0:00:02 >> Hi, this is Frank Chen.
0:00:04 Welcome to the A16Z podcast.
0:00:08 This episode is part two of a series called
0:00:10 “What’s Next for Education Startups?”
0:00:12 It originally aired as a YouTube video,
0:00:19 and you can watch all of our YouTube videos at youtube.com/a16zvideos.
0:00:21 >> Hi, this is Frank Chen.
0:00:23 Welcome to the A16Z network.
0:00:26 I’m very excited today to share a conversation I had with
0:00:29 Connie Chan, one of our general partners.
0:00:33 Connie is one of the world’s experts on trends,
0:00:36 especially consumer trends in China and tech.
0:00:40 Today, we’re going to talk about the future of lifelong learning,
0:00:46 and she’s going to share a few examples of very awesome startups in China.
0:00:50 She’s super interested in what’s happening with Gen Z consumers.
0:00:56 She’s very interested in real estate and how people are finding homes,
0:01:01 preparing their homes to be listed on Airbnb, renting homes, so on and so forth.
0:01:06 She’s also very inspired by things that entrepreneurs are doing in China
0:01:09 that might have applicability here in the United States.
0:01:15 She helped us find our investments in line and Pinterest,
0:01:20 and I think you’ll really enjoy this conversation that I had with Connie.
0:01:23 I have to tell you a funny story before we get started,
0:01:26 so we did not synchronize our sweaters.
0:01:31 We’ve known each other so long that we just knew to come in the same color family.
0:01:34 So Connie was my first hire at Andreessen Horowitz.
0:01:37 The Adam Rifkin introduced us to Adam at one time,
0:01:40 and may still be the most connected person on LinkedIn,
0:01:43 and his whole heart and mission is to connect people,
0:01:45 and so when I told Adam,
0:01:48 I was looking for the best ideal partner ever.
0:01:49 He went and found Connie,
0:01:53 and I’m so thrilled that you’ve been here for so long,
0:01:57 and now are a general partner looking to make investments, so welcome.
0:01:58 Thank you.
0:02:04 So today, we’re going to continue our series in education
0:02:06 and talk a little bit about ongoing education,
0:02:10 and we’re so excited about the things that we can do as adults
0:02:12 to continue to learn new things,
0:02:13 and for those of you that know me,
0:02:16 like learning a new thing is my favorite thing in life,
0:02:18 so I’m so excited about this episode.
0:02:20 So Connie, why don’t you set the context,
0:02:23 and let’s talk a little bit about the things that are working,
0:02:25 especially in China,
0:02:29 and I thought maybe it’d be good to just anchor on how much money
0:02:32 and how many users people spend on ongoing adult education,
0:02:34 because this is very surprising.
0:02:37 Yeah, I think about education and learning
0:02:40 in a way that goes well beyond K through 12.
0:02:44 So I’m actually hyper-focused on education for adults, people.
0:02:46 Once they’ve graduated college,
0:02:51 how can they use online education for self-improvement, for example?
0:02:53 And if you look at the dollars abroad,
0:02:57 I do a lot of studying what’s working in China and working in Asia
0:03:00 to give me inspiration for ideas here in the States.
0:03:02 It’s a massive market in Asia.
0:03:03 It’s massive in China,
0:03:06 and I think it’s because China has developed
0:03:08 all these online education platforms
0:03:10 that are specifically made for mobile
0:03:13 that unlock all these other new features and benefits.
0:03:16 And in terms of how big it is,
0:03:18 iResearch says that right now,
0:03:23 online education in China is 150 million users
0:03:28 and expected to grow to nearly 300 million by year 2020.
0:03:31 It’s a $40 billion industry,
0:03:33 expected to grow to $70 billion.
0:03:36 Of course, this is a very broad categorization
0:03:37 of what counts as education,
0:03:40 but what’s interesting is the way
0:03:42 that these research reports break it up.
0:03:44 The largest group is not K through 12.
0:03:47 It’s not even college students.
0:03:49 The largest groups of students
0:03:52 who want to do self-improvement in online education,
0:03:54 they are 26 to 35.
0:03:56 – Yeah, that’s super interesting.
0:03:58 So you would expect sort of the Asian cultures
0:04:01 that the parents sending their students
0:04:03 to after school enrichment programs,
0:04:05 and so you think that’s where all the money is going,
0:04:07 but you’re saying, look, it’s after they graduate college.
0:04:08 – Right, right.
0:04:11 And I think that’s because if you take the word education
0:04:15 and you expand it just to self-improvement, self-learning,
0:04:17 then it greatly increases the demographic
0:04:18 that you can address.
0:04:21 And yes, a lot of people just say China education,
0:04:25 it’s huge because parents spend so much money on tutoring
0:04:29 and so forth because of the way the college system works.
0:04:32 But most of that money is actually going through
0:04:34 post-college graduates.
0:04:37 – It’s really interesting ’cause in China,
0:04:39 what’s already happening is what we sort of expect
0:04:42 to happen here, which is today we have this system
0:04:44 where sort of you go through K through 12
0:04:45 and then a subset of these people go to college
0:04:49 and then basically at age 22, you’re done.
0:04:50 There’s no more formal education
0:04:54 and now it’s basically the workplace’s job to train you.
0:04:55 They’ll send you to classes and so on.
0:04:57 And we know that’s gonna change.
0:05:00 We know that the world is so dynamic now
0:05:02 that you can’t learn everything that you need
0:05:05 to be a productive worker or a citizen by age 22
0:05:07 and you’re gonna have to learn ongoing.
0:05:10 This is sort of a big part of our investment thesis
0:05:11 behind Udacity.
0:05:14 – And there’s a bunch of courses that a college curriculum
0:05:16 would likely never include.
0:05:18 How to conduct yourself at a meeting,
0:05:23 how to speak publicly, how to sleep train your kid.
0:05:24 That counts as education.
0:05:26 Parenting courses, that counts as education.
0:05:29 You would never cover that stuff in college.
0:05:31 – And in addition to sort of the evergreen stuff
0:05:33 that you mentioned, everybody needs to be
0:05:34 a good public speaker.
0:05:36 Everybody needs to know how to do it.
0:05:39 There’s also sort of topical things that emerge
0:05:41 as marketplaces emerge.
0:05:43 So I’m thinking about the Taobao sellers, right?
0:05:45 So Taobao was like eBay here.
0:05:49 And what happened in Taobao was there were sellers
0:05:51 who were experimenting with the system
0:05:54 and they kind of figured out what was working for them
0:05:57 and they would share online and in videos
0:05:59 and Taobao saw this happening and they’re like,
0:06:01 oh, let’s actually get behind this and push, right?
0:06:03 Let’s set up Taobao University
0:06:05 where we can take our very best sellers
0:06:08 and actually have them make money from their content,
0:06:10 not just their markets.
0:06:11 – Yeah, completely.
0:06:13 – Yeah, so awesome.
0:06:16 So why do you think this is happening already in Asia?
0:06:17 Why are they ahead?
0:06:22 – I think Asia is in general much more mobile first
0:06:26 and mobile only of an environment than the States.
0:06:29 Meaning that if I asked you to go buy a pair of shoes,
0:06:32 you might naturally flock to your computer
0:06:34 to get the best user experience.
0:06:36 But in Asia, you’d pull up your phone
0:06:39 and you’d open the T-Mall or the Taobao app.
0:06:41 And the idea that your PC and phone
0:06:44 are completely interchangeable
0:06:46 and you can completely rely on your phone
0:06:50 to give you everything you need is more prevalent in Asia.
0:06:52 There’s also more mobile payments
0:06:54 and the idea of paying on your phone
0:06:56 is very natural and common to people,
0:06:59 not just interior one cities, but across the country.
0:07:03 But I think there’s three core breakthroughs and insights
0:07:04 that Asia has really figured out
0:07:06 that has propelled its education market
0:07:08 so much more forward.
0:07:12 The first one is that they rely on artificial intelligence
0:07:17 and machine learning in a much more interesting way.
0:07:21 So that allows them to unlock products and features
0:07:24 and just ideas that I don’t see here in the States.
0:07:27 So for example, there’s this company called Lingo Champ
0:07:29 and it teaches you English.
0:07:32 And typically when you look at a language learning app
0:07:35 here in the States, it’s very flashcard driven
0:07:39 or it’ll give you a sentence and you can read it.
0:07:42 But in Asia, they realize that people want to learn English
0:07:44 not just to be able to read and write,
0:07:46 but more importantly to have conversations
0:07:48 to be able to visit the world,
0:07:51 to interact with other people.
0:07:54 And so they use the mobile phone and the microphone
0:07:57 to allow you to speak directly into the app
0:08:01 and read out sentences and actually carry on conversations
0:08:04 with a computer that will speak back to you.
0:08:07 And that kind of scoring using machine learning
0:08:09 and artificial intelligence allows people
0:08:14 to learn pronunciation with a standalone mobile app.
0:08:16 And I think that’s a fantastic example
0:08:18 of like leaning into artificial intelligence
0:08:21 and machine learning to dramatically reduce the cost.
0:08:24 This company, Lingo Champ, their gross margins
0:08:28 are over 70% because they don’t have the teacher cost.
0:08:30 – Right, so nobody has to sort of say,
0:08:32 “Oh, that’s a terrible accent.”
0:08:34 My funny story on this is when I was learning
0:08:38 Chinese Mandarin, my Mandarin teacher asked me one day,
0:08:39 “Are you from Hong Kong?”
0:08:41 Which, for those of you that don’t realize it,
0:08:43 is probably the most grievous insult
0:08:45 that you could hurl at somebody trying to learn Mandarin
0:08:46 ’cause it’s so bad.
