李飞飞：探索人工智能革命

感谢大家再次加入我们，收听联邦俱乐部的另一期播客节目。

Thank you for joining us for another podcast from the Commonwealth Club.

好的。感谢各位的到来。大家好，欢迎来到联邦俱乐部。我是DJ·帕蒂尔，Great Point Ventures的普通合伙人，也是联邦俱乐部理事会成员。

All right. Thank you all for being here. Hello, and welcome to the Commonwealth Club. My name is DJ Patil. I'm a general partner at Great Point Ventures and a member of the Commonwealth Club Board of Governors.

俱乐部要感谢伯纳德·奥舍尔基金会对今晚“好文学”活动的支持。现在我很荣幸向大家介绍今晚的嘉宾。被誉为现代人工智能催化剂ImageNet的创造者，李飞飞博士二十多年来一直站在该领域的最前沿。我真心认为她不仅是现代AI的一块奠基石。

The club would like to thank the Bernard Osher Foundation for supporting tonight's Good Lit event. It is now my pleasure to introduce tonight's guest. Known as the creator of ImageNet, a catalyst for modern artificial intelligence, Doctor. Fei Fei Li has spent more than two decades at the forefront of the field. I really think of her as putting not just one cornerstone down of modern AI.

她为AI奠定了多块基石，我们稍后会详细探讨。她还是斯坦福大学计算机科学系首位红杉教授，斯坦福以人为本AI研究所的联合主任及创始人。飞飞还是全国非营利组织“全民AI”的联合创始人兼主席，旨在提升AI教育的包容性与多样性。她发表了300多篇科学论文，并著有新书《我所看见的世界：AI黎明时期的好奇、探索与发现》——希望各位都已拿到，若没有，门外就有售。飞飞，感谢你的到来。

She has put down multiple cornerstones of of AI, and we're gonna talk a little bit more about that. She's also an inaugural Sequoia professor in the computer science department at Stanford University and co director and founder of the Stanford's Human Centered AI Institute. Fei Fei is also the co founder and chairperson of the national nonprofit AI for All aimed at increasing inclusion and diversity in AI education. Fei Fei has published more than 300 scientific articles and is the author of a new book, which you hopefully have all got, or if not, you can get out just outside the door here, of The Worlds I See, Curiosity, Exploration, and the Discovery at the Dawn of AI. So Fei Fei, thank you for being here.

谢谢。最重要的是，祝贺你。

Thank And most of all, congratulations.

谢谢

Thank

你。这本书太棒了。让我们从这个问题开始：如果我们回到童年时代，告诉那时的自己，像我们这样的两个移民会站在台上谈论AI。这可能吗？

you. This is an incredible book. Let me start with if we were to visit ourselves back when we were a kid and said, two immigrants like us are gonna be up on stage talking about AI. Would that have been possible?

不可能。我们那时连AI是什么都不知道。我们根本

No. We we wouldn't even know what AI was. We wouldn't

不知道AI为何物。这正印证了你取得的巨大成就——如今我们如此热烈地讨论AI及其对世界的影响。但我想我们应该退一步，从你的起点说起，因为你在书中描述的成长历程非常精彩。能否分享一下你的成长经历？

even know what AI was. And so just a testament of how much you've achieved is is that we are talking so much about AI and the impact it's having on the world. But I think we should take a step back and get to where you got your start because it's a phenomenal journey that you outline in the book itself. So maybe you could tell us a little bit about your your upbringing.

谢谢大家。DJ，其实我想分享一个关于这本书的小故事。我最初并不是按成书的样子来写的。当时受邀为大众写一本关于AI的书，正值疫情期间。

Thank you. Thank you, everyone. Actually, DJ, I wanna share with a very small story about this book. I didn't write the book the way it turned out to be. I was invited to write a book about AI for popular audience, and I it was COVID.

我花了一年时间写了一本非常学术的书。完成后，我的一位哲学家朋友兼教授读了草稿，他说我必须回去重写。听到这个我简直吓坏了。他妻子是作家，他说这就是为什么他妻子从不给他看草稿。但他接着说，Fei Fei，有些人能谈论技术性AI，我知道你可以也应该这么做。

I spent a year writing a very nerdy book. And it was my friend, a philosopher, professor who read my draft after that year I finished and who said, you have to go back and rewrite. I was, like, horrified to hear that. And his his wife is an author, and he said, that's why my wife never showed me her draft. That's but then he said, look, Fei Fei, there are people who can talk about technical AI, and I know you can and you should.

但对于所有移民、年轻女性以及各行各业的人来说，在当今的AI领域，他们找不到能产生共鸣的声音。他说服我，我的声音代表着我们社区中非常重要的一部分。于是我最终写了一本双螺旋结构的科学回忆录——一条线是一位科学家移民的成长历程，另一条线是AI的发展史，当然还有我们交织的命运。所以在后台我们聊到这点时，DJ，我和你。

But for all those immigrants out there, young women out there, people of all walks of life out there, in today's AI, they don't find a voice they can identify with. And he convinced me that my voice represents a very, very important sector of our community. And that's how I ended up writing a a double helix science memoir in the sense of one thread is a coming of age of a scientist, an immigrant. Another thread is the coming of age of AI and, of course, our intertwined history. So from that point of view, we were talking a little bit in the backstage, DJ, you and me.

这本书记录的旅程如今是独一无二的。我们国家有这么多移民，这么多不同背景的人，他们都闯出了一片天地，为国家和社会做出了巨大贡献。即使在AI这个硅谷高度技术化的领域，我也希望能为这样的声音留出空间。

This book is this journey I have taken is now unique. We have so many immigrants, so many people of all kinds of origin in this country who have, you know, who have come through, done great things, contributed to this country, to this society. And I hope that even in AI, which is a deeply, deeply tactical Silicon Valley, you know, field, we allow room for that voice.

说得好。那么请分享一下你的故事，我想这对理解你处理AI问题的方式很有启发。

Well said. Well said. Well, tell us a little bit about your story because I think it gives a lot of insight to how you approach the problems of AI.

好的，还是希望你们读这本书。我15岁随父母来到美国，落脚在新泽西州的珀尔塞福涅——不确定有人听说过这个小镇，那是90年代初一个非常小的地方。

Right. Well, I still hope you read the book. But but I I came here with my parents when I was 15, and of all places, we landed in Persephone, New Jersey. I'm not so sure anyone here would have heard of that town. It's a very small town in Persephone in early 1990s.

但机缘巧合的是，离小镇30英里处就是AT&T贝尔实验室，当时计算机科学家们正在研发第一代神经网络。我刚到新泽西州珀尔塞福涅，开始学英语，进入公立高中，那时痴迷物理——那确实是我最大的热情。我们该谈谈一位特别的人。

But very serendipitously, about 30 miles away from that little town, there was AT and T Bell Labs. And already computer scientists were making the first generation of neural network. As I just landed in Precipitate, New Jersey, started learning English, went into the public high school and I was passionate about physics. It really was my biggest passion. And we should talk about a very special person.

但在那之前，在珀尔塞福涅高中，我先概述下：我获得了普林斯顿奖学金，期间不得不养家——我父母至今不会英语。后来我用硅谷的方式解决了问题。

But before that, in Perceputing High School, before that, I'll just go over the broad stroke. I I ended up with a scholarship in Princeton. And and then in the middle of that, I we had to support our our family. My parents don't still don't speak English. And I ended up in Silicon Valley language.

我的第一个创业项目是干洗店，真的有好心人资助我。我经营了七年，贯穿本科和加州理工的博士生涯——那算是Zoom出现前的远程工作。后来我从物理转攻AI，因为本科末期，关于智能的大胆问题——什么是智能，如何制造智能机器——彻底吸引了我。

I had my first startup, which is a dry cleaner shop. And with literally some angels who funded helped me. And and I ran the dry cleaner shop for seven years throughout my undergrad as well as my PhD, which was Caltech. So there was some remote work before Zoom. And and then I switched from physics to AI because I found that the audacious question of intelligence, what intelligence is, how to make intelligent machines, captivated me towards the end of my undergrad.

我认为这就是我作为科学家的天职，从此开始了博士阶段的探索之旅。

And I thought that's just my life's calling as a scientist. And that's how I began my journey in in PhD time.

你知道吗，书中也提到，即使在中国你也是非传统学生——读的书、做的事、学习方式都很特别。这种对学习的热爱究竟从何而来？我提前点明这点是因为你现在正教计算机如何学习，这很耐人寻味。

You know, even as and this comes out a bit in the book, which you all really should read, is you were a nontraditional student even in China. The works you were reading, what you were doing, how you were approaching learning. And what is it that really where did that love and passion for learning come from? And the reason I'm going to I'm going to foreshadow this is because you're helping teach computers how to learn. That's true.

那么你的起点是在哪里呢？

So where did yours start?

你知道吗，或许归根结底，这就是人类与机器的区别所在——我的起点是爱，而电脑未必如此。

You know, maybe at the end of the day, it's the difference between humans and machines because where I start is from love, whereas computers don't necessarily start there.

展开说说。

Say more.

如果你读过我的书就会知道，我的父母以非传统的方式支持并深爱着我。他们允许一个小女孩尽情保持她的好奇与独特。书中提到的我父亲——我相信你生活中也遇到过这类人——他天生就是个充满好奇心的人，痴迷昆虫。

Because if you read my book, there was my parents who supported me, and they they loved me in a way that was unconventional. They let a little girl be as curious and as unique as she wanted to be. And my dad, which the book talked about, also is a I'm sure you get you you met these kind of people in your life. He's just an innately curious person. He he loves insects.

