker 1: only here to stay, but they're going to accelerate for 445 00:26:52,720 --> 00:26:57,399 Speaker 1: every organization. Every organization is going to adopt technologies like 446 00:26:57,560 --> 00:27:01,840 Speaker 1: AI digital capabilities a lot more quickly, and so a 447 00:27:01,880 --> 00:27:05,240 Speaker 1: lot of the lasting effects are taking what we learned 448 00:27:05,320 --> 00:27:09,359 Speaker 1: and helping organizations really scale out a lot more quickly 449 00:27:09,440 --> 00:27:12,480 Speaker 1: the use of these AI technologies and having them be 450 00:27:12,920 --> 00:27:19,320 Speaker 1: fundamental to how they operate and not just a side car. Again, 451 00:27:19,400 --> 00:27:22,440 Speaker 1: much appreciation to Ridikagoner for taking the time to speak 452 00:27:22,480 --> 00:27:24,520 Speaker 1: with us for this episode. And now we're going to 453 00:27:24,600 --> 00:27:27,359 Speaker 1: go straight into our second talk on the subject with 454 00:27:27,440 --> 00:27:34,200 Speaker 1: Jay Bellissimo. All right, well, Jay, we really appreciate you 455 00:27:34,640 --> 00:27:36,880 Speaker 1: taking time to talk to us today. Can you start 456 00:27:36,920 --> 00:27:39,359 Speaker 1: off by introducing yourself and telling us a little bit 457 00:27:39,359 --> 00:27:42,320 Speaker 1: about your background. Sure, Joe and Rob, it's great to 458 00:27:42,359 --> 00:27:44,520 Speaker 1: have the opportunity to spend some time with you today. 459 00:27:44,880 --> 00:27:48,440 Speaker 1: So my current role at IBM as general manager our 460 00:27:48,600 --> 00:27:52,200 Speaker 1: Federal and Public business UM. Prior to this role, I 461 00:27:52,280 --> 00:27:56,560 Speaker 1: spent six years focused on UM Watson and AI and 462 00:27:56,640 --> 00:28:01,399 Speaker 1: cloud excellent. So on the on this topic of AI UH, 463 00:28:01,640 --> 00:28:04,800 Speaker 1: as we you know, dive further in in this interview 464 00:28:04,880 --> 00:28:09,840 Speaker 1: and start chatting about these about the Watson, AI, Watson Assistant, etcetera. 465 00:28:10,480 --> 00:28:13,000 Speaker 1: Before we do that, would you mind just like touching 466 00:28:13,160 --> 00:28:17,960 Speaker 1: on some of the biggest UM misconceptions about AI itself 467 00:28:18,160 --> 00:28:20,760 Speaker 1: so that our listeners are properly grounded and where we 468 00:28:20,840 --> 00:28:24,840 Speaker 1: are with real world AI. Yeah, Rob, it's a great question. Again, 469 00:28:25,000 --> 00:28:28,119 Speaker 1: I started as a general manager in our Watson business 470 00:28:28,119 --> 00:28:31,040 Speaker 1: back in two thousand fourteen when it first started. And 471 00:28:31,080 --> 00:28:33,040 Speaker 1: you know, there's always been a lot of talk about AI. 472 00:28:33,520 --> 00:28:37,360 Speaker 1: AI algorithms have been around for many, many years UM, 473 00:28:37,400 --> 00:28:40,600 Speaker 1: and I'm pretty excited because since two thousand fourteen, I've 474 00:28:40,760 --> 00:28:43,160 Speaker 1: I think I've averaged about two hundred thousand miles a year, 475 00:28:43,560 --> 00:28:47,760 Speaker 1: been a thirty six countries, been evangelizing around AI UH 476 00:28:47,800 --> 00:28:50,840 Speaker 1: and I couldn't be more excited. Unlike four or five 477 00:28:50,920 --> 00:28:53,360 Speaker 1: years ago when people said, well, when's it gonna happen, 478 00:28:53,520 --> 00:28:55,920 Speaker 1: It's it's happening, right, It's not a question whether or 479 00:28:55,960 --> 00:28:58,360 Speaker 1: not it's gonna happen. It's here today, whether it be 480 00:28:58,400 --> 00:29:02,280 Speaker 1: every day you know, listening Spotify or using ways. I mean, 481 00:29:02,600 --> 00:29:05,160 Speaker 1: it's in our everyday lives. And we've also seen this 482 00:29:05,200 --> 00:29:09,680 Speaker 1: whole consumer market really blur the enterprise, you know, companies 483 00:29:09,720 --> 00:29:12,720 Speaker 1: and government organizations. So where we are this year, I 484 00:29:12,760 --> 00:29:15,920 Speaker 1: playfully say this is a year AI goes into production. 485 00:29:16,360 --> 00:29:19,360 Speaker 1: Just at IBM alone, we have over twenty thousand UH 486 00:29:19,600 --> 00:29:24,080 Speaker 1: projects locally UH in terms of you know, AI use cases, 487 00:29:24,120 --> 00:29:26,960 Speaker 1: and I think at the end, you know, Rob and Joe, 488 00:29:27,440 --> 00:29:30,040 Speaker 1: For me, the difference is what's the problem you're trying 489 00:29:30,040 --> 00:29:34,400 Speaker 1: to solve? Right, Sometimes we get out ahead and say, 490 00:29:34,440 --> 00:29:36,760 Speaker 1: you know, companies will say well, I must have a I. 491 00:29:36,920 --> 00:29:40,120 Speaker 1: But but practically it really comes back to AI is 492 00:29:40,160 --> 00:29:42,560 Speaker 1: awesome and it can do so much, but you really 493 00:29:42,600 --> 00:29:44,959 Speaker 1: have to have in mind what's the business problem you're 494 00:29:44,960 --> 00:29:46,920 Speaker 1: trying to solve. And once you can really hone in 495 00:29:46,960 --> 00:29:49,320 Speaker 1: on that, then it becomes a lot easier to look 496 00:29:49,360 --> 00:29:52,880 Speaker 1: at it because in addition to AI, another big big 497 00:29:52,880 --> 00:29:55,360 Speaker 1: piece of this is the data. Right, And you've heard 498 00:29:55,400 --> 00:29:57,760 Speaker 1: so much about the data and access of the data, 499 00:29:57,760 --> 00:29:58,960 Speaker 1: and that's one of the things. We take a lot 500 00:29:58,960 --> 00:30:01,680 Speaker 1: of pride with that ID end is you know, typically 501 00:30:01,800 --> 00:30:04,920 Speaker 1: that data is the hospital's data, or that data is 502 00:30:04,920 --> 00:30:07,560 Speaker 1: a government agency's data or its or it's a big 503 00:30:07,600 --> 00:30:11,719 Speaker 1: industrial companies data. Because that is that is really important 504 00:30:11,720 --> 00:30:14,000 Speaker 1: because ultimately that's your I P. Because when you think 505 00:30:14,040 --> 00:30:16,840 Speaker 1: about it very simply, when you think about AI, it's 506 00:30:17,440 --> 00:30:20,560 Speaker 1: you're starting with the data and working with AI. The 507 00:30:20,640 --> 00:30:23,600 Speaker 1: data is going to create insights. An insight creates knowledge. 