0:08:50 So you’re saying, look, they didn’t have to have a teacher
0:08:51 listening to you and then getting guns.
0:08:53 They’re using the machine learning to say,
0:08:55 “You don’t sound like a native and here’s where.”
0:08:58 – Right, and because their gross margins are so high,
0:09:00 their price point is so much lower
0:09:01 than having a real-life tutor
0:09:05 or even an online course instructor tutor.
0:09:07 Their price point is so affordable
0:09:11 that people all around the country can access it.
0:09:15 And that same concept of leaning into machine learning
0:09:19 is also true in music as another category.
0:09:24 There’s this company in China called BIP, Peilian.
0:09:28 And Peilian in Chinese translates to
0:09:31 they will practice piano
0:09:34 or practice an instrument alongside you.
0:09:37 And what it is is a mobile app, which is a piano teacher.
0:09:40 And this app, you put it on the stand
0:09:42 and you attach it to your piano.
0:09:47 And this teacher can help your kid age five through 16
0:09:48 learn an instrument.
0:09:49 They do piano, violin,
0:09:52 a bunch of classical Chinese instruments.
0:09:55 But again, it’s that price point that they’re able to unlock
0:09:58 because for a lot of these music instructors,
0:10:01 so much of that cost is in their travel time
0:10:03 or because you’re living in a city
0:10:06 where the cost of living is just so high.
0:10:11 But now in China, my teacher doesn’t have to live in Beijing.
0:10:13 They don’t have to live in Shanghai.
0:10:15 They can live anywhere in the country.
0:10:18 They don’t even have to live in China, right?
0:10:19 And then not only is that the case,
0:10:21 they use the machine learning aspect
0:10:24 to help the teachers with scoring the kids
0:10:26 and scoring the performance.
0:10:29 Because with music, just like with language,
0:10:31 there is an actual pitch.
0:10:34 There is an actual tempo and actual rhythm
0:10:36 that you’re supposed to play, right?
0:10:39 So they can take the composition score
0:10:41 and then hear your actual performance
0:10:42 and give you a grade,
0:10:44 which then allows one teacher
0:10:48 to teach two or three students at the same time,
0:10:50 which then unlocks even more cost savings,
0:10:54 allowing more parents to give their kids these music lessons
0:10:56 that they would typically not be able to afford.
0:11:00 – I’m flashing back to my piano learning days
0:11:03 and I’m hearing that too fast, too fast, right?
0:11:06 So now we can do that with machine learning.
0:11:06 – Right, right.
0:11:10 And imagine being able to do that during practice sessions,
0:11:13 right, and having that information feedback to the teacher.
0:11:15 There’s just a lot more we can do with machine learning,
0:11:18 especially when it comes to language and music
0:11:21 that is still, I think, very untapped here in the West.
0:11:26 – So let’s talk a little bit about sort of the efforts
0:11:27 that we’ve sort of seen here
0:11:31 and sort of how you think we get from here where we are.
0:11:33 So we have learning platforms like Masterclass,
0:11:34 we have learning platforms like Udemy,
0:11:36 we have learning companies like Udacity,
0:11:38 one of our portfolio companies.
0:11:41 What’s sort of missing from those
0:11:44 that sort of the next generation of ed tech startups
0:11:46 you’re looking for, you think we’ll have?
0:11:51 – Yeah, I think the answer is one word, it’s mobile.
0:11:56 And the reason is because mobile only is in a society
0:11:59 that I think is inevitably in our future.
0:12:00 And when you have mobile,
0:12:03 that allows for all kinds of different things.
0:12:07 It allows, again, for microphone input as an example.
0:12:09 Everyone has a camera, a front-facing
0:12:11 and a back-facing camera on their phones,
0:12:13 which allows for different kinds of input
0:12:15 and interaction with the platform.
0:12:18 Mobile allows you to have these bite-sized snacks
0:12:21 rather than opening your Instagram news feed.
0:12:23 Maybe you can take a three-minute class,
0:12:27 a five-minute class whenever you have downtime.
0:12:31 And also mobile allows people to not feel like
0:12:33 you have to be confined to a video format.
0:12:36 And I think this is really critical
0:12:39 because a lot of long-tail expertise
0:12:42 doesn’t always naturally suit video.
0:12:44 For example, you can be a math teacher
0:12:47 and, yes, you’re writing formulas on the board
0:12:49 or you can be a philosophy teacher, right?
0:12:51 And you can be sitting there giving a lecture,
0:12:53 just sitting there.
0:12:56 Or that same kind of content
0:12:58 can be also conveyed through a podcast,
0:13:00 through an audio format.
0:13:02 And once you’re focused on mobile,
0:13:05 you’re not thinking like it has to be video,
0:13:07 it has to be full-screen immersive.
0:13:10 It now can also be a podcast that you listen to
0:13:13 when you’re driving to work, when you’re walking to work.
0:13:18 And again, I think that that expansion of formats
0:13:21 is really obvious once you make something
0:13:23 that is mobile-centric.
0:13:26 – So we haven’t seen the class of mobile-first edtech
0:13:27 that you were expecting to see,
0:13:29 which is pretty surprising, right?
0:13:32 It is sort of an obvious insight
0:13:34 once you say it out loud like you did.
0:13:40 – And I think the reason is because so much of edtech
0:13:43 has been either you pay this one-time
0:13:44 very expensive tuition,
0:13:48 or honestly it’s ad-based, right?
0:13:52 Like YouTube is the biggest university in the world.
0:13:55 And most of the creators are monetizing
0:14:00 through advertisements, but because it’s ad-based,
0:14:02 a lot of the content on YouTube
0:14:04 can’t go to the depth of absurdities
0:14:06 that you need to really make a big impact
0:14:08 on your life or your career.
0:14:11 Because the creators, they have these incentives
0:14:14 to have to create content that gets lots of clicks.
0:14:17 And the reality is a lot of self-improvement
0:14:21 lifelong learning content is not all clickbait content.
0:14:25 And to go into that depth of what you need to know,
0:14:29 an ad format is not the best way to compensate
0:14:30 these creators.
0:14:31 So for example, if you’re buying a house
0:14:34 for the first time, you need to understand
0:14:36 how to think through that transaction.
0:14:38 But it doesn’t make sense for someone
0:14:40 to create these ad-based videos
0:14:42 because one, they’re not gonna get all the clicks
0:14:45 they need to justify their time and expertise.
0:14:48 But I mean, imagine a platform where someone
0:14:50 could package that in 20, 30 courses.
0:14:53 It could be a mixture of audio, a PDF,
0:14:58 video, a live stream Q&A, a paid one-on-one consultation
0:15:00 and put that all in one format
0:15:03 where that creator now can make much more money
0:15:06 and have the right incentives to create deeper,
0:15:07 better content.
0:15:09 – Yeah, so that makes perfect sense, right?
0:15:12 Which is it takes a lot of work to create this content.
0:15:14 And if you’re monetizing with advertising,
0:15:17 that means only the top 1% are gonna even break even
0:15:20 or barely break even on all of that effort, right?
0:15:22 ‘Cause you need to attract tens of millions of people.
0:15:26 – And ads reward production value, right?
0:15:28 So you need the great videographer there.
0:15:31 You need to spend an hour on your YouTube thumbnail.
0:15:33 And that’s nuts, right?
0:15:35 Because honestly, a lot of these great experts,
0:15:37 a lot of these professors, these doctors,
0:15:42 these nutritionists, they are not media experts.
0:15:44 And the fact that they have to go hire videographers
0:15:47 buy very expensive equipment, cameras, lighting,
0:15:50 what have you, learn how to edit videos themselves
0:15:54 for the first time, that’s not long-term,
0:15:55 I think, going to work.
0:15:57 Because these creators are being underpaid
0:15:58 for their knowledge.
0:16:03 So, as I think about my own sort of ongoing education
0:16:06 habits, YouTube has definitely become one of them.
0:16:08 Which is to say, I’m watching TED Talks,
0:16:09 I did something over the holidays,
0:16:13 which I’m very proud of, which is I replaced a doorknob.
0:16:15 And I’m proud of this because I’m the least handy person
0:16:17 I know, and so I watched a YouTube video
0:16:20 and went to Home Depot, and in Law’s house,
0:16:21 I replaced the little doorknob mechanism.
0:16:24 I was like, yes, I did it.
0:16:27 And so you were saying, look, I know,
0:16:28 I shouldn’t be that proud of myself.
0:16:30 But I was like, giddily proud of myself,
0:16:31 because I’m a software person,
0:16:33 and that was definitely hardware.
0:16:37 So anyway, thank you for indulging my burst
0:16:39 of enthusiasm for myself there.
0:16:41 So you’re saying, look, that type of content,
0:16:43 that’s fine for YouTube, because that’s super easy, right?
0:16:45 It’s a fine video. – And it’s visual.
0:16:48 It’s visual, you need to see which part to change out,
0:16:49 which nail to take out.
0:16:51 – So ads, for that makes sense,
0:16:53 but it doesn’t make sense for this sort of
0:16:56 highly produced package where I’m teaching you something
0:16:57 that’s a more serious life skill.
0:17:01 – Right, I mean, TED Talks are fantastic intro courses.
0:17:03 It’s a first great lecture,
0:17:05 but there should be 10 lectures
0:17:07 beyond that for every topic, right?
0:17:11 And a lot of things that are skill-based, in particular,
0:17:15 I think deserve having 10 courses,
0:17:17 20 courses, 30 courses, so on.
0:17:18 And there’s a lot of things
0:17:20 that I would be willing to pay for.
0:17:22 I would love to pay to figure out
0:17:23 how can I improve my voice.
0:17:25 I would love to pay to see
0:17:27 how can I improve parenting and so forth.
