WEBVTT - Fei-Fei Li Helped Create AI, Now She Feels the Responsibility 0:00:02.759 --> 0:00:07.240 Bloomberg Audio Studios, podcasts, radio news. 0:00:09.200 --> 0:00:12.680 I was always a curious kid. I wasn't curious about 0:00:13.119 --> 0:00:17.360 nightclubs or her or other things. I was an avid 0:00:17.480 --> 0:00:19.400 lover of science. 0:00:20.640 --> 0:00:26.560 Faith a Lee Tech CEO, Stanford professor and godmother of AI. 0:00:27.000 --> 0:00:30.720 Are you conscious of the power that you have? 0:00:31.040 --> 0:00:36.159 I'm conscious of my responsibility. I understand. I'm one of 0:00:36.200 --> 0:00:40.479 the people who brought this technology to our world. The 0:00:40.560 --> 0:00:43.919 agency is not the machines, It's ours. 0:00:45.040 --> 0:00:50.120 From Bloomberg Weekend. This is the Michelle Hussein Show. I'm 0:00:50.120 --> 0:00:59.440 Michelle Hussein. Have you ever thought about where AI came from? 0:00:59.640 --> 0:01:04.920 The origin story of tools like Chat, GPT, Claude, or Gemini, 0:01:04.959 --> 0:01:08.679 which are part of many people's everyday lives. I don't 0:01:08.720 --> 0:01:13.040 mean the principle of training large language models on data, 0:01:13.480 --> 0:01:17.240 but the long, often lonely work of the scientists who 0:01:17.280 --> 0:01:23.440 believed years ago that you could teach machines to be intelligent. Well, 0:01:23.480 --> 0:01:26.800 today's guest has lived that story because she was at 0:01:26.800 --> 0:01:31.440 the forefront of unlocking this world for us all, and 0:01:31.520 --> 0:01:36.199 she has a deeply resonant personal story. She was fifteen 0:01:36.280 --> 0:01:39.200 when she came to the US from China with her parents. 0:01:39.640 --> 0:01:44.120 She had little English life was hard, money incredibly tight, 0:01:44.920 --> 0:01:49.200 but she excelled at school and set herself audacious problems 0:01:49.240 --> 0:01:53.160 to solve. Her name is fay fe Lee, and by 0:01:53.160 --> 0:01:56.440 the end of our conversation, I was asking myself, why 0:01:56.440 --> 0:01:59.840 do most people know the name Sam Aortman and not hers. 0:02:00.560 --> 0:02:04.840 She's the one whose breakthrough was critical, No wonder they 0:02:04.880 --> 0:02:07.120 call her the godmother of AI. 0:02:08.440 --> 0:02:09.160 We ended up. 0:02:09.080 --> 0:02:13.400 Talking about everything from her own childhood to yes, whether 0:02:13.480 --> 0:02:16.600 AI will do away with us humans, and what we 0:02:16.600 --> 0:02:20.320 should teach the kids of today. Her answer on that 0:02:20.760 --> 0:02:24.320 might just surprise you, but I hope you enjoy our 0:02:24.400 --> 0:02:30.720 conversation as much as I did. Doctor fay fe Lee, 0:02:30.720 --> 0:02:34.280 thank you for coming on the show. I'd love to 0:02:34.360 --> 0:02:39.000 start with this remarkable period for your industry. It's three 0:02:39.080 --> 0:02:41.520 years since chat GPT was released to the public, and 0:02:41.639 --> 0:02:45.040 since then there have been new vehicles, new apps, huge 0:02:45.080 --> 0:02:49.040 amounts of investment flowing towards the industry. How does this 0:02:49.160 --> 0:02:50.360 moment feel to you? 0:02:50.960 --> 0:02:53.280 Well, first of all, Michelle, thank you for inviting me 0:02:53.400 --> 0:02:57.200 to this show. I'm very excited. It's a good question 0:02:57.560 --> 0:03:00.040 because AI is not new to me. I've been in 0:03:00.120 --> 0:03:04.080 this field for twenty five years. I live and breathe 0:03:04.120 --> 0:03:08.639 it every day since the beginning of my career. Yet 0:03:08.880 --> 0:03:13.440 this moment is still as daunting and almost surreal to 0:03:13.560 --> 0:03:18.760 me in terms of its massive, profound impact. This is 0:03:18.800 --> 0:03:24.760 a civilizational technology, and it's surreal personally to me because 0:03:25.240 --> 0:03:28.880 I'm part of the group of scientists that made this happen, 0:03:29.680 --> 0:03:34.079 and clearly I did not expect it'll be this massive. 