508 00:30:24,040 --> 00:30:26,400 Speaker 1: And with all that rob coming back to your question, 509 00:30:27,040 --> 00:30:30,000 Speaker 1: that's where I get really excited because in the end, 510 00:30:30,400 --> 00:30:33,600 Speaker 1: this is really a partnership between man and machine. It's 511 00:30:33,600 --> 00:30:37,560 Speaker 1: not man verse machine. And and that's important because let's 512 00:30:37,560 --> 00:30:40,600 Speaker 1: face it, there's so many menial tasks that can be 513 00:30:40,640 --> 00:30:43,000 Speaker 1: automated and that's no different if you look back over 514 00:30:43,040 --> 00:30:46,000 Speaker 1: the industrial revolutions over the last hundred years and so 515 00:30:46,120 --> 00:30:49,520 Speaker 1: really you can use this technology to do so much, 516 00:30:49,680 --> 00:30:53,440 Speaker 1: but it's there's this misnomer that is displacing jobs and 517 00:30:53,680 --> 00:30:55,960 Speaker 1: that couldn't be further from the truth when you really 518 00:30:55,960 --> 00:30:59,600 Speaker 1: think through how this can be applied. Is it automating, Absolutely, 519 00:31:00,120 --> 00:31:03,480 Speaker 1: but there's so many jobs that are being created that 520 00:31:03,600 --> 00:31:06,640 Speaker 1: we need to make sure that we not just our company, 521 00:31:06,640 --> 00:31:09,960 Speaker 1: but any of the company's academia governments. We need to 522 00:31:10,040 --> 00:31:13,640 Speaker 1: come together as stewards of this next generation of opportunities 523 00:31:13,960 --> 00:31:16,040 Speaker 1: and all of these new jobs and make sure we 524 00:31:16,160 --> 00:31:20,800 Speaker 1: ushered in together responsibly. Because I'm not worried about the jobs. 525 00:31:20,920 --> 00:31:23,480 Speaker 1: What I'm worried about back to all the clients have 526 00:31:23,520 --> 00:31:25,960 Speaker 1: been at hundreds and hundreds of clients over the last 527 00:31:26,200 --> 00:31:29,480 Speaker 1: six years. Um, it's how do I equip my workforce 528 00:31:30,000 --> 00:31:33,400 Speaker 1: to to stay current. And that's the part people really 529 00:31:33,400 --> 00:31:35,360 Speaker 1: need to double click on and make sure that we're 530 00:31:35,400 --> 00:31:40,680 Speaker 1: all responsibly making sure that everyone has can transition with 531 00:31:40,760 --> 00:31:44,000 Speaker 1: into this new era with the required skills. And the 532 00:31:44,080 --> 00:31:47,400 Speaker 1: last point is just on as we do this, there 533 00:31:47,400 --> 00:31:52,120 Speaker 1: are still some challenges around ethical, ethics and bias, and 534 00:31:52,280 --> 00:31:55,680 Speaker 1: that's something again everyone, academia companies like you know, we 535 00:31:55,720 --> 00:31:58,320 Speaker 1: all need to continue to work together to make sure 536 00:31:58,400 --> 00:32:01,200 Speaker 1: again we ushered this in and if very responsible way. 537 00:32:01,360 --> 00:32:04,320 Speaker 1: Maybe we should focus for a second on specifically what 538 00:32:04,480 --> 00:32:07,080 Speaker 1: some of those challenges are. You mentioned of course, ethics 539 00:32:07,080 --> 00:32:09,720 Speaker 1: and bias. We can maybe circle back to that if 540 00:32:10,240 --> 00:32:12,560 Speaker 1: if we want. But another thing that I saw you 541 00:32:12,600 --> 00:32:15,440 Speaker 1: mentioned I watched part of a talk that you gave 542 00:32:15,480 --> 00:32:18,840 Speaker 1: where you mentioned the idea of information architecture that you 543 00:32:18,840 --> 00:32:21,960 Speaker 1: know that you can't have AI without I A. Could 544 00:32:21,960 --> 00:32:23,680 Speaker 1: you talk a bit about that and what kind of 545 00:32:23,720 --> 00:32:28,720 Speaker 1: challenges are still present there? Sure? Sure, But fundamentally, what 546 00:32:28,760 --> 00:32:32,719 Speaker 1: we're saying with that statement is basically, the data is 547 00:32:32,760 --> 00:32:35,360 Speaker 1: going to be critical and when you think about the 548 00:32:35,480 --> 00:32:38,480 Speaker 1: data and what you do with the data and you analyze, right, 549 00:32:38,600 --> 00:32:41,560 Speaker 1: you collect the data, then you analyze it, you organize it, 550 00:32:41,640 --> 00:32:44,200 Speaker 1: and then you infuse it with a I. That's really 551 00:32:44,240 --> 00:32:46,840 Speaker 1: what we're saying is when you have a I, you 552 00:32:46,920 --> 00:32:51,720 Speaker 1: really also need to have that information architecture because they 553 00:32:51,760 --> 00:32:55,480 Speaker 1: go hand in hand. Because again, we've had the data, 554 00:32:55,480 --> 00:32:58,000 Speaker 1: but we've never had the data at the levels we have. 555 00:32:58,240 --> 00:33:00,880 Speaker 1: It's it's really outstripped human of the ability to keep 556 00:33:00,920 --> 00:33:03,520 Speaker 1: pace with it. So with that we have all this 557 00:33:03,600 --> 00:33:08,360 Speaker 1: on structured data of the world's data is unstructured. On 558 00:33:08,440 --> 00:33:11,880 Speaker 1: top of that, only of the data is access via 559 00:33:11,920 --> 00:33:14,560 Speaker 1: the web, so a lot of that data sits behind 560 00:33:14,560 --> 00:33:18,560 Speaker 1: the corporate firewalls, right, and so there's so much opportunity. 