1 00:00:00,160 --> 00:00:02,840 Speaker 1: Hey everyone, it's Robert and Joe here. Today we've got 2 00:00:02,880 --> 00:00:04,760 Speaker 1: something a little bit different to share with you. It 3 00:00:04,840 --> 00:00:08,039 Speaker 1: is a new season of the Smart Talks with IBM 4 00:00:08,160 --> 00:00:09,160 Speaker 1: podcast series. 5 00:00:09,640 --> 00:00:11,639 Speaker 2: Today we are witnessed to one of those rare moments 6 00:00:11,680 --> 00:00:14,400 Speaker 2: in history, the rise of an innovative technology with the 7 00:00:14,400 --> 00:00:18,680 Speaker 2: potential to radically transform business and society forever. The technology, 8 00:00:18,760 --> 00:00:22,240 Speaker 2: of course, is artificial intelligence, and it's the central focus 9 00:00:22,239 --> 00:00:24,840 Speaker 2: for this new season of Smart Talks with IBM. 10 00:00:25,320 --> 00:00:28,400 Speaker 1: Join hosts from your favorite Pushkin podcasts as they talk 11 00:00:28,480 --> 00:00:31,680 Speaker 1: with industry experts and leaders to explore how businesses can 12 00:00:31,720 --> 00:00:35,400 Speaker 1: integrate AI into their workflows and help drive real change 13 00:00:35,400 --> 00:00:38,240 Speaker 1: in this new era of AI. And of course, host 14 00:00:38,280 --> 00:00:40,479 Speaker 1: Malcolm Gladwell will be there to guide you through the 15 00:00:40,479 --> 00:00:42,640 Speaker 1: season and throw in his two cents as well. 16 00:00:43,120 --> 00:00:46,200 Speaker 2: Look out for new episodes of Smart Talks with IBM 17 00:00:46,400 --> 00:00:49,559 Speaker 2: every other week on the iHeartRadio app, Apple Podcasts, or 18 00:00:49,600 --> 00:00:53,360 Speaker 2: wherever you get your podcasts. And learn more at IBM 19 00:00:53,479 --> 00:00:55,480 Speaker 2: dot com slash smart Talks. 20 00:00:56,120 --> 00:00:59,760 Speaker 3: Hey, it's Jacob Goldstein for Smart Talks with IBM. Last 21 00:00:59,840 --> 00:01:02,280 Speaker 3: year I had the pleasure of sitting down with doctor 22 00:01:02,360 --> 00:01:07,399 Speaker 3: David Cox, VP of AI Models at IBM Research. We 23 00:01:07,480 --> 00:01:11,200 Speaker 3: explored the fascinating world of AI foundation models and their 24 00:01:11,240 --> 00:01:15,919 Speaker 3: revolutionary potential for business automation and innovation. When we first 25 00:01:16,000 --> 00:01:19,319 Speaker 3: aired this episode, the concept of foundation models was just 26 00:01:19,400 --> 00:01:23,280 Speaker 3: beginning to capture our attention. Since then, this technology has 27 00:01:23,400 --> 00:01:27,480 Speaker 3: evolved and redefined the boundaries of what's possible. Businesses are 28 00:01:27,520 --> 00:01:30,959 Speaker 3: becoming more savvy about selecting the right models and understanding 29 00:01:30,959 --> 00:01:34,360 Speaker 3: how they can drive revenue and efficiency. As I listened 30 00:01:34,360 --> 00:01:37,280 Speaker 3: back to the conversation, it was interesting to reflect on 31 00:01:37,319 --> 00:01:41,120 Speaker 3: some new developments and ideas that have emerged, and many 32 00:01:41,160 --> 00:01:44,320 Speaker 3: of these we will continue to explore throughout the season, 33 00:01:44,840 --> 00:01:46,920 Speaker 3: like how to play an active role in choosing the 34 00:01:46,920 --> 00:01:50,440 Speaker 3: best model for your needs. Whether you're a longtime listener 35 00:01:50,520 --> 00:01:52,920 Speaker 3: or tuning in for the first time, I'm certain you'll 36 00:01:52,920 --> 00:01:56,680 Speaker 3: find doctor Cox's insights as thought provoking as ever. Thanks 37 00:01:56,680 --> 00:01:59,760 Speaker 3: as always for joining us. Now let's dive in. 38 00:02:01,440 --> 00:02:05,240 Speaker 4: Hello, Hello, Welcome to Smart Talks with IBM, a podcast 39 00:02:05,240 --> 00:02:10,720 Speaker 4: from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell. Our 40 00:02:10,760 --> 00:02:15,440 Speaker 4: guest today is doctor David Cox, VP of AI Models 41 00:02:15,480 --> 00:02:20,480 Speaker 4: at IBM Research and IBM Director of the MIT IBM 42 00:02:20,560 --> 00:02:24,800 Speaker 4: Watson AI Lab, a first of its kind industry academic 43 00:02:24,919 --> 00:02:30,000 Speaker 4: collaboration between IBM and MIT focused on the fundamental research 44 00:02:30,400 --> 00:02:35,160 Speaker 4: of artificial intelligence. Over the course of decades, David Cox 45 00:02:35,240 --> 00:02:39,880 Speaker 4: watched as the AI revolution steadily grew from the simmering 46 00:02:39,919 --> 00:02:43,760 Speaker 4: ideas of a few academics and technologists into the industrial 47 00:02:43,840 --> 00:02:48,160 Speaker 4: boom we are experiencing today. Having dedicated his life to 48 00:02:48,240 --> 00:02:51,640 Speaker 4: pushing the field of AI towards new horizons, David has 49 00:02:51,680 --> 00:02:55,600 Speaker 4: both contributed to and presided over many of the major 50 00:02:55,720 --> 00:03:01,320 Speaker 4: breakthroughs in artificial intelligence. In today's episode, you'll hear David 51 00:03:01,320 --> 00:03:06,440 Speaker 4: explain some of the conceptual underpinnings of the current AI landscape, 52 00:03:06,600 --> 00:03:11,240 Speaker 4: things like foundation models, in surprisingly comprehensible terms, I might add, 53 00:03:11,800 --> 00:03:14,959 Speaker 4: we'll also get into some of the amazing practical applications 54 00:03:14,960 --> 00:03:18,000 Speaker 4: for AI in business, as well as what implications AI 55 00:03:18,120 --> 00:03:21,640 Speaker 4: will have for the future of work and design. David 56 00:03:21,680 --> 00:03:25,400 Speaker 4: spoke with Jacob Goldstein, host of the Pushkin podcast What's 57 00:03:25,440 --> 00:03:29,720 Speaker 4: Your Problem. A veteran business journalist, Jacob has reported for 58 00:03:29,760 --> 00:03:32,560 Speaker 4: The Wall Street Journal, the Miami Herald, and was a 59 00:03:32,560 --> 00:03:38,200 Speaker 4: longtime host of the NPR program Planet Money. Okay, let's 60 00:03:38,240 --> 00:03:39,120 Speaker 4: get to the interview. 61 00:03:41,480 --> 00:03:43,240 Speaker 3: Tell me about your job at IBM. 62 00:03:43,840 --> 00:03:47,160 Speaker 5: So I wear two hats at IBM. So one, I'm 63 00:03:47,200 --> 00:03:50,080 Speaker 5: the IBM Doctor of the MIT IBM Watson AI Lab. 64 00:03:50,560 --> 00:03:53,600 Speaker 5: So that's a joint lab between IBM and MIT where 65 00:03:53,640 --> 00:03:56,200 Speaker 5: we try and invent what's next in AI. It's been 66 00:03:56,280 --> 00:03:59,120 Speaker 5: running for about five years, and then more recently I 67 00:03:59,160 --> 00:04:02,400 Speaker 5: started as Vice president for AI Models, and I'm in 68 00:04:02,480 --> 00:04:07,040 Speaker 5: charge of building IBM's foundation models, you know, building these 69 00:04:07,120 --> 00:04:09,240 Speaker 5: these big models, generative models that allow us to have 70 00:04:09,240 --> 00:04:11,320 Speaker 5: all kinds of new exciting capabilities in AI. 71 00:04:11,840 --> 00:04:13,760 Speaker 3: So, so I want to talk to you a lot 72 00:04:13,840 --> 00:04:17,240 Speaker 3: about foundation models, about genitive AI. But before we get 73 00:04:17,279 --> 00:04:19,360 Speaker 3: to that, let's just spend a minute on the on 74 00:04:19,440 --> 00:04:25,080 Speaker 3: the IBM MIT collaboration. Where did that partnership start, How 75 00:04:25,080 --> 00:04:25,920 Speaker 3: did it originate? 76 00:04:27,080 --> 00:04:29,880 Speaker 5: Yeah, So, actually it turns out that MIT and IBM 77 00:04:30,120 --> 00:04:33,200 Speaker 5: have been collaborating for a very long time in the 78 00:04:33,240 --> 00:04:37,280 Speaker 5: area of AI. In fact, the term artificial intelligence was 79 00:04:37,360 --> 00:04:41,000 Speaker 5: coined in a nineteen fifty six workshop that was held 80 00:04:41,040 --> 00:04:43,200 Speaker 5: at Dartmouth. It was actually organized by an IBM or 81 00:04:43,279 --> 00:04:46,560 Speaker 5: Nathaniel Rochester, who led the development of the IBM seven 82 00:04:46,600 --> 00:04:49,840 Speaker 5: and one. So we've really been together in AIS since 83 00:04:49,880 --> 00:04:54,719 Speaker 5: the beginning and as AI kept accelerating more and more 84 00:04:54,760 --> 00:04:58,360 Speaker 5: and more, I think there was a really interesting decision 85 00:04:58,360 --> 00:05:01,080 Speaker 5: to say, let's make this a formal partnership. So IBM 86 00:05:01,120 --> 00:05:02,960 Speaker 5: in twenty seventeen and now, so it'll be committing close 87 00:05:02,960 --> 00:05:05,840 Speaker 5: to a quarter billion dollars over ten years to have 88 00:05:05,920 --> 00:05:09,799 Speaker 5: this joint lab with MIT, and we we located ourselves 89 00:05:09,839 --> 00:05:12,000 Speaker 5: right on the campus and we've been developing very very 90 00:05:12,080 --> 00:05:14,359 Speaker 5: deep relationships where we can you know, really get to 91 00:05:14,400 --> 00:05:17,479 Speaker 5: know each other, work shoulder to shoulder, conceiving what we 92 00:05:17,480 --> 00:05:20,120 Speaker 5: should work on next, and then executing the projects. And 93 00:05:20,200 --> 00:05:24,159 Speaker 5: it's really you know, very few entities like this exist 94 00:05:24,640 --> 00:05:27,640 Speaker 5: between academia