0:03:34.600 --> 0:03:36.520 When was the moment it changed? Because I know you've 0:03:36.560 --> 0:03:39.880 talked about the years when it was like an AI 0:03:40.040 --> 0:03:44.200 winter and what you're describing now, how the moment feels 0:03:44.240 --> 0:03:46.640 extraordinary even to you. Is it because of the pace 0:03:46.680 --> 0:03:49.040 of developments or is it because the world has woken 0:03:49.120 --> 0:03:51.800 up to it and therefore turned the spotlight on people 0:03:51.840 --> 0:03:52.160 like you. 0:03:52.680 --> 0:03:57.240 I think it's intertwined, right, But for me to define 0:03:57.280 --> 0:04:01.480 this as a civilizational technology is not about spotlight. It's 0:04:01.560 --> 0:04:06.080 not even about how powerful it is, is about how 0:04:06.120 --> 0:04:12.320 many people it impacts. Everyone's life work, while being future 0:04:13.000 --> 0:04:18.760 will somehow be touched by or impacted by AI. 0:04:18.520 --> 0:04:20.120 In bad ways as well as good. 0:04:20.480 --> 0:04:22.320 What technology is a dubba sword? 0:04:22.520 --> 0:04:22.719 Right? 0:04:22.960 --> 0:04:27.160 Yes? I think both ways, Because since the dawn of 0:04:27.240 --> 0:04:32.440 human civilization, we create tools we call technologies and these 0:04:32.480 --> 0:04:36.600 tools are meant in general for doing good things, but 0:04:36.720 --> 0:04:39.800 along the way we might intentionally use it in the 0:04:39.839 --> 0:04:43.120 wrong way, or we might have unintended consequences. 0:04:43.320 --> 0:04:45.760 And strands of this I know will emerge in different 0:04:45.800 --> 0:04:48.320 ways over the course of this conversation. But you said 0:04:48.360 --> 0:04:52.240 the word power, and I'm struck by the fact that 0:04:52.920 --> 0:04:56.640 the power of this technology is in the hands of 0:04:57.320 --> 0:05:00.680 a very small number of companies, most of them American. 0:05:01.040 --> 0:05:03.120 How does that sit with you? 0:05:03.120 --> 0:05:10.000 You're right, the major tech companies, through their massively reaching products, 0:05:10.760 --> 0:05:15.960 they are impacting globally our society the most. I would 0:05:16.000 --> 0:05:20.160 personally like to see this technology being much more democratized. 0:05:20.920 --> 0:05:24.640 I would like to see no matter who builds or 0:05:24.680 --> 0:05:28.800 holds the profound impact of this technology, do it in 0:05:28.839 --> 0:05:33.480 a responsible way. And I also do believe every individual 0:05:34.320 --> 0:05:38.159 in this era should feel they have the agency to 0:05:38.320 --> 0:05:39.720 impact this technology. 0:05:39.920 --> 0:05:41.680 We'll talk a bit more in a moment. I think 0:05:41.720 --> 0:05:45.240 about the democratization and how that might be achieved. But you, 0:05:45.320 --> 0:05:47.760 of course, in terms of companies in this field, you're 0:05:47.839 --> 0:05:50.240 part of that because you are a tech CEO as 0:05:50.279 --> 0:05:52.920 well as an academic. In fact, I think you're very 0:05:52.960 --> 0:05:55.720 young company, little more than a year old is reportedly 0:05:55.720 --> 0:05:57.320 already worth a billion dollars. 0:05:57.880 --> 0:06:03.279 Yes, co founder CEO of war Laps, and we are 0:06:03.320 --> 0:06:08.240 building the next frontier of AI, which is spatial intelligence, 0:06:08.320 --> 0:06:11.839 which people don't hear too much about today because we're 0:06:11.880 --> 0:06:16.000 all about large language models. Yet I believe spatial intelligence 0:06:16.080 --> 0:06:20.320 is as critical and complimentary to language intelligence. 0:06:20.760 --> 0:06:24.520 Releasing this idea of virtual world, which again we'll dig into. 0:06:24.600 --> 0:06:27.240 But before we do that, I want you to take 0:06:27.320 --> 0:06:29.320 us back again to the fact that you've seen the 0:06:29.360 --> 0:06:32.159 whole trajectory of this industry. You've been in it for 0:06:32.200 --> 0:06:35.360 twenty five years. I know that your first academic love 0:06:35.440 --> 0:06:38.359 was physics. Yes, what was it in the life or 0:06:38.400 --> 0:06:40.960 work of the physicists you most admire that made you 0:06:41.120 --> 0:06:42.839 think beyond that particular field. 