561 00:33:18,680 --> 00:33:22,480 Speaker 1: So in the end, you've you've traditionally had information architecture. 562 00:33:22,560 --> 00:33:25,360 Speaker 1: So even though it's this new way of AI. They 563 00:33:25,400 --> 00:33:29,160 Speaker 1: really go hand in hand. So obviously today we were 564 00:33:29,160 --> 00:33:33,120 Speaker 1: gonna end up focusing a good bit on how AI 565 00:33:33,280 --> 00:33:35,480 Speaker 1: is being used to respond in the wake of the 566 00:33:35,520 --> 00:33:38,960 Speaker 1: COVID nineteen pandemic. But first just to establish, you know, 567 00:33:39,840 --> 00:33:41,960 Speaker 1: what what the needs are, what the problem is. To 568 00:33:41,960 --> 00:33:43,640 Speaker 1: begin with, can you talk about some of the ways 569 00:33:43,720 --> 00:33:48,680 Speaker 1: that the current pandemic has affected the how people interact 570 00:33:48,760 --> 00:33:52,760 Speaker 1: with businesses and institutions it as you well know, I mean, 571 00:33:52,960 --> 00:33:55,680 Speaker 1: none of us expected what we're living in right now. 572 00:33:56,080 --> 00:33:59,040 Speaker 1: And the really cool thing is the way everyone's coming 573 00:33:59,040 --> 00:34:02,080 Speaker 1: together across the untes. Recently, I had a call with 574 00:34:02,120 --> 00:34:05,680 Speaker 1: the general manager of IBM Italy and IBM Spain and 575 00:34:05,720 --> 00:34:08,960 Speaker 1: we were just comparing stories IBM China as an example, 576 00:34:09,000 --> 00:34:11,239 Speaker 1: and we were just exchanging over the last sixty days 577 00:34:11,280 --> 00:34:14,160 Speaker 1: what we've all seen and it's pretty powerful, the way 578 00:34:14,200 --> 00:34:17,000 Speaker 1: everyone in the face of all this adversity has really 579 00:34:17,000 --> 00:34:20,279 Speaker 1: come together. And what was interesting a thread that we 580 00:34:20,320 --> 00:34:24,760 Speaker 1: saw across all these different companies was we start with empathy, 581 00:34:25,080 --> 00:34:28,080 Speaker 1: right and as much as you know we we we 582 00:34:28,160 --> 00:34:31,719 Speaker 1: talked about business, this this goes far beyond business, and 583 00:34:31,760 --> 00:34:34,960 Speaker 1: this is really being empathetic. Every country is going through 584 00:34:34,960 --> 00:34:38,800 Speaker 1: a different, different phase of this pandemic, but this common 585 00:34:38,840 --> 00:34:43,160 Speaker 1: threat is just really listening, talking to clients, talking to 586 00:34:43,400 --> 00:34:47,640 Speaker 1: government organizations h HS, Health and Human Services and others, 587 00:34:48,040 --> 00:34:51,360 Speaker 1: and just really listening what what are those problems that 588 00:34:51,360 --> 00:34:55,480 Speaker 1: we're trying to solve together across corporate boundaries. And so 589 00:34:55,560 --> 00:34:58,320 Speaker 1: a great example I would use I was personally involved 590 00:34:58,760 --> 00:35:02,840 Speaker 1: with the Learn's Healthcare of Atlanta, and I remember it 591 00:35:02,880 --> 00:35:04,759 Speaker 1: was a Sunday, and I had spoken to the c 592 00:35:04,920 --> 00:35:08,680 Speaker 1: i O and their issue, which is I think pretty 593 00:35:08,920 --> 00:35:13,000 Speaker 1: um common across the law of the hospital and healthcare 594 00:35:13,080 --> 00:35:16,600 Speaker 1: providers is in their case, their nurse station was being 595 00:35:16,640 --> 00:35:20,600 Speaker 1: overwhelmed understandingly by a concerned citizens. Maybe it's a parent 596 00:35:21,040 --> 00:35:23,800 Speaker 1: calling about they're worried about maybe their child has a fever, 597 00:35:24,239 --> 00:35:26,719 Speaker 1: or there could be other symptoms. So they were just 598 00:35:26,880 --> 00:35:30,200 Speaker 1: overwhelming the nurse station. So within forty eight hours from 599 00:35:30,239 --> 00:35:33,440 Speaker 1: that first conversation with the c i O, we were 600 00:35:33,480 --> 00:35:39,799 Speaker 1: able to stand up Watson Citizen Engagement UM bought a 601 00:35:39,880 --> 00:35:42,839 Speaker 1: virtual agent we like to call them and and basically 602 00:35:43,160 --> 00:35:47,040 Speaker 1: helped them so when they those calls came in, this 603 00:35:47,520 --> 00:35:51,880 Speaker 1: virtual agent could take a lot of the questions and 604 00:35:51,920 --> 00:35:55,480 Speaker 1: in a very natural way, very interactive way, engage with 605 00:35:55,600 --> 00:35:58,840 Speaker 1: citizens who were concerned. And the way we trained it 606 00:35:58,920 --> 00:36:02,680 Speaker 1: is the hospital heads to civic protocols, and protocol would 607 00:36:02,719 --> 00:36:05,279 Speaker 1: be as an example, my my child has a fever, 608 00:36:05,440 --> 00:36:08,279 Speaker 1: what do I do right? So the protocol is just 609 00:36:08,320 --> 00:36:10,880 Speaker 1: an encapsulation of all the different variations. So in the 610 00:36:11,000 --> 00:36:14,239 Speaker 1: end it really is consistent with the hospital's procedures. Every 611 00:36:14,280 --> 00:36:17,319 Speaker 1: hospital could be different with the protocols they run. The 612 00:36:17,360 --> 00:36:21,080 Speaker 1: exciting thing is the problem was that the nurses needed 613 00:36:21,120 --> 00:36:23,719 Speaker 1: to spend more time with patients as much as they 614 00:36:23,719 --> 00:36:28,560 Speaker 1: wanted to spend times consulting and using this technology to 615 00:36:28,680 --> 00:36:32,319 Speaker 1: be able to interface with and being being with the 616 00:36:32,400 --> 00:36:35,480 Speaker 1: citizens as they called. And then at the right time 617 00:36:35,600 --> 00:36:40,160 Speaker 1: if a nurse or another medical professional need to be engaged, 618 00:36:40,440 --> 00:36:46,000 Speaker 1: then this citizen engagement. But can then help what we 619 00:36:46,040 --> 00:36:49,239 Speaker 1: call our Watson assistant for citizens. Um, what this can 620 00:36:49,280 --> 00:36:53,400 Speaker 1: do now is now we directed to a person as needed. 