industry. It's been really fun of the last 95 00:05:27,680 --> 00:05:28,920 Speaker 5: five years to be a part of it. 96 00:05:29,560 --> 00:05:31,080 Speaker 3: And what do you think are some of the most 97 00:05:31,080 --> 00:05:34,600 Speaker 3: important outcomes of this collaboration between IBM and MIT. 98 00:05:36,000 --> 00:05:38,680 Speaker 5: Yeah, so we're really kind of the tip of the 99 00:05:38,720 --> 00:05:43,640 Speaker 5: sphere for for IBM's the I strategy. So we're really looking, 100 00:05:43,880 --> 00:05:46,680 Speaker 5: you know, what's coming ahead, and you know, in areas 101 00:05:46,720 --> 00:05:50,480 Speaker 5: like foundation models, you know, as the field changes, MIT 102 00:05:50,640 --> 00:05:53,279 Speaker 5: people are interested in working on you know, faculty, students 103 00:05:53,279 --> 00:05:55,440 Speaker 5: and staff are interested in working on what's the latest thing, 104 00:05:55,480 --> 00:05:58,520 Speaker 5: what's the next thing. We at IBM Research are very 105 00:05:58,560 --> 00:06:00,839 Speaker 5: much interested in the same. We can kind of put 106 00:06:00,839 --> 00:06:03,719 Speaker 5: out feelers, you know, interesting things that we're seeing in 107 00:06:03,760 --> 00:06:06,480 Speaker 5: our research, interesting things we're hearing in the field. We 108 00:06:06,520 --> 00:06:09,640 Speaker 5: can go and chase those opportunities. So when something big comes, 109 00:06:09,760 --> 00:06:12,719 Speaker 5: like the big change that's been happening lately with foundation models, 110 00:06:12,720 --> 00:06:15,120 Speaker 5: we're ready to jump on it. That's really the purpose, 111 00:06:15,200 --> 00:06:18,160 Speaker 5: that's that's the lab functioning the way it should. We're 112 00:06:18,200 --> 00:06:21,680 Speaker 5: also really interested in how do we advance you know, 113 00:06:21,839 --> 00:06:24,560 Speaker 5: AI that can help with climate change or you know, 114 00:06:24,680 --> 00:06:27,320 Speaker 5: build better materials and all these kinds of things that 115 00:06:27,360 --> 00:06:30,320 Speaker 5: are you know, a broader aperture sometimes than what we 116 00:06:30,400 --> 00:06:33,400 Speaker 5: might consider just looking at the product portfolio of IBM, 117 00:06:33,720 --> 00:06:35,480 Speaker 5: and that that gives us again a breadth where we 118 00:06:35,520 --> 00:06:38,160 Speaker 5: can see connections that we might not have seen otherwise. 119 00:06:38,520 --> 00:06:41,080 Speaker 5: We can you know, think things that help out society 120 00:06:41,200 --> 00:06:42,680 Speaker 5: and also help out our customers. 121 00:06:43,480 --> 00:06:47,920 Speaker 3: So the last whatever six months, say, there has been 122 00:06:47,960 --> 00:06:53,680 Speaker 3: this wild rise in the public's interest in AI right 123 00:06:53,720 --> 00:06:57,160 Speaker 3: clearly coming out of these generative AI models that are 124 00:06:57,160 --> 00:07:01,279 Speaker 3: really accessible you know, certainly chat GPT language models like that, 125 00:07:01,320 --> 00:07:04,400 Speaker 3: as well as models that generate images like mid journey. 126 00:07:05,000 --> 00:07:08,480 Speaker 3: I mean, can you just sort of briefly talk about 127 00:07:07,920 --> 00:07:11,920 Speaker 3: the breakthroughs in AI that have made this moment feel 128 00:07:12,000 --> 00:07:15,520 Speaker 3: so exciting, so revolutionary for artificial intelligence. 129 00:07:16,560 --> 00:07:21,280 Speaker 5: Yeah. You know, I've been studying AI basically my entire 130 00:07:21,320 --> 00:07:23,440 Speaker 5: adult life. Before I came to IBM, I was a 131 00:07:23,440 --> 00:07:26,000 Speaker 5: professor at Harvard. I've been doing this a long time, 132 00:07:26,240 --> 00:07:28,520 Speaker 5: and I've gotten used to being surprised. It sounds like 133 00:07:28,560 --> 00:07:31,640 Speaker 5: a joke, but it's serious, Like I'm getting used to 134 00:07:31,680 --> 00:07:35,600 Speaker 5: being surprised at the acceleration of the pace again. It 135 00:07:35,680 --> 00:07:38,560 Speaker 5: tracks actually a long way back. You know, there's lots 136 00:07:38,600 --> 00:07:41,160 Speaker 5: of things where there was an idea that just simmered 137 00:07:41,840 --> 00:07:45,200 Speaker 5: for a really long time. Some of the key math 138 00:07:45,680 --> 00:07:49,280 Speaker 5: behind the stuff that we have today, which is amazing. 139 00:07:50,000 --> 00:07:52,880 Speaker 5: There's an algorithm called back propagation, which is sort of 140 00:07:52,960 --> 00:07:55,600 Speaker 5: key to training neural networks that's been around, you know, 141 00:07:55,640 --> 00:07:59,520 Speaker 5: since the eighties in wide use. And really what happened 142 00:07:59,640 --> 00:08:03,320 Speaker 5: was it simmered for a long time, and then enough 143 00:08:03,480 --> 00:08:07,239 Speaker 5: data and enough compute came so we had enough data 144 00:08:07,280 --> 00:08:11,240 Speaker 5: because you know, we all started carrying multiple cameras around 145 00:08:11,240 --> 00:08:13,640 Speaker 5: with us. Our mobile phones have all you know, all 146 00:08:13,640 --> 00:08:16,400 Speaker 5: these cameras and this we put everything on the Internet, 147 00:08:16,520 --> 00:08:18,760 Speaker 5: and there's all this data out there. We caught a 148 00:08:18,840 --> 00:08:21,320 Speaker 5: lucky break that there was something called the graphics processing unit, 149 00:08:21,400 --> 00:08:23,640 Speaker 5: which turns out to be really useful for doing these 150 00:08:23,720 --> 00:08:26,360 Speaker 5: kinds of algorithms, maybe even more useful than it is 151 00:08:26,400 --> 00:08:30,360 Speaker 5: for doing graphics. They're greater graphics too, And things just 152 00:08:30,480 --> 00:08:33,080 Speaker 5: kept kind of adding to the snowball. So we had 153 00:08:33,240 --> 00:08:37,439 Speaker 5: deep learning, which is sort of a rebrand of neural 154 00:08:37,480 --> 00:08:39,960 Speaker 5: networks that I mentioned from the eighties, and that was 155 00:08:40,040 --> 00:08:43,199 Speaker 5: enabled again by data because we digitalized the world and 156 00:08:43,679 --> 00:08:46,520 Speaker 5: compute because we kept building faster and faster and more 157 00:08:46,520 --> 00:08:49,760 Speaker 5: powerful computers, and then that allowed us to make this 158 00:08:49,760 --> 00:08:53,440 Speaker 5: this big breakthrough. And then, you know, more recently, using 159 00:08:53,520 --> 00:08:57,600 Speaker 5: the same building blocks, that inexorable rise of more and 160 00:08:57,640 --> 00:09:02,160 Speaker 5: more and more data, that technology called self supervised learning. 161 00:09:02,640 --> 00:09:07,360 Speaker 5: Where the key difference there in traditional deep learning, you know, 162 00:09:07,400 --> 00:09:10,040 Speaker 5: for classifying images, you know, like is this a cat 163 00:09:10,120 --> 00:09:13,320 Speaker 5: or is this a dog? And a picture those technologies 164 00:09:13,800 --> 00:09:17,120 Speaker 5: require supper visions, so you have to take what you 165 00:09:17,200 --> 00:09:18,560 Speaker 5: have and then you have to label it. So you 166 00:09:18,600 --> 00:09:19,920 Speaker 5: have to take a picture of a cat and then 167 00:09:19,960 --> 00:09:22,640 Speaker 5: you label it as a cat, and it turns out that, 168 00:09:22,800 --> 00:09:25,000 Speaker 5: you know, that's very powerful, but it takes a lot 169 00:09:25,000 --> 00:09:27,920 Speaker 5: of time to label gats and to label dogs, and 170 00:09:28,360 --> 00:09:30,280 Speaker 5: there's only so many labels that exist in the world. 171 00:09:30,679 --> 00:09:34,240 Speaker 5: So what really changed more recently is that we have 172 00:09:34,320 --> 00:09:36,800 Speaker 5: self supervised learning where you don't have to have the labels. 173 00:09:36,800 --> 00:09:39,360 Speaker 5: We can just take unannotated data. And what that does 174 00:09:39,400 --> 00:09:42,480 Speaker 5: is it allows you use even more data. And that's 175 00:09:42,520 --> 00:09:46,120 Speaker 5: really what drove this this latest sort of rage. And 176 00:09:46,120 --> 00:09:48,280 Speaker 5: then and then all of a sudden we start getting 177 00:09:48,320 --> 00:09:52,199 Speaker 5: these these really powerful models. And then really, this has 178 00:09:52,240 --> 00:09:57,040 Speaker 5: been simmering technologies, right, this has been happening for a 179 00:09:57,080 --> 00:10:01,280 Speaker 5: while and progressively getting more and more powerful. One of 180 00:10:01,280 --> 00:10:05,560 Speaker 5: the things that really happened with CHATGBT and technologies like 181 00:10:06,000 --> 00:10:09,079 Speaker 5: Stable Diffusion and mid Journey was that they made it 182 00:10:09,640 --> 00:10:12,319 Speaker 5: visible to the public. You know, you put it out 183 00:10:12,360 --> 00:10:14,600 Speaker 5: there the public can touch and feel and they're like, wow, 184 00:10:14,920 --> 00:10:18,480 Speaker 5: not only is there palpable change, and wow this you know, 185 00:10:18,520 --> 00:10:20,000 Speaker 5: I can talk to this thing. Wow, this thing can 186 00:10:20,080 --> 00:10:22,959 Speaker 5: generate an image. Not only that, but everyone can touch 187 00:10:23,000 --> 00:10:27,280 Speaker 5: and feel and try. My kids can use some of 188 00:10:27,280 --> 00:10:32,720 Speaker 5: these AI art generation technologies, and that's really just launched. 