0:06:43.279 --> 0:06:46.159 I grew up in a it's not a small town, 0:06:46.240 --> 0:06:50.279 but it's a less well known city in China, and 0:06:50.400 --> 0:06:54.560 I come from a small family, so you could say 0:06:54.720 --> 0:06:58.000 life was small in the sense it was in the 0:06:58.240 --> 0:07:02.720 eighties my childhood, which is fairly simple and isolated. You're 0:07:02.720 --> 0:07:08.880 a child I was, and my family in general is 0:07:08.960 --> 0:07:16.240 just small. Physics is almost the opposite. It's vast, it's audacious, 0:07:16.760 --> 0:07:20.600 The imagination is unbounded. You look up in the sky, 0:07:21.280 --> 0:07:25.080 you can ponder about the beginning of the universe. You 0:07:25.160 --> 0:07:28.240 look at a piece of a snowflake, you can zoom 0:07:28.280 --> 0:07:32.640 into the molecular structure of the matters you think about time, 0:07:32.760 --> 0:07:37.520 you think about magnetic field, you think about nuclear It 0:07:37.600 --> 0:07:41.280 takes my imagination to places that you can never be 0:07:41.920 --> 0:07:45.960 in this world. And what really fundamentally fascinates me to 0:07:46.040 --> 0:07:49.480 this day about physics is not be afraid of asking 0:07:49.560 --> 0:07:56.000 the most boldest audacious question about our physical world, of 0:07:56.080 --> 0:07:58.560 our universe, of where we come from. 0:07:58.720 --> 0:08:02.280 But your audacious question, I think, was about what is intelligence? 0:08:02.440 --> 0:08:06.240 Yes, And I think it's rooted in my love for 0:08:06.320 --> 0:08:10.760 physics and my training in physics that I look at 0:08:10.960 --> 0:08:15.360 each physicist I admire, look at their audacious question, right 0:08:15.400 --> 0:08:23.040 from Newton to Maxwell, to Schrodinger to i Instell, my 0:08:23.160 --> 0:08:27.760 favorite physicist, and I wanted to find my own audacious question. 0:08:28.160 --> 0:08:32.840 And somewhere in the middle of college, my audacious question 0:08:33.200 --> 0:08:37.360 shifted from physical matters to intelligence. 0:08:38.120 --> 0:08:38.679 What is it? 0:08:38.760 --> 0:08:43.160 How does it come about? And most fascinatingly, how do 0:08:43.240 --> 0:08:48.440 we build intelligent machines? And that became my quest Minor Star. 0:08:48.520 --> 0:08:53.520 And that's quantumly because from machines that were doing calculations 0:08:53.520 --> 0:08:56.760 and computations, you're really talking about machines that learn and 0:08:56.800 --> 0:08:58.040 that are constantly learning. 0:08:58.280 --> 0:09:01.960 I like you use the physics pun the quantum leap. 0:09:03.760 --> 0:09:07.120 Crossing over into into ordinary parlance. But I think, didn't 0:09:07.120 --> 0:09:09.120 you have this light bulb moment as you're thinking about 0:09:09.160 --> 0:09:12.840 intelligence you realize that a critical part of that is 0:09:12.880 --> 0:09:17.000 our ability to recognize objects. The fact that around us 0:09:17.120 --> 0:09:19.559 right now there are multiple objects, we know what they are. 0:09:19.679 --> 0:09:24.200 The ability that humans have about recognizing objects in the 0:09:24.240 --> 0:09:28.240 world is foundational. So I decided that was my first 0:09:28.360 --> 0:09:33.479 love start my PhD dissertation is to build machine algorithms 0:09:33.559 --> 0:09:37.400 to recognize as meaning objects as possible. 0:09:37.720 --> 0:09:39.920 I think the key question is how right how do 0:09:40.000 --> 0:09:42.280 you teach the machines? And what I found really interesting 0:09:42.360 --> 0:09:46.360 about your background is that you are reading very widely 0:09:46.679 --> 0:09:52.040 and there are two key breakthroughs that make your ultimate 0:09:52.080 --> 0:09:55.800 breakthrough possible. And one is that you start thinking and 0:09:55.920 --> 0:09:59.720 learning about what psychologists and linguists are saying that it's 0:09:59.760 --> 0:10:01.000 really to your field. 