0:17:30 And there aren’t great platforms right now
0:17:34 that make it as easy as creating like a Shopify website
0:17:37 for these creators to monetize their knowledge.
0:17:39 And these creators, typically,
0:17:41 one, they’re not media experts,
0:17:43 two, they’re not technologists,
0:17:46 so they don’t have time to build their own blogs
0:17:48 or their own websites and integrate PayPal
0:17:50 or credit card payments into them.
0:17:54 And the biggest problem is they’re truly underpaid right now
0:17:57 for the knowledge that they’re freely sharing on YouTube.
0:17:58 – Yeah.
0:18:00 If you think about sort of an example
0:18:02 that is in this ecosystem,
0:18:03 you think about masterclass, right?
0:18:06 Where the entrepreneur’s doing a great job
0:18:09 of sort of hoovering up all of the top experts
0:18:11 in their fields.
0:18:13 And I think part of the reason he went top down
0:18:15 is sort of the same reason that Elon Musk
0:18:19 went to the Roadster first and then the X
0:18:21 and then the, or the S and then the X and then the three,
0:18:23 right, sort of he’s working his way down.
0:18:26 And I think part of that is because
0:18:29 I wonder if there’s enough cultural support
0:18:34 in the West for paying for education of this kind, right?
0:18:35 So it sounds like in China,
0:18:36 you already have that cultural support.
0:18:39 So like what is education amongst household expenses?
0:18:41 Is it like number three or number four
0:18:43 after housing and medical, right?
0:18:48 So you have this inbred sort of support, cultural support.
0:18:50 Like of course I’m paying for education, right?
0:18:54 And so once mobile sort of content sources sprung up,
0:18:56 the money just went, right?
0:18:58 And so what do you think is gonna happen here?
0:18:59 Do we need more cultural support?
0:19:01 How does that interaction happen?
0:19:05 – I think the way that we Silicon Valley
0:19:08 and platforms can help encourage this shift
0:19:10 for more lifelong learning and self-improvement
0:19:14 is really breaking away from just the ad-based model
0:19:16 and finding the right incentives for creators
0:19:18 to be able to monetize.
0:19:20 Because I think a lot of creators,
0:19:23 when they have an ability to make a significant amount
0:19:26 of income from sharing their expertise,
0:19:28 they will create better content.
0:19:30 And as there’s better content out there,
0:19:33 users will say, hey, this is a fantastic way
0:19:38 to put a small investment into myself, right?
0:19:40 And right now the platforms, I think,
0:19:44 are not doing enough to help these creators monetize.
0:19:46 And for a platform that doesn’t just mean
0:19:48 changing their business model,
0:19:51 it also means monetizing their own brand
0:19:55 and becoming a mainstream app, a mainstream website.
0:19:58 And that’s really important because for a lot
0:20:02 of these platforms, they shouldn’t have just one teacher
0:20:03 teaching you how to sing.
0:20:05 There should be 20, 30 teachers.
0:20:07 And then there should be rankings based off
0:20:09 of student reviews or based off people
0:20:13 who actually completed the course, right?
0:20:14 And repeat students and so forth.
0:20:18 And all those things should help bubble up the best teacher.
0:20:21 And these platforms need to do a lot to invest
0:20:23 in building out their own brands
0:20:24 to become mainstream in order to do that.
0:20:28 And I love the master class content.
0:20:32 I think once they expand, they’re gonna have to include
0:20:33 more teachers for the same categories.
0:20:34 – They sort of go down in market.
0:20:36 It doesn’t have to be Steve Martin teaching you
0:20:37 how to do comedy.
0:20:39 It’ll be your local comedy genius.
0:20:41 – Or it could be all of them together, right?
0:20:44 And they could be priced at different price points.
0:20:47 And then when you go the level beyond Steve Martin,
0:20:50 you can have them ranked differently, right?
0:20:52 And I’d love to be able to figure out
0:20:55 what are the rankings of the classes that people finished?
0:20:56 What are the rankings where people gave
0:20:58 the highest reviews, right?
0:21:00 What are the rankings based off price?
0:21:02 What have you, right?
0:21:08 And all of that kind of data is totally presentable right now.
0:21:12 It’s just not being surfaced by the platforms.
0:21:15 – Another sort of age old challenge in sort of building
0:21:18 these pervasive education marketplaces in the past
0:21:20 has been you sort of have very broad categories
0:21:21 of education.
0:21:23 There’s sort of, let’s call it hobby entertainment, right?
0:21:24 I’m learning the piano.
0:21:26 I want to sing better, right?
0:21:28 And then there’s sort of business self-improvement.
0:21:31 Like I want to learn how to use Excel better
0:21:33 or I want to be a better offer up seller
0:21:34 or something like that.
0:21:37 So do you think that there’s gonna be one platform
0:21:39 that sort of wins both?
0:21:41 Do you think there will be more specialty things
0:21:42 that sort of cater to each of these
0:21:44 ’cause it feels like they have different dynamics?
0:21:46 – I think it’s possible.
0:21:50 But it’s unclear how the future will shake out.
0:21:51 I mean, for example, I think there’s a lot
0:21:54 of great workout apps today already.
0:21:56 That put a bunch of fitness instructors
0:21:57 or nutritionists up against each other
0:22:00 and you can choose which instructor you want.
0:22:02 And they have that category down pretty well.
0:22:05 But I also think it’s very possible
0:22:07 if there was a platform that created the right tools.
0:22:11 I know that’s like Shopify in a box where I can say,
0:22:14 here are my podcasts, here are my blog posts.
0:22:18 These are the times where I’m gonna do a live stream Q&A.
0:22:22 This is the PDF of the book I’m willing to sell, right?
0:22:24 If it gave creators these options
0:22:27 to just turn on these modules
0:22:31 and create their own knowledge store,
0:22:32 I think it’s possible also
0:22:36 to have one major platform as well.
0:22:39 It serves both sort of the hobbyist entertainment market
0:22:42 as well as the serious self-improvement market.
0:22:44 – And it’s possible that it’s not a new startup.
0:22:47 It could eventually be something that YouTube goes into
0:22:48 or something that Twitter goes into.
0:22:50 Twitter has a ton of influencers too
0:22:52 and lots of long-tail experts.
0:22:56 But I think the opportunity is still there
0:22:59 and still so early enough that a new startup could take it.
0:23:02 – Great.
0:23:06 If you were to give, if you had one or two pieces of advice
0:23:08 that you have for entrepreneurs in this space,
0:23:09 what would it be?
0:23:15 – This is probably a contrarian view
0:23:18 even in Silicon Valley, but I would build for mobile first.
0:23:21 And I would build your app before you build your website
0:23:24 because it will drastically unlock different ways of thinking.
0:23:28 You’ll be able to use your GPS, your microphone,
0:23:30 you’ll be able to use the camera.
0:23:33 And if those new additions of features
0:23:34 don’t help you brainstorm new things,
0:23:37 then that’s a problem actually.
0:23:40 You’ll be able to use in-app payments, right?
0:23:42 You might be able to use Apple Pay and so forth.
0:23:45 So I think one big thing I would say is
0:23:47 if you’re building for the future,
0:23:50 consider building this platform first on mobile
0:23:52 even before you go to the PC.
0:23:55 And now I know that’s a very contrarian view
0:23:58 ’cause a lot of investors will also say
0:24:00 go to the PC first, get your brand
0:24:02 and then go to the app.
0:24:05 But I think when you start at least brainstorming
0:24:07 at the very least on a mobile platform first,
0:24:09 it unlocks this idea,
0:24:11 how can I use a microphone differently?
0:24:15 And then now that I have microphone and audio input,
0:24:18 how can I use machine learning differently, right?
0:24:21 And that allows you to unlock ideas
0:24:23 like the lingo champ for English learning
0:24:25 or like Paylian for piano teaching
0:24:28 that honestly someone building for a PC
0:24:30 would never get to that insight.
0:24:31 Right, right.
0:24:33 And then presumably your next piece of advice
0:24:36 would be experiment on the business model, right?
0:24:38 So we’ve got mobile, we’ve got machine learning
0:24:40 and now it’s like let’s do something other than ads.
0:24:43 Yeah, for sure, for sure.
0:24:48 I am not a fan of strictly ad-based models
0:24:52 mostly because the ones that do succeed,
0:24:53 say like a Facebook or Google,
0:24:55 I mean the reason their ads succeed
0:24:57 is not because of the massive page views
0:25:00 it’s also because of all the information
0:25:01 they have on that user.
0:25:04 So the ads are highly targeted, right?
0:25:06 And if you’re a platform where you don’t have
0:25:10 such detailed information on your end users,
0:25:11 your ads are not as valuable
0:25:14 and they’re not gonna convert as well.
0:25:17 So focusing on just building up page views
0:25:18 and hoping they’ll monetize with ads
0:25:21 to me is a scary strategy in general
0:25:24 for any consumer app.
0:25:27 Yeah, but I think business model experimentation
0:25:29 in the education space is huge
0:25:31 because I mean a lot of these categories
0:25:33 like your skills for doorknob, right?
0:25:37 Maybe it could sell you a similar doorknob.
0:25:38 They should have sold me the doorknob.
0:25:40 They should have sold you other doorknob.
0:25:42 They should have sold you other home projects
0:25:43 that hey, if you take this course
0:25:46 buy the components for it at a discount
0:25:49 and that can be a partnership with your local Home Depot
0:25:52 ’cause geographically they know the local Home Depot
0:25:54 is only two, three miles away from where you are.
0:25:55 Right.
0:25:57 And those ideas are very possible
0:25:59 and not being implemented today.
0:26:02 Kind of thinking if this person learned this course
0:26:06 what else can I sell them beyond just another course?