189 00:10:32,800 --> 00:10:36,040 Speaker 5: You know, it's like a propelled slingshot at us into 190 00:10:36,360 --> 00:10:38,400 Speaker 5: a different regime. In terms of the public awareness of 191 00:10:38,400 --> 00:10:39,239 Speaker 5: these technologies. 192 00:10:39,920 --> 00:10:43,040 Speaker 3: You mentioned earlier in the conversation foundation models, and I 193 00:10:43,080 --> 00:10:44,920 Speaker 3: want to talk a little bit about that. I mean, 194 00:10:44,960 --> 00:10:48,360 Speaker 3: can you just tell me, you know, what are foundation 195 00:10:48,600 --> 00:10:51,360 Speaker 3: models for AI and why are they a big deal? 196 00:10:52,520 --> 00:10:56,240 Speaker 5: Yeah, So this term foundation model was coined by a 197 00:10:56,280 --> 00:10:59,960 Speaker 5: group at Stanford, and I think it's actually a really 198 00:11:00,080 --> 00:11:03,600 Speaker 5: apt term because remember I said, you know, one of 199 00:11:03,600 --> 00:11:06,920 Speaker 5: the big things that unlocked this latest excitement was the 200 00:11:06,920 --> 00:11:10,520 Speaker 5: fact that we could use large amounts of unannotated data. 201 00:11:11,080 --> 00:11:12,440 Speaker 5: We could train a model. We don't have to go 202 00:11:12,480 --> 00:11:16,000 Speaker 5: through the painful effort of labeling each and every example. 203 00:11:16,559 --> 00:11:18,800 Speaker 5: You still need to have your model do something you 204 00:11:18,800 --> 00:11:21,000 Speaker 5: wanted to do. You still need to tell it what 205 00:11:21,040 --> 00:11:22,600 Speaker 5: you want to do. You can't just have a model 206 00:11:22,640 --> 00:11:25,160 Speaker 5: that doesn't, you know, have any purpose. But what a 207 00:11:25,200 --> 00:11:29,040 Speaker 5: foundation models that provides a foundation, like a literal foundation. 208 00:11:29,280 --> 00:11:31,320 Speaker 5: You can sort of stand on the shoulders of giants. 209 00:11:31,360 --> 00:11:34,079 Speaker 5: You can have them these massively trained models, and then 210 00:11:34,120 --> 00:11:36,280 Speaker 5: do a little bit on top. You know, you could 211 00:11:36,480 --> 00:11:38,560 Speaker 5: use just a few examples of what you're looking for 212 00:11:39,360 --> 00:11:41,520 Speaker 5: and you can get what you want from the model, 213 00:11:42,040 --> 00:11:44,080 Speaker 5: So just a little bit on top now gets to 214 00:11:44,240 --> 00:11:46,240 Speaker 5: the results that a huge amount of effort used to 215 00:11:46,280 --> 00:11:48,199 Speaker 5: have to put in, you know, to get from the 216 00:11:48,320 --> 00:11:50,360 Speaker 5: ground up to that level. 217 00:11:50,640 --> 00:11:54,560 Speaker 3: I was trying to think of of an analogy for 218 00:11:54,679 --> 00:11:57,720 Speaker 3: sort of foundation models versus what came before, and I 219 00:11:57,760 --> 00:12:00,240 Speaker 3: don't know that I came up with a good one, 220 00:12:00,280 --> 00:12:01,959 Speaker 3: but the best I could do was this. I want 221 00:12:01,960 --> 00:12:04,920 Speaker 3: you to tell me if it's plausible. It's like before 222 00:12:05,000 --> 00:12:08,520 Speaker 3: foundation models, it was like you had these sort of 223 00:12:08,600 --> 00:12:11,800 Speaker 3: single use kitchen appliances. You could make a waffle iron 224 00:12:11,840 --> 00:12:14,360 Speaker 3: if you wanted waffles, or you could make a toaster 225 00:12:14,520 --> 00:12:17,079 Speaker 3: if you wanted to make toast. But a foundation model 226 00:12:17,160 --> 00:12:19,720 Speaker 3: is like like an oven with a range on top. 227 00:12:19,800 --> 00:12:21,559 Speaker 3: So it's like this machine and you could just cook 228 00:12:21,640 --> 00:12:23,480 Speaker 3: anything with this machine. 229 00:12:24,120 --> 00:12:28,600 Speaker 5: Yeah, that's a great analogy. They're very versatile. The other 230 00:12:28,720 --> 00:12:31,280 Speaker 5: piece of it, too, is that they dramatically lower the 231 00:12:31,400 --> 00:12:34,560 Speaker 5: effort that it takes to do something that you want 232 00:12:34,600 --> 00:12:37,600 Speaker 5: to do. And sometimes I used to say about the 233 00:12:37,640 --> 00:12:39,600 Speaker 5: old world of AI, would say, you know, the problem 234 00:12:39,640 --> 00:12:43,400 Speaker 5: with automation is that it's too labor intensive. H sounds 235 00:12:43,440 --> 00:12:44,400 Speaker 5: like I'm making a joke. 236 00:12:44,640 --> 00:12:49,200 Speaker 3: Indeed, famously, if automation does one thing, it substitutes machines 237 00:12:49,320 --> 00:12:52,520 Speaker 3: or computing power for labor. Right, So what does that 238 00:12:52,600 --> 00:12:56,880 Speaker 3: mean to say AI is or automation is too labor intensive. 239 00:12:57,360 --> 00:12:59,320 Speaker 5: It sounds like I'm making a joke, but I'm actually serious. 240 00:12:59,559 --> 00:13:02,800 Speaker 5: What I mean is that the effort it took the 241 00:13:02,880 --> 00:13:06,719 Speaker 5: old regime to automate something was very, very high. So 242 00:13:06,920 --> 00:13:09,800 Speaker 5: if I need to go and curate all this data, 243 00:13:09,840 --> 00:13:13,040 Speaker 5: collect all this data, and then carefully label all these examples, 244 00:13:13,440 --> 00:13:17,360 Speaker 5: that labeling itself might be incredibly expensive and time. So 245 00:13:17,760 --> 00:13:20,360 Speaker 5: and we estimate anywhere between eighty to ninety percent of 246 00:13:20,400 --> 00:13:23,240 Speaker 5: the effort it takes to feel an AI solution actually 247 00:13:23,360 --> 00:13:26,959 Speaker 5: is just spent on data, so that that has some consequences, 248 00:13:27,240 --> 00:13:32,600 Speaker 5: which is the threshold for bothering. You know, if you're 249 00:13:32,600 --> 00:13:34,800 Speaker 5: going to only get a little bit of value back 250 00:13:35,040 --> 00:13:37,280 Speaker 5: from something, are you going to go through this huge 251 00:13:37,280 --> 00:13:40,800 Speaker 5: effort to curate all this data and then when it 252 00:13:40,800 --> 00:13:43,240 Speaker 5: comes time to train the model, you need highly skilled 253 00:13:43,240 --> 00:13:47,280 Speaker 5: people expensive or hard to find in the labor market. 254 00:13:47,440 --> 00:13:48,959 Speaker 5: You know, are you really going to do something that's 255 00:13:49,000 --> 00:13:50,920 Speaker 5: just a tiny, little incremental thing. Now, you're going to 256 00:13:50,960 --> 00:13:54,600 Speaker 5: do the only the highest value things that weren't at 257 00:13:54,800 --> 00:13:56,000 Speaker 5: level because. 258 00:13:55,679 --> 00:13:59,080 Speaker 3: You have to essentially build the whole machine from scratch, 259 00:13:59,280 --> 00:14:02,200 Speaker 3: and there aren't many things where it's worth that much 260 00:14:02,240 --> 00:14:04,280 Speaker 3: work to build a machine that's only going to do 261 00:14:04,400 --> 00:14:05,600 Speaker 3: one narrow thing. 262 00:14:06,040 --> 00:14:09,000 Speaker 5: That's right, and then you tackle the next problem and 263 00:14:09,080 --> 00:14:11,400 Speaker 5: you basically have to start over. And you know, there 264 00:14:11,440 --> 00:14:14,240 Speaker 5: are some nuances here, like for images, you can pre 265 00:14:14,280 --> 00:14:16,800 Speaker 5: train a model on some other tasks and change it around. 266 00:14:16,800 --> 00:14:19,760 Speaker 5: So there are some examples of this like non recurring 267 00:14:19,880 --> 00:14:22,480 Speaker 5: cost that we have in the old world too, But 268 00:14:22,520 --> 00:14:25,040 Speaker 5: by and large, it's just a lot of effort. It's hard. 269 00:14:25,320 --> 00:14:29,600 Speaker 5: It takes, you know, a large level of skill to implement. 270 00:14:30,400 --> 00:14:33,160 Speaker 5: One analogy that I like is, you know, think about 271 00:14:33,200 --> 00:14:35,320 Speaker 5: it as you know, you have a river of data, 272 00:14:35,720 --> 00:14:39,080 Speaker 5: you know, running through your company or your institution. Traditional 273 00:14:39,080 --> 00:14:41,600 Speaker 5: AI solutions are kind of like building a dam on 274 00:14:41,600 --> 00:14:45,080 Speaker 5: that river. You know, dams are very expensive things to build. 