0:10:01.240 --> 0:10:04.319 Well, that's the beauty of doing science at the forefront, 0:10:04.520 --> 0:10:06.880 because it's you. No one knows how to do it. 0:10:07.559 --> 0:10:11.040 As an AI scientist, you look for possible answers, and 0:10:11.080 --> 0:10:14.840 it's pretty natural to look at human brain and human 0:10:14.880 --> 0:10:19.880 mind and try to understand or be inspired by what 0:10:20.080 --> 0:10:23.760 humans can do. And one of the inspirations I got 0:10:23.880 --> 0:10:27.559 in my early days of studying or trying to unlock 0:10:27.640 --> 0:10:31.600 this visual intelligence problem is look at how our visual 0:10:31.920 --> 0:10:36.360 semantic space is structured. There are so many tens and 0:10:36.480 --> 0:10:39.880 thousands and millions of objects in the world. How are 0:10:39.920 --> 0:10:44.280 they organized? Are they organized by alphabets or are they 0:10:44.440 --> 0:10:47.199 organized by size or colors? 0:10:47.480 --> 0:10:49.880 And you're asking that because you have to understand how 0:10:49.920 --> 0:10:53.640 our brains organize in order to have something to teach 0:10:53.720 --> 0:10:54.439 the computers. 0:10:54.679 --> 0:10:57.280 That's one way to think about it. And I come 0:10:57.400 --> 0:11:02.440 upon this linguistic work call word net word that is 0:11:02.480 --> 0:11:07.320 a way to organize semantic concepts, not visual, just semantic 0:11:08.000 --> 0:11:11.200 or words in a particular taxonomy. 0:11:11.360 --> 0:11:12.160 Give me an example. 0:11:12.280 --> 0:11:16.760 An example is in a dictionary, an apple and an 0:11:16.800 --> 0:11:21.080 appliance are very close together, but in real life, an 0:11:21.120 --> 0:11:24.600 apple and a pair a much closer to each other, 0:11:25.040 --> 0:11:29.800 and an apple and pair both belong to fruits, whereas 0:11:29.880 --> 0:11:34.400 an appliance belong to a whole different family of objects 0:11:34.800 --> 0:11:38.560 and I came upon this taxonomy called word net that 0:11:39.320 --> 0:11:44.000 describes this organization of objects, and I made a connection 0:11:44.360 --> 0:11:47.640 in my head. Is that two things is very clear. 0:11:48.280 --> 0:11:51.800 One is that this could be the way visual concepts 0:11:51.920 --> 0:11:55.880 are organized, because apples and pairs are much more connected 0:11:55.960 --> 0:12:01.040 than apples and wash machines, for example. But also even 0:12:01.120 --> 0:12:05.800 more importantly is the scale. Is that if you look 0:12:05.880 --> 0:12:11.280 at the number of objects described by language, realize how 0:12:11.520 --> 0:12:16.080 vast it is, how big it is. And that was 0:12:16.240 --> 0:12:21.920 a particular epiphany for me. We as intelligent animals, as humans, 0:12:22.480 --> 0:12:26.360 we experienced the world with massive amount of data, and 0:12:26.400 --> 0:12:28.760 we need to endow machines with that ability. 0:12:28.960 --> 0:12:32.120 Hence the big data sets needed. And actually it's worth 0:12:32.200 --> 0:12:34.959 noting that at that time, I think this is early 0:12:35.040 --> 0:12:37.719 part of the century, the idea of big data two 0:12:37.760 --> 0:12:39.520 thousand and six exists. 0:12:40.040 --> 0:12:44.960 No big data was not this phrase did not even exist. 0:12:45.280 --> 0:12:48.800 The kind of scientific data sets we were playing with 0:12:49.600 --> 0:12:54.760 were tiny. For example, in images, most of the graduate 0:12:54.800 --> 0:12:59.160 students in my era were playing with data sets of 0:12:59.600 --> 0:13:04.439 four or six or at most twenty object classes each 0:13:04.640 --> 0:13:08.400 class there were a couple of hundred examples and most 0:13:08.600 --> 0:13:12.040 and that's how tiny it is. Whereas fast forward three 0:13:12.120 --> 0:13:17.400 years later, after we created image that it had twenty 0:13:17.440 --> 0:13:23.720 two thousand classes of objects and fifteen million annotated image. 