0:26:09 What physical things can I sell them?
0:26:11 What other services can I sell?
0:26:12 Yeah.
0:26:15 It seems inevitable that as we continue into a world
0:26:18 that rapidly changes, therefore needs new skills
0:26:22 all of the time that the spending pattern here
0:26:24 on education will flatten, right?
0:26:25 Which is the way I think about education spending
0:26:28 over a lifetime today is kind of like
0:26:30 there’s an elephant inside a python, right?
0:26:31 Which is you spend a lot of money
0:26:34 and then you get to college, right?
0:26:36 Where you have the 529 plan, right?
0:26:38 To help subsidize, right?
0:26:40 Tax deferred dollars to go to university
0:26:42 and then basically it drops to zero, right?
0:26:45 It’s sort of like a very small proportion
0:26:47 of the population spends money on ongoing training
0:26:49 and if you spend it, it’s mostly like,
0:26:52 oh, work had me do it and then I expensed it, right?
0:26:54 But it’s nowhere approaching college tuition.
0:26:56 So it’s sort of this big sort of college expense
0:27:00 in the middle, it feels like as we move into a new world
0:27:01 like we want to flatten that out, right?
0:27:05 We want to give access to piano teaching for kids,
0:27:06 smooth it out earlier in life
0:27:08 and then we sort of smooth it out later in life too.
0:27:12 And that’s gonna require this business model experimentation.
0:27:14 Business model experimentation
0:27:17 and just making that information more accessible.
0:27:21 Like if I told you you could spend $15
0:27:23 and get 10 courses on how to improve your voice,
0:27:24 would you consider it?
0:27:28 I am looking for voice instructors right now, right?
0:27:29 – You would do it, right, right, right.
0:27:31 – I’d pay far more than that.
0:27:32 – Right.
0:27:35 – But I feel like so many of these ideas
0:27:38 or these instructors, people oftentimes just forget
0:27:42 that they exist because it’s not so in their face.
0:27:45 And it’s not also done in bite-sized stacks
0:27:48 on their own schedule, on their own timeframe, right?
0:27:50 And when you’re on a mobile platform,
0:27:52 when you’re doing these bite-sized lessons,
0:27:55 you can do it every morning.
0:27:56 – Yeah.
0:27:58 Well, I personally can’t wait for a lot of this stuff.
0:28:00 As I mentioned, I love learning new things
0:28:03 and I can’t wait to have very compelling products
0:28:05 that are teaching me new things
0:28:08 a little more sophisticated than how to replace a doorknob.
0:28:10 Maybe a little less sophisticated than how to do it.
0:28:12 – A platform should have also sold you services
0:28:14 of a handyman nearby in case you failed.
0:28:16 – That’s true, right?
0:28:17 – Luckily.
0:28:19 – There’s just so many ways you can monetize
0:28:21 a simple video like replacing a doorknob
0:28:22 that’s not being done today.
0:28:24 – Yeah, yeah, totally true.
0:28:27 – Like all these how-to home fixes,
0:28:30 a good chunk of people who attempt them can’t do it, right?
0:28:32 And they’re willing to pay for the local
0:28:34 handyman to do it. – Surrounded by an explosion
0:28:35 of tools, right?
0:28:36 Oh, I give up.
0:28:37 – Right, or I may be missing this part,
0:28:39 maybe I’m missing this wrench or hey,
0:28:41 the power tool, right?
0:28:42 You could have done this in half the time
0:28:44 if you had this power tool.
0:28:47 These ideas aren’t being thought of right now
0:28:51 because it’s a business model innovation, right?
0:28:54 Think about not just selling them the next course
0:28:56 and stuffing more ads into your course
0:28:58 and then therefore making your video much longer
0:29:00 than it needs to be, which is the game
0:29:02 that a lot of these influencers have to play.
0:29:05 They’re being forced to play that game right now.
0:29:07 Give them better ways to monetize what they’re selling.
0:29:10 – Yeah, and that business model would be good for,
0:29:12 ’cause another age old problem with these education markets
0:29:14 places that try to get broad, right?
0:29:16 I want to have all of this content
0:29:19 is that the repeat usage is never as good
0:29:20 as the entrepreneur hopes, right?
0:29:24 So you kind of hope that I sell you the piano playing class
0:29:26 and then you’ll come to me for filmmaking
0:29:27 or whatever it is.
0:29:29 And it turns out in a lot of these
0:29:33 that you’re almost capturing that customer again
0:29:36 for the first time, even though they bought a class from you,
0:29:37 right? – Yeah.
0:29:39 And to this point is actually,
0:29:42 I wouldn’t say I’m never a fan of subscription models,
0:29:45 but for this category,
0:29:48 I don’t think subscription is necessarily the best model.
0:29:51 Because for me to sign up for a subscription,
0:29:54 I have to think I’m gonna take more than one class, right?
0:29:57 I mean, why not instead let me pay per course?
0:30:00 And for other courses, if you want to push discovery,
0:30:04 allow me to sample the first 10 minutes of a class for free,
0:30:07 right, or do some other kind of incentive
0:30:09 to get me to see the value.
0:30:11 And then maybe after two, three courses,
0:30:13 then sell me something like a subscription.
0:30:14 Or I’m like, yes, for sure,
0:30:17 I’m gonna use this multiple times over.
0:30:20 But the idea of jumping from day one
0:30:21 to push you a subscription, I think,
0:30:24 is a hard business model for this category.
0:30:25 – Yeah, it’s hard.
0:30:26 You have to capture the people
0:30:28 who would pay upfront for health clubs, right?
0:30:30 Which is the sort of, it’s the aspirational me
0:30:33 that will go to the gym all the time, right?
0:30:34 – Right.
0:30:35 – Well, thanks for joining us.
0:30:39 We’re so excited for the future of EdTech
0:30:41 that is mobile first and AI enabled
0:30:43 and isn’t just advertising,
0:30:44 ’cause I wanna learn new stuff,
0:30:45 I wanna learn it all the time.
0:30:47 The next thing I think in our house
0:30:48 will be clearing clutter.
0:30:51 And so it’s funny, Marie Kondo
0:30:52 has that series on Netflix now.
0:30:56 And so maybe I should watch it
0:30:58 and maybe there will be a tailor-made startup
0:31:00 for that type of stuff.
0:31:02 They can offer me help when I get stuck
0:31:04 clearing my own crap.
0:31:06 So, all right, thanks, YouTube.
0:31:08 We’ll see you next episode.
0:31:09 If you liked what you saw,
0:31:13 go ahead and comment and subscribe on the bottom.
0:31:14 And we’ll see you next episode.
0:31:15 – Bye.

0:00:03 Hi, and welcome to the A16Z podcast. I’m Hannah.
0:00:07 Today’s episode features a special conversation with renowned scientist George Church,
0:00:11 known for his groundbreaking work and methods used for the first genome sequence
0:00:14 and for his work in genome editing, writing, and recoding.
0:00:17 Church’s innovations have become an essential building block
0:00:21 for most of the DNA sequencing methods and companies we see today.
0:00:25 He has joined in this conversation with A16Z bio-general partner Jorge Conde,
0:00:29 who, among other things, founded a company with Church out of the church lab.
0:00:32 The two take us on a wild journey into the scientist’s mind and work,
0:00:37 starting with what the leading pioneer in this space makes of where we are today with CRISPR,
0:00:39 especially given recent news about CRISPR babies in China,
0:00:43 then moving on to the broader implications of all that on a cultural level
0:00:48 to finally what it takes to go from science fiction to lab to reality.
0:00:49 So, let’s start at the beginning.
0:00:55 If we were to bet 10 years ago whether we’d have a CRISPR baby, a mammoth baby,
0:01:00 or a Neanderthal baby, which would you have bet would have come first?
0:01:05 Oh, and questionably, a CRISPR baby, you know, I mean, it was not a huge technical leap.
0:01:10 They all involved societal and ethical questions,
0:01:14 but that one probably had the clearest path, you know,
0:01:19 because there was such divergence of opinion, somebody was going to do it.
0:01:24 And would you have expected that it would have been essentially a rogue effort
0:01:28 versus a solo effort, as it seems to have been the case in the China CRISPR baby news?
0:01:30 I wouldn’t characterize it as a solo effort.
0:01:32 I’ve seen the author list, it’s quite long.
0:01:38 And I also find it unlikely that a government as technically astute
0:01:44 and as engaged in observation would be unaware of such an important thing.
0:01:48 If I were a technically astute government,
0:01:51 there are very limited number of topics I would be paying attention to.
0:01:58 And these would be things like, you know, nuclear, biological, encryption, and CRISPR.
0:02:01 It’s a short list, so I don’t think it’s solo.
0:02:05 So let’s talk a little bit about the way it’s been sort of positioned, at least publicly.
0:02:11 Can you describe a little bit what the experiment actually was?
0:02:15 What did the scientists or scientists do in this particular case for the CRISPR baby?
0:02:19 I’ve actually seen a lot of the data and the preprints.
0:02:25 And this was a simple, in a certain sense, application of CRISPR
0:02:31 to alleviate a potential for HIV infection.
0:02:37 You know, 900,000 people die every year of HIV, and this was an approach to it.
0:02:44 And they did it by knocking out the gene that encodes the HIV receptor on the surface of T cells.
0:02:45 This is CCR5.
0:02:52 This is CCR5, which has already been approved for FDA clinical trials for Sangamo
0:02:56 and for editing in adults that have AIDS.
0:03:00 That’s a different scenario, but vets many of the issues that come up
0:03:04 as to whether this is a reasonable editing strategy.
0:03:08 So first of all, people have described it as knocking out the gene.