275 00:14:45,440 --> 00:14:49,680 Speaker 5: They require highly specialized skills and lots of planning. And 276 00:14:49,880 --> 00:14:51,560 Speaker 5: you know, you're only going to put a dam on 277 00:14:51,960 --> 00:14:54,680 Speaker 5: a river that's big enough that you're gonna get enough 278 00:14:54,800 --> 00:14:57,200 Speaker 5: energy out of it that it was worth trouble. You're 279 00:14:57,200 --> 00:14:58,640 Speaker 5: gonna get a lot of value out of that dam. 280 00:14:58,680 --> 00:15:00,320 Speaker 5: If you have a river like that, you know, a 281 00:15:00,400 --> 00:15:03,960 Speaker 5: river of data, but it's actually the vast majority of 282 00:15:04,160 --> 00:15:06,520 Speaker 5: the water you know in your kingdom actually isn't in 283 00:15:06,560 --> 00:15:10,560 Speaker 5: that river. It's in puddles and greeks and babid brooks, 284 00:15:10,640 --> 00:15:14,080 Speaker 5: And you know, there's a lot of value left on 285 00:15:14,120 --> 00:15:16,680 Speaker 5: the table because it's like, well, I can't there's nothing 286 00:15:16,720 --> 00:15:18,520 Speaker 5: you can do about it. It's just that that's too 287 00:15:19,480 --> 00:15:22,600 Speaker 5: low value. So it takes too much effort, so I'm 288 00:15:22,640 --> 00:15:24,200 Speaker 5: just not going to do it. The return around investment 289 00:15:24,560 --> 00:15:27,120 Speaker 5: just isn't there, so you just end up not automating 290 00:15:27,160 --> 00:15:29,960 Speaker 5: things because it's too much of a pain. Now what 291 00:15:30,000 --> 00:15:32,440 Speaker 5: foundation models do is they say, well, actually, no, we 292 00:15:32,480 --> 00:15:35,680 Speaker 5: can train a base model a foundation that you can 293 00:15:35,720 --> 00:15:37,360 Speaker 5: work on that we don't we don't care. We don't 294 00:15:37,400 --> 00:15:39,240 Speaker 5: specify what the task is ahead of time. We just 295 00:15:39,280 --> 00:15:42,440 Speaker 5: need to learn about the domain of data. So if 296 00:15:42,440 --> 00:15:45,320 Speaker 5: we want to build something that can understand English language, 297 00:15:45,640 --> 00:15:48,920 Speaker 5: there's a ton of English language text available out in 298 00:15:48,960 --> 00:15:53,040 Speaker 5: the world. We can now train models on huge quantities 299 00:15:53,040 --> 00:15:56,200 Speaker 5: of it, and then it learned the structure. It learned 300 00:15:56,280 --> 00:15:59,040 Speaker 5: how language, you know, good part of how language works 301 00:15:59,120 --> 00:16:01,400 Speaker 5: on all that unlabeled data. And then when you roll 302 00:16:01,480 --> 00:16:04,440 Speaker 5: up with your task, you know, I want to solve 303 00:16:04,440 --> 00:16:07,560 Speaker 5: this particular problem, you don't have to start from scratch. 304 00:16:07,600 --> 00:16:11,040 Speaker 5: You're starting from a very very very high place. So 305 00:16:11,080 --> 00:16:13,560 Speaker 5: that just gives you the ability to just you know, now, 306 00:16:13,600 --> 00:16:16,440 Speaker 5: all of a sudden, everything is accessible. All the puddles 307 00:16:16,440 --> 00:16:19,200 Speaker 5: and greeks and babbling brooks and kettlepons, you know, those 308 00:16:19,200 --> 00:16:23,960 Speaker 5: are all accessible now. And that's that's very exciting. But 309 00:16:24,040 --> 00:16:26,520 Speaker 5: it just changes the equation on what kinds of problems 310 00:16:26,640 --> 00:16:27,840 Speaker 5: you could use AI to solve. 311 00:16:27,960 --> 00:16:33,400 Speaker 3: And so foundation models basically mean that automating some new 312 00:16:33,520 --> 00:16:36,760 Speaker 3: task is much less labor intensive, The sort of marginal 313 00:16:36,840 --> 00:16:39,840 Speaker 3: effort to do some new automation thing is much lower 314 00:16:39,880 --> 00:16:43,120 Speaker 3: because you're building on top of the foundation model rather 315 00:16:43,200 --> 00:16:47,560 Speaker 3: than starting from scratch. Absolutely, so that is that is 316 00:16:47,680 --> 00:16:51,280 Speaker 3: like the exciting good news. I do feel like there's 317 00:16:52,080 --> 00:16:54,680 Speaker 3: a little bit of a countervailing idea that's worth talking 318 00:16:54,720 --> 00:16:57,080 Speaker 3: about here, and that is the idea that even though 319 00:16:57,080 --> 00:17:01,120 Speaker 3: there are these foundation models that are really powerful, that 320 00:17:01,160 --> 00:17:04,200 Speaker 3: are relatively easy to build on top of, it's still 321 00:17:04,240 --> 00:17:07,439 Speaker 3: the case, right that there is not some one size 322 00:17:07,480 --> 00:17:11,159 Speaker 3: fits all foundation model. So you know, what does that 323 00:17:11,320 --> 00:17:13,520 Speaker 3: mean and why is that important to think about in 324 00:17:13,560 --> 00:17:14,320 Speaker 3: this context? 325 00:17:14,880 --> 00:17:18,679 Speaker 5: Yeah, so we believe very strongly that there isn't just 326 00:17:18,800 --> 00:17:21,680 Speaker 5: one model to rule them all. There's a number of 327 00:17:21,720 --> 00:17:24,720 Speaker 5: reasons why that could be true. One which I think 328 00:17:24,800 --> 00:17:28,800 Speaker 5: is important and very relevant today is how much energy 329 00:17:29,119 --> 00:17:33,880 Speaker 5: these models can consume. So these models, you know, can 330 00:17:33,920 --> 00:17:39,360 Speaker 5: get very very large. So one thing that we're starting 331 00:17:39,400 --> 00:17:42,120 Speaker 5: to see or starting to believe, is that you probably 332 00:17:42,160 --> 00:17:47,280 Speaker 5: shouldn't use one giant sledgehammer model to solve every single problem, 333 00:17:47,480 --> 00:17:49,400 Speaker 5: you know, like we should pick the right size model 334 00:17:49,440 --> 00:17:52,239 Speaker 5: to solve the problem. We shouldn't necessarily assume that we 335 00:17:52,280 --> 00:17:56,840 Speaker 5: need the biggest, baddest model for every little use case. 336 00:17:57,320 --> 00:17:59,520 Speaker 5: And we're also seeing that, you know, small models that 337 00:17:59,520 --> 00:18:03,439 Speaker 5: are trained to like to specialize on particular domains can 338 00:18:03,480 --> 00:18:07,600 Speaker 5: actually outperform much bigger models. So bigger isn't always even better. 339 00:18:07,720 --> 00:18:10,280 Speaker 3: So they're more efficient and they do the thing you 340 00:18:10,320 --> 00:18:11,960 Speaker 3: want them to do better as well. 341 00:18:12,480 --> 00:18:15,760 Speaker 5: That's right. So Stanford, for instance, a group of Stanford 342 00:18:15,800 --> 00:18:18,919 Speaker 5: trained a model. It is a two point seven billion 343 00:18:18,960 --> 00:18:22,080 Speaker 5: parameter model, which isn't terribly big by today's standards. They 344 00:18:22,080 --> 00:18:24,359 Speaker 5: trained it just on the biomedical literature, you know, this 345 00:18:24,400 --> 00:18:26,800 Speaker 5: is the kind of thing that universities do, and what 346 00:18:26,840 --> 00:18:30,320 Speaker 5: they showed was that this model was better at answering 347 00:18:30,400 --> 00:18:32,920 Speaker 5: questions about the biomedical literature than some models that were 348 00:18:33,440 --> 00:18:37,159 Speaker 5: one hundred billion parameters, you know, many times larger. So 349 00:18:37,320 --> 00:18:39,880 Speaker 5: it's a little bit like you know, asking an expert 350 00:18:40,320 --> 00:18:43,600 Speaker 5: for help on something versus asking the smartest person. You know, 351 00:18:44,160 --> 00:18:46,720 Speaker 5: the smartest person you know may be very smart, but 352 00:18:46,800 --> 00:18:49,679 Speaker 5: they're not going to be expertise. And then as an 353 00:18:49,720 --> 00:18:52,199 Speaker 5: added bonus, you know, this is now a much smaller model, 354 00:18:52,280 --> 00:18:54,159 Speaker 5: it's much more efficient to run. We are you know, 355 00:18:54,760 --> 00:18:58,640 Speaker 5: you know, it's cheaper. So there's lots of different advantages there. 356 00:18:58,680 --> 00:19:02,280 Speaker 5: So I think we're going to see attention in the 357 00:19:02,320 --> 00:19:05,600 Speaker 5: industry between vendors that say, hey, this is the one, 358 00:19:05,800 --> 00:19:08,159 Speaker 5: you know, big model, and then others that say, well, actually, 359 00:19:08,440 --> 00:19:10,960 Speaker 5: you know, there's there's you know, lots of different tools 360 00:19:10,960 --> 00:19:13,000 Speaker 5: we can use that all have this nice quality that 361 00:19:13,040 --> 00:19:15,680 Speaker 5: we outlined at the beginning, and then we should really 362 00:19:15,680 --> 00:19:17,200 Speaker 5: pick the one that makes the most sense for the 363 00:19:17,560 --> 00:19:18,280 Speaker 5: task at hand. 364 00:19:19,560 --> 00:19:23,960 Speaker 3: So there's sustainability basically efficiency, another kind of set of 365 00:19:23,960 --> 00:19:27,880 Speaker 3: issues that come up a lot with AI A are bias, hallucination. 366 00:19:28,600 --> 00:19:31,200 Speaker 3: Can you talk a little bit about bias and hallucination, 367 00:19:31,320 --> 00:19:34,240 Speaker 3: what they are and how you're working to mitigate those problems. 368 00:19:34,640 --> 00:19:37,520 Speaker 5: Yeah, so there are lots of issues still as amazing 369 00:19:37,520 --> 00:19:40,440 Speaker 5: as these technologies are, and they are amazing, let's let's 370 00:19:40,480 --> 00:19:42,960 Speaker 5: be very clear, lots of great things we're going to 371 00:19:43,080 --> 00:19:46,920 Speaker 5: enable with these kinds of technologies. Bias isn't a new problem. 