他热爱动物，热爱自然。这份热爱极具感染力，我就像他的小灵媒，童年时就拥抱自然并找到了自己的路。可能我比他更书呆子气些吧。

He loves animals. He loves nature. And his love of those things were infectious, and I kind of become his little psychic. And and just embraced the nature when I was a kid and and found my own path. I was a little nerdier than him, I guess.

所以我更多是在书籍而非昆虫中找到方向。但正是他们出于爱为我创造的空间——在那个教育体系相当严苛的环境里，父母竭尽全力为我开辟出这片自由的天地。空间虽小，但对那个环境中的小女孩而言无比珍贵。这就是一切的开始。

So I I found my path in the books a little more than just bugs. And but it's it's the space they gave me out of their love for me. It it was in that educational system, it's pretty tight education system, but yet my parents, with all their strength, created a space for me and let me be free in that space. It's not a huge space, but it's a very precious space for a young kid, especially a young girl in in that kind of environment. And and that's where it began.

但我觉得你暗示的其中一个人是你的数学老师？能聊聊他吗？

But I think you're alluding to one of the people you're talking about is is your math teacher. Could you talk tell us about him?

是的。如果说这本书部分献给一个特别的人，他的特别之处在于他只是位普通的美国公立高中教师。他叫鲍勃·萨贝拉，是我在珀耳塞福涅高中的首位数学老师。作为数学系主任，他必须处理我这个不会英语的ESL问题学生，后来成了我的微积分老师。

Yeah. So if this book is partially dedicated to a very special person, the reason he's so special is he's your everyday American public high school teacher. He's not unique in that sense. His name is Bob Sabella, and he was my first math teacher at Persephone High School because he was the head of math department, I had to take placement tests. I didn't speak English, so he had to deal with the troubled ESL kid.

在此期间，他是最慷慨、富有同情心又严厉的老师。他理解我作为ESL学生的困境，察觉我的孤独与需要帮助。

And eventually, he became my calculus teacher. But in the meantime, he was the most generous, compassionate, and also tough love teacher. He kind of recognized my challenge as a as a ESL English as second language student. He recognized my loneliness. He recognized I needed help.

他真正将我庇护在他的羽翼下。渐渐地我们成为朋友，不仅讨论数学，还聊他热爱的文学和科幻。他带我回家，他同为高中数学老师的妻子也接纳了我，他们成了我在美国的家人。

And he he he really just took me under his wings. And one thing led to to another, we became friends. And we talk about math, but we also talk about literature and science sci fi, which he he loves. And he took me to his home and his wife, also a high school math teacher, just embraced me. And they became my American family.

他们成为我首次触及这个国家对我核心意义的体验。在硅谷这里，我们许多人同样来自东西海岸接受教育，我们就像沿海居民。对我们而言，美国代表着光鲜亮丽的事物——初创企业、曼哈顿的摩天大楼。

They become my first encounter to the very core value of what this country means to me. Here in Silicon Valley, where many of us also are educated from the other coasts, we're like coastal people. We we look at America means shiny things. It means startups. It means, you know, big skyscrapers in in the Manhattan.

但对我来说，美国意味着那位公立高中数学老师所体现的最根本价值观：慷慨、善良与正直。鲍勃·萨贝拉先生和他的家庭成为了我的美国家人，他们贯穿我的学业与职业生涯始终，甚至是帮我买下干洗店的天使之一。

But for me, America means public high school math teacher who had the most fundamental value of generosity, of kindness, and of integrity. And that is that really Mr. Bob Sabella and his family became my American family. I they stayed with me throughout my course of my study and career. They were one of the angels who helped me to even purchase the strike cleaner shop.

他陪伴我度过大学和研究生时光。遗憾的是，当我成为斯坦福助理教授时他离世了，但我仍与他家人保持亲密联系。我深深感恩生命中有鲍勃·萨贝拉先生，相信每个美国人心中都有一位萨贝拉，我想用这本书来致敬他们。

And, you know, he stayed with me throughout college, throughout graduate school. Unfortunately, he passed away when I actually became a Stanford assistant professor, but I stayed very close with his family. And I'm so grateful that I have I have mister Bob Sabella in my life. And I think many of us in America have a mister Sabella in our life, and I wanna use this book to celebrate them and to

现场有老师吗？有的。感谢你们从事教育。这很奇妙，因为你书中贯穿的主题正是人们给你设置挑战时，你几乎带着种狂妄说'这还不够好'。

Do have any teachers here? We do. Thank you for being a teacher. You know, it's it's amazing because you talk also about that the this teacher, and and this is sort of themed through the book, how people put challenges in front of you, And and you have this almost audacity to say that's that that's not good enough.

DJ，我觉得那个词应该是'妄想'。

DJ, I think that word is delusion.

我更倾向用'狂妄'，这让我想到ImageNet——能否请你谈谈这个构想是如何诞生的？从你设定的规模与目标来看，这已超越狂妄达到近乎荒谬的程度。能否带我们回顾这段历程？

I'm I'm gonna go with Adassie because it it is like and and I think I think this is a good place to talk about ImageNet because and so I'll ask you to tell us a little bit about it, about how it how we got to how you got to the idea of ImageNet. But this is, I mean, scope and magnitude of where you started with and what you decided to get to. I mean, there's audacious and there's fay fay audacious. So maybe you could take us through a little bit of that journey.

确实。这个词虽不适用于我，但物理学吸引我的正是那种敢于追问本质的狂妄——时空起源、宇宙最小粒子与边界。这些疯狂问题最终引向我认为最狂妄的命题：何为智能？机器能思考吗？当我毕业时，我们正深陷AI寒冬。

Yeah. Well, I do like that word not about me, but about that's what physics captured me, is the audacity to dare to ask the most fundamental question of our nature, where beginning of space time or smallest particle of the universe and end of the the boundary of the universe. These are crazy, crazy questions that's so audacious. And they eventually led me to the most audacious question in my mind, which is what is intelligence, and can we make machines think? So by the time I finished the graduate school, we were smack in the middle of AI winter.

公众对此知之甚少。2005-2006年我毕业时，我们步入一个有趣的时代：机器学习作为数学工具已引入AI领域，互联网时代也刚起步。作为助理教授，我研究着AI界最烧脑的问题——如何让计算机像人类一样'看见'。

Right? The public doesn't know about much about what this is. But we are starting to this is end of my grad school is 02/2005, 02/2006. So we we entered a very interesting age that machine learning as a mathematical tool is already being introduced to the field of AI, we also are at the beginning of the Internet age. And I was a assistant professor working on one of the most mind boggling question in computer in AI, which is make computers see, see as humans see.

人类的视觉远非如此，Didier。此刻我看到的不是黑色团块或圆柱体，而是一个鲜活的人、你的表情、你喝水的动作。

Humans don't see Didier, you sit here. I don't see a black blob or a cylindrical shape. I see a person. I see your expression. I see you were drinking water.

你握着笔等等。我们以如此丰富的意义感知世界，却不知如何让计算机做到。我想创造能识别世间万物的计算机，但当时领域的标准做法只是用四类物体的小数据集。幸运的是，我的跨学科博士教育给了我不同视角。

You're holding a pen and all that. So we see the world with so much richness and meaning, and we don't know how to make computers do that. And I wanted to create this computer that can see all the object in the world. But in the meantime, my field, at that time, the standard practice is take a dataset of four different objects and just work on it. And I was also fortunate enough to have a education, to have my PhD that was interdisciplinary.

我曾涉猎神经科学和认知科学领域，试图从人类智能中汲取灵感。我们知道人类确实具备洞察意义、感知世界丰富性的能力。而且人们从小就能观察到很多事物——你就是在不断体验这个世界的过程中成长的。大约在2006年2月到2007年2月期间，我和学生们产生了一个既大胆又近乎妄想的想法：为什么我们不将视觉世界的全部丰富性绘制出来，输入计算机，并用海量数据对其进行训练呢？

I did some neuroscience, cognitive science, to be inspired by human intelligence. And we know humans are really capable of seeing meaning, seeing richness in the world. And they also see a lot as a kid. You grow up just experiencing this world. And I think that was the audacious, delusional idea that I had with my students around 02/2006, 02/2007, which is, why do we map out the whole world of visual richness and give it to the computer and train it with the vast amount of data.

2006年2月那会儿，要知道我们连像样的云基础设施都没有。那时候连存储那么多视觉数据的硬盘和内存都很难凑齐。

02/2006, we're talking still, you know, we didn't have really much cloud infrastructure No. To be able to to even get enough hard disks and memory to store that much visual energy.

没错。实际上这正是我提出这个想法时遇到的阻力之一。他们说我们连容纳你设想数据的硬盘都没有。但我们选择相信这个理念。我查阅了最大的词汇分类词典WordNet，从中提取了数万个视觉类别。

Yes. Actually, was precisely some of the pushback when I talked about this idea. They're like, we don't even have hard drive to contain what you're trying to do. But I guess we believed in it. I went to the biggest lexicon taxonomy dictionary, it's called WordNet, and took tens of thousands of visual categories from WordNet.

能举个具体类别的例子吗？

What's an example of a category to help?