0:03:11 Other people have described it as editing the CCR5 gene.
0:03:15 Having seen the data, what exactly was done to CCR5?
0:03:21 Right, so what CRISPR does well is often described as editing, it really is damaging.
0:03:24 It’s not really that good at precision editing.
0:03:28 Hopefully there will be a good way in the future.
0:03:30 And so what it does is it knocks out genes.
0:03:34 And in this case, that’s exactly what you want, is you want to knock out the CCR5 gene.
0:03:35 And there’s precedent for it.
0:03:40 About up to 10% of certain parts of Europe have a double null.
0:03:43 To double null in this case, basically, two non-functional CCR5s.
0:03:48 And you need really both non-functional in order to be resistant to the virus.
0:03:50 And it doesn’t make you resistant to all viruses.
0:03:55 It doesn’t even make you resistant to all HIV viruses, but that’s not the point.
0:03:58 It’s like a vaccine, it makes you resistant to whatever you’re vaccinated against.
0:04:03 And analogies were made in the consenting for this between this and vaccination.
0:04:08 There is no good vaccine, there’s no cure for HIV/AIDS.
0:04:14 And right now, if you get it, and there are 37 million people who have been affected,
0:04:19 if you get it, you’re doomed to a lifetime of combined antiretroviral therapy,
0:04:24 which is not the thing that you would wish to have if you had any choice.
0:04:24 Sure.
0:04:28 Well, vaccines, if there did exist one, would be quite a good choice.
0:04:32 And so this is as close as you can get to a vaccine.
0:04:37 So I read that the double nulls for CCR5 have increased predisposition to West Nile.
0:04:38 That is correct.
0:04:42 So there’s a risk for almost every preventative antiretroviral therapy.
0:04:43 And this is the risk in this case.
0:04:47 In most populations, that’s considered a smaller medical risk.
0:04:51 It’s obviously a case by case for populations and individuals.
0:04:54 And there are undoubtedly other advantages and disadvantages.
0:04:58 And I may be taking you a little bit out of context here, but I’ve heard you describe
0:05:02 CRISPR as genetic vandalism.
0:05:06 So do you think that that’s a good application for germline editing?
0:05:10 Well, it’s vandalism in the sense that it can add or delete a small number of base
0:05:16 pairs, typically, in the range of 1 to hundreds.
0:05:22 It’s not going to do something really wacky, except it may be some incredibly low frequency.
0:05:24 Again, no drug is without its side effects.
0:05:28 And that’s why there’s all the fine print that accompanies all the approved drugs.
0:05:31 So I think in this case, it is what you want.
0:05:32 It’s exactly what you want.
0:05:36 You want to destroy the CCR5 gene without destroying any adjacent genes.
0:05:41 And that’s every allele that I’ve seen in the literature for CCR5, whether it’s done
0:05:46 in adults or done in tissue culture, is what you would want.
0:05:50 So I’ve read in this case in the Chinese CRISPR baby publication that there is some
0:05:55 mosaicism that he might not have functionally knocked out all of the CCR5.
0:06:00 So is there any worry that after the post-experiment that this particular child might still be
0:06:02 at risk for HIV infection?
0:06:08 So first of all, in the approved clinical trials on adults that have HIV/AIDS, there
0:06:11 is a lot of mosaicism.
0:06:13 It’s considered part of the clinical trial.
0:06:20 And maybe as little as 20% are properly edited, meaning double nulls.
0:06:24 That’s enough, though, because all the rest are wiped up by the virus, and then the ones
0:06:28 that are edited dominate the T-cell population.
0:06:29 So it’s one way of thinking about it.
0:06:34 So it’s basically selection for the edited T-cells so you don’t develop immunosuppression.
0:06:35 Right.
0:06:37 So as long as there’s a fair number of properly edited ones.
0:06:43 Now on the other hand, looking at the data, I don’t see that much evidence for mosaicism.
0:06:49 It’s quite possible that what you see in the pre-implantation embryo when you select a
0:06:57 few cells out of that blastocyst is not representative of the final, and the final is less mosaic,
0:06:59 or maybe even non-mosaic.
0:07:04 So when they talk about a baby having mosaicism in the case of the CRISPR baby, essentially
0:07:08 what they’re referring to is that there are some cells that will have edits, and some
0:07:10 cells that won’t.
0:07:16 And so essentially that child may grow up to be a mosaic of two different or multiple
0:07:17 different cell types.
0:07:18 Correct.
0:07:22 And the same thing I should note is true for adult gene therapies is that whether they’re
0:07:27 done ex vivo or in vivo, it usually results in a high level of mosaicism because the delivery
0:07:29 is inefficient.
0:07:33 And it may even be the case that the germline has lower mosaicism.
0:07:34 We need more data.
0:07:35 Great.
0:07:42 So the amount of off-target and mosaicism so far for these two babies seems to be low,
0:07:43 but time will tell.
0:07:49 You know, it could be that we’re just lucky the same way that the first in vitro fertilization,
0:07:51 Louise Brown, turned out just fine.
0:07:55 And so that greatly influenced, it shouldn’t have, I mean, it’s an N of one.
0:07:59 We shouldn’t have all said, “Oh, IVF is perfect because we have one perfect baby.”
0:08:03 To your referencing the test tube hysteria around the first IVF.
0:08:11 1978, which subsided, it grew too much and it subsided too quickly based on N of one.
0:08:15 And I think here we have an N of two, maybe an N of three, and there’s going to be a lot
0:08:20 of attention paid to the actual outcomes rather than how we got there, hopefully.
0:08:25 If you had been in charge of the project, would you have done CCR five or is there another
0:08:28 different obvious application that you’d have gone after first?
0:08:29 Very preclinical.
0:08:33 In other words, I create technologies that’s been used by companies that I found, and they
0:08:34 do the clinical trial.
0:08:39 So I probably would not be doing a clinical trial at all, just to put it in context.
0:08:44 But in terms of choice of target, I have said publicly already that targets that have been
0:08:49 championed by the critics to the extent they champion anything, or the ones that they present
0:08:56 as possibilities or as higher priority, although with great reservations, even for those, are
0:09:03 things that are typical Mendelian diseases, that is to say diseases that are very severe
0:09:11 and are predictably heritable, which are things like hemoglobinopathy, salicylium, sickle
0:09:18 cell, cystic fibrosis, and so forth, ignoring the fact that if you’re in an IVF-PGD clinic
0:09:25 anyway, to do your CRISPR editing of your Mendelian disease, you could just do selection
0:09:26 for most of these things.
0:09:33 So I think it’s kind of like they’re rationalizing their choice, which in the same sense that
0:09:37 they might feel is rationalizing to pick a more prominent disease.
0:09:40 But also, I think in all the examples you just cited, you would actually need to edit the
0:09:45 gene to create function as opposed to knocking out, as was the case with CCR five.
0:09:49 And in some sense, the critics might think that that’s attractive, that CRISPR is inappropriate
0:09:53 at this moment because it gives us more time to think about it.
0:09:58 But in any case, yeah, I think that we want an example of a disease that is very common,
0:10:04 and most of the gene therapies are rare, whether editing or not, and we want something that’s
0:10:09 very serious, and certainly HIV falls in that category.
0:10:15 So it struck me as a plausibly justifiable choice, possibly more justifiable and something
0:10:21 that you can avoid with genetic counseling or with PGD IVF or both.
0:10:27 So IVF-PGD stands for In vitro fertilization with prenatal genetic diagnosis.
0:10:32 So the diagnosis can essentially be done before you implant the embryo from an in vitro fertilization
0:10:33 into the mother.
0:10:39 And so by some people’s definition, that’s still kind of a lab resource rather than a
0:10:40 baby.
0:10:45 And those are typically used for Mendelian diseases, meaning that you can see in the
0:10:49 parents, for example, both parents could be unaffected carriers.
0:10:55 You can predict that 25% of their children will or their embryos in vitro fertilization
0:10:58 could be affected with a very serious disease.
0:11:03 So now that the gene is out of the bottle, we have the first CRISPR babies born.
0:11:09 First of all, what was the role of ethicists in the first project in the CCR-5 Chinese
0:11:11 CRISPR baby project?
0:11:14 And what do you see as the role of ethicists going forward?
0:11:19 Well so the National Academy of Sciences in the US and with participation from China and
0:11:24 other countries in February 2017 came out with a report where they listed 10 items that
0:11:31 would be recommendations for prerequisites for doing germline editing in children.
0:11:38 I mean obviously you can do germline editing in animals or you can do it in cells in culture
0:11:42 or even embryos in culture but actually in planting and having children.
0:11:44 And a lot of these had ethical components.
0:11:50 Many of them were very similar to what you would expect the FDA or the CFDA or the EMA
0:11:56 to be, these are all regulatory agencies around the world, would recommend for any therapeutic
0:11:57 clinical trial.
0:12:03 We should all be focused on safety and efficacy and ethics and that’s what these 10 items
0:12:06 look like for germline as well.
0:12:10 Do you suspect or do you expect, I should say, that we’re going to see more and more
0:12:15 of these experiments going forward or do you think that after this first one, going back
0:12:18 to the IVF example, do you think there will be a pause?
0:12:23 Well there probably will be something that looks like a pause but it will probably be
0:12:25 an acceleration.
0:12:29 So the same thing happened with the prominent DNA, there was supposedly a moratorium but
0:12:34 during that time, I mean I was a first-hand observer, my research went faster because
0:12:39 people were building incredible facilities for containment and they had just state-of-the-art
0:12:42 equipment that helped everything go faster in my opinion.