372 00:19:47,240 --> 00:19:51,840 Speaker 5: So you know, basically we've seen this since the beginning 373 00:19:51,880 --> 00:19:54,800 Speaker 5: of AI. If you train a model on data that 374 00:19:55,200 --> 00:19:57,320 Speaker 5: has a bias in it, the model is going to 375 00:19:57,359 --> 00:20:01,920 Speaker 5: recapitulate that bias when it provides it's answers. So every time, 376 00:20:02,119 --> 00:20:04,639 Speaker 5: you know, if all the text you have says, you know, 377 00:20:04,680 --> 00:20:07,760 Speaker 5: it's more likely to refer to female nurses and male scientists, 378 00:20:07,800 --> 00:20:09,879 Speaker 5: then you're going to you know, get models that you know. 379 00:20:09,960 --> 00:20:13,040 Speaker 5: For instance, there was an example where a machine learning 380 00:20:13,040 --> 00:20:17,480 Speaker 5: based translation system translated from Hungarian to English. Hungarian doesn't 381 00:20:17,480 --> 00:20:20,800 Speaker 5: have gendered pronouns. English does, and when you ask them 382 00:20:20,800 --> 00:20:23,119 Speaker 5: to translate, it would translate they are a nurse to 383 00:20:23,560 --> 00:20:26,520 Speaker 5: she is a nurse, translate they are a scientist. To 384 00:20:26,600 --> 00:20:29,720 Speaker 5: he is a scientist. And that's not because the people 385 00:20:29,720 --> 00:20:32,520 Speaker 5: who wrote the algorithm were building in bias and coding 386 00:20:32,560 --> 00:20:34,080 Speaker 5: in like, oh, it's got to be this way. It's 387 00:20:34,119 --> 00:20:36,359 Speaker 5: because the data was like that. You know, we have 388 00:20:36,480 --> 00:20:40,920 Speaker 5: biases in our society and they're reflected in our data 389 00:20:40,960 --> 00:20:44,600 Speaker 5: and our text and our images everywhere. And then the 390 00:20:44,640 --> 00:20:47,600 Speaker 5: models they're just mapping from what they what they've seen 391 00:20:47,600 --> 00:20:50,120 Speaker 5: in their training data to to the result that you're 392 00:20:50,200 --> 00:20:51,800 Speaker 5: trying to get them to do and to give, and 393 00:20:51,840 --> 00:20:56,280 Speaker 5: then these biases come out. So there's a very active 394 00:20:57,119 --> 00:21:00,320 Speaker 5: program of research and you know, we do quite a 395 00:21:00,320 --> 00:21:03,880 Speaker 5: bit at IBM research and my T but also all 396 00:21:03,920 --> 00:21:06,639 Speaker 5: over the community and industry and academia trying to figure 397 00:21:06,640 --> 00:21:09,840 Speaker 5: out how do we explicitly remove these biases, how do 398 00:21:09,840 --> 00:21:12,000 Speaker 5: we identify them, how do you know, how do we 399 00:21:12,040 --> 00:21:14,679 Speaker 5: build tools that allow people to audit their systems to 400 00:21:14,680 --> 00:21:17,000 Speaker 5: make sure they aren't biased. So this is a really 401 00:21:17,040 --> 00:21:20,200 Speaker 5: important thing. And you know, again this was here since 402 00:21:20,240 --> 00:21:24,000 Speaker 5: the beginning, you know, of machine learning and AI, but 403 00:21:24,680 --> 00:21:28,439 Speaker 5: foundation models and large language models and generative AI just 404 00:21:28,480 --> 00:21:31,199 Speaker 5: bring it into sharper even sharper focus because there's just 405 00:21:31,240 --> 00:21:34,600 Speaker 5: so much data and it's sort of building in baking 406 00:21:34,680 --> 00:21:37,560 Speaker 5: in all these different biases we have, so that that's 407 00:21:37,680 --> 00:21:41,680 Speaker 5: that's absolutely a problem that these models have. Another one 408 00:21:41,680 --> 00:21:45,480 Speaker 5: that you mentioned was hallucinations. So even the most impressive 409 00:21:45,520 --> 00:21:49,720 Speaker 5: of our models will often just make stuff up. You know, 410 00:21:49,920 --> 00:21:52,919 Speaker 5: the technical term that the field has chosen is hallucination. 411 00:21:53,520 --> 00:21:56,439 Speaker 5: To give you an example, I asked chat tbt to 412 00:21:56,720 --> 00:22:00,480 Speaker 5: create a biography of David Cox at IBM, and you know, 413 00:22:00,720 --> 00:22:03,320 Speaker 5: it started off really well. You know, the identified that 414 00:22:03,359 --> 00:22:05,800 Speaker 5: I was the director of the MNT IBM Watson and 415 00:22:05,800 --> 00:22:08,200 Speaker 5: said a few words about that, and then it proceeded 416 00:22:08,200 --> 00:22:12,760 Speaker 5: to create an authoritative but completely fake biography of me 417 00:22:12,800 --> 00:22:15,320 Speaker 5: where I was British, I was born in the UK, 418 00:22:16,680 --> 00:22:19,640 Speaker 5: I went to British university, you know universities in the UK. 419 00:22:19,720 --> 00:22:21,320 Speaker 5: I was professor the authority. 420 00:22:21,400 --> 00:22:24,960 Speaker 3: Right, it's the certainty that that is weird about it, right, 421 00:22:24,960 --> 00:22:28,240 Speaker 3: It's it's dead certain that you're from the UK, et cetera. 422 00:22:28,840 --> 00:22:31,879 Speaker 5: Absolutely, yeah, it has all kinds of flourishes like I 423 00:22:31,920 --> 00:22:36,639 Speaker 5: want awards in the UK. So yeah, it's it's problematic 424 00:22:36,720 --> 00:22:39,560 Speaker 5: because it kind of pokes a lot of weak spots 425 00:22:39,560 --> 00:22:44,760 Speaker 5: in our human psychology where if something sounds coherent, We're 426 00:22:44,880 --> 00:22:47,640 Speaker 5: likely to assume it's true. We're not used to interacting 427 00:22:47,640 --> 00:22:52,360 Speaker 5: with people who eloquently and authoritatively, you know, emit complete nonsense, 428 00:22:52,440 --> 00:22:55,280 Speaker 5: like yeah, you know, we can debate about that, but. 429 00:22:55,240 --> 00:22:57,600 Speaker 3: Yeah, we can debate about that. But yes, it's the 430 00:22:58,520 --> 00:23:02,159 Speaker 3: sort of blive confidence throws you off when you realize 431 00:23:02,200 --> 00:23:03,119 Speaker 3: it's completely wrong. 432 00:23:03,240 --> 00:23:06,000 Speaker 5: Right, that's right. And we do have a little bit 433 00:23:06,040 --> 00:23:09,240 Speaker 5: of like a great and powerful oz sort of vibe 434 00:23:09,280 --> 00:23:11,600 Speaker 5: going sometimes where we're like, well, you know, the AI 435 00:23:11,800 --> 00:23:15,560 Speaker 5: is all knowing and therefore whatever it says must be true. 436 00:23:15,800 --> 00:23:20,040 Speaker 5: But these things will make up stuff, you know, very aggressively, 437 00:23:20,760 --> 00:23:23,199 Speaker 5: and you know, you everyone can try asking it for 438 00:23:23,200 --> 00:23:26,720 Speaker 5: their their bio. You'll you'll get something that You'll always 439 00:23:26,720 --> 00:23:29,040 Speaker 5: get something that's of the right form, that has the 440 00:23:29,119 --> 00:23:32,040 Speaker 5: right tone. But you know, the facts just aren't necessarily there. 441 00:23:32,359 --> 00:23:34,760 Speaker 5: So that's obviously a problem. We need to figure out 442 00:23:34,760 --> 00:23:37,959 Speaker 5: how to close those gaps, fix those problems. There's lots 443 00:23:38,000 --> 00:23:40,080 Speaker 5: of ways we can use them much more easily. 444 00:23:40,600 --> 00:23:43,320 Speaker 4: I'd just like to say, faced with the awesome potential 445 00:23:43,359 --> 00:23:46,360 Speaker 4: of what these technologies might do, it's a bit encouraging 446 00:23:46,440 --> 00:23:49,960 Speaker 4: to hear that even chat GPT has a weakness for 447 00:23:50,080 --> 00:23:55,200 Speaker 4: inventing flamboyant, if fictional versions of people's lives, and while 448 00:23:55,320 --> 00:23:58,879 Speaker 4: entertaining ourselves with chat GPT and mid journey is important, 449 00:23:59,320 --> 00:24:03,840 Speaker 4: the way lpeople use consumer facing chatbots and generative AI 450 00:24:04,359 --> 00:24:08,240 Speaker 4: is just fundamentally different from the way an enterprise business 451 00:24:08,320 --> 00:24:11,960 Speaker 4: uses AI. How can we harness the abilities of artificial 452 00:24:12,000 --> 00:24:15,040 Speaker 4: intelligence to help us solve the problems we face in 453 00:24:15,119 --> 00:24:18,959 Speaker 4: business and technology. Let's listen on as David and Jacob 454 00:24:19,119 --> 00:24:20,440 Speaker 4: continue their conversation. 455 00:24:21,200 --> 00:24:24,160 Speaker 3: We've been talking in a somewhat abstract way about AI 456 00:24:24,280 --> 00:24:27,040 Speaker 3: in the ways it can be used. Let's talk in 457 00:24:27,040 --> 00:24:30,400 Speaker 3: a little bit more of a specific way. Can you 458 00:24:30,440 --> 00:24:34,240 Speaker 3: just talk about some examples of business challenges that can 459 00:24:34,280 --> 00:24:37,640 Speaker 3: be solved with automation, with this kind of automation we're 460 00:24:37,640 --> 00:24:38,560 Speaker 3: talking about. 461 00:24:39,119 --> 00:24:42,520 Speaker 5: Yeah, so the really really, this guy's the limit. There's 462 00:24:42,560 --> 00:24:46,480 Speaker 5: a whole set of different applications that these models are 463 00:24:46,520 --> 00:24:49,359 Speaker 5: really good at. And basically it's a superset of everything 464 00:24:49,400 --> 00:24:52,480 Speaker 5: we used to use AI for in business. So you know, 465 00:24:53,080 --> 00:24:54,760 Speaker 5: the simple kinds of things are like, hey, if I 466 00:24:54,760 --> 00:24:58,240 Speaker 5: have text and I have product reviews, and I want 467 00:24:58,240 --> 00:25:00,000 Speaker 5: to be able to tell if these are positive or negative. 