德国牧羊犬是一个类别，暹罗猫是一个类别，微波炉是一个类别，轿车也是一个类别。

A German shepherd is a category. A Siamese cat is a category. A microwave is a category. A sedan is a category.

所以你们收集了1万个这类物品。

So you have 10,000 of these.

是2万2千个。然后我们转向互联网，下载了数十亿张图片。

22,000. 2,000. And then we went to the Internet. We downloaded billion images.

其中不少是猫咪照片吧。

Lots of cat photos.

确实有很多猫。但网上也有狗的照片。我们获取了数十亿张图片后决定进行整理——也就是清洗、分类和标注。现在回想起来这简直疯狂，当时我居然打算雇佣本科生来做这件事。

Lots of cat. But the Internet also has dogs. So we took billions of images and we decided to curate, which means to clean them and sort them and label them. That was really truly kind of crazy. In hindsight, I don't know what I was thinking, because I thought I would be hiring undergrads to do this.

但普林斯顿的本科生们似乎不太买账，他们拒绝了这个工作。2007年2月左右，机缘巧合下，有个从硅谷斯坦福转学到普林斯顿的学生告诉我——这真是个典型的硅谷故事——有家初创公司正在使用亚马逊Mechanical Turk平台。

And the Princeton undergrads were just not that great, I guess. They refused to do it. And around 02/2007, serendipitously, a student that I was a faculty at Princeton. A student who moved from Silicon Valley, from Stanford to Princeton said, in Silicon Valley, this is such a great story of Silicon Valley, there is a startup. I heard they're using this platform called Amazon Mechanical Turk.

然后那天我去查亚马逊 Mechanical Turk 是什么时，发现它是一个在线劳动力市场，人们在那里完成数字任务。你可以发布一些小型的数字任务，那里有数百万的工人。接着你基本上达成协议，比如你发布一个任务并支付10美元。如果我是线上工人，如果我喜欢这个任务和报酬，我就会报名。

And and then that day when I went to check what Amazon Mechanical Turk was, it turned out to be a online market of workers trying to do digital tasks. You put little digital tasks there, and there are, you know, millions of workers. And then you kind of agree. You you say, well, I put up a task, and I'll pay $10. If I'm a worker online, if I like that task and I like your pay, I sign up.

所以这有点像是一个市场。我研究后发现，我那十亿张不知道由谁来整理和分类的图片找到了归宿。于是我们建立了整个亚马逊 Mechanical Turk 系统。长话短说，我们花了三年时间。最终在2009年2月，我们整理出了当时人工智能领域最大的数据集——ImageNet。

So it's kind of a marketplace. I I checked that out, and I realized my a billion images that I don't know who is gonna cling and sort for me have found a marketplace. So we set it up this whole Amazon Mechanical Turk system. We took three years, long story short. And eventually, we curated ImageNet dataset in 02/2009, the largest dataset in AI at that time.

这也是第一个真正的大数据。这是大数据首次进入AI领域，并迫使整个领域开始用大数据思维思考。然后我们将其开源，并创建了ImageNet挑战赛。我们

It's also the first real big data. It's the first time big data entered AI and forced forced the field to think with big data. And then we open sourced it. We created a ImageNet challenge. We

但在谈到挑战赛之前——嗯。我认为有必要指出，学术界人士当时的看法是怎样的？当你最终宣布‘看，这是ImageNet’时，发生了什么？

But before you get to the challenge Uh-huh. I think it's important to note, what was the perspective of people from academia? When you finally are like, Behold, ImageNet, what happens?

沉默。反响不大。你觉得他们为什么没理解？他们说这太超前了，因为用大数据来研究AI在当时还不是一个概念。现在有了ChatGPT这些，大家都觉得理所当然。但当时并非如此。

Silence. There wasn't much Why don't you think they got it? They said it was ahead of its time because to approach AI with big data was not a concept. Now with ChatGPT and all this, everybody said, of course. But it wasn't like that.

当时人们关注的是数学模型。这些模型非常优雅，但他们没有从数据中心的视角思考。这需要等待合适的时机。我很幸运。科学史上有很多超前于时代的赌注，可能它们等待的时间不够长，或者发生了其他事情。

There were mathematical models people were paying attention to. These are very elegant models, but they were not thinking from a data centric point of view. And it has to wait for the right opportunity. I was lucky. I mean, are many bets in the history of science that were ahead of their time and maybe they didn't wait long enough or something happened.

但幸运的是，当这个想法开始结出果实时，我还活着。我们在2012年发起了这个挑战赛。加拿大的一位教授杰夫·辛顿，带着两位非常聪明的学生。一个叫亚历克斯·克里泽夫斯基，另一个叫伊利亚·苏茨克维。

But I was lucky to be alive, I guess, when this idea start to, you know, to to take fruition. And we put out this challenge in 2012. A professor in Canada, professor Jeff Hinton, with two very smart student. One is called Alex Krusevskiy. The other one is called Ilya Sorskever.

最近新闻上见过。好吧。那么，

Been in the news. Okay. So, like

他们看到了ImageNet，还发现了一种叫GPU的东西。他们用了两块GPU，因为

And they saw image that and they also saw there is this thing called GPU. They took two GPUs because

GPU？

GPUs?

显卡是由英伟达制造的图形处理器。它比CPU（即2012年左右我们大多数人使用的经典英特尔芯片）更擅长并行处理。他们用ImageNet数据和一种已有的经典算法为这两块GPU编程。记得我说过珀尔塞福涅吗？当他们正在30英里外开发神经网络时，我降落在了珀尔塞福涅。

Graphics a graphic processing unit made by NVIDIA. It is better at parallel processing than CPUs, which is the classic Intel chip that most of us were using around 2012. And they programmed that two GPUs with ImageNet data with a classic algorithm that has been around. Remember I said Persepany? I landed in Persephone while they were developing neural network 30 miles away.

而在九十年代，这三个元素的交汇就是人们所说的深度学习革命的开始。

And in the nineties, the convergence of these three elements was what people call the the beginning of deep learning revolution.

嗯，或许你知道，让我印象深刻的是，就在同一时期，我们数据社区的人都在惊叹：天哪，她是怎么整合这个数据集的？这将解锁魔法。这是来自领英的观点。你们确实赋能了一支难以置信的大军去尝试算法、测试和竞赛。

Well, maybe what you know, one of the things that sort of strikes me is, like, at that same time, we were on the data side of the community going, oh my gosh. How did she put this data set together? This is gonna unlock magic. You know, this is from the LinkedIn side. And so you really were able to empower an unbelievable army of people to try algorithms, to test, to compete.

我很好奇，当你观察这些团队工作时，有多少是近身肉搏式的合作？当时社区内的互动本质是怎样的？我这么问是因为看当今AI领域，我们看到的是赤裸裸的

I'm curious, you know, as you were watching these groups work, was it how how much hand to hand combat was it? Was there a lot of collaboration going on? What was the nature of the dynamics in the community? And the reason I ask that is we look at today's AI and we see this flat out

竞赛。但

race. But

从你的书中我感受到，那时截然不同。

I get the sense from your book, it was very different.

确实完全不同。那是AI非常学术化的时代。我觉得自己成长在AI的黄金时代，因为它如此学术化，如此友好。当然人们会有不同观点——比如他们认为图像数据不重要，这是他们的看法。但我们协作、分享、开源。

It was very different. It was the time where AI was very academic. I feel I grew up in the golden era of AI because it was so academic, it was so friendly. Of course, people have opinions like they didn't think image data was a thing and it's their opinion. But we collaborate, we share, we open source.

我经常去MIT、伯克利，还有斯坦福和加州理工。那是个完全不同的时代。但也是AI尚未奏效的时代，对吧？所以没人在乎AI。

I was visiting MIT all the time and Berkeley and it's just Stanford, Caltech. It was a very different era. But it also is an era that AI didn't work. Right? So nobody cared about AI.

所以所以当时

So so There were

没有风投来敲门。

no VCs knocking on the door.

当时没有风投机构。另外，DJ，我想说的是在人工智能时代，我认为女性得到了蓬勃发展。作为一名年轻的女计算机科学家，我认为自己实际上受益于AI领域尚未被权力和金钱侵蚀的时期。

There is no VCs. And also, DJ, I wanna say that era of AI, I think women flourished. As a young woman computer scientist, I think I actually benefited from a period of AI where power and money didn't come in yet.

很有意思。说得对。那么我们来谈谈其他方面吧，因为你接手了ImageNet项目后见证了这些突破，你开始聚集那些通常在其他地方得不到关注的极具创造力的博士生。如今这个领域许多知名人物都曾在此。但你还决定涉足不同领域，包括医疗保健。

Interesting. Hear, hear. Well, let's talk a little bit about some of the other aspects because you took a you've taken which ImageNet and then you saw these breakthroughs, you were able to start putting together really creative PhD students that typically weren't getting attention anywhere else. Some of the very popular names today, the well known names of people in the field. But you also decided to take on different areas, including health care.

其中让我印象特别深刻的一个主题是尊严。这让我感触很深，因为书中也贯穿了你母亲与严重慢性疾病抗争的故事。所以我想请你展开谈谈，你如何看待尊严？如何看待与这个领域人们的关系？因为你提供的叙事实在太丰富了。

And one of the lines that I think you really that stood out to me is the theme in particular is dignity. And it strikes me because also woven through this book is the story of your mother and very serious chronic health conditions. So I'd love to tease that out a little bit about how you see dignity, how you see this relationship with people in this space, because there's so much, it's just such a rich narrative that you provided.