0:13:23.440 --> 0:13:27.000 And image net was a huge breakthrough. It's the reason 0:13:27.120 --> 0:13:30.000 that you've been called the godmother of AI. I'd love 0:13:30.080 --> 0:13:34.200 to understand more about what it is about you that 0:13:34.480 --> 0:13:37.240 enabled you to make these connections and see things that 0:13:37.280 --> 0:13:40.400 other scientists didn't. And one that immediately comes to mind 0:13:40.600 --> 0:13:44.040 is that you acquired English as a second language after 0:13:44.080 --> 0:13:46.840 you moved to the United States. Is there something in 0:13:46.880 --> 0:13:50.880 that that led to what you're describing. 0:13:51.600 --> 0:13:54.760 Michelle, that's a great question. I've been asked this question, 0:13:54.840 --> 0:13:59.160 and my true answers I don't know. Human creativity is 0:13:59.280 --> 0:14:02.840 still such a mystery. People talk about AI doing everything, 0:14:02.880 --> 0:14:05.960 and I disagree. There's so much about the mystery of 0:14:06.040 --> 0:14:09.000 human mind that we don't know. So I can only 0:14:09.160 --> 0:14:14.760 conjecture that a combination of my interest and my experiences 0:14:14.840 --> 0:14:19.320 led to this, including the curiosity and the hunger for 0:14:19.560 --> 0:14:24.440 audacious problem definition I find myself. I'm not afraid of 0:14:24.640 --> 0:14:31.840 asking crazy questions in science, not afraid of seeking solutions 0:14:31.880 --> 0:14:37.360 that's out of box. And maybe my appreciation for the 0:14:37.480 --> 0:14:42.520 linguistic link with vision might be accenterated by my own 0:14:42.600 --> 0:14:45.200 journey of learning the different language. But I don't know 0:14:45.280 --> 0:14:45.800 the answer. 0:14:46.560 --> 0:14:49.920 What was it like coming to the United States as 0:14:49.920 --> 0:14:54.280 a teenager, because I think that's a particularly difficult stage 0:14:54.320 --> 0:14:57.400 in life to move and to have to make new friends, 0:14:57.480 --> 0:15:00.200 even if you weren't battling a language barrier. 0:15:00.680 --> 0:15:06.400 That was hard, I think being an immigrant. And I 0:15:06.440 --> 0:15:11.160 came to this country when I was fifteen in New Jersey. Parcipitately, 0:15:11.200 --> 0:15:14.600 in New Jersey, none of us, my parents nor I 0:15:14.880 --> 0:15:19.040 spoke much English at all. I was young enough to 0:15:19.120 --> 0:15:22.800 learn more quickly. It was very hard for my parents, 0:15:23.000 --> 0:15:26.200 and we were not financially very well at all. My 0:15:26.280 --> 0:15:29.600 parents were doing cashier jobs and I was doing Chinese 0:15:29.640 --> 0:15:34.840 restaurant jobs. Eventually, by the time I entered college, also, 0:15:35.120 --> 0:15:38.320 my mom's health was not so good. So my family 0:15:38.360 --> 0:15:41.240 and I decided we have to run this little dry 0:15:41.320 --> 0:15:45.800 cleaner shop in New Jersey to make some money to survive, right, 0:15:46.000 --> 0:15:49.520 And you got involved yourself, I joked, I was the CEO. 0:15:50.800 --> 0:15:54.120 I ran the dry cleaner shop for seven years from 0:15:54.160 --> 0:15:58.480 what age I was eighteen nineteen to the middle of 0:15:58.520 --> 0:16:01.680 my graduate school, so all the way through college and 0:16:01.720 --> 0:16:04.640 graduate school, seven years, even when you're at a distance 0:16:04.680 --> 0:16:07.960 you're running your parents strike cleaning business. Yes, because I 0:16:08.080 --> 0:16:10.000 was the one who speak English, so I take all 0:16:10.040 --> 0:16:12.240 the customer phone calls, I deal with the building, I 0:16:12.360 --> 0:16:16.560 deal with inspections or the business. What did it teach 0:16:16.600 --> 0:16:21.280 you resilience? As a scientist, you have to be resilient 0:16:21.360 --> 0:16:27.720 because science is a nonlinear journey. Nobody has one conjecture 0:16:27.840 --> 0:16:30.240 and all the solutions in front of you. You have 0:16:30.320 --> 0:16:34.040 to go through such a challenge to arrive and find 0:16:34.080 --> 0:16:38.640 an answer. And as an immigrant you learn to be resilient. 