468 00:25:00,240 --> 00:25:02,040 Speaker 5: You know, like let's look at all the negative reviews, 469 00:25:02,040 --> 00:25:03,399 Speaker 5: so we can have a human look through them and 470 00:25:03,720 --> 00:25:07,560 Speaker 5: see what was up. Very common business use case. You 471 00:25:07,600 --> 00:25:11,760 Speaker 5: can do it with traditional deep learning based AI. So so 472 00:25:11,640 --> 00:25:13,240 Speaker 5: there's things like that that are you know, it's very 473 00:25:13,440 --> 00:25:15,439 Speaker 5: prosaic sort of we were already doing it. We've been 474 00:25:15,440 --> 00:25:18,960 Speaker 5: doing it for a long time. Then you get situations 475 00:25:19,000 --> 00:25:21,159 Speaker 5: that are that were harder for the old day. I like, 476 00:25:21,600 --> 00:25:24,879 Speaker 5: if I'm I want to impress something like I want 477 00:25:24,920 --> 00:25:27,200 Speaker 5: to I have like say I have a chat transcript, 478 00:25:27,240 --> 00:25:30,200 Speaker 5: Like a customer called in and they had a complaint. 479 00:25:30,880 --> 00:25:34,359 Speaker 5: They called back. Okay, now a new you know, a 480 00:25:34,520 --> 00:25:36,480 Speaker 5: person on the line needs to go read the old 481 00:25:36,520 --> 00:25:39,320 Speaker 5: transcript to catch up. Wouldn't it be better if we 482 00:25:39,359 --> 00:25:41,800 Speaker 5: could just summarize that, just condense it all down a 483 00:25:41,920 --> 00:25:44,119 Speaker 5: quick little paragraph. You know, customer called they we upset 484 00:25:44,119 --> 00:25:46,160 Speaker 5: about this, rather than having to read the blow by blow. 485 00:25:46,600 --> 00:25:49,760 Speaker 5: There's just lots of settings like that where summarization is 486 00:25:49,800 --> 00:25:52,840 Speaker 5: really helpful. Hey, you have a meeting and I'd like 487 00:25:52,920 --> 00:25:55,600 Speaker 5: to just automatically you know, have had that meeting or 488 00:25:55,640 --> 00:25:57,480 Speaker 5: that email or whatever. I'd like to just have a 489 00:25:57,520 --> 00:25:59,679 Speaker 5: condensed down so I can really quickly get to the 490 00:25:59,680 --> 00:26:02,360 Speaker 5: heart of the matter. These models are are really good 491 00:26:02,359 --> 00:26:05,000 Speaker 5: at doing that. They're also really good at question answering. 492 00:26:05,320 --> 00:26:07,640 Speaker 5: So if I want to find out, what's how many 493 00:26:07,720 --> 00:26:11,320 Speaker 5: vacation days do I have? I can now interact in 494 00:26:11,440 --> 00:26:14,879 Speaker 5: natural language with a system that can go and that 495 00:26:14,920 --> 00:26:17,320 Speaker 5: it has access to our HR policies, and I can 496 00:26:17,359 --> 00:26:20,120 Speaker 5: actually have a you know, a multi turn conversation where 497 00:26:20,119 --> 00:26:22,600 Speaker 5: I can, you know, like I would have with you know, somebody, 498 00:26:22,760 --> 00:26:27,800 Speaker 5: you know, actual HR professional or customer service representative. So 499 00:26:28,240 --> 00:26:31,320 Speaker 5: a big part, you know, of what this is doing 500 00:26:31,440 --> 00:26:33,960 Speaker 5: is it's it's putting an interface. You know, when we 501 00:26:33,960 --> 00:26:37,600 Speaker 5: think of computer interfaces, we're usually thinking about UI user 502 00:26:37,640 --> 00:26:40,600 Speaker 5: interface elements where I click on menus and there's buttons 503 00:26:40,640 --> 00:26:44,520 Speaker 5: and all this stuff. Increasingly, now we can just talk, 504 00:26:44,680 --> 00:26:46,879 Speaker 5: you know, you just in words. You can describe what 505 00:26:46,920 --> 00:26:49,600 Speaker 5: you want, you want to answer, ask a question, you 506 00:26:49,640 --> 00:26:51,840 Speaker 5: want to sort of command the system to do something, 507 00:26:52,640 --> 00:26:54,720 Speaker 5: rather than having to learn how to do that clicking buttons, 508 00:26:54,760 --> 00:26:56,520 Speaker 5: which might be inefficient, Now we can just sort of 509 00:26:56,800 --> 00:26:57,400 Speaker 5: spell it out. 510 00:26:57,960 --> 00:27:00,960 Speaker 3: Interesting, right, the graphical user interface that we all sort 511 00:27:00,960 --> 00:27:04,280 Speaker 3: of default to, that's not like the state of nature, Right, 512 00:27:04,359 --> 00:27:06,879 Speaker 3: that's a thing that was invented and just came to 513 00:27:06,920 --> 00:27:09,320 Speaker 3: be the standard way that we interact with computers. And 514 00:27:09,359 --> 00:27:13,800 Speaker 3: so you could imagine, as you're saying, like chat essentially 515 00:27:14,000 --> 00:27:17,240 Speaker 3: chatting with the machine could could become a sort of 516 00:27:17,320 --> 00:27:20,560 Speaker 3: standard user interface, just like the graphical user interface, did 517 00:27:20,760 --> 00:27:22,160 Speaker 3: you know over the past several decades. 518 00:27:22,600 --> 00:27:26,040 Speaker 5: Absolutely, And I think those kinds of conversational interfaces are 519 00:27:26,040 --> 00:27:30,280 Speaker 5: going to be hugely important for increasing our productivity. It's 520 00:27:30,280 --> 00:27:32,159 Speaker 5: just a lot easier if I if I have to 521 00:27:32,200 --> 00:27:33,879 Speaker 5: learn how to use a tool, or I don't have 522 00:27:33,880 --> 00:27:36,960 Speaker 5: to kind of have awkward, you know, interactions from the computer. 523 00:27:36,960 --> 00:27:38,159 Speaker 5: I can just tell it what I want and I 524 00:27:38,200 --> 00:27:41,240 Speaker 5: can understand it. Could you know, potentially even ask questions 525 00:27:41,240 --> 00:27:45,200 Speaker 5: back to clarify and have those kinds of conversations that 526 00:27:45,240 --> 00:27:48,320 Speaker 5: can be extremely powerful. And in fact, one area where 527 00:27:48,320 --> 00:27:51,000 Speaker 5: that's going to I think be absolutely game changing is 528 00:27:51,040 --> 00:27:55,200 Speaker 5: in code. When we write code. You know, programming languages 529 00:27:55,680 --> 00:27:59,320 Speaker 5: are a way for us to sort of match between 530 00:28:00,000 --> 00:28:03,439 Speaker 5: a very sloppy way of talking and the very exact 531 00:28:03,480 --> 00:28:05,360 Speaker 5: way that you need to command a computer to do 532 00:28:05,560 --> 00:28:08,360 Speaker 5: what you wanted to do. They're cumbersome to learn, they 533 00:28:08,359 --> 00:28:10,800 Speaker 5: can you know, create very complex systems that are very 534 00:28:10,800 --> 00:28:13,800 Speaker 5: hard to reason about. And we're already starting to see 535 00:28:14,119 --> 00:28:15,840 Speaker 5: the ability to just write down what you want and 536 00:28:16,160 --> 00:28:18,680 Speaker 5: AI will generate the code for you. And I think 537 00:28:18,680 --> 00:28:20,439 Speaker 5: we're just going to see a huge revolution of like 538 00:28:20,680 --> 00:28:22,800 Speaker 5: we just converse you and we can have a conversation 539 00:28:23,160 --> 00:28:25,280 Speaker 5: to say what we want, and then the computer can 540 00:28:25,320 --> 00:28:29,000 Speaker 5: actually not only do fixed actions and do things for us, 541 00:28:29,000 --> 00:28:31,359 Speaker 5: but it can actually even write code to do new things, 542 00:28:31,359 --> 00:28:35,000 Speaker 5: you know, and generate software itself. Given how much software 543 00:28:35,040 --> 00:28:37,480 Speaker 5: we have, how much craving we have for software, like 544 00:28:37,520 --> 00:28:41,240 Speaker 5: we'll never have enough software in our world. Uh, you know, 545 00:28:41,360 --> 00:28:44,520 Speaker 5: the ability to have AI systems as a helper in that, 546 00:28:45,120 --> 00:28:46,680 Speaker 5: I think we're going to see a lot of a 547 00:28:46,720 --> 00:28:47,520 Speaker 5: lot of value there. 548 00:28:48,720 --> 00:28:51,360 Speaker 3: So if you if you think about the different ways 549 00:28:52,000 --> 00:28:54,200 Speaker 3: AI might be applied to business, I mean you've talked 550 00:28:54,200 --> 00:28:56,520 Speaker 3: about a number of the sort of classic use cases. 551 00:28:57,240 --> 00:29:00,600 Speaker 3: What are some of the more out there use cases. 552 00:29:00,640 --> 00:29:03,520 Speaker 3: What are some you know, unique ways you could imagine 553 00:29:03,560 --> 00:29:05,320 Speaker 3: AI being applied to business. 554 00:29:06,960 --> 00:29:09,520 Speaker 5: You know, there's really disguised the limit. I mean, we 555 00:29:09,600 --> 00:29:11,520 Speaker 5: have one project that I'm kind of a fan of 556 00:29:11,600 --> 00:29:15,760 Speaker 5: where we actually were working with a mechanical engineering professor 557 00:29:15,920 --> 00:29:18,720 Speaker 5: at MIT working on a classic problem, how do you 558 00:29:18,800 --> 00:29:22,200 Speaker 5: build linkage systems which are like you imagine bars and 559 00:29:22,480 --> 00:29:24,800 Speaker 5: joints and overs. 