是的。也要特别提到你，DJ。你是我们国家数字健康领域的先锋思想家和领导者对吧？这个话题我们可以深入探讨。没错，你说得对。

Yeah. And also a shout out to you, DJ. You're one of the forefront thinkers and leaders in digital health in our country, right? So this is something we can talk a lot about. Yeah, so you're right.

我一直是个基础科学研究者，专注于基础科学。但大约在二月份——感觉就在ImageNet取得突破的2012年2月到2013年2月期间——如果你住在硅谷，会知道另一个热门话题是自动驾驶汽车。那时我刚成为斯坦福人工智能实验室主任。

So I have always been a basic scientist, basic science scientist, which is we focus on the fundamental science. But around the time of February, I feel like it was right around ImageNet breakthrough. So 02/2012, 02/2013. If you live in the Silicon Valley, there is another hot topic, and that's self driving cars. And I just became Stanford's director of Stanford AI Lab.

这是个历史悠久的AI实验室，我是那里唯一的女性，也是首位女主任。当时我正与多家对自动驾驶技术感兴趣的公司接洽。我们开始研究这个关乎人命的高风险技术难题：处理复杂视觉数据并做出决策。与此同时，我的个人生活是频繁往返急诊室、手术室，呼叫救护车，进行家庭护理——因为我是父母（尤其母亲）唯一的看护人。这让我突然意识到自动驾驶与医疗护理（特别是病患照护）存在诸多共性。

It's a very historical AI Lab, and I was the only woman and the first woman director there. And I was in the middle of, you know, talking to different companies at that time, now interested in self driving car and all that. And then we're starting to look at that technical problem of high stake human lives, very complex visual data, and decision making. And in the meantime, I live a parallel life of going to ER, going to operation room, calling ambulance, home caring, case managing, because because I'm the only caretaker for for my for my parents, especially my mom. And then it really dawned on me that the problem of self driving car and health care, especially care delivery, how to take care of the patient, has a lot of commonality.

两者都关乎人命健康，具有高风险性；病患行为复杂，护理者行为同样复杂；而且美国面临护理人员短缺——你比我更清楚——护士短缺，老年护理人员更是严重不足。

It's high stake because human lives and health. It's complex because the behavior of patients are very complex. The behavior of caretakers are very complex. And also it needs help because in America, you know this better than me, we have labor shortage for caretakers. We have nurse shortage, and we have even bigger shortage for caretakers for elderlies.

于是我开始思考：计算机视觉技术或许能帮助照护病患和老人。

So I started to think there is a possibility to use computer vision as, you know, a technology to help taking care of patients and elderlies.

我能稍微打断一下吗？正如你所说，这种'辅助而非取代'的理念正是你的独特之处。能否谈谈这种理念形成的原因？

Can I just interrupt you for a quick second? Because I think you this is what makes you unique, as you said, to help take care of, not replace. And you you talk a little bit about so could you tell us a little bit about why you also have that ethos?

我们必须彻底摒弃'取代'这个动词，用'增强'和'提升'来替代。书里写过一个故事——DJ你说过喜欢那个故事——简而言之，我召集了包括Arne Milstein教授在内的临床医生和医学研究者，开始了这项合作。

We need we absolutely need to replace the verb replace and to replace it with augment and enhance. Absolutely. I I wanna tell one story that I did write in the book. And I think, DJ, you say you like that story. So so long story short, I got a group of clinician and researcher, medical researchers, especially professor Arne Milstein, and I start collaborating on this.

与此同时，我母亲经历了一场大手术，我开始在ICU和后来的普通病房照顾她。术后医生要求她使用呼吸训练器来扩张肺部，这对她这样的心脏病患者至关重要。你知道吗，医生刚下医嘱，我这个学计算机的女儿菲菲就立刻进入了高度紧张状态。我给她制定了时间表，心想：好，医生说要每小时做一次这个训练。

In the meantime, my mom went into a major surgery, and then I started taking care of her in ICU and then in the in the patient room. And the doctor after major surgery, the doctor ordered her to use spirometer to expand her lung because that's very important for her as a cardio patient. You know, once as soon as the doctor made that order, me, Fei Fei, the daughter, the computer science, I'm very intense. I put her on a schedule. I'm like, Okay, the doctor said every hour you have to do this.

我当时肩负着不让妈妈生病、避免肺炎的使命，完全进入了工作狂模式。结果一天后，我妈彻底崩溃放弃了。她就像反抗似的说：我不管了。

And I was on a mission to make my mom not sick, not have pneumonia. I was just in that workaholic mode. And after a day, my mom just completely broke down and gave up. She like rebelled. She's like, I don't care.

她说我再也不做呼吸训练了，巴拉巴拉...我受到了很大冲击。我觉得自己很受伤，心想：我明明是在帮忙，这很重要啊，巴拉巴拉...

I'm not gonna do spirometer. Blah blah blah blah blah. I was very traumatized. I felt like I was traumatized. And I was like, I'm trying to help and this is important, blah blah blah.

我就是无法理解。几周后我们回家休养时，我问妈妈：为什么当时你要抗拒用呼吸训练器？我是想帮你啊。我知道你当时半梦半醒，但我是清醒的。

I I just don't understand. And then several weeks passed, we were back home recovering. I asked my mom, I said, why did you fight with me on that spirometer? Because I was trying to help you. You know, I know you were like half half medicated, but I was clear minded.

我知道怎么帮你。结果她说：知道吗？你在控制我。这让我失去了尊严和自尊。这大概是我作为技术从业者一生中最深刻的对谈，尽管只有短短一句话。

I know how to help you. And then she said, you know what? You are controlling me. I don't have my dignity and self respect. It was probably one of the most profound conversation even though it was like one sentence in my entire life as a technologist.

我意识到，如果在那次对话前让我开发AI，我会造出一个菲菲的复制品。我会让AI按时间表提醒她，就会是那种模式。而人类要复杂得多——尊严、我们如何通过他人或技术建立帮扶关系，这些要微妙深刻得多。

I realized if I were to make AI before that conversation, I would make a copy of Fei Fei. I would make AI to work with that schedule to remind her. It would just be like that. Whereas humans are much more complex. The dignity, how we create a relationship with the help, whether it's through another human or through technology, is much more nuanced, much more profound.

无论发生什么，自尊与尊严都是人性的核心。技术工作者的职责是增强、扩展并尊重这种人性尊严，而非剥夺它。这段经历发生在2016年左右，具体年份记不清了。

No matter what happens, it's the self respect and dignity that's at the center of our humanity. And technology technologist's role is to enhance and augment and respect that dignity in humanity, not to take it away. And I think that was early. That was like 2016. I don't remember exactly the year.

我认为这段经历（当然还有其他许多经历）真正塑造了我作为技术工作者对人与科技关系的思考。这就是为什么我总强调：工具是来辅助的，不是来取代的。嗯，说得很好。

And I think that episode, of course, there's many other episodes, really informed me as a technologist of how to think about the relationship between technology and and people. This is why I focus on saying, tools are here to help, not to replace. Well, let's well said.

让我们延伸讨论这个工作方向，因为你不仅创办了非营利组织，还在斯坦福成立了新研究院，真正注入人文关怀并帮助弱势群体。或许我们可以先聊聊'全民AI'，谈谈这个项目的使命，再留些时间讨论其他内容。

Let's talk let's take an extension of that work because you not only started a nonprofit, but you also started this new institute at Stanford to really insert humanity and also help the underserved. So maybe let's first start with AI for All and tell us about that and hold some space for what the mission is there.

'全民AI'是全国性非营利组织。2015年我和斯坦福前学生奥尔加·泽科夫斯基、同事里克·萨默将其创立为暑期项目，邀请多元背景的高中生来大学校园进行两到三周的AI学习。但我们聚焦包容性，服务传统上被技术忽视的群体，并注重将人文使命融入技术教育。这个项目始于2015年2月。

AI for All is a national prophet. My former student Olga Zekofsky and colleague Rick Summer from Stanford started in 2015 as a summer program to welcome diverse high school students to come to university campuses to spend two or three weeks to learn about AI. But we focus on inclusion, serving community that has been traditionally underserved by technology. And we also focus on infusing that human mission into technology education. And that started in 02/2015.

实际上，准备工作始于2014年2月，因为我注意到当时的播客和电波中充斥着比尔·盖茨、埃隆·马斯克等人谈论人工智能多么危险，仿佛在召唤恶魔。就连史蒂芬·霍金也在讨论这个，说终结者要来了，这是一场危机。而我当时正生活在另一个无人提及的危机中——那时我是斯坦福人工智能实验室唯一的女性教员。

Well, actually, the preparation started in 02/2014 because I was noticing that the podcast, the airwave of even then was filled with people like, you know, Bill Gates, Elon Musk talking about AI being so dangerous, summoning the demon. Even Stephen Hawkins was talking about this, like it's Terminators are coming. It's a crisis. And then I was living in a different crisis that nobody talks about. I was, at that time, the only female faculty in Stanford AI lab.

我们有多少？不到15%的女性研究生。我们的研究生项目中几乎没有非裔美国背景的学生。我当时就觉得，这两个危机，一个如此科幻，另一个如此真实。它们之间是否存在联系？

We have what? Less than 15% women graduate student. We have barely any students of, you know, African American background in in our graduate pro program. And I was like, these two crises, one is the so sci fi, the other one is so real. Is there any connection?