621 00:36:53,680 --> 00:36:56,720 Speaker 1: So the beauty is it helps front end it and 622 00:36:57,040 --> 00:36:59,000 Speaker 1: answer those questions. But if they get to a point 623 00:36:59,040 --> 00:37:01,560 Speaker 1: where they're not sure, then enhance it off to a 624 00:37:01,640 --> 00:37:04,560 Speaker 1: person and that would probably be again a great example. 625 00:37:04,600 --> 00:37:07,560 Speaker 1: And then another really good one that has come up 626 00:37:07,560 --> 00:37:09,839 Speaker 1: with a lot of states that we're we're working with, 627 00:37:10,160 --> 00:37:13,560 Speaker 1: like the state of Pennsylvania is unemployment insurance as you 628 00:37:14,120 --> 00:37:16,280 Speaker 1: Joe and Rob no. I mean, some of these systems 629 00:37:16,360 --> 00:37:19,600 Speaker 1: very public, uh in some of the states like New Jersey, 630 00:37:19,880 --> 00:37:23,040 Speaker 1: you know, the call for COBAL programmers or others where 631 00:37:23,080 --> 00:37:25,320 Speaker 1: they're just overwhelmed. I think there's as many as thirty 632 00:37:25,360 --> 00:37:28,360 Speaker 1: million unemployed over the last six months. I think the 633 00:37:28,560 --> 00:37:31,680 Speaker 1: unemployment rate is around four to five percent right now. 634 00:37:32,160 --> 00:37:35,400 Speaker 1: And and so again, using some of the same Watson 635 00:37:35,520 --> 00:37:39,960 Speaker 1: Assistant for um citizens technology, we've been able to help 636 00:37:40,239 --> 00:37:44,759 Speaker 1: states like Pennsylvania implement these types of solutions. So maybe 637 00:37:44,840 --> 00:37:48,719 Speaker 1: could we imagine UM walking through what one of these 638 00:37:48,760 --> 00:37:51,560 Speaker 1: experiences might be like from the caller's point of view, 639 00:37:51,560 --> 00:37:55,080 Speaker 1: because we all, you know, have experience with probably not 640 00:37:55,239 --> 00:37:59,200 Speaker 1: a I powered but more deterministic traditional call routing systems, 641 00:37:59,239 --> 00:38:01,279 Speaker 1: like when you call your edit card company or your 642 00:38:01,320 --> 00:38:04,200 Speaker 1: internet provider or something, and it it can we all know, 643 00:38:04,320 --> 00:38:07,520 Speaker 1: be a kind of frustrating experience. How does an AI 644 00:38:07,600 --> 00:38:10,520 Speaker 1: powered experience differ from that? How in what ways could 645 00:38:10,520 --> 00:38:16,680 Speaker 1: it actually be more practically helpful and potentially more emotionally reassuring. Yeah, 646 00:38:16,719 --> 00:38:19,359 Speaker 1: it's a very good question and and and ultimately it's 647 00:38:19,400 --> 00:38:22,120 Speaker 1: interesting and this is all about you know, serving the 648 00:38:22,120 --> 00:38:25,680 Speaker 1: people in the communities. Um. Having said that, you know, 649 00:38:25,760 --> 00:38:29,000 Speaker 1: some of the technology solutions are different. Some are programmable, 650 00:38:29,360 --> 00:38:32,520 Speaker 1: you know, chatbots, and they're really light in terms of 651 00:38:32,560 --> 00:38:35,440 Speaker 1: real intelligence of what they do, and in those cases 652 00:38:35,480 --> 00:38:38,960 Speaker 1: they can frustrate people calling into these when you really 653 00:38:39,040 --> 00:38:42,360 Speaker 1: factor in machine learning, as we do with our Watson 654 00:38:43,080 --> 00:38:46,239 Speaker 1: technologies and solutions, again, it comes back to the four 655 00:38:46,640 --> 00:38:50,440 Speaker 1: areas I talked about before. It's understanding, right, and there 656 00:38:50,480 --> 00:38:53,120 Speaker 1: are nuances in the language and how we communicate with 657 00:38:53,120 --> 00:38:56,000 Speaker 1: each other, so it understands the context of the words 658 00:38:56,040 --> 00:39:00,120 Speaker 1: and the phrases. And to your point, Joe, Um, these 659 00:39:00,160 --> 00:39:04,000 Speaker 1: systems are probabilistic, not deterministic, and that's a game changer 660 00:39:04,040 --> 00:39:06,200 Speaker 1: in my view. And that's why we're pretty excited about 661 00:39:06,239 --> 00:39:10,279 Speaker 1: our Watson in our AI technologies because it really to 662 00:39:10,360 --> 00:39:14,759 Speaker 1: your point, it's it's more empathetic. It engages genuinely with 663 00:39:14,840 --> 00:39:18,080 Speaker 1: people that understands the words they use in the language 664 00:39:18,440 --> 00:39:20,600 Speaker 1: and they know how to respond. And then to that 665 00:39:20,719 --> 00:39:24,600 Speaker 1: third point, they continuously learn and they get smarter and 666 00:39:24,640 --> 00:39:28,280 Speaker 1: smarter with every interaction. So they're not perfect, but every 667 00:39:28,320 --> 00:39:32,360 Speaker 1: time you learn from past experiences, then it's only going 668 00:39:32,400 --> 00:39:34,319 Speaker 1: to get better and smarter, and it will be more 669 00:39:34,360 --> 00:39:38,719 Speaker 1: engaging with with citizens or consumers depending on how you 670 00:39:38,840 --> 00:39:42,120 Speaker 1: use the technology. This is a fascinating thing to think 671 00:39:42,160 --> 00:39:46,800 Speaker 1: about because UM essentially we're talking about having a more 672 00:39:46,880 --> 00:39:52,040 Speaker 1: human experience with the technology. Uh I. Like you said, 673 00:39:52,040 --> 00:39:55,279 Speaker 1: most of us probably have experience with UM with the 674 00:39:55,320 --> 00:40:00,400 Speaker 1: other model of of automated UM, you know, machine tech chnology. 