It's great to be here in the HP garage. Some say the birthplace of Silicon Valley. Your birthplace is Israel and at the age of 12 your dad, a botanist, took a sabbatical to this very area and you were given a Radio Shack computer. Talk about that moment. What was this child prodigy thinking? How did it shape you into who you are today?
Well, I was not given a computer because people did not have personal computers at the time, but the school I was in, which was Palo Alto High School, which is like literally around the corner from here, had a computer lab where two people time-shared a TRS-80 computer.
Well, the kind of the pyramid and the cathode ray mono screen, green or amber?
Green. It was, yeah, it was...
Do you have an audio cassette to program it or...
No, no, I... yeah, yeah, a desk and it was, yeah, it was very early in the day and it was my first exposure to computers and it was, you know, pretty amazing. It was just as punched cards were coming off and people were actually able to program interactively and we learned to program in BASIC and, you know, there was this magic box that did what I said and, you know, we got to write cool things like little games and such and even back then I saw the potential.
So, you go back to Israel. You graduate from Hebrew University at 17. You get your PhD at 21. You're like a child prodigy that could grow up and be an evil character in a James Bond movie.
I didn't get my PhD at 21. I got my master's at 18 and then I did my military service where I was actually an intelligence analyst which was really cool because it was my first exposure to like Bayesian reasoning. You got to have all these little bits of information, many of which were irrelevant, many of which were missing. You had to put a puzzle together. And then...
It's like the game of Clue, you know, it doesn't matter.
Much, much harder than the game of Clue because so much irrelevant information and so many missing pieces and you had to kind of create a picture of what was going on. And so, I think that together with my early experience of computers really led me down the path of can we use computers to model the complexity of the real world, which subsequently was what turned into my career.
Yeah. So, where I was going with that was you were ahead of your peers in terms of academic accomplishments, even though you had 2 years in the military.
2 and 1/2 years. Even though you had that period, it sounds like that period also helped empower you, embolden you, and gave you some perspective. Um, what was that moment like where you made certain decisions to become the person you are because you could have gone different ways. Like I said, James Bond, you know, uh, bad guy. But, uh, you know, you use your brains for good. So, what was it like at that juncture?
So, you know, it was actually an evolving journey, although bad was never part of the option space. Uh, but initially I think I was enamored by the intellectual beauty of mathematical models of the real world. And it was really more about that that I was focused. So, like the beauty of taking something that was complicated and vague and really building something that was mathematical and predictive and allowed you to make smart decisions. So, that's where I started. It was only through the course of multiple steps in my journey, and I'll tell you about a pivotal moment in just a second, that I became excited about using it to do good in the world. And so, my journey started out with, um, really just elegance and then, um, at various steps along the way, it was kind of like, well, elegance is great, but does it help anybody? Does it help actual people? And do I perceive, can I actually viscerally feel the impact that it's having on people's lives? So, one of the first of those junctures was actually when I began my postdoc at UC Berkeley following my PhD. And I'd gotten, you know, thesis award and so on, and my postdoc advisor, Stuart Russell, took me to lunch right after I started and said, you know, you had this great thesis, and if I give you three really talented undergrads who could help you program anything from your thesis that you want, then what would you have them work on? And I think my jaw literally dropped because no one had ever asked me that question. And the truth was nothing. It was a work of mathematical elegance, but it didn't have any real-world applications. And I think that was a transformative moment that really started me on the journey of asking that question more and more pointedly over time. So, I started to work on, you know, machine learning, and then machine learning in service of computer vision, and then machine learning and AI in service of biomedical data. Um, it was the turning point that eventually took me to the Coursera detour because of the opportunity to have so much impact in the world in such a short amount of time. And it was also that North Star that eventually brought me back to using AI for the purpose of human health, which is ultimately what landed me in what I'm doing today.
So, before Coursera, you get a call from David Sakolow saying, 'Hi, you're a MacArthur genius fellow.' One, what did they see that they celebrated in you? And how did that change your life? You know, people understand that genius award, but I don't know how many people kind of dissect what does that do to someone? I mean, you already had all these achievements. You were ahead of the class and other things. But what was that moment like for you?
So, honestly, it was a bit terrifying. So, yes, you're right. I'd received a whole bunch of academic awards and accolades, but those were always among what seemed like a relatively small group of qualified individuals, people who were, whatever, had gotten this Edge of Life Computers and Thought Award, which was for, you know, AI scientists under the age of 35. And yeah, I was not the only AI scientist under the age of 35, but it was a relatively small set. This was in the world for which any resident of the United States was eligible to receive that award. And I was like, why me? What have I done to deserve being selected among 250 million people, which is what it was at the time. And...
We're now at over 330, but back then that's how good she is. She knows.
And so, I felt like I hadn't actually earned that recognition. And I spent much of my career following that trying to kind of earn it retroactively, which I think was another pivotal moment on my journey to do something that is truly meaningful. And I actually told Daniel Sokolow afterwards that the Coursera decision, my decision to kind of put my career on hold and go found a company that had, frankly, no substantive technology and certainly no scientific component.
Wasn't essentially your life science work.
No, and it's not even of my AI and machine learning work. It was basically a media company.
Yeah, you're filling a void that you thought needed to be filled.
Absolutely, but it was 100% about impact and not at all about scientific curiosity or conceptual elegance. And it was partly because of that MacArthur Award.
Yeah. So, the, I don't know, 500,000 or whatever they gave you. Did you put it under your mattress? What did you do with that money?
I pretty much put it under my mattress. Yeah.
You didn't gamble it. You put it all on number six or horse.
I, you know, I've never been much of a gambler except I guess when it comes to life decisions like going out and founding Coursera. Uh, when I the first time I went to Vegas, I think I allocated myself $2 in nickels and when it ran out I stopped my gambling spree. So...