我想在2014年的某个时候，我建立了这个联系：如果你真的担心AI的未来，就应该关注今天是谁在创造AI。谢谢。当我意识到这点后，我和当时的研究生奥尔加、朋友里克讨论，我们都认为必须做些什么。我们需要审视人才输送渠道，需要从K12教育阶段开始培养。

And and I I think somewhere in twenty fourteen, I made that connection is that if you are truly worried about AI's future, you should worry about who is making AI today. Thank you. And once I made that connection, I talked to my graduate student at that time, Olga, and my friend Rick, and we're like, we gotta do something. We need to look at the pipeline. We need to work with K 12 students.

就这样我们创立了非营利组织AFROL。头几年只有斯坦福参与，后来伯克利也加入了。梅琳达·盖茨和黄仁勋家族办公室支持了我们初期启动。项目至今仍在运行，说实话我希望它能发展得更快些。

And that's how we started this AFROL nonprofit organization. We started the first couple of years was just Stanford and then Berkeley joined and all that. And Melinda Gates and Jensen Huang family office supported our initial launch. And it's still going. Honestly, I wish it could grow faster.

怎样才能加速发展？需要什么条件？现在每个学生都想用ChatGPT写论文，而这才是真正培养建设者的项目。如何确保那些可能受技术影响最深的群体也能参与构建这类技术？

What would help it grow faster? What would take it? Because we're seeing the transformation of every student wants to use ChatTPT to write their essays, yet this is the actual program for builders. What would it take to make sure that we have this type of technology built by the communities that might be impacted the most?

你想听真实答案吗？

You want the real answer?

我相信他们肯定想听。

I think they do for sure.

是资金。作为非营利组织，我们需要支持。

It's money. We need it's a nonprofit. We we need we need support.

是啊，归根结底是资金问题。

Yeah. It's a money question.

作为非营利机构，这确实是个资金问题。说来话长，但我们仍在非常努力地争取，绝不会放弃。

As a nonprofit, it is a money question. So but that's a long story. It it is there. We're still trying very hard, so we're not giving up.

很好。我很高兴你在考虑这件事。如果人们想了解更多信息，他们该如何获取？

Good. I'm glad you are considering it. If people want to find out more about it, how did they find out?

Ai4haud.org。

Ai4haud.org.

对。没错。或许你能跟我们聊聊为什么要在斯坦福创办一个新研究所？

Yeah. Right. And maybe talk to us a little bit about why start a new institute at Stanford?

是的。我们之前聊过ImageNet，后来AI深度学习革命爆发了。时间快进到2017年，我当时已是教授——那时我当了大概十二年教授。在学术界你可以申请学术休假对吧？

Yeah. So we talk about ImageNet and then the AI deep learning revolution happened. And then fast forward to 2017, I was a professor that at that time, I was a professor for, like, twelve years. And in academia, you can take sabbatical. Right?

我很幸运收到多家公司邀请。谷歌联系我时，希望我担任谷歌云首席科学家。真正吸引我的是云服务覆盖所有行业——从医疗保健到金融服务，农业、能源，还有娱乐媒体零售等等。作为学者，我觉得这是走进现实世界、了解技术影响力的好机会，也能学习如何将技术转化为产品服务。

So I was very lucky to be invited by multiple companies. And Google called me, and they want me to be chief scientist of Google Cloud. What really captured my attention for this opportunity is cloud serves all businesses, from health care to financial services, to agriculture, to energy, to, you know, entertainment, media, retail, whatever. So I felt as a academic, it would be a good opportunity for me to go into the real world and learn about the impact of my technology. And also to try to learn how to deliver that in products and services.

加入谷歌后简直大开眼界。有几点让我震惊：首先斯坦福已是条件优渥的地方，但到了谷歌才真正见识到技术制造的规模。

So I went to Google and it was like amazing. Right? Several things shocked me. First is Stanford is a pretty privileged place. Yet going to Google, I realized the scale of this technology making, you know.

咖啡吧台。没错。还有免费餐饮。但最惊人的是规模——谷歌研究员能调用的计算资源量，那些软件工程基础设施。

Coffee bars. Exactly. And the free food. But the scale was just amazing. Amount of compute the Google researchers can use, the the kind of software engineering infrastructure.

你在书里提到过这个

I mean, you talk about this in the book of

对。

Yeah.

当初要东拼西凑才能搞到几块GPU，现在打个响指就能在...

How many GPUs you had this basically scrounge dollars for, and then then you snap a finger. And how many you can get at

我记得书中提到过800这个数字。800对比...是的，这是Transformer和GPT技术出现之前的事。总之，这是让我震惊并深受启发的一点。另一点启发并激励我的是AI的普及范围。

I think in the book, was talking about 800. 800 versus Yeah. This is pre transformer, pre GPT technology. Anyway, so that was one thing that shocked me and really illuminated me. Another thing that illuminated me and also inspired me was AI's reach.

我们当时真的在和各类开发者交流——从使用计算机视觉辅助经营的日本黄瓜种植户，到试图优化供应链和客户体验的财富五百强企业。我突然意识到，这项我曾参与创造的技术，曾经只是我个人私下里的好奇探索。那时只有我和我的AI，因为没人在意。而现在全世界都在使用它，我们这些创造者就有了责任。这个认知开始在我心中萌芽。

We were literally talking to developers like a Japanese cucumber farmer using computer vision to help his business all the way to Fortune five companies, you know, trying to optimize supply chain and customer experience and everything in between. And it dawned on me that this technology that I participated in creating used to be a private curiosity for me. It's just me and my AI because nobody else cared. Now it's the whole world is using it, and there is a responsibility for those of us who created it. And that started to dawn on me.

与此同时，2017到2018年正值第一波科技冲突浪潮。相信很多人都记得，这紧随剑桥分析公司和2016年大选事件之后。我们首次见证了自动驾驶汽车造成的伤亡，首次发现带有偏见的面部识别系统，首次陷入...

And in the meantime, 2017 and 2018 is the first wave of tech clash. I'm sure a lot of you remember this is right at the heel of Cambridge Analytica, the twenty sixteen election. I'm sure you have a strong memory of, DJ. And and the first time we have self driving car injuries and death, the first time we see biased system, facial recognition bias, The first time we got into

刑事司法保释评估系统从根本上就带有种族歧视。

Criminal justice bail calculators assessors being fundamentally racist.

没错。我们还首次见证了关于AI武器化的激烈辩论。这一切让我明白一件事：AI是混乱的。作为影响社会的技术，AI是混乱的；但作为数学公式，AI并不混乱。

Exactly. And we also see the first intense debate of AI and weaponization. And all this taught me one thing, that AI is messy. AI as a technology social impact is messy. AI as a math is not messy.

数学就是一加一等于二。但作为技术，它是混乱的。而我们将引领人类进入这个新时代，我们的责任是什么？正是在这一点上，尽管我深爱谷歌，感激这段经历，但我决定重返公共领域，建立创造和治理AI的新框架——以人为本的AI框架。

Math is one plus one equals two. But as a technology, it's messy. And we are the people who are gonna usher the humanity, human race, into this new era. What is our responsibility? And that's where I really decided as much as I love Google, I love the experience, I appreciate it, I have a responsibility to return to the public sector to create a new framework of creating AI and governing AI, and that is the human centered AI framework.

在斯坦福领导层的支持下，我在休假结束后回到斯坦福，创建了世界上第一个人本AI研究院。

And with the support of Stanford leadership, I returned to Stanford after my sabbatical and created the first human centered AI institute in the world, I guess.

是的。我们收到了许多优质提问，联邦俱乐部的优秀工作人员正在收集问题。让我们把话题转向当下和未来走向。

Yes. Well, we've had a number of really good questions. Our great staff here at Commonwealth Club are collecting questions. Keep bringing them in. A number of the questions, let's shift gears into today and where we're going.

或许可以从这里开始：现在打开广播或网页就能看到AI新闻，最近最热的就是OpenAI的事件。关于末日论者与技术乐观主义者的争论，您持什么立场？我们该如何看待这个问题？

And maybe the place to start is we're seeing you can't turn on the radio or open a web browser without hearing some news about AI. The most recent one being everything that's happened at OpenAI most recently. And there there's a question of doomers versus techno optimists. Where do you stand in the spectrum? And how should we be thinking about it?

是的。我推荐大家看看一个月前我和杰夫·辛顿在多伦多的公开讨论，视频在YouTube上。AI问题太过微妙复杂，不能简单二元划分。就像所有技术一样，人类文明史就是一部工具创造史。

Yeah. I actually recommend a public discussion I had with Jeff Hinton a month ago in Toronto. It's on YouTube. AI is too nuanced and too complex to go binary. And as all technologies, You know, humanity, human civilization is a history of tool making.

我们一直在创造越来越强大的工具，与工具的关系也始终复杂。而AI将把这种关系推向新高度。因此从学术角度探讨‘AI是否具有意识？它是否有感知？’是合理的问题。作为一名生活在大学校园的学者，没有问题是错误的。

And we've always created more and more powerful tools, and we've always always have had a complex relationship with tools. And AI is gonna kick this into another gear. So it is intellectually a fair question to talk about is AI conscious? Is it sentient? As a as a scholar who lives on a university campus, no question is wrong.