675 00:40:00,480 --> 00:40:03,080 Speaker 1: When we call a credit card company or whatever the 676 00:40:03,080 --> 00:40:06,480 Speaker 1: case may be, we feel ourselves just thrown into those brackets, 677 00:40:06,600 --> 00:40:10,120 Speaker 1: and it feels it can feel dehumanizing, it can feel 678 00:40:10,200 --> 00:40:15,120 Speaker 1: very frustrating. UM. I wonder as as we as a 679 00:40:15,120 --> 00:40:20,360 Speaker 1: population begin to experience more and more of these AI models, 680 00:40:20,920 --> 00:40:24,440 Speaker 1: UM I imagine people are are going to go into them, 681 00:40:24,480 --> 00:40:29,719 Speaker 1: perhaps expecting that frustrating, UM sometimes dead ended situation, but 682 00:40:29,760 --> 00:40:33,279 Speaker 1: instead they're going to encounter something that is reacting to 683 00:40:33,360 --> 00:40:37,520 Speaker 1: them more that may even in these cases be exhibiting 684 00:40:37,600 --> 00:40:42,000 Speaker 1: something like empathy. UM. How do you see that that 685 00:40:42,120 --> 00:40:46,319 Speaker 1: shift going with us as a technology using culture? And 686 00:40:46,360 --> 00:40:48,360 Speaker 1: then is there a is there a potential that we 687 00:40:48,440 --> 00:40:52,200 Speaker 1: overshoot then and we start expecting more empathy than is 688 00:40:52,239 --> 00:40:55,719 Speaker 1: possible from the machine. It's a great question, Robin, and 689 00:40:55,719 --> 00:40:57,920 Speaker 1: in the way I look at it as anything you 690 00:40:57,960 --> 00:41:02,759 Speaker 1: can do to meaningfully engage with citizens and consumers is 691 00:41:02,800 --> 00:41:06,400 Speaker 1: a very good thing. We have different technologies with our 692 00:41:06,440 --> 00:41:10,640 Speaker 1: Watson technology, we have personal personality insights and other types 693 00:41:10,680 --> 00:41:14,000 Speaker 1: of capabilities that it's trying to enrich that interaction. Um 694 00:41:14,120 --> 00:41:15,880 Speaker 1: But at the end of the day, you know, every 695 00:41:15,880 --> 00:41:19,520 Speaker 1: company will have their approach and our approach has always 696 00:41:19,520 --> 00:41:22,759 Speaker 1: been again man and machine, and we're always trying to 697 00:41:22,840 --> 00:41:26,160 Speaker 1: make sure we have the most meaningful engagement between man 698 00:41:26,200 --> 00:41:30,040 Speaker 1: and machine. And to your exact point, uh, the empathy, 699 00:41:30,120 --> 00:41:32,520 Speaker 1: you know, how are people feeling that the way they 700 00:41:32,640 --> 00:41:35,760 Speaker 1: use certain words? What does that mean? Can we drive 701 00:41:35,960 --> 00:41:39,960 Speaker 1: more insight from that data? And ultimately for us, the 702 00:41:40,000 --> 00:41:43,319 Speaker 1: game changers knowledge from the insight and that's when you 703 00:41:43,360 --> 00:41:46,640 Speaker 1: really get into this I think a much higher level 704 00:41:46,719 --> 00:41:50,600 Speaker 1: of interaction. But you also govern that in terms of 705 00:41:50,680 --> 00:41:53,280 Speaker 1: how far do you want to to go down that path. 706 00:41:53,400 --> 00:41:56,960 Speaker 1: So in example, just to extend this conversation so you 707 00:41:57,040 --> 00:42:02,280 Speaker 1: have that engagement, right, it's understanding, reasoning, learning, interacting with people, 708 00:42:02,680 --> 00:42:06,560 Speaker 1: in natural language ways. But now when you look at that, um, 709 00:42:06,719 --> 00:42:08,919 Speaker 1: then you get into the point of okay, so you've 710 00:42:08,920 --> 00:42:11,880 Speaker 1: got this trust and you have this new way of engagement. 711 00:42:12,320 --> 00:42:15,640 Speaker 1: You can you can interact by the use of words 712 00:42:15,680 --> 00:42:19,040 Speaker 1: because you're getting smarter. And again it's systems like Watson 713 00:42:19,080 --> 00:42:21,520 Speaker 1: never forget, and that's the power to get smarter and smarter. 714 00:42:21,960 --> 00:42:25,280 Speaker 1: But then you extend it more and you say, okay, maybe, UM, 715 00:42:25,320 --> 00:42:28,000 Speaker 1: we could take an example. Let's say Rob, you and 716 00:42:28,040 --> 00:42:32,480 Speaker 1: I are applying for a mortgage at a bank, and 717 00:42:32,480 --> 00:42:35,440 Speaker 1: and maybe you and I interacting with one of these systems. 718 00:42:35,719 --> 00:42:38,239 Speaker 1: And this is a fictitious example, but there are examples 719 00:42:38,280 --> 00:42:40,719 Speaker 1: like this. But maybe you and I are interacting with 720 00:42:40,760 --> 00:42:43,040 Speaker 1: a with a system in AI system, not our system, 721 00:42:43,120 --> 00:42:45,879 Speaker 1: just any system. Right now, what if you and I 722 00:42:46,080 --> 00:42:49,040 Speaker 1: check all the boxes but in the end you get approved, 723 00:42:49,040 --> 00:42:52,120 Speaker 1: I get rejected. So you could say, well, wait a minute, 724 00:42:52,120 --> 00:42:54,440 Speaker 1: this doesn't feel right. I know he's my friend and 725 00:42:54,480 --> 00:42:57,319 Speaker 1: we have similar capability, so why so you get into 726 00:42:57,360 --> 00:43:00,759 Speaker 1: this whole explainability of a I know what happened in 727 00:43:00,760 --> 00:43:03,560 Speaker 1: those neural networks. You get into bias. How do I 728 00:43:03,640 --> 00:43:07,160 Speaker 1: know Rob? It wasn't biased against me? For some odd reason, 729 00:43:07,680 --> 00:43:10,640 Speaker 1: and so you need to have explainability. And so again 730 00:43:10,760 --> 00:43:14,360 Speaker 1: we have some great capabilities in IBM and our Watson 731 00:43:14,440 --> 00:43:18,080 Speaker 1: solutions where we we trace that we can flag where 732 00:43:18,200 --> 00:43:22,000 Speaker 1: there's potential issues with bias or explainability, because we want 733 00:43:22,040 --> 00:43:24,960 Speaker 1: you to have that full traceability across end to end. 734 00:43:25,480 --> 00:43:27,840 Speaker 1: And that's the other kind of evolution. So back to 735 00:43:27,880 --> 00:43:31,440 Speaker 1: your your original question, Rob, we absolutely want to enrich 736 00:43:31,520 --> 00:43:33,440 Speaker 1: that experience. We want to learn from the words you 737 00:43:33,560 --> 00:43:37,600 Speaker 1: use and and tighten that communication and loyalty and trust. 