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