我们理应提出这些疑问。但这需要与更现实的社会风险——我称之为灾难性社会风险——并列考量，比如虚假信息与民主危机、职业变革、武器化应用、偏见问题、版权争议、知识产权侵犯、隐私侵害。这些都是如此真实的议题，如果我们把所有注意力都转向最极端的末日论调，反而会错过针对这些紧迫社会风险的关键讨论与行动。同时，作为参与创造AI的科学家，我也看到它在医疗健康、医学研究、科学发现、气候解决方案等诸多领域的机遇。

We we should ask this. But that is juxtapositioned with much more real social risks that I call them catastrophic social risks, such as disinformation and democracy, job change, weaponization, bias, copyright, intellectual property, privacy infringement. These are such real issues that I would worry if we shift all of our attention and and fill our airwaves with the most extreme doomers conversation. We are actually missing the critical conversation and action towards these really important pressing urgent social risks. In the meantime, as a scientist who created participating creating AI, I see its opportunity in healthcare, in medicine, in scientific discovery, in climate solutions, in many things.

所以我同样看到了教育等领域的惊人潜力。但我不认为我们可以只谈好处而回避弊端。你们能听出来，我不会给出非黑即白的极端答案，因为真正的工作、讨论和思考都存在于混沌的中间地带。

So I also see amazing opportunity, education. But I don't think we can only say it is only good. There's nothing bad. So you can hear, I'm I'm I'm just not gonna give you black and white extreme answers because I think the real work and the real discussion and real thinking is the messy middle.

你怎么看待——这些都是绝妙的问题，我觉得足够你写下一本书了。所以祝贺...

How do you know, one of the questions these are fantastic questions. I think you have enough for your next book here. So kudos to

CBP会替我写下一本

CBP will write my next

这里有个问题是：人们如何跟上节奏、保持前沿认知并参与其中？有些人可能不懂技术，有些人可能懂。对于想要学习这些复杂议题并发出声音的人，你有什么建议？

One of the questions that's in here is, how do people get up to speed, stay on top of things, be engaged, either because they may not be technical, some may be technical. What's your advice for people on how they can best learn about these complex subjects and then have a voice?

首先，你必须相信自己有发言权——尤其在民主社会中，作为公民、社区成员和你所在领域的专家，你天然拥有话语权。AI只是工具，我们需要学会使用它或理解它。至于如何了解当今的AI技术，其实有很多资源。如果你有技术背景，可以阅读大量论文。

Yeah, first of all, I think we start with you have to believe you have a voice, especially you're in a democratic society and as a citizen, as a member of a community, and as an expert in your own field, you have a voice. And this is a tool. Our relationship with this tool is that we need to learn how to use it or we need to learn what this is. In terms of how to get to know today's technical AI, there are actually a lot of resources. If you are technically you know, you have a technical background, then there is a lot of papers to read.

互联网上也有很多资源，比如YouTube或其他社交媒体，许多人在解读最新论文。虽然信息量庞大，但有技术背景的人总能找到途径。如果没有技术背景，也有优质资源——例如订阅斯坦福HAI的简报，我们会提供...

There's a lot of you know, even on the Internet, whether it's YouTube or other social media, you've got a lot of people who are communicating this and digesting the latest papers. It's a little overwhelming, but if you do have a tech background, there are resources. If you don't have a tech background, there are actually also good resources. For example, if you subscribe to Stanford HAI's newsletter, we put a lot of

他们怎么找到这个？

How how do they find it?

访问hai.stanford.edu，页面底部有订阅入口。我们有很多公开研讨会资源。

Hai.stanford.edu. You can scroll down. There is subscribe. Just go subscribe. Because we have a lot of public seminars.

我们为政策制定者、行业决策者等提供了大量简报材料，还有Coursera等平台上的课程资源。这些资料都是公开的。说实话，内容确实有点让人应接不暇。但只要你怀有学习意愿，根据你的专业背景和切入角度，其实有很多学习入口——特别是在我们所在的这个领域。

We have a lot of briefings for policy makers, for industry decision makers, and all that. And there are also courses that you can take from Coursera and all that. So the materials are out there. I'm not gonna lie, it's a bit overwhelming. But if you have the will to learn, depending on where your expertise come from, your angle is, it's actually a very there is a lot of access point, especially where we are.

我比较担心全球范围内的可及性问题，比如南半球国家。如果那里的学生想学习这些，我认为我们面临更大挑战。但在我们这个社区里，学习资料非常丰富。

I worry about global access, for example, our global south. If our students wanna learn there, I think we have a bigger issue. But in this community, there is a lot of materials.

那么让我们把话题提升到国家和国际层面。你会对总统说些什么？

Well, let's take this to the national and international stage. What what are you telling the president?

我确实见过他。是这样的，当时我...

I so I met him. Right? And in I

我是说，这非常值得注意。你是少数几位被选中向拜登总统进行相关教育的人士之一。

mean, you were it's it's very notable. You were one of the featured people to to up to to educate president Biden on this.

我告诉拜登总统，这项技术至关重要，我们应该以登月计划般的决心来配置公共部门AI资源。请允许我解释：美国在科技领域乃至全球更全面的领导地位，始终源于一个虽不完美但相对健康的创新生态系统——包括创业精神、产业界，以及政府在其应发挥作用时提供的积极支持。但当前这种平衡已被严重打破。

So I told president Biden that this technology is so important that we should adapt a moonshot mentality in resourcing public sector AI. And let me thank you. Let me explain. America's leadership in our technology and also in in actually a more well rounded leadership in the world has always stemmed from a fairly, it's not perfect, fairly healthy ecosystem of innovation, entrepreneurship, industry, and with the government playing when it's done right, playing a positive role in resourcing this. But right now, we are so imbalanced.

全美没有一所大学能训练CHAT GPT模型。我曾怀疑所有大学的GPU加起来能否做到。最近听说东海岸有些大学开始增购GPU，这让我非常羡慕。但无论如何...

Not a single university in America can train a CHAT GPT model. I I used to wonder if all universities GPU combined can train a chat GPT model. Recently, I heard there are some universities on the East Coast who are starting to buy more GPUs. So I'm I'm very envious of them. But anyway but but the

有点讽刺的是，你们距离总部只有几英里距离。是啊，甚至都...

It's kinda there's an irony that you're only so many miles from the headquarters Yeah. Don't even

别提了。确实。说到校友...

get me started. Yeah. We yeah, and speaking of alumni.

你还在以他们命名的大厅里授课呢。是的。

You give your lectures in there. Yes. The hall named after them.

是的。但话说回来，我们为何需要公共部门？公共部门？有两个至关重要的原因。首先，公共部门创造公共产品。

Yeah. But, so why do we need public sector? Sector? There are two very important reasons. First of all, public sector creates public goods.

科学发现就是公共产品。我们可以利用AI从攻克癌症到治疗罕见病，发现新材料，加速聚变研究，绘制全球生物多样性图谱，创造气候解决方案。这些都是公共产品，它们诞生于公共部门。如果我们剥夺公共部门使用这一关键工具的权利，就等于剥夺了人类获取这些知识与创新的机会。这是第一个原因。

Scientific discovery is public good. We can use AI to cure from cancer to rare diseases, to discover new materials, to accelerate fusion research, to map out the whole world's biodiversity, to create climate solutions. These are all public goods and they happen in public sector. If we're depriving public sector from this tool, this critical tool, we're depriving our humanity or our our species from knowing this knowledge and having this innovation. So that's one reason.

第二个原因是公共部门作为可信来源，负责解释、评估这项技术的本质。斯坦福HAI Mind研究所是首个也是目前唯一评估多家公司大语言模型的机构。我们刚刚发布了根据欧盟《人工智能法案》建议的透明度标准评估大语言模型的工作。但公众往往更信任公共部门的评估。

The second reason is public sector serves as a trusted source to explain and to evaluate and to assess what this technology is. Right? So Stanford HAI Mind Institute was the first one and the only one who is evaluating the large language models created by multiple companies. We just put out a work of evaluating the large language model against European Union's AI Act suggested transparency measure. But the public tends to trust what the public sector does.

若不能充分支持公共部门，我们将无法真正理解这项技术，无法深入探究其原理，也无法共同开发这项重要技术。因此基于这两点，我们必须大力支持公共部门，尤其要鼓励其致力于安全AI、伦理AI以及跨学科AI研究。

Now, if we don't resource public sector well, we're not gonna be able to know what this technology is. We're not gonna be able to look under the hood and also to co develop this important technology. So for these two reasons, it is so important we resource public sector, And we we especially encourage public sector to work on safe AI, ethical AI, and AI for for for many, many disciplines.

国际层面该如何看待？毕竟当前存在与中国的激烈经济竞争，还有俄罗斯、伊朗等国使用AI引发的真实安全问题。你曾在谷歌经历'马文计划'争议时期，我很好奇你如何看待这种国际复杂性？

How should we think about it internationally? Because, you know, there is this abject conflict that is taking place economically with China. There's real security questions with countries like Russia and Iran using AI. And and you were also at Google while the whole debate around Project Maven took place. And I am curious through your eyes, how do you see about the international complexity?

确实，DJ，这就是为什么我说情况很复杂。技术本身复杂，工具复杂，人类社会更复杂。我不会假装自己通晓所有答案。

Yeah, so the truth is, look, DJ, this is why I say it's messy, right? Technology is messy. Tools are messy. Human world is messy. And I'm not going to pretend I know the answer of all of this.