738 00:43:37,840 --> 00:43:40,400 Speaker 1: But with that trust becomes a big responsibility to make 739 00:43:40,440 --> 00:43:42,680 Speaker 1: sure a you're protecting the data. You've got to be 740 00:43:42,719 --> 00:43:44,960 Speaker 1: secure to the core in these systems, and then to 741 00:43:45,080 --> 00:43:48,839 Speaker 1: extend it back to explainability, back to removing bias. A 742 00:43:48,880 --> 00:43:51,680 Speaker 1: lot of that last part on biases. How are you 743 00:43:51,719 --> 00:43:55,200 Speaker 1: training the systems if you're putting in uh, let's say 744 00:43:55,560 --> 00:43:58,359 Speaker 1: nurses and doctors from you know, data dumped from last 745 00:43:58,400 --> 00:44:02,680 Speaker 1: thirty years um today you know there's a great inequality 746 00:44:02,719 --> 00:44:05,400 Speaker 1: with women and men, and and we're all doing our 747 00:44:05,480 --> 00:44:09,120 Speaker 1: part to make sure we conn accelerate the turn or 748 00:44:09,120 --> 00:44:11,120 Speaker 1: the pivot we need to make. But in that case, 749 00:44:11,200 --> 00:44:14,279 Speaker 1: what if you pump into a system thirty years where 750 00:44:14,320 --> 00:44:17,960 Speaker 1: it's been skewed where nurses are typically women, right, and 751 00:44:18,000 --> 00:44:21,600 Speaker 1: now you're feeding out data into some models um and 752 00:44:21,600 --> 00:44:24,200 Speaker 1: and these algorithms, and at the end of the day 753 00:44:24,440 --> 00:44:27,120 Speaker 1: it might give a different result, but you might scratch 754 00:44:27,160 --> 00:44:28,680 Speaker 1: your head and said, well, over the last five years 755 00:44:28,680 --> 00:44:30,640 Speaker 1: there's a lot more male nurses than there were thirty 756 00:44:30,719 --> 00:44:34,000 Speaker 1: years ago. So that's a point of again a responsibility. 757 00:44:34,160 --> 00:44:37,239 Speaker 1: It's so important to have that end to end process 758 00:44:37,680 --> 00:44:39,960 Speaker 1: on how again, what's the problem you're trying to solve. 759 00:44:40,239 --> 00:44:41,960 Speaker 1: And then as you go through that end to end 760 00:44:42,000 --> 00:44:44,960 Speaker 1: you've got the data responsibly. You have to load the data, 761 00:44:45,160 --> 00:44:47,280 Speaker 1: you have to understand the data, you have to protect 762 00:44:47,280 --> 00:44:50,839 Speaker 1: the data again that's IP intellectual property. And then as 763 00:44:50,880 --> 00:44:54,759 Speaker 1: you feed into the models, you want to have that explainability, traceability, 764 00:44:54,880 --> 00:44:57,600 Speaker 1: and ultimately the ethics which equates in large part to 765 00:44:57,960 --> 00:45:01,160 Speaker 1: the bias and and so you have that responsibility and 766 00:45:01,160 --> 00:45:02,600 Speaker 1: that at the end of the day, that's what IDEM 767 00:45:02,680 --> 00:45:05,319 Speaker 1: is very proud of. We have tools and capabilities to 768 00:45:05,360 --> 00:45:08,440 Speaker 1: make sure that there's integrity throughout the whole end to 769 00:45:08,560 --> 00:45:13,839 Speaker 1: end process. So obviously the COVID nineteen pandemic has has 770 00:45:13,880 --> 00:45:18,520 Speaker 1: forced UH companies and institutions to to take up AI 771 00:45:18,600 --> 00:45:21,480 Speaker 1: technology to implement it. Um, you know, just to to 772 00:45:21,640 --> 00:45:26,960 Speaker 1: get through this time period. But what lasting changes do 773 00:45:27,000 --> 00:45:30,160 Speaker 1: you do you see, um really sticking with us from 774 00:45:30,200 --> 00:45:31,880 Speaker 1: this that are really going to benefit us in the 775 00:45:31,920 --> 00:45:36,080 Speaker 1: long run. Yeah, it's a great question, Rob, and I think, 776 00:45:36,400 --> 00:45:38,839 Speaker 1: you know, I don't think anyone has the exact answer, right, 777 00:45:38,880 --> 00:45:41,520 Speaker 1: And we are in a new norm and it's to 778 00:45:41,640 --> 00:45:44,080 Speaker 1: be defined as we move forward. But when I look 779 00:45:44,120 --> 00:45:48,360 Speaker 1: at a lot of the AI and other technologies, blockchain included, 780 00:45:48,920 --> 00:45:53,239 Speaker 1: and I look at organizations that maybe are you know, 781 00:45:53,360 --> 00:45:56,839 Speaker 1: in a function. It could be a government organization and 782 00:45:56,920 --> 00:46:01,879 Speaker 1: let's say they're they're um, they're remit is to make 783 00:46:01,920 --> 00:46:06,680 Speaker 1: sure that critical supplies move across their supply chains. Right. 