237. Mistake It Till You Make It: Learn Faster and Fail Smarter

[00:00:00] Matt Abrahams: The best communication is architected for understanding. My name is Matt Abrahams, and I teach strategic communication at Stanford Graduate School of Business. Welcome to Think Fast Talk Smart, the podcast. Today I am really excited to speak with Astro Teller. Astro is a computer scientist, entrepreneur, and inventor. He serves as Captain of Moonshots at X, Alphabet's Moonshot Factory, where he oversees audacious, high impact technology projects. He has written two novels, and the nonfiction book, Sacred Cows: The Truth About Divorce and Marriage that he co-wrote with his wife. He's also a new podcaster, hosting the really fun and insightful The Moonshot Podcast. Welcome Astro. I am super excited to be chatting with you. I've been excited ever since we arranged this. Thank you for being here.

[00:00:51] Astro Teller: Thanks for having me.

[00:00:52] Matt Abrahams: Excellent. Shall we get started?

[00:00:53] Astro Teller: Yeah, let's do it.

[00:00:54] Matt Abrahams: To begin, I'd love for you to define for our audience what you mean by a moonshot.

[00:00:59] Astro Teller: Sure. So we think of this as the blueprint for moonshots, and in order for us to have something that counts as a moonshot, it has to have three components. One, there has to be a huge problem with the world that you can name and you wanna solve. Number two, there has to be some kind of radical proposed solution that, however unlikely it is, you could make that thing. This is like a science fiction sounding product or service. We could agree ahead of time if you could make it, it would resolve that huge problem with the world. And then three, there has to be some kind of breakthrough technology that gives us at least a glimmer of a hope that we could make that science fiction sounding product or service that would solve that huge problem with the world. Once you have those three things, we're not done. That means you have a moonshot story hypothesis, you have a testable hypothesis, and from there the question is how quickly can we verify that you're wrong so we can move on to the next idea? Because anything that fits those three criteria that I just named, exciting as they are, it is exactly because it is so unlikely to work that we have to be constantly pursuing the reality that each one of them is likely to be wrong.

[00:02:18] Matt Abrahams: So the goal is actually to come up with these audacious ideas and strivings and then to as quickly as possible negate them.

[00:02:26] Astro Teller: Yeah, of course, we want any particular moonshot we come up with to turn out to be a once in a generation opportunity for the world, but since most of them aren't, wanting to win, wanting to get it to be right, each time leads to sadness after sadness, and denial kicks in that slows up the efficiency of being the learning machine that we would aspire to be, verifying which of the many that we've started is in fact worth doubling down on. 

[00:02:57] Matt Abrahams: You said several things there I wanted to dive deeper into. So you see this as learning, and it's a learning machine, as you said, which leads me to wonder, is there a particular type of mindset that you try to bring about in your organization that looks at it this way? Because many people don't think of projects this way. They start a project, they want it to be successful, not let's start a project and figure out all the reasons why it won't be successful. Can you talk a little bit about that mindset?

[00:03:21] Astro Teller: Sure. Lemme give you one or two examples. I mean, the first is, if you were working at X, I would say, can we pre agree, before you've come up with something and then fallen in love with it, that if it's a one in a hundred chance of working, it has a ninety-nine percent chance of not working. Which means that if you tie your sense of self-worth to getting a yes, you're just lying to yourself. You think that you are the one percent and you're always gonna win. But this can't be Lake Wobegon, where we're all above average. So that's one way to help put some intellectual guardrails around what we're about to go do. Here's another way of doing it. Think of the last really hard thing that you and a team did. I'd ask each of your listeners to think about that.