作为斯坦福这样的顶尖学府，我们的优势在于拥有各领域专家。HAI作为技术研究所，与胡佛研究所、弗里曼·斯波格利国际研究所、经济政策研究所CEPIR深度合作，集合政策专家、政治学家、经济学家共同应对这些挑战。同时我们坚信合作的力量。

One of the great things about being a wonderful university like Stanford is that we have experts. In fact, HAI as a technology institute, we deeply, deeply collaborate. We collaborate with the Hoover Institute. We collaborate with the Freeman Spogli Institute for International Studies, and we collaborate with CEPIR, which is our economic policy institute because we've got policy thinkers, political scientists, economists to to help us to to grapple this together. And we also I believe in partnership.

作为学术机构，我们致力于与包括各国政府在内的多方建立伙伴关系。我们与欧盟、澳大利亚等价值观相近的政府保持密切合作，这极其重要。在我的领导岗位上，我认为自己可以搭建桥梁，帮助政策专家理解技术本质，进而运用他们的专长为政府提供建议。

First of all, as a academic institute, we believe in partnership with, you know, many communities, including the government and governments of the world. We have been especially close with the EU and Australian and those government that have a shared value system. And we think that's very, very important. And we also play a role. I feel one of my role I can play in my leadership position is to create create the bridge to help our deep policy experts, political scientists to be educated about the technology, and then they can apply their expertise to to go advise the governments or or or do Do you

你觉得他们真正理解了吗？最近你们其实——既然已是公开报道——举办了一个训练营

think they get it? Because recently you ran actually well, I can say it because it's publicly reported. You ran a boot camp

是的。

Yeah.

关于国会。你知道，我们已经看到国会的一些评论。或者当你和他们交谈时，

For Congress. You know, we've seen some of the commentary out of Congress. Or when you talk to them,

什么政府比我强，DJ。实际上我认为他们变得更好了，因为我们的研究所成立于2019年，我第一次出差就是去华盛顿特区，人们不理解我在说什么。比如，他们几乎连AI都拼不对。我们实际上与特朗普政府合作过，其中有一些这方面的专家。是的。

what what's it government better than me, DJ. I actually think they have changed for the better because our institute was established in 2019, and the first trip I took was Washington DC, and people don't understand what I was talking about. Like, they can barely spell AI and not. We we were we actually we actually work with Trump administration, and there were some experts in this. Yeah.

所以你可能甚至认识其中一些人。但总的来说，即使是两党人士，他们也没有关注。然而过去这一年，来自华盛顿的关注度是惊人的。现在，我确实希望他们能找到更多专家。他们也会与公共部门的人交流。

So you might know even know some of them. But by and large, even bipartisan, they didn't pay attention. Well, this past year, the amount of attention from the DC world is phenomenal. Now, I do wish they find more experts. They do talk to people of public sector as well.

某一类人。是的。而且

Of one category. Yes. And also

对。我非常严肃地说这个，因为我们刚刚看到这个。你们中有些人可能也看到了。就在这个周末。《纽约时报》有个大新闻。

Right. I I say this with all seriousness because we just saw this. Some of you may have seen this. Also this weekend. There was a big news story about The New people Times.

《纽约时报》说这些是推动AI发展的核心人物。如果我可以说的话，名单上全是白人男性，忽略了你书中也提到的那些历史上指导你、帮助你、你参与他们旅程的大多数人。他们也忽略了你。所以这也引出了一个问题：我们几乎有一个被ChatGPT幻觉重写的单一版本的AI历史正在上演？

New York Times who said these are the fundamental people who have driven AI. And if I just may call it, it was all white men and ignored most of the historical people that you talk about also in your book who mentored you, helped you, you were part of their journey. They ignored you as well. And so this gets to a question also of we have a singular we almost have a rewritten Chad GPT hallucinated version of A. I.

历史正在上演吗？

History going on right now?

是的。我认为我们需要指出这一点。我不知道你们有多少人看过——我知道你看过——那篇《纽约时报》的文章，里面100%是男性。其中一些人虽然可能是商业领袖或投资者，但根本不是AI创造的核心人物。所以，我不知道，我们该给《纽约时报》写三封信吗？

Yes. I I think we do need to call this out. I think I don't know how many of you saw, I know you did, that New York Times article, and it was 100% men. And some of them, while they might be business leaders or investors, were not central to the creation of AI at all. So, I don't know, should we send three letters to New York Times?

确实。我认为我们需要公开批评他们。这是不可接受的。有那么多女性以及他们甚至没提到的男性做出了根本性贡献，他们却完全忽视了。这简直不可接受。

It is. I do think we need to call them out. This is unacceptable. There are so many women of and also even men they didn't mention who have made fundamental contribution, and they just completely ignored it. And it's just unacceptable.

我同意。我们这里有一些学生提出的问题。其中一个是，也许可以从这个开始，将话题拉回你的书——正如我们之前谈到的，你向你的导师和学生们表达了难以置信的敬意。所以或许把几个问题结合起来：对于那些需要导师指导或可能成为导师的人，你有什么建议来确保他们提携所有人？高中生如何获得实践经验？

I agree. One of the questions that we've got in here is from a number of students who are here. And one of the things that they ask is, maybe to start with this, tying this back to your book, you really pay, as we talked about earlier, unbelievable tributes to your mentors and your students. So maybe putting this in a couple of questions together, what advice do you have for people who are in need of mentorship or could be mentors to make sure that they're lifting everyone up? How can high school students get hands on experience?

对他们来说有哪些机会？还有一个问题是，你对那些即将高中毕业、进入大学的人有什么建议？他们应该考虑些什么？

What opportunities are for them? And also another one was, you know, what do you advise people who are graduating from high school, going into college, should they be thinking about?

对老师好一点。开个玩笑。如果说我写这本书有一个主题的话，那就是北极星的主题。当然，我在AI领域找到了我的北极星，我并不认为每个人都必须在AI领域找到他们的北极星。但我认为，关于我的旅程、我的导师的旅程以及我的学生的旅程的真实人类故事是：你要无所畏惧，勇敢，充满热情和好奇心，找到你的北极星。

Be nice to your teachers. Just kidding. If there's one theme I was writing this book with, it's the theme of north stars. It's I of course, I found my north star in AI, and I don't think everybody needs to find their North Star in AI. But I think the true human story of my journey and my mentor's journey, my student's journey is you be fearless, be courageous, and be passionate, and be curious, and find the North Star.

北极星可能会在人生的不同阶段发生变化，但无论你处于哪个阶段，无论是指导他人、接受指导还是考虑大学专业，我都强烈鼓励你找到那个让你感到好奇、充满热情和使命感的北极星。我认为拥有这样的北极星是一件美妙的事情。这是一种美好的感觉，它能让你最具创造力。坦白说，如果你在做让你兴奋的事情，你不会感到疲惫。

And they might change in different stages of life, but no matter what stage you are in, whether you're mentoring someone or being mentored or thinking about your majoring college, I really encourage you to find that North Star that makes you feel curious and passionate and filled with a sense of mission. And I think that is just a wonderful thing to have. It's a beautiful feeling, and it makes you most creative. And frankly, you you don't feel tired if you are working on something you're excited about.

对于那些处于可以成为导师位置的人呢？你见过哪些对拥有权力或特权的人特别有效的方法？

What about for people who are in positions here who could be mentors? What have you seen that works really well for people who are in positions of power or privilege?

多年来我一直担任导师的角色。在与学生共事的过程中，我不断学到的一点是：我们如何帮助他们找到他们的北极星。这不是关于我们，而是关于我们指导的人。最好的指导方式是了解学生是谁，年轻人是谁，他们想要什么。

So that's my role for many years now as a mentor. And as I work with students, the thing that I keep learning over and over again is how do we help them to find their North Star. It's not about us. It's about the people we mentor. And the best way to mentor is to learn who the the students are, who the young people are, what they want.

有时候，他们可能需要一位导师帮助他们看到自己的北极星。有时候年轻人并不一定能看清他们的使命是什么。如果一个导师能扮演这个角色，至少对我来说，这是非常令人满意和有回报的。

And maybe sometimes they it just takes a mentor to help them to see their North Star. Sometimes it's young people are not necessarily seeing what they they their their calling is. And if a mentor can play that role, at least for me, that's so satisfactory and rewarding.

很好。有几个人问到量子计算机的角色以及它可能对AI产生的影响。

Great. Several people asked about the role of quantum computers and the impact it may have on AI.

这是个很好的问题。我不知道。我认为我对量子计算的了解还不够，我们也不清楚量子计算的未来会怎样。现在还非常非常早期。如果量子计算能如我们所愿取得成功，它将解锁当今计算机所不具备的那种计算能力。

Well, that's a great question. I don't know. I don't think I know enough, and I don't think we know enough about where quantum computing is going yet. It's still very, very early. If quantum computing in the way that we hope it is succeeds, it'll unlock the kind of computational power that today's computers don't have.

由于AI是一个极度依赖计算的领域，它由大数据和大计算驱动，AI的实现方式可能会发生改变。但就目前而言，至少在我有限的知识范围内，这还不是实质性的影响。你明白吗？

And since AI is a deeply, deeply computational field, it's driven by big data, big compute, there might be a shift in how AI is done. But right now, just at least in my limited knowledge, it's not material yet. You know?