784 00:46:06,760 --> 00:46:10,000 Speaker 1: It could be let's just say, ventilators, instead of redirecting 785 00:46:10,680 --> 00:46:13,480 Speaker 1: a hundred ventilators to Atlanta, maybe you redirect them to 786 00:46:13,520 --> 00:46:16,360 Speaker 1: New York City. Right. Um. So we know there's a 787 00:46:16,400 --> 00:46:18,840 Speaker 1: lot of great technology out there that does that, but 788 00:46:18,880 --> 00:46:22,600 Speaker 1: a lot of that beyond the predictive analytics, it's gonna 789 00:46:22,600 --> 00:46:28,080 Speaker 1: be situational awareness, that whole operational situational awareness. And so 790 00:46:28,200 --> 00:46:30,640 Speaker 1: that's just an example of supply chain. So right now 791 00:46:31,120 --> 00:46:34,560 Speaker 1: it's pertinent because it's COVID nineteen. But what if you 792 00:46:34,920 --> 00:46:37,840 Speaker 1: put that solution in there, and we are exploring some 793 00:46:37,920 --> 00:46:41,480 Speaker 1: things with one of the organizations UM, and you have 794 00:46:41,520 --> 00:46:43,600 Speaker 1: a short term solution, but over the longer term, what 795 00:46:43,680 --> 00:46:47,600 Speaker 1: if it's hurricanes And this next chapter you're gonna be 796 00:46:47,719 --> 00:46:50,840 Speaker 1: using a lot of these same technologies and potential solutions, 797 00:46:50,880 --> 00:46:54,160 Speaker 1: and you'll just iterate on these solutions. So I don't 798 00:46:54,280 --> 00:46:56,440 Speaker 1: view a lot of these solutions. Going back to the 799 00:46:56,480 --> 00:47:00,160 Speaker 1: Children's Hospital of Atlanta. Yes, the immediate need is to 800 00:47:00,160 --> 00:47:04,799 Speaker 1: to be that UM Watson assistant for citizens, but over 801 00:47:04,920 --> 00:47:08,120 Speaker 1: time that could evolve to other areas where they want 802 00:47:08,120 --> 00:47:10,720 Speaker 1: to use AI. Back to my earlier point about scaling 803 00:47:11,040 --> 00:47:14,040 Speaker 1: in other use cases across their hospital, or back to 804 00:47:14,080 --> 00:47:17,520 Speaker 1: this organization. And then the only other point I would 805 00:47:17,600 --> 00:47:20,840 Speaker 1: highlight is UM Again, there's a lot of great technology 806 00:47:20,840 --> 00:47:23,399 Speaker 1: out there today around data and analytics, and our view 807 00:47:23,400 --> 00:47:27,759 Speaker 1: AI is very complementary. But let's go back to hybrid cloud, right, 808 00:47:27,840 --> 00:47:30,440 Speaker 1: and let's go back to Kubernetes and red hat and 809 00:47:30,440 --> 00:47:32,839 Speaker 1: and how all these things are coming together. These are 810 00:47:32,920 --> 00:47:38,080 Speaker 1: game changing, transformational solutions and and so beyond the covid 811 00:47:38,640 --> 00:47:42,040 Speaker 1: um period here at the pandemic um, there's gonna be 812 00:47:42,040 --> 00:47:44,400 Speaker 1: a lot more opportunity to continue to take those and 813 00:47:44,440 --> 00:47:48,280 Speaker 1: go faster and further. And and so my point being 814 00:47:48,360 --> 00:47:51,239 Speaker 1: is when you look at coming back to the logistics, 815 00:47:51,320 --> 00:47:53,799 Speaker 1: I'm gonna shift gears just for a quick moment. Look 816 00:47:53,800 --> 00:47:56,520 Speaker 1: at what we're doing around food safety with blockchain, right, 817 00:47:56,520 --> 00:48:00,400 Speaker 1: We've been partnered with Walmart, um and others around making 818 00:48:00,400 --> 00:48:03,520 Speaker 1: sure that you've got better traceability across that whole supply 819 00:48:03,600 --> 00:48:06,040 Speaker 1: chain because people get sick and some are definitely sick 820 00:48:06,600 --> 00:48:10,640 Speaker 1: um over some of these uh scenarios. So if you 821 00:48:10,680 --> 00:48:14,600 Speaker 1: can take blockchain, in AI, in cloud and move put 822 00:48:14,600 --> 00:48:18,080 Speaker 1: all these together to provide these solutions, this is there 823 00:48:18,120 --> 00:48:20,279 Speaker 1: for the long term, right. And so again we come 824 00:48:20,280 --> 00:48:23,720 Speaker 1: back to point solutions right now in this immediate crisis. 825 00:48:24,040 --> 00:48:26,359 Speaker 1: But I think a lot of these solutions live on 826 00:48:26,600 --> 00:48:28,720 Speaker 1: and I think they get better. I think they scale. 827 00:48:29,000 --> 00:48:31,560 Speaker 1: And back to the point about maybe now it's worried 828 00:48:31,560 --> 00:48:34,800 Speaker 1: about ventilators, Tomorrow it could be worried about a hurricane 829 00:48:34,840 --> 00:48:38,480 Speaker 1: disaster and moving other critical supplies across the US or 830 00:48:38,520 --> 00:48:41,480 Speaker 1: even globally. You know, we we cover a lot of 831 00:48:41,600 --> 00:48:46,640 Speaker 1: inventions on this show. We discuss how different technologies emerge 832 00:48:46,680 --> 00:48:50,560 Speaker 1: and how they're rolled out, And it really is fascinating 833 00:48:50,600 --> 00:48:54,120 Speaker 1: to think about about AI UM and and these examples 834 00:48:54,360 --> 00:48:57,319 Speaker 1: you've you've brought up already, uh and and just how 835 00:48:57,360 --> 00:48:59,640 Speaker 1: different it is from other technologies. So many of these 836 00:48:59,640 --> 00:49:02,239 Speaker 1: technol have come out and and it's it's you know, 837 00:49:02,320 --> 00:49:06,040 Speaker 1: it's instantly going to be used for our uh you know, 838 00:49:06,080 --> 00:49:08,279 Speaker 1: our our baser instincts, so or it is going to 839 00:49:08,360 --> 00:49:12,120 Speaker 1: be misused in some fashion because there's not you know, 840 00:49:12,160 --> 00:49:16,120 Speaker 1: how how ethical, how how how how ethical can can 841 00:49:16,160 --> 00:49:19,000 Speaker 1: an invention be for the most part when you're talking 842 00:49:19,040 --> 00:49:23,840 Speaker 1: about some sort of you know, basic energy um UM technology, 843 00:49:23,920 --> 00:49:27,040 Speaker 1: But with AI you're you're talking about the the the 844 00:49:27,080 --> 00:49:31,319 Speaker 1: ethical use of the thing being rolled up, in its development, 845 00:49:31,400 --> 00:49:35,879 Speaker 1: in its actual existence. That's exactly right. And again, as 846 00:49:35,920 --> 00:49:37,680 Speaker 1: I said at the outstart, I mean we all have 847 00:49:37,760 --> 00:49:43,799 Speaker 1: a responsibility, right IBM competitors, governments, UM institutions. In that 848 00:49:43,880 --> 