[00:04:03] Now, imagine you lose all of the hardware, all of the software, anything except what's in your and your team's heads. How long, once you've succeeded, would it take for you to go back and rebuild the solution? Having already verified what it is. Most people say somewhere between five and twenty percent of the time that it took. And for moonshots, it's much closer to five percent than twenty percent. Let's call it ten percent on average. What are we gonna call that other ninety percent of the work? It's learning. Most of the work isn't the making of stuff, it's the learning what to make. And the more you are shooting over the horizon, which is what a moonshot is, the more of the journey is exploration, not the settling that happens afterwards.

[00:04:49] So if you agree conceptually, intellectually that we're on a learning journey, then scary as it is to be wrong, we need to focus on the moments where we learn something. And when do you learn something? You learn something when you have a model about the world, and then in some way you get some data that tells you you are wrong. You learn nothing when you're right. At best, you deepen grooves in your brain. If you are wrong, whatever else we wanna call it, that feels bad. That's a failure of a kind. If you hate failure, you will emotionally avoid that moment, which means you're emotionally avoiding learning. If we just agreed we're gonna spend ninety plus percent of our time learning, then we need to destigmatize failure so we can have our learning loops be as tight as possible.

[00:05:36] Matt Abrahams: So a moonshot mindset is really about seeing learning as the goal. How do you inculcate and support that, and how do you produce anything if, for many companies, productivity is what's the goal? So how do you produce the amazing things that you've done and help people see learning as the goal, not productivity.

[00:05:56] Astro Teller: It's very hard to prevent people from chasing progress. Even at X, there's a decent amount of it, even though I'm constantly going around trying to stamp it out and redirect people towards learning. So it's not hard for the end result of a lot of learning to turn into something very valuable that looks ultimately like great progress. During the journey, the thing that is hardest to fight is you feel, especially, once you've done some learning and you've found something that works, it's so hard for you as an X'er not to feel like I found it. I'm right. Now, just let me build it and then I'll give it to the world and we're gonna be great.

[00:06:35] And I just have to say, we, the leadership at X, have to say over and over again, I know it feels like you found it and you know more about the teleporter, if that's what you're working on, than I do. But I'm telling you from induction, having watched this hundreds of times, now you are not correct yet. You have a lot to learn. And what you think is the answer isn't the answer. It's not that I don't want you to be right, it's just my experience tells me you have a lot more to learn, so I need you to stay in learning mode. Keep your humility and your curiosity really high for another two or three years. It's helping people to do that that's really hard. 

[00:07:11] Matt Abrahams: I find it really interesting that humility and curiosity are what keep the flames going for learning. How do you capture the learning across the different teams? We had Amy Edmondson on, and she talks a lot about the right kind of failure, and learning from the failure, and the processes you can put in place. What are some of the best practices you've implemented? So if somebody learns something, that learning is cascaded so others can benefit from that learning. 

[00:07:35] Astro Teller: There's a bunch of different ways. The truth is, some of it is just institutional knowledge, and that's sort of a background radiation that's very real at X. There are lots of things that we capture, we actually have a document called headwinds and tailwinds to remind us of some of these learnings. These kinds of things tend to be tailwinds in our experience. These kinds of things tend to be headwinds. That doesn't mean you can't ignore one of them if you want to or if you think it's worth it. But these are good reminders from the past that we've learned. We have maybe twenty percent of the people who work at X aren't on one of the projects.

[00:08:13] They're in these central teams. So this is like finance, legal, public policy, some business development, a lot of sort of hardware prototyping. And what happens is they go over here and they help the teleporter team, and then they go over here and they help the time machine team. And so they act as a kind of vector transplanting interesting ideas and learnings from one team to another. So even if those teams don't talk to each other very much, because they share these central teams, the central teams can move good ideas back and forth.

[00:08:46] Matt Abrahams: So you actually have a structural element where there are people spoke and hub and the those folks in the hub bring those learnings across. I find that really helpful for other people to think about in their organizations, how they might do that. 