560 00:29:24,480 --> 00:29:26,800 Speaker 3: You know, the things that are building a thing, building 561 00:29:26,800 --> 00:29:28,640 Speaker 3: a physical machine of some kind. 562 00:29:29,240 --> 00:29:34,400 Speaker 5: Like real like metal and you know nineteenth century just 563 00:29:34,600 --> 00:29:37,520 Speaker 5: old school industrial revolution. Yeah yeah, yeah, but you know 564 00:29:37,560 --> 00:29:40,320 Speaker 5: the little arm that's that's holding up my microphone in 565 00:29:40,360 --> 00:29:42,800 Speaker 5: front of me. Cranes, get build your buildings, you know, 566 00:29:42,880 --> 00:29:45,400 Speaker 5: parts of your engines. This is like classical stuff. It 567 00:29:45,440 --> 00:29:47,720 Speaker 5: turns out that you know, humans, if you want to 568 00:29:47,720 --> 00:29:50,920 Speaker 5: build an advanced system, you decide what like curve you 569 00:29:50,960 --> 00:29:53,600 Speaker 5: want to create, and then a human together with a 570 00:29:53,640 --> 00:29:57,400 Speaker 5: computer program, can build a five or six bar linkage 571 00:29:57,560 --> 00:29:58,960 Speaker 5: and then that's kind of where you top out it 572 00:29:59,040 --> 00:30:01,920 Speaker 5: because it gets too comp replicated to work more than that. 573 00:30:02,560 --> 00:30:05,080 Speaker 5: We built a generative AI system that can build twenty 574 00:30:05,080 --> 00:30:08,440 Speaker 5: bar linkages, like arbitrarily complex. So these are machines that 575 00:30:08,480 --> 00:30:12,840 Speaker 5: are beyond the capability of a human to design themselves. 576 00:30:13,360 --> 00:30:16,320 Speaker 5: Another example, we have an AI system that can generate 577 00:30:16,480 --> 00:30:18,880 Speaker 5: electronic circuits. You know, we had a project where we're 578 00:30:18,880 --> 00:30:22,000 Speaker 5: working where we're building better power converters which allow our 579 00:30:22,800 --> 00:30:25,920 Speaker 5: computers and our devices to be more efficient, save energy, 580 00:30:26,720 --> 00:30:29,680 Speaker 5: you know, less less carbone. But I think the world 581 00:30:29,720 --> 00:30:32,640 Speaker 5: around us has always been shaped by technology. If we 582 00:30:32,680 --> 00:30:34,959 Speaker 5: look around, you know, just think about how many steps 583 00:30:35,000 --> 00:30:37,280 Speaker 5: and how many people and how many designs went into 584 00:30:37,320 --> 00:30:41,120 Speaker 5: the table and the chair and the lamp. It's really 585 00:30:41,160 --> 00:30:44,520 Speaker 5: just astonishing. And that's already you know, the fruit of 586 00:30:45,440 --> 00:30:47,800 Speaker 5: automation and computers and those kinds of tools. But we're 587 00:30:47,800 --> 00:30:51,120 Speaker 5: going to see that increasingly be product also of AI. 588 00:30:51,200 --> 00:30:53,240 Speaker 5: It's just going to be everywhere around us. Everything we 589 00:30:53,400 --> 00:30:56,560 Speaker 5: touch is going to have been helped in some way 590 00:30:56,640 --> 00:30:58,240 Speaker 5: to get to you by. 591 00:30:58,080 --> 00:31:01,560 Speaker 3: A you know, that is a pretty profound transformation that 592 00:31:01,600 --> 00:31:04,400 Speaker 3: you're talking about in business. How do you think about 593 00:31:04,400 --> 00:31:07,600 Speaker 3: the implications of that both for the sort of you know, 594 00:31:07,880 --> 00:31:11,080 Speaker 3: business itself and also for employees. 595 00:31:12,760 --> 00:31:16,880 Speaker 5: Yeah, so I think for businesses this is gonna cut costs, 596 00:31:17,160 --> 00:31:21,040 Speaker 5: make new opportunities to like customers, you know, like there's 597 00:31:21,080 --> 00:31:23,880 Speaker 5: just you know, it's sort of all upside right like 598 00:31:24,720 --> 00:31:26,720 Speaker 5: for the for the workers, I think the story is 599 00:31:26,760 --> 00:31:29,720 Speaker 5: mostly good too. You know, like how many things do 600 00:31:29,760 --> 00:31:33,239 Speaker 5: you do in your day that you'd really rather not? 601 00:31:33,440 --> 00:31:33,560 Speaker 2: Right? 602 00:31:34,080 --> 00:31:36,040 Speaker 5: You know, and we're used to having things we don't 603 00:31:36,160 --> 00:31:39,440 Speaker 5: like automated away. You know, we we didn't. You know, 604 00:31:39,440 --> 00:31:42,040 Speaker 5: if you didn't like walking many miles to work, then 605 00:31:42,080 --> 00:31:43,840 Speaker 5: you know, like you can have a car and you 606 00:31:43,840 --> 00:31:46,320 Speaker 5: can drive there. Or we used to have a huge 607 00:31:46,320 --> 00:31:49,720 Speaker 5: fraction over ninety percent of the US population engaged in agriculture, 608 00:31:49,800 --> 00:31:52,400 Speaker 5: and then we mechanized it. How very few people work 609 00:31:52,440 --> 00:31:54,360 Speaker 5: in agriculture, a small number of people can do the 610 00:31:54,360 --> 00:31:56,760 Speaker 5: work of a large number of people. And then you know, 611 00:31:56,880 --> 00:31:59,960 Speaker 5: things like email, and yeah, they've led to huge productor 612 00:32:00,160 --> 00:32:02,640 Speaker 5: the enhancements because I don't need to be writing letters 613 00:32:02,680 --> 00:32:04,720 Speaker 5: and sending them in the mail. I can just instantly 614 00:32:04,720 --> 00:32:09,000 Speaker 5: communicate with people. We just become more effective, Like our 615 00:32:09,080 --> 00:32:13,480 Speaker 5: jobs have transformed, whether it's a physical job like agriculture 616 00:32:13,560 --> 00:32:16,040 Speaker 5: or whether it's a knowledge worker job where you're sending 617 00:32:16,080 --> 00:32:19,960 Speaker 5: emails and communicating with people and coordinating teams. We've just 618 00:32:20,000 --> 00:32:22,680 Speaker 5: gotten better. And you know, the technology has just made 619 00:32:22,720 --> 00:32:25,800 Speaker 5: us more productive. And this is just another example. Now, 620 00:32:26,080 --> 00:32:28,200 Speaker 5: you know, there are people who worry that you know, 621 00:32:28,880 --> 00:32:31,320 Speaker 5: will be so good at that that maybe jobs will 622 00:32:31,320 --> 00:32:35,200 Speaker 5: be displaced, and that's a legitimate concern. But just like 623 00:32:36,560 --> 00:32:38,720 Speaker 5: how in agriculture, you know, it's not like suddenly we 624 00:32:38,800 --> 00:32:41,880 Speaker 5: had ninety percent of the population unemployed. You know, people 625 00:32:41,880 --> 00:32:46,040 Speaker 5: transitioned to other jobs. And the other thing that we 626 00:32:46,160 --> 00:32:50,160 Speaker 5: found too, is that our appetite for doing more things 627 00:32:50,800 --> 00:32:54,120 Speaker 5: is as humans is sort of insatiable. So even if 628 00:32:54,480 --> 00:32:57,680 Speaker 5: we can dramatically increase how much one human can do, 629 00:32:58,480 --> 00:33:00,720 Speaker 5: that doesn't necessarily mean we're going to do fixed amount 630 00:33:00,760 --> 00:33:02,960 Speaker 5: of stuff. There's an appetite to have even more, so 631 00:33:02,960 --> 00:33:05,200 Speaker 5: we're going to you can continue to grow, grow the pie. 632 00:33:05,640 --> 00:33:08,040 Speaker 5: So I think at least certainly in the near term, 633 00:33:08,280 --> 00:33:09,640 Speaker 5: you know, we're going to see a lot of drudgery 634 00:33:09,680 --> 00:33:12,400 Speaker 5: go away from work. We're going to see people be 635 00:33:12,400 --> 00:33:15,520 Speaker 5: able to be more effective at their jobs. You know, 636 00:33:15,520 --> 00:33:19,440 Speaker 5: we will see some transformation in jobs and like we've 637 00:33:19,480 --> 00:33:24,520 Speaker 5: seen that before, and the technology least has the potential 638 00:33:24,720 --> 00:33:26,040 Speaker 5: to make our lives a lot easier. 639 00:33:27,280 --> 00:33:32,280 Speaker 3: So IBM recently launched Watson X, which includes Watson X 640 00:33:32,360 --> 00:33:35,320 Speaker 3: dot AI. Tell me about that, Tell me about you 641 00:33:35,320 --> 00:33:37,400 Speaker 3: know what it is and the new possibilities that it 642 00:33:37,440 --> 00:33:38,000 Speaker 3: opens up. 643 00:33:38,920 --> 00:33:42,640 Speaker 5: Yeah. So so Watson X is obviously a bit of 644 00:33:43,160 --> 00:33:46,120 Speaker 5: a new branding on the Watson brand. 645 00:33:46,760 --> 00:33:46,960 Speaker 2: T J. 646 00:33:47,120 --> 00:33:50,720 Speaker 5: Watson that was the founder of IBM and our EI 647 00:33:50,800 --> 00:33:54,640 Speaker 5: technologies have had the Watson brand. Watson X is a 648 00:33:54,680 --> 00:33:58,280 Speaker 5: recognition that that there's something new, there's something that actually 649 00:33:58,280 --> 00:33:59,200 Speaker 5: has changed the game. 650 00:33:59,440 --> 00:33:59,560 Speaker 3: You know. 651 00:33:59,800 --> 00:34:03,480 Speaker 5: We gone from this old world of automation is to 652 00:34:03,600 --> 00:34:07,160 Speaker 5: labor intensive to this new world of possibilities where it's 653 00:34:07,240 --> 00:34:11,840 Speaker 5: much easier to use AI. And what watsonex does it 654 00:34:11,880 --> 00:34:16,799 Speaker 5: brings together tools for businesses to harness that power. So 655 00:34:17,120 --> 00:34:21,439 Speaker 5: whatsonex dot AI foundation models that our customers can use. 656 00:34:21,560 --> 00:34:24,600 Speaker 5: It includes tools that make it easy to run, easy 657 00:34:24,680 --> 00:34:29,040 Speaker 5: to deploy, easy to experiment. There's a watsonex dot Data 658 00:34:29,360 --> 00:34:32,800 Speaker 5: component which allows you to sort of organize and access 659 00:34:32,840 --> 00:34:34,920 Speaker 5: to your data. So what we're really trying to do 660 00:34:35,000 --> 00:34:39,919 Speaker 5: is give our customers a cohesive set of tools to 661 00:34:39,960 --> 00:34:43,200 Speaker 5: harness the value of these technologies and at the same 662 00:34:43,239 --> 00:34:46,239 Speaker 5: time be able to manage the risks and other things 663 00:34:46,280 --> 00:34:48,160 Speaker 5: that you have to keep an eye on in an 664 00:34:48,280 --> 00:34:49,239 Speaker 5: enterprise context. 