0:16:39.000 --> 0:16:42.680 Were your parents pushing you because they clearly wanted a 0:16:42.880 --> 0:16:46.760 better life for you, That's why they left China for 0:16:46.800 --> 0:16:49.600 the United States. So how much of this was coming 0:16:49.640 --> 0:16:51.480 from them and how much from your own sense of 0:16:51.560 --> 0:16:53.280 responsibility to them. 0:16:53.720 --> 0:16:57.440 To their credit, they did not push me much. They're 0:16:57.480 --> 0:17:01.640 not tiger parents in today's language. I think part of 0:17:01.680 --> 0:17:04.600 it is they're just trying to survive, to be honest, 0:17:05.440 --> 0:17:10.160 and especially my mom, she is an intellectual at heart, 0:17:10.359 --> 0:17:16.800 she loves reading, but the combination of a tough, survival 0:17:16.920 --> 0:17:21.400 driven immigration life plus her health issues, she was not 0:17:21.400 --> 0:17:24.960 pushing me at all. As a teenager, I kind of 0:17:25.040 --> 0:17:27.560 had no choice. I either have to make it or 0:17:27.640 --> 0:17:30.920 not make it, and the stakes are pretty high. So 0:17:31.359 --> 0:17:34.919 I was pretty self motivated. I was just curious. I 0:17:35.040 --> 0:17:39.560 was always a curious kid, and my curiosity had an outlet, 0:17:40.000 --> 0:17:44.000 which was science, and that really grounded me. I wasn't 0:17:44.080 --> 0:17:48.320 curious about nightclubs or other things. I was an avid 0:17:48.480 --> 0:17:49.880 lover of science. 0:17:50.040 --> 0:17:52.479 But I also going to avid reader that from your 0:17:52.520 --> 0:17:57.040 early childhood, you're putting away the children's books, tanning towards 0:17:57.080 --> 0:17:59.919 the classics, the drawn up books. 0:18:00.480 --> 0:18:05.240 My mom liked reading, and she probably thought that you're 0:18:05.240 --> 0:18:09.399 talking about when I was nine ten. At that time, 0:18:09.480 --> 0:18:13.640 she probably did have influence more influenced. She thought I 0:18:13.800 --> 0:18:18.160 was precocious enough that just read some grown up books. 0:18:19.520 --> 0:18:22.119 You also had a teacher who is really important in 0:18:22.160 --> 0:18:23.280 your life. Tell me about him. 0:18:23.760 --> 0:18:28.879 I excelled in math, and I liked math, and I 0:18:28.960 --> 0:18:34.840 befriended the head teacher, mister Bob Sabella. And the math teacher, 0:18:35.400 --> 0:18:41.119 we became friends through the mutual love of science fiction reading, and. 0:18:41.080 --> 0:18:42.440 We start read in Chinese. 0:18:43.359 --> 0:18:46.240 At that time, I was reading in Chinese. Eventually I 0:18:46.280 --> 0:18:52.240 started reading English. He's a remarkable person because he probably 0:18:52.280 --> 0:18:56.399 saw in me this desire to learn, so he went 0:18:57.080 --> 0:19:00.399 out of his way to create opportunity for me to 0:19:00.960 --> 0:19:06.480 continue to excel in math. I remembered I placed out 0:19:06.520 --> 0:19:11.040 of the highest curriculum of math, and so there was 0:19:11.080 --> 0:19:14.439 no more courses for me. He would use his lunch 0:19:14.480 --> 0:19:18.880 hours to create a one one class for me to do. 0:19:19.600 --> 0:19:22.000 Now that I'm grown up, I know he was not 0:19:22.200 --> 0:19:26.200 paid extra. It was really out of a teacher's love 0:19:26.359 --> 0:19:31.800 and sense of responsibility. And he really became such an 0:19:32.200 --> 0:19:33.960 important person in my. 0:19:33.880 --> 0:19:35.960 Life because he still alive mister Sabella. 0:19:36.119 --> 0:19:40.159 Mister Isabella passed away when I was assistant professor at Stanford, 0:19:40.320 --> 0:19:43.800 but his family, his two sons, his wife, and I 0:19:43.800 --> 0:19:45.680 I think they are my new Jersey family. 