0:12:47 And I think the same thing has gone with almost every major ethical debate is it attracts
0:12:54 attention, attracts money, whatever is ethical at the time is accelerated and then so whenever
0:13:00 we become comfortable with it, all that acceleration clicks into place and it’s as if there’s been
0:13:02 a steady growth.
0:13:07 That doesn’t mean we should be incautious, on the contrary, I’m very much pro-regulation.
0:13:14 I think that regulation is what saves us from phalidomide and Vioxx and hormone replacement
0:13:17 therapy and so forth, long-term.
0:13:21 So I think we need to support our regulatory agencies around the world.
0:13:25 They are not agents of slowing things down, they’re actually agents of smoothing things
0:13:26 out.
0:13:28 Yeah, and I think it’s pretty clear we’re seeing that today in the regulatory environment,
0:13:29 certainly here in the US.
0:13:34 I mean, we’ve got the first cell therapies, the first gene therapies, the first digital
0:13:35 therapies.
0:13:38 It’s a pretty remarkable moment from a regulatory standpoint for a new therapy.
0:13:41 To some extent, I think they like new technologies more than like the old ones.
0:13:47 The old ones tend to fail because they’re so incremental that they’re no longer compared
0:13:50 to the placebo, they’re compared to whatever they’re an increment over or whatever therapy
0:13:57 already works and they often fail, but brand new category, monoclonal antibodies or cell
0:14:04 therapies or gene therapies, those just like blow past and create all sorts of new improvements,
0:14:06 traumatic improvements in safety and efficacy.
0:14:11 So the FDA is, that’s their mandate, is to cure people, not to stop people from practicing
0:14:12 medicine.
0:14:18 So just to take that vein, if we look forward, what do you see as sort of the next non-incremental
0:14:23 sort of step function change in the way we treat disease or manage disease or even diagnose
0:14:24 disease?
0:14:28 Well, first of all, if we started diagnosing, that would be a really big thing.
0:14:33 It’s really, we’re as a population, even worldwide, we’re under-diagnosed.
0:14:38 There’s a lot of very cost-effective diagnoses that partly because they’re cost-effective,
0:14:45 they’re undervalued and the care providers are not compensated as much as some less effective
0:14:47 but expensive medicine.
0:14:51 So that’s one thing, diagnosis would be terrific and that’s part of preventative medicine.
0:14:56 So we talk a lot about precision medicine, but the preventative part gets kind of swept
0:14:57 under the rug a bit.
0:15:01 If you look at the pie charts for a number of government agencies, including the NCI,
0:15:09 NIH in general, is preventative is sort of in the 1 to 5% of the pie chart, but its payback
0:15:10 is enormous.
0:15:14 And so basically you’re saying misaligned incentives and human behavior has sort of
0:15:17 mitigated how much prevention we actually do.
0:15:21 That’s right, but that would be a huge breakthrough so we could do more diagnosis and more prevention.
0:15:26 Now the ultimate diagnosis for genetics is whole genome sequencing and environmental
0:15:30 monitoring with sequencing as well for pathogens, allergens and so forth.
0:15:39 The therapeutic cognate of that is preventing serious Mendelian diseases that are very predictive
0:15:44 and very often single gene or have enough of a single gene component that they’re ready
0:15:47 for medical practice, thousands of them.
0:15:48 And those can be prevented.
0:15:52 We often talk about gene therapy, actually that’s a million dollar drug.
0:15:57 It is once and done so you don’t have a lifetime of dosing, but it is expensive, we need to
0:15:58 acknowledge that.
0:16:00 Partly because a lot of them are rare.
0:16:06 If you get a common gene therapy, like let’s say aging reversal or some major infectious
0:16:11 agent that everybody wants to be vaccinated against, it’s like most infectious ages have
0:16:16 potentially billions of customers, then that will bring the price down radically.
0:16:22 But in addition to gene therapy, either in adults, children, fetuses or germ line, there
0:16:28 is the option of doing IVFPGD that we already mentioned and even earlier in matchmaking.
0:16:36 So if you never meet or fall in love with someone who is predisposed to create heavily
0:16:43 disease, genetically diseased children, that’s very both cost effective and humane.
0:16:45 So you’re describing 23andMe meets 10andMe.
0:16:47 No, I am not actually.
0:16:52 I’m describing a whole genome sequencing, which is not, there are a very small number
0:16:56 of companies that provide whole genome sequencing because everything else, anything less than
0:17:00 whole genome sequencing is not medically powerful enough.
0:17:04 Anything less than that misses because you false assurance.
0:17:09 That combined with some sort of dating that is an odd combination and possibly further
0:17:14 combined with whoever is paying for the Mendelian costs right now, which are about a million
0:17:18 dollars per person, doesn’t have to be gene therapy, which happens to be a million.
0:17:21 It can be just caregiving.
0:17:27 It adds up and somebody is paying for that, typically insurers and employment benefits
0:17:34 and they could be saving this money if they could encourage their clients, patients to
0:17:38 avoid falling in love, marrying and having children when they have incompatibility.
0:17:39 Their carriers of something.
0:17:40 And this actually works.
0:17:48 So Dorya Sharim has eliminated significant menial disease like Tasex by practicing a
0:17:53 version of this that probably isn’t perfectly generalizable, but there are versions that
0:17:59 could keep a great deal of privacy and allow people to just never know whether they’re
0:18:02 affected or not, or whether they’re carriers or not, never know if anybody else is affected,
0:18:05 but still avoid meeting.
0:18:09 I mean, the analog version of this was back in the day in certain communities, Jewish
0:18:12 communities where there was disease, the rabbi would essentially replace function.
0:18:13 That’s what Dorya Sharim was.
0:18:14 Exactly.
0:18:20 It was started by an individual who had five children in a row that were affected by Tasex,
0:18:25 which is a terrible burden on the child and the family that typically died before they’re
0:18:27 four years old and very painful.
0:18:34 And so he correctly determined that you could do this very inexpensively and mainly.
0:18:35 Via matchmaking.
0:18:36 Via matchmaking.
0:18:37 Right.
0:18:39 So let’s take another blast back to the past.
0:18:44 So about 10 years ago, you and I started a company in whole genome sequencing.
0:18:45 Called Nome.
0:18:46 Thank you.
0:18:47 Called Nome.
0:18:48 Not Nome.
0:18:49 Called Nome.
0:18:52 We used to have this constant back and forth that you thought it should be called Nome.
0:18:53 I would call it Nome.
0:18:54 Yeah.
0:18:55 I thought it should be called Nome.
0:18:56 Yeah.
0:18:57 And this was the market test.
0:18:59 It was the youth into me, which is incredibly frustrating.
0:19:00 Yeah.
0:19:01 And now I listen to you.
0:19:02 Thank you.
0:19:03 Like I said, no.
0:19:04 Okay.
0:19:08 My rejoiner on that always was, if you want to call it know me, then I want to call you
0:19:12 Jorge Iglesias and you are never a big fan of that one.
0:19:13 Okay.
0:19:14 I don’t have no problem with that name.
0:19:15 I think it’s a better name.
0:19:16 It’s a nice name.
0:19:17 It’s going to increase the brand.
0:19:20 It just has more syllables, that’s all.
0:19:22 But it just, it rolls off the tongue.
0:19:23 Does.
0:19:24 Yes.
0:19:28 It basically made the bet that whole genome sequencing was important.
0:19:32 That interpretation of that data would be relevant, that it would be meaningful.
0:19:36 Ten years hence, there still are not many people walking around that have had their
0:19:41 whole genomes sequenced despite the fact that the cost has now fallen arguably below $1,000
0:19:44 or at least we’re at that $1,000 threshold.
0:19:46 So I had two questions for you.
0:19:52 Number one is, is the $1,000 threshold for this to be useful for everyone to get sequenced
0:19:53 too high a dollar number?
0:19:57 In other words, does it need to be $100 or $10?
0:20:02 And number two, to the extent that this hasn’t happened yet, why hasn’t it happened yet if
0:20:03 it’s not cost?
0:20:06 I would say there’s three reasons why it hasn’t happened yet.
0:20:10 And I’ve been living this reality for most of my career.
0:20:15 I’m convinced that it would be valuable for the world, it costs effective medicine, preventative.
0:20:18 And I think the three reasons are one is cost.
0:20:20 The cost should probably be $0.
0:20:22 And secondly, it’s privacy.
0:20:27 We should have a convincing mechanism of people getting benefit from their genome without
0:20:29 necessarily knowing their genome or anybody else knowing their genome.
0:20:34 You can have something where it’s only an encrypted form, not available to anybody, including
0:20:36 insurance and government.
0:20:37 That’s second.
0:20:41 And the third is most people don’t understand the value proposition.
0:20:44 It’s either misrepresented, but by both extremes.
0:20:48 So some people say, oh, it’s so valuable that you’re going to whip out your cell phone and
0:20:50 look at your genome twice a day.
0:20:54 And at the other extreme, they say, I can’t imagine ever using it.
0:20:58 And the reality is somewhere in between, and I think the analogy is seatbelts.
0:21:00 So seatbelts were essentially free.
0:21:01 They were standard equipment.
0:21:04 They were required by law that you buckle.
0:21:09 And there were a lot of ad campaigns to get you to do so, kind of like smoking.
0:21:15 And none of those were effective because people did, you know, the kind of ordinary math,
0:21:19 which is, hey, I’ve got a less than 1% chance of ever using a seatbelt, ever needing one.
0:21:21 So I’m not going to bother.
0:21:26 And then the thing that made the difference was technology to sense the buckling and turning
0:21:28 off an annoying sound.
0:21:29 So that’s what made the difference.
0:21:30 And we need an equivalent thing.
0:21:32 It’s a public health issue.
0:21:35 It’s not an individual health issue.
0:21:37 So I don’t benefit from being sequenced, the collective.