665 00:34:50,880 --> 00:34:53,600 Speaker 3: So we talk about the guests on this show as 666 00:34:54,080 --> 00:34:58,200 Speaker 3: new creators, by which we mean people who are creatively 667 00:34:58,239 --> 00:35:03,120 Speaker 3: applying technology in businesiness to drive change. And I'm curious 668 00:35:03,640 --> 00:35:08,319 Speaker 3: how creativity plays a role in the research that you do. 669 00:35:08,920 --> 00:35:13,520 Speaker 5: I Honestly, I think the creative aspects of this job, 670 00:35:13,960 --> 00:35:17,279 Speaker 5: this is what makes this work exciting. You know, I 671 00:35:17,280 --> 00:35:19,200 Speaker 5: should say, you know, the folks who work in my 672 00:35:19,320 --> 00:35:24,160 Speaker 5: organization are doing the creating, and I guess you're. 673 00:35:24,000 --> 00:35:26,480 Speaker 3: Doing the managing so that they could do the creator. 674 00:35:27,400 --> 00:35:30,799 Speaker 5: I'm helping them be their best, and I still get 675 00:35:30,840 --> 00:35:33,719 Speaker 5: to get involved in the weeds of the research as 676 00:35:33,800 --> 00:35:36,560 Speaker 5: much as I can. But you know, there's something really 677 00:35:36,560 --> 00:35:40,480 Speaker 5: exciting about inventing, you know, like one of the nice 678 00:35:40,480 --> 00:35:44,680 Speaker 5: things about doing invention and doing research on AI. In industries, 679 00:35:45,040 --> 00:35:47,960 Speaker 5: it's usually grounded and a real problem that somebody is having. 680 00:35:48,040 --> 00:35:50,680 Speaker 5: You know, a customer wants to solve this problem. It's 681 00:35:51,239 --> 00:35:54,239 Speaker 5: losing money, or there wuld be a new opportunity. You 682 00:35:54,280 --> 00:35:58,799 Speaker 5: identify that problem and then you build something that's never 683 00:35:58,840 --> 00:36:01,120 Speaker 5: been built before to do that. And I think that's 684 00:36:01,400 --> 00:36:05,279 Speaker 5: honestly the adrenaline rush that keeps all of us in 685 00:36:05,320 --> 00:36:07,880 Speaker 5: this field. How do you do something that nobody else 686 00:36:08,120 --> 00:36:11,520 Speaker 5: on earth has has done before or tried before, so 687 00:36:11,560 --> 00:36:14,880 Speaker 5: that that kind of creativity and there's also creativity as 688 00:36:14,920 --> 00:36:17,680 Speaker 5: well and identifying what those problems are, being able to 689 00:36:17,760 --> 00:36:24,120 Speaker 5: understand the places where the technology is close enough to 690 00:36:24,400 --> 00:36:28,400 Speaker 5: solving a problem and doing that matchmaking between problems that 691 00:36:28,440 --> 00:36:31,120 Speaker 5: are now solvable, you know, and an AI where the 692 00:36:31,120 --> 00:36:34,759 Speaker 5: field is moving so fast, this is constantly growing horizon 693 00:36:35,480 --> 00:36:37,600 Speaker 5: of things that we might be able to solve. So 694 00:36:37,719 --> 00:36:41,920 Speaker 5: that matchmaking, I think, is also a really interesting creative problem. 695 00:36:42,120 --> 00:36:44,640 Speaker 5: So I think I think that's that's that's why it's 696 00:36:44,640 --> 00:36:47,359 Speaker 5: so much fun. And it's a fun environment we have 697 00:36:47,480 --> 00:36:50,520 Speaker 5: here too. It's you know, people drawing on whiteboards and 698 00:36:50,960 --> 00:36:54,960 Speaker 5: writing on pages of math and like in a movie, 699 00:36:55,400 --> 00:36:58,239 Speaker 5: like in a movie, yes, straight from sexual casting. 700 00:36:58,080 --> 00:37:00,320 Speaker 3: Drawing, the drawing on the window, writing on the window, 701 00:37:00,400 --> 00:37:06,160 Speaker 3: and sharp absolutely so, so let's close with the really 702 00:37:06,400 --> 00:37:12,080 Speaker 3: long view. How do you imagine AI and people working 703 00:37:12,120 --> 00:37:14,319 Speaker 3: together twenty years from now? 704 00:37:16,400 --> 00:37:21,040 Speaker 5: Yeah, it's really hard to make predictions. The vision that 705 00:37:21,600 --> 00:37:27,640 Speaker 5: I I like, actually this came from an MIT economist 706 00:37:27,760 --> 00:37:33,040 Speaker 5: named David Ottur, which was imagine AI almost as a 707 00:37:33,120 --> 00:37:37,880 Speaker 5: natural resource. You know, we know how natural resources work, right, 708 00:37:38,040 --> 00:37:39,719 Speaker 5: Like there's an or we can dig up out of 709 00:37:39,760 --> 00:37:42,320 Speaker 5: the earth that comes from kind of springs from the earth, 710 00:37:42,440 --> 00:37:45,400 Speaker 5: or we usually think of that in terms of physical stuff. 711 00:37:46,040 --> 00:37:47,640 Speaker 5: With AI, you can almost think of it as like 712 00:37:47,719 --> 00:37:50,719 Speaker 5: there's a new kind of abundance potentially twenty years from now, 713 00:37:50,800 --> 00:37:53,479 Speaker 5: where not only can we have things we can build 714 00:37:53,560 --> 00:37:56,160 Speaker 5: or eat or use or burn or whatever. Now we have, 715 00:37:56,400 --> 00:37:59,200 Speaker 5: you know, this ability to do things and understand things 716 00:37:59,200 --> 00:38:02,399 Speaker 5: and do intellectual work, and I think we can get 717 00:38:02,400 --> 00:38:06,719 Speaker 5: to a world where automating things is just seamless. We're 718 00:38:06,760 --> 00:38:11,759 Speaker 5: surrounded by capability to augment ourselves to get things done. 719 00:38:12,480 --> 00:38:15,239 Speaker 5: And you could think of that in terms of like, oh, 720 00:38:15,280 --> 00:38:17,399 Speaker 5: that's going to displace our jobs, because eventually the AI 721 00:38:17,480 --> 00:38:19,560 Speaker 5: system is going to do everything we can do. But 722 00:38:19,920 --> 00:38:22,120 Speaker 5: you could also think of it in terms of like, wow, 723 00:38:22,160 --> 00:38:24,480 Speaker 5: that's just so much abundance that we now have, and 724 00:38:24,520 --> 00:38:27,759 Speaker 5: really how we use that abundance is sort of up 725 00:38:27,800 --> 00:38:30,360 Speaker 5: to us, you know, like when you can writing software 726 00:38:30,440 --> 00:38:32,799 Speaker 5: is super easy and fast and anybody can do it. 727 00:38:33,200 --> 00:38:35,279 Speaker 5: Just think about all the things you can do now, like, 728 00:38:35,640 --> 00:38:37,719 Speaker 5: think about all the new activities and go about all 729 00:38:37,760 --> 00:38:39,880 Speaker 5: the ways we could use that to enrich our lives. 730 00:38:40,320 --> 00:38:43,359 Speaker 5: That's where I'd like to see us in twenty years. 731 00:38:43,400 --> 00:38:46,000 Speaker 5: You know, we can we can do just so much 732 00:38:46,160 --> 00:38:49,399 Speaker 5: more than we were able to do before abundance. 733 00:38:50,200 --> 00:38:53,040 Speaker 3: Great, thank you so much for your time. 734 00:38:53,800 --> 00:38:55,839 Speaker 5: Yeah, It's been a pleasure. Thanks for inviting me. 735 00:38:57,320 --> 00:39:01,400 Speaker 4: What a far ranging, deep conversation. I'm mesmerized by the 736 00:39:01,440 --> 00:39:05,360 Speaker 4: vision David just described. A world where natural conversation between 737 00:39:05,360 --> 00:39:09,960 Speaker 4: mankind and machine can generate creative solutions to our most 738 00:39:10,040 --> 00:39:13,799 Speaker 4: complex problems. A world where we view AI not as 739 00:39:13,880 --> 00:39:17,919 Speaker 4: our replacements, but as a powerful resource we can tap 740 00:39:17,960 --> 00:39:23,440 Speaker 4: into and exponentially boost our innovation and productivity. Thanks so 741 00:39:23,560 --> 00:39:26,920 Speaker 4: much to doctor David Cox for joining us on smart Talks. 742 00:39:27,360 --> 00:39:31,080 Speaker 4: We deeply appreciate him sharing his huge breadth of AI 743 00:39:31,160 --> 00:39:35,160 Speaker 4: knowledge with us and for explaining the transformative potential of 744 00:39:35,239 --> 00:39:38,600 Speaker 4: foundation models in a way that even I can understand. 745 00:39:39,200 --> 00:39:43,719 Speaker 4: We eagerly await his next great breakthrough. Smart Talks at 746 00:39:43,719 --> 00:39:48,239 Speaker 4: IBM is produced by Matt Romano, David jaw nishe Venkat 747 00:39:48,280 --> 00:39:52,720 Speaker 4: and Royston Preserve with Jacob Goldstein. We're edited by Lydia 748 00:39:52,760 --> 00:39:57,080 Speaker 4: Jean Kott. Our engineers are Jason Gambrel, Sarah Bouguer and 749 00:39:57,160 --> 00:40:01,799 Speaker 4: Ben Holliday. Theme song by Gramma's Scope. Special thanks to 750 00:40:01,880 --> 00:40:06,040 Speaker 4: Carli Megliori, Andy Kelly, Kathy Callahan and the Eight Bar 751 00:40:06,160 --> 00:40:10,200 Speaker 4: and IBM teams, as well as the Pushkin marketing team. 752 00:40:10,480 --> 00:40:13,800 Speaker 4: Smart Talks with IBM is a production of Pushkin Industries 753 00:40:14,040 --> 00:40:18,160 Speaker 4: and iHeartMedia. To find more Pushkin podcasts, listen on the 754 00:40:18,200 --> 00:40:23,359 Speaker 4: iHeartRadio app, Apple Podcasts, or wherever you listen to podcasts. 755 00:40:23,800 --> 00:40:40,840 Speaker 4: Him Malcolm Gladwell. This is a paid advertisement from IBM.