0:19:45.760 --> 0:19:48.479 At this point, you used the word love about what 0:19:48.600 --> 0:19:51.800 he did for you, and I wonder if they also 0:19:51.880 --> 0:19:55.320 He and his family introduced you to American society with 0:19:55.880 --> 0:19:59.119 your first friends and an entrance to the whole world 0:19:59.160 --> 0:20:00.679 of America beyond your school. 0:20:00.880 --> 0:20:08.720 Absolutely, they introduced me to the quintessential middle class American family. 0:20:09.720 --> 0:20:13.480 They live in a suburban house, they have two lovely kids. 0:20:13.960 --> 0:20:17.040 Of course they're all married and have the grand children 0:20:17.160 --> 0:20:20.240 generation now, but it was a great window for me 0:20:20.400 --> 0:20:23.760 to know the society, to be grounded, to have friends 0:20:23.800 --> 0:20:25.200 and have a teacher who cared. 0:20:46.760 --> 0:20:49.359 Do you think you could have had the career that 0:20:49.440 --> 0:20:53.239 you have had in China because now there are like 0:20:53.320 --> 0:20:56.760 significant advances happening in AI in China. 0:20:57.440 --> 0:20:59.520 I don't think I'm able to answer this question because 0:20:59.520 --> 0:21:03.600 I think life is so serendipitous, right the journey would 0:21:03.600 --> 0:21:08.080 be very different, and in a way we could have 0:21:08.240 --> 0:21:13.360 simulated all possibilities. But what is timeless or what is 0:21:13.520 --> 0:21:19.480 invariant for anyone is the sense of curiosity, the pursuit 0:21:19.520 --> 0:21:23.760 of north stars. So if I were me, I will 0:21:23.800 --> 0:21:26.240 still be doing AI somehow, I believe. 0:21:27.840 --> 0:21:29.480 Do you still feel connected to China? 0:21:30.040 --> 0:21:33.280 It's part of my heritage. I feel very lucky that 0:21:33.520 --> 0:21:36.800 my career has been in America, higher education and in 0:21:36.920 --> 0:21:40.560 Silicon Valley then as also in and out of industry 0:21:41.160 --> 0:21:45.560 and being in tech. The combination of all these ingredients 0:21:46.240 --> 0:21:50.919 is very global. The environment my family right now is in, 0:21:51.080 --> 0:21:55.840 which is Stanford, San Francisco, Silicon Valley, is very international, 0:21:55.920 --> 0:21:59.680 so I feel very connected. And the discipline in which 0:21:59.720 --> 0:22:04.399 is A is so horizontal it touches people everywhere that 0:22:04.520 --> 0:22:09.120 I do feel much more like a global citizen at 0:22:09.119 --> 0:22:09.960 this point. 0:22:09.800 --> 0:22:11.520 And of course it is a global industry. But there 0:22:11.520 --> 0:22:14.040 are some really striking advances in China, not least the 0:22:14.119 --> 0:22:17.480 number of patents, in the number of AI public papers 0:22:17.520 --> 0:22:19.880 coming out the Deep Seek moment earlier this year. As 0:22:19.880 --> 0:22:22.639 you look ahead in this century, do you think China 0:22:22.680 --> 0:22:25.520 will catch up with the US in the way that 0:22:25.600 --> 0:22:28.399 it has in other fields like manufacturing. 0:22:28.920 --> 0:22:32.280 I do think China is a powerhouse in AI. In 0:22:32.320 --> 0:22:36.960 this moment, most people would recognize the two leading countries 0:22:37.119 --> 0:22:42.320 in AI are China and US. The excitement, the energy, 0:22:42.920 --> 0:22:46.440 and also frankly, the ambition of many regions and many 0:22:46.520 --> 0:22:49.960 countries in the world of wanting to have a roll 0:22:50.040 --> 0:22:54.560 in AI, wanting to catch up or even come ahead 0:22:54.600 --> 0:22:58.320 in certain areas AI. That is pretty universal. 0:22:58.160 --> 0:23:02.840 And your own next frontier spatial intelligence. Tell me what 0:23:03.160 --> 0:23:06.040 you mean when you use the word spatial intelligence. What 0:23:06.080 --> 0:23:07.359 are you working on right now? 0:23:07.560 --> 0:23:13.960 Yeah? So, spatial intelligence is the ability for AI or frankly, 0:23:14.119 --> 0:23:20.640 any intelligence to understand, perceive, reason and interact and also 0:23:20.800 --> 0:23:28.800 create spaces worlds. It comes from a continuation of visual intelligence. 