0:21:43 These people, 95, 96% will get a blank sheet.
0:21:50 They should get a blank sheet in terms of really actionable, very serious Mendelian diseases.
0:21:51 And that should be the expectation.
0:21:56 Not the two extremes that you’ll use it every day, or that everybody will use it every day,
0:21:57 or the other extreme, which is totally useless.
0:22:04 It’s this strange thing where 1% to 4% of the population will have a very big impact
0:22:05 on their life.
0:22:12 And the bottom line for their care providers, millions of dollars, huge impact on the whole
0:22:17 family, if you’re one of the unlucky 4%.
0:22:19 And we need to get that message out there.
0:22:24 And I think that bringing the price down to $0 and showing that it’s protectable, encrypted
0:22:29 and so that nobody can get access to it except for things that benefit you or your family
0:22:33 or society, that will get their attention.
0:22:36 But it’s going to take a little bit more than that’s going to take some anecdotes.
0:22:40 You would think that data would be better than anecdotes, but you need both.
0:22:43 And I think it’s going to happen very soon now, because we finally have the $0 genome
0:22:50 and the encryption, and we’re starting to get communication of this rare advantage where
0:22:54 you’re not exempt, even though the odds are that you’re exempt, you don’t know that you’re
0:22:57 exempt until you get your genome sequenced.
0:22:59 So two questions for you on the three ones you’ve laid out.
0:23:04 The first one is, in the early days of Nome, I remember when we would think about this
0:23:10 question of security, you correctly pointed out that if you really wanted my genome,
0:23:13 you would just wait for me to leave the room and collect it from all of the genome.
0:23:14 Exactly.
0:23:15 That is even more true than it was back in 2007.
0:23:16 Right.
0:23:18 So you collect all of this chair off this table, and you’ve got me.
0:23:19 Got it.
0:23:24 So why is security and privacy, is it even a meaningful thing to think about if it’s
0:23:25 an impossible thing to achieve?
0:23:30 Well, the point is, if it’s preventing people from getting their own genome sequenced, if
0:23:36 they think that them seeing their own genome puts them at risk for somebody like hacking
0:23:41 or requesting it or subpoenaing it, then yes, it’s a problem, because there is a difference
0:23:46 between me woefully getting my genome and looking at it and somebody surreptitiously
0:23:47 taking it.
0:23:48 Okay?
0:23:51 So we can pass laws that punish people for surreptitiously taking my DNA.
0:23:56 We do have the Genetic Information Non-Discrimination Act of 2008 that is along those lines.
0:24:00 It’s not perfect, but it shows the intention of the public.
0:24:05 So that can kind of handle the abandoned DNA problem, and we could keep shoring that
0:24:07 up and building up those laws.
0:24:11 But then there’s the question, if I look at my genome, if I have my genome available
0:24:16 in text format, unprotected, then anybody can come along and demand it, right?
0:24:20 Insurance companies say, “I know you know it, so I want to see it.”
0:24:24 And they can say, “I want to see it, so I can convict your brother.”
0:24:29 And if it’s encrypted so that even you can’t hack it, then you can just say, “Sorry, it’s
0:24:30 out of my hands.
0:24:31 I don’t have my genome.
0:24:33 If you want my genome, you’re going to have to steal it from me.”
0:24:34 Right?
0:24:35 Got it.
0:24:36 And I think that’s where we are today, finally.
0:24:39 By the way, you may not remember this, but we were laying out the risk factors and all
0:24:46 of the other things for the consent form on all the things that a potential recipient
0:24:48 of their genome data would have to think about.
0:24:53 By far and away, my favorite one that you contributed was the potential risk that someone
0:24:55 could plant your DNA at a crime scene.
0:24:56 Right.
0:24:57 Yep.
0:24:58 High risk or low risk?
0:25:03 So that was also in a personal genome project consent form, which started around that same
0:25:04 time as Noam did.
0:25:06 Is it high risk or low risk?
0:25:14 I’d say that we’re getting more and more sophisticated at sequencing and methylation analysis.
0:25:17 You’d have to have the whole genome now rather than back then, it might be just the CODIS
0:25:18 parts.
0:25:26 CODIS is just a few handful of simple sequence repeats that are used in criminal investigations
0:25:27 like —
0:25:28 Take forensics.
0:25:31 Yeah, forensics and CSI type stuff.
0:25:35 Now you’d need the whole genome because if somebody felt it was being hacked, they’d
0:25:37 say, “Well, let’s check the whole genome.”
0:25:39 A defense attorney could ask for the whole genome.
0:25:42 Further, you could ask for methylation to show that it’s the right age.
0:25:47 For example, I can have my DNA from 20 years ago, and you’d have to show — or you could
0:25:48 check the immune status.
0:25:53 So you could say, “Oh, does the immune status coincide with what the –” which that should
0:25:59 be an argument for you to be constantly sequencing your immune, your blood DNA, so you can date
0:26:01 whatever samples you’re taking.
0:26:03 For every hack, there is a counter hack.
0:26:08 So I think I’m glad that we’re not at that stage right at the moment, even though we
0:26:11 predicted it back in 2005.
0:26:16 So going back to 2005, can you describe briefly what the Personal Genome Project was?
0:26:19 Because it was the first effort to really start to think through these issues.
0:26:25 The Personal Genome Project was one of the first recognitions about how identifiable
0:26:31 both your genome is and also even parts of it and your medical records.
0:26:37 And people were starting to want to share genomic data and medical records, ideally integrated,
0:26:43 so that you could see what an individual, what we would now call precision medicine,
0:26:46 record would look like, right, back in 2005.
0:26:51 And I wrote an editorial saying that this was a risk, that the data could leak out,
0:26:55 and once it leaked out, the people could be re-identified, and all of their diseases could
0:26:59 be determined from either the medical record or the genome or both.
0:27:00 And this is played out.
0:27:05 I mean, there’s many examples of millions of people being their medical records and/or
0:27:08 their genome leaking out in various ways.
0:27:12 And of course, now, since then, WikiLeaks has occurred, which is just an example of
0:27:17 how they can be officially stored publicly after leaking.
0:27:20 So I think that was what we were concerned about, and we started the Personal Genome
0:27:25 Project so that we could get people properly consented so they knew these risks, they accepted
0:27:26 them.
0:27:27 And you had to take a quiz, right?
0:27:28 Exactly.
0:27:34 Up to that point, many of the consent forms were long, written in legalese, a lot of
0:27:39 language protected the institution rather than the person, and you would sign them often
0:27:43 under course of circumstances where you were afraid you weren’t going to get the best medical
0:27:45 care if you didn’t sign it.
0:27:50 So we added to that a simple multi-choice exam where you kind of simultaneously got
0:27:55 educated if you didn’t get a perfect score until you got a perfect score.
0:28:00 So it wasn’t like we wanted 90% comprehension, we wanted 100% that you knew all of the risks
0:28:03 and all the benefits, and we had a record of that.
0:28:06 So those were some of the key points of the Personal Genome Project, but the other key
0:28:08 point is we really wanted to share it.
0:28:13 Not just what a lot of people call sharing, even to this day, 13 years later, they call
0:28:19 sharing medical data for research is really a silo that’s hard to get into.
0:28:22 Now, unfortunately, it’s not impossible to get into, it’s not really encrypted the way
0:28:27 you would want it to be, and so there’s a lot of potential for leakage, but it’s hard
0:28:33 enough for regular scientists of good intention to get access to it legitimately.
0:28:37 So we wanted something that was more like Wikipedia where you didn’t have to agree to
0:28:42 be a co-author on a paper, you didn’t have to pay a lot of money, you literally could
0:28:47 use it for whatever you wanted to use it for, commercial, private, teaching, whatever, just
0:28:53 by clicking on it, and that project still exists today in many countries now with high-level
0:28:56 enthusiasm among the participants.
0:29:01 So you were obviously participant one, 001 of the Personal Genome Project.
0:29:02 You’re an open book.
0:29:06 If you go to your lab website, you have everything you’re working on, everything you’ve ever
0:29:11 worked on, you’ve described your phenotype in detail, which I think is fascinating.
0:29:15 Did you learn anything from having access to your own genome that you found particularly
0:29:17 interesting or enlightening?
0:29:21 So I didn’t expect to because I felt that I was likely to be in the 96% that would get
0:29:23 a blank report.
0:29:27 As it turned out, it did learn a couple of things.
0:29:31 So one of them my family was very concerned about because I had a family history of cognitive
0:29:36 decline was that I had no risk factors for Alzheimer’s.
0:29:38 To this ApoE4 status.
0:29:39 APP.
0:29:40 Precinalin 1 and 2.
0:29:41 Every known factor.
0:29:46 So that was reassuring, although I try to tell people not to be reassured that there’s always
0:29:48 something new to learn.
0:29:54 Secondly, I’m an alpha 1 antitrypsin compound heterozygote, which just means I have two
0:29:59 different risk factors that result in a risk for lung disease.
0:30:02 So I should probably avoid pollution, which is probably not a bad thing for everybody
0:30:04 to avoid and smoking.
0:30:08 And those were the two main things that I learned.
0:30:13 So it’s not that different from getting a blank sheet, quite frankly, but probably more
0:30:18 importantly was having my medical records publicly available meant the hematologist
0:30:22 gave me personal advice on my incorrect use of statin.
0:30:26 So it turned out that I was not being properly diagnosed.
0:30:31 Going back to where we’re under diagnosed, and I was having poor reactions of statin
0:30:32 as well as low efficacy.
0:30:34 It wasn’t doing its job.
0:30:38 And so we tried a little bit of nudging them around and finally gave up when I showed and
0:30:43 determined that a vegan diet, strict vegan diet, was enough to bring me down from almost
0:30:46 300 to almost 200.