[00:08:57] Astro Teller: Yeah, it is hub and spoke from a help perspective, but it is not, as the concept of the visualization of hub and spoke might imply, a highly top down process. Discovery is mostly a bottoms up process. It's very hard to dictate, so you have to create some structure and guardrails, but then let people be pretty free range within those guardrails, or they won't ultimately find things that are unexpected.

[00:09:25] Matt Abrahams: So you have to have just enough boundary setting to keep things moving forward, but not so much that people can't be creative. And we've seen that a lot. We've talked a lot about improvisation and how that mindset helps. And in improv, they have some rules and that's what allows for that creativity.

[00:09:40] Astro Teller: Right. Let me give you an example. Obviously we have to have a performance management system. It'd be weird if we didn't. The temptation across the whole world, so including at the Moonshot Factory, is to have our performance management system be one that rewards people for outcomes. You rang the bell. You got the million dollars or the big deal you signed. Good for you. Bonus promotion, whatever that is. That is the death of radical innovation. Because if I'm asking you to do things that have a one in a hundred chance of success, and then you get rewarded on the basis of a yes of a success, you'll very quickly learn to pretend you're doing radical innovation, but find ways to get the safety of like, okay, maybe it's not as radical, maybe it's not as innovative, but I'll dress it up so it looks like that, and I'm pretty sure I can make this work even if it's not that exciting. And so I'm gonna focus on this thing because I know I'm only gonna get rewarded when I get a yes. All humans will do that. And so the performance management system we have directs people back towards the habits, not towards the outcomes. What are the kinds of things like humility and curiosity that tend to have wild success as a long-term side effect? And then you have to trust, and this is a big scary trust fall, that those outcomes will happen when you focus on the habits.

[00:11:06] Matt Abrahams: And you have good evidence that will happen. It's harder from a management point of view to measure those kind of things versus did you ring the bell or not? That's a very binary, easy thing to do, so you have to have more flexibility and openness as managers, I guess as well.

[00:11:20] Astro Teller: Yeah, and this is a struggle to this day at X. It is so much harder to train managers to hold people accountable to habits rather than to outcomes that they routinely will try to find ways to hack the systems so that they can reward people for outcomes, or they'll gripe about their performance management system. I get it, and that's an example where we're trying to create some guardrails. Let them use some creativity within those guardrails, but also make sure that they're pointed in this case towards the habits rather than the outcomes. 

[00:11:53] Matt Abrahams: I really like this idea of rewarding people for habits, not for outcomes. I want to take a step back. We haven't actually heard from you an example or two of what you guys have created. Can you share one moonshot that was successful so we can have an appreciation of the different types of things you work on? 

[00:12:08] Astro Teller: Let me give a few examples. Google Brain, which is one of the places that caused the explosion of industrialized machine learning that is now the sort of hot topic in the world, that came from X. That's an example. Uh, the self-driving cars now called Waymo. Those came from X. Uh, Wing, the drones for package delivery, which are actually doing almost as well as Waymo and people haven't caught on that it's going to be as big a deal, that came from X. And I will give some more examples, but let me use Waymo as an example. So this is now fifteen years ago. Huge problem with the world. More than a million people a year die in car accidents in the world because of human error.

[00:12:52] More than a trillion dollars is wasted between sitting in traffic and all of the costs of the accidents, even leaving death aside. Between a million lives and a trillion dollars, that's a problem worth fixing. Radical proposed solution. What if, just like we got to the place with elevators where we realized you didn't have to have a human driving the elevator. You could just trust the elevator to drive itself. We could get behind the idea that if they could drive themselves, we could get to the place where these metal boxes, like an elevator, take the passengers where they need to go, and you just push a button and say, here's where I want to go. And it takes you there much safer.

[00:13:37] And there's all kinds of benefits, including, maybe particularly, the lives that are being saved because those metal boxes can now take themselves there much, much safer than a human driving the car. So that's the radical proposed solution. And then there was a set of, at the time, new technologies, how to coordinate lidar, radar, and cameras around a vehicle. There had been some evidence, before we started on what is now called Waymo, from the DARPA Grand Challenges. The first of which in 2006 was one by Stanford here. That car was called stanley.