So I want to come back to Coursera in a moment but before we get there, 5 years ago a lot of the best AI minds were saying what's happening now was 90 years away. But they thought, you know, Stuart Russell here, my children's children will be dealing with what we're dealing with right now. 5 years ago, did you see it that way? Or did you think the world was moving pretty fast and large language models and other things were kicking in and the person that got the, you know, that was a mentee of Stuart Russell and got the MacArthur. Did you think that we were on this trajectory?
Um, yes and no. And so I did for example, when I made the decision to go back to AI in service of biology and medicine, it was because I went to NeurIPS, the big AI conference, for the very first time in a number of years and, um, if you think back on the Coursera trajectory, I left Stanford to start Coursera in late 2011 which is just before the AI revolution really started with ImageNet and AlexNet and so on and so forth. So given that I was like heads down in the trenches building a company which was a very, um, disconcerting experience for me because I'd never even been at a company, I'd never started a company and I was trying to do that from the ground up. I wasn't really paying attention. And so when I went to that NeurIPS conference for the first time for a number of years and I was like, whoa, we are on an exponential curve. And so I recognized that exponential curve. I saw that for example in the numbers having to do with the ability to understand images, where when I left Stanford, we were doing slightly better than random at saying there is a bear in this image or not. And then, at that point, we were suddenly starting to write entire sentences about images. And so, that exponential curve was clear to me, but I think exponential curves are really misleading things to people. And if you get the base of the exponent just a little bit wrong, things could take...
You could go to Alpha Centauri.
Or you could take 20 years. And so, I knew we were on the exponential curve, but I did not predict large language models within whatever this was 2016, within less than a decade.
So, talk to me about Coursera. You know, you're an academic. You, you know, Stanford's the pinnacle. Was it a statement to say I want to democratize education? Or did you and Andrew Ng say, you know, maybe there's another way? And when you started it, did your peers back at the farm, Stanford, say, what are you doing? Did they feel threatened? Or they was like, oh, that's a nice project. Good luck. Um, and how did that play out?
I mean, so, first of all, from a personal level, this was, like I said, um, like jumping off a cliff where you have no idea where the bottom is.
More like in Bourne Identity, when Jason Bourne jumps out of a window and jumps through another window in another building. You were doing something like that.
Yeah, except that you don't know where the bottom is, and there may not be a bottom, because it was kind of like, okay, I'm leaving Stanford behind, I'm going to do this thing where I had no idea how to build a company. Uh, it was definitely driven by what I saw was, as an incredible opportunity to have massive impact, which I think we have indeed had. I mean, the company has over 150 million learners worldwide, and, you know, I don't know where this...
I know you didn't say learners, not educated people, because we're lifelong learners.
And you're almost saying making a statement that people should not stop at high school. They should keep learning forever.
And a student is, by association, someone who's in an academic institution. And the vast majority of our learners were not that. And so, and over 40% were in developing countries and emerging economies where, frankly, many of them had no access to the kind of education that we were just like handing out to Stanford students. And I felt like there was just this incredible opportunity to transform the lives of people. And so I felt that as almost like a moral imperative to not just let this lie on the shelf. And this actually dates back to some of the work that I'd done at Stanford on AI for biology and medicine where some of my papers that even today can be considered to be the most impactful, which, for example, really were the first AI-enabled digital pathology for cancer patients. And we tried really hard to get people to use the technology and really make an impact to patients. And it's very hard to build a product within an academic environment. You can write papers, but people don't take papers and turn them into actual products. And so I really felt like if I was going to make an impact I needed to roll up my sleeves and do this myself. I couldn't count on just putting it out there and assuming someone will pick up the torch and run with it.
So there were these things called MOOCs. And the dirty little secret, you look under the hood, people weren't staying around till the end. Did that inform Coursera how it built its product and how it engaged with its customers? And how were you better than that average? Or were you in the same area?
So honestly, uh, so we were MOOCs and yes, those statistics applied to us in the same way that they did to the other MOOCs. I think the statistics were frankly a little bit misinterpreted because a lot of the people who came to take the MOOCs weren't actually planning to finish them. They viewed it akin to picking up a non-fiction library book and reading a few chapters and walking away feeling like they'd learned something, and they felt like they'd achieved what they wanted to achieve. Having said that, even among the people whose completion rates, who came in with the declared intent to complete, our completion rate was, you know, depending on the course between 20 and 50% and that was lower than we would have liked. It did certainly shape a little bit of our product strategy. For example, uh, you know, it turned out the people who paid, even if it was a modest amount of money, felt like there was skin in the game, and so the completion rates were much higher. So, that was one of the rationales for the decision to charge for the certificate on those courses and actually get them the certificate, which was also kind of a reward at the end. Um, but at the end of the day, I think a lot of learning is still an interactive social experience, and we had a dilemma back then. We could either offer the courses less frequently and have more of a community because more people were taking them simultaneously and there was like a level of engagement between them, or offer them more, but then people would have to wait and then they would drop out before they ever made it into the class. And we kind of debated what is the right frequency. We made a decision not to invest in unscalable technologies like TAs and that would require an operational load and more of like an engagement among learners as part of it because it was simpler and more scalable. Uh, the next company that I founded in the education space, which is a company called Engageli, is actually in some ways the other half of what the Coursera experience should have been, which is the interactive experience with a much deeper engagement between learners and between learners and the instructor.
Yeah. So, the planet has like 195 countries, plus or minus. When you looked at the data, um, you know, the national support in India is really engineering. Who were taking your MOOCs and what countries were over indexing? And did that inform you on how to, you know, distribute it? Or were you just focused on the content and making sure it was a good experience?