00:49:46,160 Speaker 1: earlier example, Rob, when we were we were going through 849 00:49:46,160 --> 00:49:50,359 Speaker 1: that fictitious example of you and I applying for for loan, 850 00:49:50,480 --> 00:49:52,439 Speaker 1: what if I got kicked out? You got to prove 851 00:49:52,840 --> 00:49:55,319 Speaker 1: just think internal to that bank, they have to go 852 00:49:55,360 --> 00:49:58,600 Speaker 1: through their own validation and due diligence to make sure 853 00:49:58,640 --> 00:50:00,560 Speaker 1: it's clear. And then you might have on top of 854 00:50:00,560 --> 00:50:03,080 Speaker 1: that regulators that we want to look at this. So 855 00:50:03,200 --> 00:50:06,120 Speaker 1: it does bear a lot of responsibility, but the point 856 00:50:06,160 --> 00:50:08,120 Speaker 1: is we all have to usher this in in a 857 00:50:08,200 --> 00:50:11,080 Speaker 1: very responsible way. And UM I have the good fortune 858 00:50:11,120 --> 00:50:14,440 Speaker 1: of sitting in on our part of our AI Ethics 859 00:50:14,440 --> 00:50:17,040 Speaker 1: board with an IBM, and I can tell you we 860 00:50:17,040 --> 00:50:20,400 Speaker 1: we have great focus. We meet literally every week and 861 00:50:20,400 --> 00:50:23,720 Speaker 1: we're always pushing ourselves to think about these types of things, 862 00:50:24,080 --> 00:50:28,480 Speaker 1: whether it be UM you know, UH facial detection or 863 00:50:28,800 --> 00:50:31,480 Speaker 1: just the way we talked about some of the uses 864 00:50:32,000 --> 00:50:34,799 Speaker 1: of this technology. We we have to be very responsible 865 00:50:34,840 --> 00:50:38,360 Speaker 1: and make sure that again we're working together and ultimately 866 00:50:38,440 --> 00:50:41,200 Speaker 1: with the governments. Every country is different, right in Europe. 867 00:50:41,640 --> 00:50:44,880 Speaker 1: You've seen a lot over the last year, especially with 868 00:50:45,000 --> 00:50:47,360 Speaker 1: g d r P and other types of requirements, you know, 869 00:50:47,360 --> 00:50:51,080 Speaker 1: whether it be the cloud, the the AI data protection 870 00:50:51,480 --> 00:50:53,279 Speaker 1: UM so every country is a little different. I learned 871 00:50:53,400 --> 00:50:55,759 Speaker 1: a lot of that just being in countries and Singapore 872 00:50:55,760 --> 00:51:00,000 Speaker 1: and Japan, in Thailand, and everyone does it differently. But regardless, 873 00:51:00,040 --> 00:51:03,120 Speaker 1: the great thing about IBM. From my perspective is that 874 00:51:03,520 --> 00:51:05,920 Speaker 1: we're in all those countries and we're very active with 875 00:51:06,000 --> 00:51:08,759 Speaker 1: the governments, and we're very active with the companies, were 876 00:51:08,840 --> 00:51:11,920 Speaker 1: very active with academia like M I T S, and 877 00:51:11,960 --> 00:51:16,000 Speaker 1: we're all working together to find out and explore what 878 00:51:16,000 --> 00:51:18,520 Speaker 1: what are those ways because as you know Rob and Joe, 879 00:51:18,520 --> 00:51:21,640 Speaker 1: it's an evolution. There is no set answer. It's always 880 00:51:21,719 --> 00:51:24,440 Speaker 1: changing and this technology is only going to go faster 881 00:51:24,840 --> 00:51:27,680 Speaker 1: and be better, and that needs to We need to 882 00:51:27,760 --> 00:51:30,480 Speaker 1: ensure everyone's on their toes, so to speak, because it's 883 00:51:30,480 --> 00:51:35,080 Speaker 1: a big responsibility. I'm I excited about AI, super excited. 884 00:51:35,239 --> 00:51:39,319 Speaker 1: It's a game changer, but only if we all make 885 00:51:39,360 --> 00:51:42,959 Speaker 1: sure that we're developing it in a very responsible way 886 00:51:43,000 --> 00:51:49,000 Speaker 1: and we're working together to make it happen. All right, 887 00:51:49,040 --> 00:51:51,200 Speaker 1: So there you have it. Thanks once more to Ridka 888 00:51:51,239 --> 00:51:53,759 Speaker 1: Gunner and Jay Bellissimo for taking time out of their 889 00:51:53,840 --> 00:51:56,080 Speaker 1: day to chat with us here. And if you'd like 890 00:51:56,160 --> 00:51:59,399 Speaker 1: to learn more about Watson Assistant, just go to IBM 891 00:51:59,440 --> 00:52:03,600 Speaker 1: dot com slash Watson, slash COVID dash Response and you 892 00:52:03,640 --> 00:52:06,920 Speaker 1: can also check out IBM dot com slash smart Talks 893 00:52:07,320 --> 00:52:10,680 Speaker 1: for more information about the topics were discussing here and 894 00:52:10,719 --> 00:52:13,040 Speaker 1: if you would like to listen to additional episodes of 895 00:52:13,040 --> 00:52:15,440 Speaker 1: Stuff to Blow Your Mind, you can find us wherever 896 00:52:15,480 --> 00:52:17,839 Speaker 1: you get your podcast. We just asked the you rate, 897 00:52:17,920 --> 00:52:22,160 Speaker 1: review and subscribe huge thanks as always to our heroic 898 00:52:22,200 --> 00:52:24,960 Speaker 1: audio producer Seth Nicholas Johnson. If you would like to 899 00:52:25,000 --> 00:52:27,240 Speaker 1: get in touch with us with feedback on this episode 900 00:52:27,320 --> 00:52:29,360 Speaker 1: or any other, to suggest a topic for the future, 901 00:52:29,480 --> 00:52:31,919 Speaker 1: or just to say hello, you can email us at 902 00:52:32,000 --> 00:52:42,560 Speaker 1: contact at stuff to Blow your Mind dot com. Stuff 903 00:52:42,560 --> 00:52:44,760 Speaker 1: to Blow Your Mind is production of I Heart Radio. 904 00:52:45,120 --> 00:52:47,440 Speaker 1: For more podcasts for my heart Radio, visit the iHeart 905 00:52:47,520 --> 00:52:50,239 Speaker 1: Radio app, Apple Podcasts, or wherever you listening to your 906 00:52:50,280 --> 00:52:55,640 Speaker 1: favorite shows. B b b b b bla bla bla 907 00:52:55,640 --> 00:53:03,480 Speaker 1: Bliss Greeted by part