[00:14:20] Matt Abrahams: And DARPA, just so people understand, is something that the US Federal Government, it's a funding source for creative ideas that can be used for defense and other things.

[00:14:28] Astro Teller: Exactly. And so there was this worldwide, or at least countrywide, grand challenge that was announced about cars that could drive themselves. And while even in the second time it was run, there were three groups that completed the hundred and fifty miles, they were out in a desert. It was one one thousandth the difficulty of what Waymo currently has to do, but it was enough evidence that maybe it just, maybe it was time. 

[00:14:52] Matt Abrahams: I love that example and many people listening perhaps have not seen a Waymo or been in one, I have, and where I live is where a lot of them were tested. So I'm a little more used to it. It is freaky to be in the back of a car that has no driver, but it is absolutely cool as well. I want to come to something that you just did. You just told us, not just about Waymo, but you told us a story. How important is storytelling in defining and realizing moonshots?

[00:15:20] Astro Teller: I think storytelling is phenomenally important. But I watch people at the Moonshot Factory get a little bit confused about this. At least when I say storytelling, I don't primarily mean marketing or just getting someone excited. I mean, if you worked at X and you're proposing that we start a moonshot for teleportation, let's say, I need from you an architecture of understanding. Where are we trying to go? What makes us think we might be able to get from where we are to there? And how should we interpret things along the way? What's that set of lily pads that you can envision? You'll be wrong. I'm fine with the fact that you're wrong, but if you can't paint that picture so that we have a hypothesis to test, it's very hard to get behind the idea that the upside of getting there will be worth the risks if you can't even paint an idealized picture of how we would get there in a controlled way, where the costs aren't out of control, where nobody gets hurt, et cetera.

[00:16:22] Matt Abrahams: The idea of an architecture of understanding is what a story provides. And the outcome of the story for you that's important is a hypothesis that can be tested, and that's a really unique way, I think, of looking at storytelling. So you've added podcast host to your list of titles and accomplishments. What do you think about the hosting gig and what motivated you to start the show? I listen and I love it. I've learned a lot. It's really cool for me to be in person with you after having listened to you in my headset. What brought it about? 

[00:16:50] Astro Teller: First, if people are interested, they can just look up Moonshot Factory, Moonshot Podcast. We were turning fifteen years old and at least one of the things was, you know, we're trying to get a little bit less secret and we wanted to start exposing some of how we work and what we've done, what we've learned. Partly because we want to empower other people to take their own moonshots and giving them that learning opportunity, hopefully gives them a boost. Rather than me, just blah, blahing about moonshot taking, what we've done is we've sliced it up into about fifty minute episodes, ten of them in the first season, where we've given people who've come through The Moonshot Factory a chance to tell their stories, what they've learned.

[00:17:36] So you get to hear in the first person what it was like to be taking those moonshots. And these are some people who are now luminaries in their field, Sebastian Thrun, or Andrew Ng, or Jeff Dean, and others. And how did we get to something like Google Brain? How did we get to something like Waymo, the self-driving cars? And when you hear from them the mistakes they made or those initial things that gave them the faith to start, that, that seed crystal that they thought was worth building on, I hope it then helps other people, not only understand us and the moonshots we take, but inspire them to go do some of their own. 

[00:18:13] Matt Abrahams: It's very inspirational and it's really cool to hear in their own words what they've done. And there are learnings that come. I have listened and thought differently about ways I team with people and the ways that I think about the decisions I need to make. So I'm excited that you're doing it and I'm excited to listen to season two. Before we end, I ask all my guests three questions. One I create just for you, and then the other two are similar. Are you up for that?

[00:18:37] Astro Teller: Sure.

[00:18:38] Matt Abrahams: I'd be very curious, what's a current moonshot you're working on that has you really excited?