0:23:29.200 --> 0:23:32.040 The first half of my career around the image that 0:23:32.040 --> 0:23:35.160 at time was trying to solve the fundamental problem of 0:23:35.760 --> 0:23:39.520 understanding what we are seeing, and that was a very 0:23:39.560 --> 0:23:44.040 important problem. But it's not enough because that's a very 0:23:44.080 --> 0:23:47.800 passive act. It's just receiving information and being able to 0:23:48.560 --> 0:23:51.879 understand this is a cop, this is a beautiful lady, 0:23:52.280 --> 0:23:55.520 this is a microphone. But if you look at evolution, 0:23:55.720 --> 0:24:01.840 you look at human intelligence, perception is profound linked to action. 0:24:02.720 --> 0:24:07.040 We see because we move, we move, therefore we need 0:24:07.080 --> 0:24:10.640 to see better. And how do you create that connection 0:24:10.840 --> 0:24:14.320 has a lot to do with space, because you need 0:24:14.359 --> 0:24:17.119 to see the three D space. You need to understand 0:24:17.160 --> 0:24:20.720 how things move. You need to understand when I touch 0:24:20.840 --> 0:24:26.600 this cup, how do I mathematically organize my fingers so 0:24:26.680 --> 0:24:30.600 that it creates a space that would allow me to 0:24:30.800 --> 0:24:36.080 grab this cup. All this intricacy is center around this 0:24:36.240 --> 0:24:38.520 capability of spatial intelligence. 0:24:38.560 --> 0:24:40.880 And I've looked on your website and I've seen the 0:24:40.920 --> 0:24:46.399 preview that you've released Marble. You can see essentially a 0:24:46.480 --> 0:24:51.080 virtual world which one goes into and from room to room, 0:24:51.119 --> 0:24:53.200 and one door opens and you go from one place 0:24:53.240 --> 0:24:56.640 to another. But I'm not sure how you use it. 0:24:56.680 --> 0:25:00.439 Is it essentially to you a tool for training AI. 0:25:00.480 --> 0:25:04.840 It's a different way to train AI, rather than for example, 0:25:04.920 --> 0:25:07.359 meta saying this is in the metaverse. This is a 0:25:07.400 --> 0:25:09.439 world you can go into and spend time and as 0:25:09.480 --> 0:25:10.080 a human being. 0:25:10.320 --> 0:25:13.640 Right, So let's just be clear with the definition. Marble 0:25:14.080 --> 0:25:18.680 is a frontier model. It's a frontier world model. What 0:25:18.720 --> 0:25:21.080 it does is not just to see and go into 0:25:21.119 --> 0:25:26.200 a world. What's really remarkable is it generates a three 0:25:26.320 --> 0:25:31.120 D world at a simple prompt. A prompt could be 0:25:32.080 --> 0:25:36.080 give me a modern looking kitchen, or the prompt could 0:25:36.160 --> 0:25:39.880 be here's a picture of a modern looking kitchen. Make 0:25:39.960 --> 0:25:43.920 it a three D world. The ability to create a 0:25:44.040 --> 0:25:48.880 three D world is a fundamental ability. It's fundamental to humans, 0:25:48.880 --> 0:25:51.800 and it's I hope one day, is fundamental to AI. 0:25:52.359 --> 0:25:57.920 If you are a designer or architect, you can use 0:25:58.000 --> 0:26:02.439 this three D world to ida to design. If you 0:26:02.520 --> 0:26:06.200 are a game developer, you can use it to obtain 0:26:06.280 --> 0:26:09.720 these three D worlds so that you can design games. 0:26:09.920 --> 0:26:13.440 If you want to do robotics simulation, these worlds will 0:26:13.480 --> 0:26:17.320 become very useful as training data for robots. If you 0:26:17.480 --> 0:26:23.800 want to create immersive educational experiences in ARVR, this model 0:26:23.840 --> 0:26:25.040 would help you to do that. 0:26:25.119 --> 0:26:27.959 Interesting I'm imagining girls in Afghanistan, Maybe you could do 0:26:28.480 --> 0:26:32.240 virtual classrooms in a very challenged place. 0:26:32.200 --> 0:26:35.800 Yes, Or I'm imagining, for example, how do you explain 0:26:35.880 --> 0:26:40.920 to an eight year old what is a cell? One day, 0:26:41.119 --> 0:26:45.080