好的。你本来就睡得很少。是什么让你夜不能寐？

Okay. You barely sleep as it is. What keeps you up at night?

《纽约时报》那篇文章。你是

That New York Times article. Are you

在担心通用人工智能吗？

worried about AGI?

这是个好问题。我不会说我不担心。我想我已经在担忧人工智能了。如我所说，我已经在担心灾难性的社会风险。我担心2024年。

That's a great question. I would not say I'm not worried. I think I'm already worried about AI. As I said, I'm already worried about the catastrophic social risks. I'm worried about twenty twenty four.

我担心当下正被AI颠覆的创作者经济。我担心经济格局的转变。有件事我想重点强调——我确实相信AI是提升整体生产力的绝佳工具。这个机遇我看得非常清楚。

I'm worried about the creator economy that is being disrupted by AI right now. I'm worried about the shifting of economy. There's one thing I want to double click on that. I do believe AI is a great tool to increase the total productivity. I see that opportunity so clearly.

但生产力提升不意味着繁荣的公平分配。全球化进程中我们已见证这点，AI时代可能也会重演。所以仅讨论生产力提升是不够的，我们还必须探讨繁荣分配。这是AI时代从政策角度和社会科学角度都应思考的重大议题。

But increase of productivity doesn't mean there is a fair distribution of prosperity. And we have seen that in globalization and we might be seeing this in AI. So it's not enough to just talk about productivity increase. We have to also talk about prosperity distribution. That's a huge topic in the age of AI we should be thinking about from policy point of view, as well as from just social science point of view.

这些确实让我夜不能寐。正因如此，我对从事以人为本的AI工作感到兴奋。目前我还没为悲观者的末日场景彻夜难眠。

So all these do keep me up at night. So that's why I'm excited to work in human centered AI. I'm not yet staying up at night thinking about the gloomers, doomsday scenario.

是的。很有趣的是，早先你提到比尔·盖茨等人讨论末日场景时，我们正受奥巴马总统委托制定国家AI战略。

Yeah. It's interesting because when you were talking earlier about when Bill Gates and others were talking about the doomsday scenario, we were tasked by president Obama to come up with a national strategy for AI.

我们就是这么认识的？对吧。

That's how we met? Right.

那是我们初次相识。我认为我们当时试图整合的内容——有些做对了，但也有不少误判——用「哪些岗位将被取代或转移」可能更准确。那么展望未来五年，最大冲击会发生在哪里？

That's how we first met. And one of the things that I think we we tried to put together, and we got, I think, a lot right, but also a fair amount wrong, of which jobs are going to be replaced or displaced is probably a more accurate phrasing. So as you look forward over the next five years, where does the biggest impact happen?

嗯，我认为需要先定义这个「冲击」。如果针对就业，我们已看到知识工作者受影响对吧？连软件工程都受波及。因为原先我们以为会是卡车司机和...

Yeah, well, I think the impact, we have to define it. If it's really on jobs, I think we're seeing knowledge workers, right? Even software engineering being impacted. Then Because before, we thought it was going to be truck drivers and

短期厨师和其他高度依赖手工的劳动，但我们在灵活性方面搞砸了。这很难被替代。

short cooks and other very manual labor, but we messed up on dexterity. It's hard to replace.

没错。实际上不止是灵活性问题。我们还缺乏良好的世界模型。我们有的是优秀的语言模型。

Yeah. Exactly. There's actually more than dexterity. We also don't have a good world model. We have a good language model.

这就涉及到技术话题了。鉴于我们在语言模型上取得突破，我认为知识工作——特别是关于生产力提升的方式——如何让软件开发更高效？如何更好地服务客户？如何优化搜索和推荐？当然，我说的每件事都有两面性，对吧？

And this is getting to the technical topics. Given we have a breakthrough in language models, I think knowledge work, especially the way, again, about productivity, how do we make software engineering more productive? How do we service customers better? How do we search better, recommend things better? Of course, everything I said has both good and bad, right?

但我认为影响会非常深远。大语言模型技术不是炒作。它确实能对行业产生深刻而深远的影响。

But I think there is going to be a lot of impact. The large language model technology is not a hype. It really can have deep, deep profound industry impact.

你允许自己的孩子使用它吗？

Do you let your kids use it?

实际上，我允许。

I do, actually.

具体怎么做的？

How so?

好的。这个问题我被问过很多次，实际上我想说的是——教育工作者确实经常问我，尤其是K12阶段的。我的孩子正好在这个年龄段，问题是孩子们该不该用这个？我的答案是肯定的，但有前提条件。对吧？

Okay. This is a question I get a lot, and I I actually wanna talk about it is educators do ask me, especially K-twelve. So my kids are in that age, is that should kids use this? And my answer is yes, but there is a but. Right?

首先，听着，这是个工具。是我们创造了它。就像精灵不会回到瓶子里。而这代人生来就要面对这个工具已然存在的世界。但我们应该做的是让他们通过学会负责任地使用来接纳它。

Like, first of all, look, this is a tool. We created it. It's that the genie is not gonna go back to the bottle. And this is a generation that will be embracing a world with this tool already created. But I think what we should do is let them embrace it by learning to use it responsibly.

通过教导他们使用工具来增强自身能动性、创造力和生产力，而不是把它当作拐杖导致丧失自主能力。如何调节这个度就是教育工作者的任务，也需要我们所有人协作。但我确实认为这对K12学习都是绝佳工具。

By, you know, learning to use it so that they can enhance and augment their own agency, their own creativity, their own productivity, not to use it as a crutch that let them lose their agency. And how to dial that knob is the the the task of an educator and then also in collaboration with all of us. But I I do think this is a great tool for for even k 12 learning.

好的，我们时间快到了，总是难以相信时间飞逝得如此之快。最后一个问题：你收到过最好的职业建议是什么？是谁给出的？你如何将这一建议转化为对当今人们的指导？

Well, we're we're almost coming up on time, which is always hard to believe how fast it flies here. One last question. What's the best piece of professional advice that you've received? Who was it from? And how do you translate that for today's people going forward?

我想回到我的高中数学老师鲍勃·萨贝拉身上。我不认为他实际上说过‘这是我给你的建议’这样的话。他只是身体力行地展现了‘传递善意’这一点。你知道，他对我没有任何期待。我当时是个懵懂无知的孩子。

I guess I'm gonna go back to my high school math teacher, Bob Sabella. I don't think he actually said, Here is a piece of advice I'm giving you. He just embodied it, which is paying forward. You know, he had nothing to expect of me. I was this kid who didn't know what's going on.

他的善良、同情心、慷慨以及正直——他是位严格的老师——帮助我看到了未来，帮我铺就了一条道路。这种‘传递善意’的方式，让我不断反思自己有幸获得的不仅来自他，还有所有导师和支持者的帮助。

And his kindness, his compassion, his generosity, and also his integrity. He was a tough teacher. And created a helped me to see a future, helped me to pave a path. And that, you know, is a way of paying forward. And I over and over again, I think about what I have had the privilege to benefit from all the support of not only him, but my mentors and my supporters.

而我想将这份善意传递下去。这就是我正在从事这项事业的原因。我认为这对每个人都是件极有意义的事。

And I wanna pay forward. This is why I'm doing what I'm doing. And I think it's just a great thing to do for for everybody.

我觉得这是个完美的结束语——你带来的这份爱与激情。李飞飞，我个人想感谢你成为AI社区的北极星，感谢你为引领我们朝向这个方向所付出的所有精力与努力。谢谢您，李飞飞博士。请大家和我一起向她致谢。

I think that's a great note to end on, which is love and the the the passion that you bring to this. Fei Fei, I wanna personally thank you for being such a north star to the community around AI and all the energy and effort you do to keep us headed towards that north star. So thank you, Doctor. Fei Fei Li. Please join me in thanking her.

李飞飞是《我能看见的世界：AI黎明时期的好奇、探索与发现》的作者。我鼓励大家在会场外或当地书店购买李博士的著作。若您想支持俱乐部实现线上线下活动规划，请访问我们的网站www.commonwealthclub.org。我是DJ·帕蒂尔。

Fei Fei is the author of Worlds I Can See, Curiosity, Exploration, and Discovery at the Dawn AI. I encourage you to pick up a copy of Doctor. Lee's book here outside or at your local bookstore. And if you'd like to support the club's efforts in making a virtual and in person programming possible, please visit our website at www.commonwealthclub.org. I'm DJ Patil.

感谢各位到场，我们下次再见。谢谢你，飞飞。

Thank you for being here, and we'll see you next time. Thank you, Fei Fei.

谢谢你DJ，谢谢大家。

Thank you, DJ. Thank you, everyone.

加利福尼亚。您可在Apple Podcasts、Google Play和Stitcher收听我们数千期播客节目。若喜欢我们的内容，请考虑支持我们的工作，帮助我们每年为听众带来500场节目。请访问commonwealthclub.org/donate。通过我们的旅行项目，用思维环游国内外精彩目的地。

California. You've Hear thousands of our podcasts on Apple Podcasts, Google Play, and Stitcher. If you like what you've heard, please consider supporting our work and help us bring 500 programs a year to listeners like you. Go to commonwealthcommonwealthclub.org slash donate. Think your way around the world with our travel programs to exciting domestic and international destinations.

当您身处湾区时，欢迎亲临我们的现场活动。感谢您的收听与支持。

And when you're in the Bay Area, please join us live at our events. Thank you for listening and for your support.