[00:18:42] Astro Teller: We started this about eight years ago. This is very typical of the moonshots we take. They seem crazy when we start them, and then at least occasionally, much later, they turn out to be in the right place at the right time, is a moonshot for the electric grid. It turns out that everyone would like to make software that helps the electric grid get better, including us. And you can't. You can't until somebody has a circuit diagram for the grid itself. If you don't have a digital twin for the grid where every wire is, where every inverter is, where every transformer is, no software on top of that will help you manage what you have or plan for the future or optimize it in real time.

[00:19:26] And so what our team, Tapestry, did over the last seven years was build out the tools for taking in a wide variety of data, including the very noisy and imperfect data that grid operators have about their own grid, 'cause they don't have a map of their own grid. And then using lots of different sources, street view cars, drone data, satellite data, lots of different inputs, we can then use induction and deduction to do the detective work and back out the exact grid that a particular grid operator has. And then we can help them manage the grid they have, plan for the future of their grid, and optimize their grid in real time. 

[00:20:11] Matt Abrahams: So it's as if you've decoded the genome for the grid and now we can do some work with it. That sounds very useful and hopefully very beneficial. Question number two, who is a communicator that you admire and why?

[00:20:22] Astro Teller: I enjoy giving public talks, and so oration has always been very interesting to me. And I have watched a wide variety of public speakers, not because I'm going to be any one of them, but to understand how each of their patters functions. You know, Martin Luther King Jr., great speaker, his preacher style, there are things you can learn from that. James Baldwin, amazing speaker, not a preacher, somewhat professorial, but with a calm angriness because of the specifics of the life that he led. Barack Obama very professorial. My grandfather, Edward Teller, actually, was another great public speaker, and I learned a lot from watching each of these people.

[00:21:10] Matt Abrahams: Yeah, they each bring a really interesting approach that you can then synthesize, and by the way, your TED talk is one that everybody should listen to because you deliver it very well and it's very exciting to learn from. Final question, what are the first three ingredients that go into a successful communication recipe?

[00:21:26] Astro Teller: I personally don't love the start with a personal story. I would encourage people to get real, as in honest, get specific, and you can be human. So when we tell a story, like in our podcast, we use visuals from the field, from the actual people who were doing it. We explain what we were trying to do and how where we actually ended up wasn't that, getting real. We show people how we harvest value, even from being wrong, and can have fun in the process of learning. And when you get a cycle of those three things, while that doesn't feel like Mary Jane in Idaho start of the sort of New York Times article, it is deeply human. You can feel what it feels like for these people to go through it. It's exciting, it's meaningful. You can see yourself in it. It's memorable because of those things. So that's how we try to do it. 

[00:22:27] Matt Abrahams: Be real, be specific, be human, and all of that allows you to connect through emotion. I appreciate that recipe, and I appreciate this conversation, Astro. This was fantastic. You've talked about architecting to understanding, and you certainly did that for us, and I hope all of us can build these habits and reward ourselves for these habits and not just outcomes. Thank you for your time.

[00:22:46] Astro Teller: Thank you. That was a lot of fun.

[00:22:50] Matt Abrahams: Thank you for joining us for another episode of Think Fast Talk Smart, the podcast. To learn more about creativity and innovation, please listen to episode 70 and 20 with Jeremy Utley and Tina Seelig. This episode was produced by Katherine Reed, Ryan Campos, and me, Matt Abrahams. Our music is by Floyd Wonder. With special thanks to Podium Podcast Company. Please find us on YouTube and wherever you get your podcasts. Be sure to subscribe and rate us. Also follow us on LinkedIn, TikTok ,and Instagram. And check out fastersmarter.io for deep dive videos, English language learning content, and our newsletter. Please consider our premium offering for extended Deep Thinks episodes, AMAs, Ask Matt Anythings, and much more at fastersmarter.io/premium.