0:30:52 So it’s not generic advice, it’s something that’s very personal and precision and empirical.
0:30:55 So that was another advantage of having people look on.
0:30:59 And then there was an advantage to the project of me being guinea pig number one.
0:31:04 The IRB, Harvard Medical School IRB, asked me to be an institution review board.
0:31:10 There’s sort of an ethics and protocol reviews of human subjects research.
0:31:16 They wanted me to participate as initially the only subject, or at least part of the
0:31:18 first 10.
0:31:24 And that was beneficial in that when we were developing the skin biopsy for induced pleuripotent
0:31:29 stem cells, the skin biopsy, first one we tried out in me, was ridiculously painful.
0:31:30 I remember that.
0:31:32 And in retrospect, it was crazy.
0:31:39 It was like a six millimeter punch, 12 stitches, no anesthetic, or at least not in the right
0:31:40 place.
0:31:45 And then we switched over to a cream anesthetic, which is instead of 12 injections in the wrong
0:31:48 place, it was cream in the right place.
0:31:51 And then a simple bandage rather than stitches, and a one millimeter punch.
0:31:57 So that was an example for me being eyewitness or guinea pig.
0:32:01 I said, no, that’s not an acceptable protocol immediately.
0:32:06 And I might not have said that if I were detached and I just said to one of the staff physicians,
0:32:08 oh, just go do it.
0:32:14 So that’s a summary of why sometimes it’s important for the top researcher to also
0:32:16 be a guinea pig in the study.
0:32:21 And I don’t think it supplies all studies, but it certainly applied to the Personal Genome
0:32:22 Project.
0:32:24 So switching gears to the church lab.
0:32:28 So if you go on your website, you have a list of sort of the active projects that you’re
0:32:29 working on.
0:32:32 And I mean, it almost reads like screenwriters for like coming up with the next, you know,
0:32:34 great movie.
0:32:35 Talk to me about the church lab.
0:32:37 How do you think about what you work on?
0:32:41 And even one step before that, how does one get into the church lab?
0:32:44 Because from an external standpoint, I mean, this is like Willy Wonka’s chocolate factory
0:32:45 for science.
0:32:49 So what do you look for in incoming students for the church lab?
0:32:50 And then let’s go from there.
0:32:51 Yeah.
0:32:55 So a lot of it looks like science fiction and most people would run away from that, not
0:32:56 run towards it.
0:33:01 And they did when I was starting out, but now we have a track record, same level of creativity
0:33:03 and risk taking.
0:33:08 But actually many of the things we do are they look hard from the outside, but from
0:33:12 the inside, they look like they’re low hanging fruit and they happen way ahead of schedule.
0:33:18 So for example, things that did look like science fiction were fluorescent next generation
0:33:20 sequencing and Nanopore sequencing.
0:33:24 Both of those were wacky when I started them in the 1980s.
0:33:27 And the whole idea that you could bring down the price of the genome from three billion
0:33:33 down to now sub thousand dollars also seemed science fiction.
0:33:36 But now that we’ve done it, now it becomes a beacon for people to say, Oh, whoever did
0:33:38 that, we should go there.
0:33:43 And if, oh, if at the same lab also helped bring in multiple ways of doing genome editing,
0:33:48 including CRISPR, if you just do one, you could be lucky, but if you do several different
0:33:52 ways of doing next-gen sequencing, several different ways of doing editing, then that’s
0:33:55 an attractant to get in.
0:33:56 Self selection is another major filter.
0:34:00 We do such quirky stuff that people don’t even bother to apply unless they’re kind of
0:34:02 already on our wavelength.
0:34:06 So then the biggest filter for me, and I tell us in the first interview, the first conversation
0:34:09 I have is we’re looking for people that are nice.
0:34:11 We’re not necessarily looking for geniuses.
0:34:13 We got plenty of geniuses.
0:34:15 We’re looking for people that are nice.
0:34:16 And how does one demonstrate niceness?
0:34:19 Well, you know, I think, to some extent, just having that conversation, if they want to
0:34:21 be cutthroat, they’re not going to come back.
0:34:26 If they’re kind of sitting on the fence, then they’re going to rise to the occasion.
0:34:31 They’re going to be influenced by that conversation and by all the people that have already passed
0:34:33 through that filter that are in the lab.
0:34:39 And you create a culture where you try not to compete with other labs if you can avoid
0:34:40 it.
0:34:44 Sometimes it’s unavoidable, but you can avoid it by inviting them to work with them, leaving
0:34:49 alone fields where there’s plenty of momentum and a lack of interest in collaboration, making
0:34:53 sure there’s a diverse enough set of ideas going on the lab so that everybody gets to
0:34:59 leave with a subset of those ideas as a parting gift or continue to collaborate if they want
0:35:01 to as long as they want to.
0:35:06 So I think you build up this momentum of knocking off things that look like science fiction,
0:35:12 turning them into science fact and create a culture of ability to fail and to jump back
0:35:15 and to be nice to your colleagues within and outside the lab.
0:35:19 So if I go through the list of things that you’re working on, it’s a pretty broad array
0:35:20 of things.
0:35:25 So you are crispering dogs to keep them young, you are crispering pig organs or had been
0:35:30 working on editing pig organs to make them useful for transplantation.
0:35:34 And then you run the other end of the spectrum, you’re re-engineering biology to create a
0:35:39 mirror universe of things that would be essentially immune to all known viruses or microbes.
0:35:41 How do you pick the projects?
0:35:44 What is it about what’s in the water in the, well, in the Iglesias lab, formerly known
0:35:45 as the church lab?
0:35:52 Like, what’s in the water that gets this lab to produce so many startups and spinouts?
0:35:56 What is that entrepreneurial energy that’s sort of been fostered and created here?
0:35:57 Right.
0:36:02 It may look like an averse set of projects, but they’re actually have common thread that
0:36:08 is surprisingly focused, meaning most people wouldn’t fit in this lab because we’re sort
0:36:12 of into radical transformative technologies, not incremental.
0:36:17 A lot of labs don’t even want to touch technology until it’s working in a company.
0:36:23 We work years before that company and we create the company and then the company has another
0:36:29 few years before it’s sufficiently worked out that it can be adopted by a technology
0:36:32 adoption lab, which is before most biologists.
0:36:37 So anyway, that’s one thing that we’re a little bit on the edge and it’s an acquired taste
0:36:39 or maybe even a rare taste.
0:36:41 What’s an idea that was pitched to you that you said, “Wow, that’s too crazy?”
0:36:43 Well, I’m usually the one pitching the crazy ideas.
0:36:47 I mean, not to say that we don’t have a lot of creativity in the lab.
0:36:48 That’s pretty rare.
0:36:51 In fact, we’ve sort of banned the word impossible.
0:36:55 We certainly try to behave ethically, but I think that many things, there’s a technological
0:36:59 solution to some of the ethical components, not all of them.
0:37:03 And we try to explore creative solutions to ethical problems.
0:37:06 Personal Genome Project was one of those creative solutions.
0:37:10 Surveillance for synthetic biology is another one that I suggested in 2004.
0:37:16 Biocontainment using recoding is a way that we can make any organism resistant to all
0:37:19 viruses and horizontal transfer.
0:37:20 Most of these things now work.
0:37:21 And many of these things are now companies.
0:37:22 That’s correct.
0:37:27 And their foundation was some sort of safety ethics component to the company.
0:37:30 And part of the secret sauce is hidden in plain view.
0:37:31 Like you say, we’re quite transparent.
0:37:35 We can keep other people’s secrets, but our own, we try to get people to adopt them.
0:37:39 Part of the thing that we do that is, instead of saying failure is not an option, which
0:37:42 was one of the Apollo slogans, we say, “Fail fast.
0:37:44 Just pick yourself up quickly.
0:37:46 Have a bunch of things going in parallel.
0:37:50 Find the low-hanging fruit empirically as well as theoretically.”
0:37:54 And just a lot of things people reject too easily.
0:37:57 They either don’t think of it at all, or they think about and reject it.
0:38:02 And so if we see something that looks a little hard, we’ll put it up on the shelf or in plain
0:38:06 view so we can keep reminding ourselves whenever a new technology makes that possible, we pull
0:38:08 it back off the shelf and we do it.
0:38:12 And so we have that culture of constantly reevaluating things that are on the edge of
0:38:14 science fiction.
0:38:18 Do you recruit entrepreneurs that happen to be scientists, or are you turning scientists
0:38:19 into entrepreneurs?
0:38:26 I mainly recruit people who are multilingual, multidisciplinary, because I found it’s hard
0:38:31 to build a multidisciplinary team from disciplinarians.
0:38:36 You have to have a lot of people who already have done two things, and even if you get two
0:38:39 people who have done two things each, they don’t have to overlap, but they’ve done enough
0:38:42 translation that they can start talking to each other.
0:38:46 And if you have enough of those multidisciplinary individuals, then you can sprinkle in a few
0:38:49 disciplinarians and you have an amazing team.
0:38:53 So the church lab, you were pioneers in sequencing, so reading DNA.
0:38:58 You were pioneers in CRISPR, so writing DNA.
0:38:59 What comes next?
0:39:02 Well, so there’s three-dimensional structure of living organisms, so we’d really like
0:39:08 to know every voxel, every volume, element of every pixel in the body of an embryo or
0:39:09 a larger section of tissue.
0:39:13 We’d like to know every molecule here, and we now have tools for doing DNA, RNA, and
0:39:17 protein in 3D at super resolution, finally.
0:39:18 So that’s one thing.
0:39:22 We would like to be able to do higher levels of multiplexing in the terms of editing synthesis
0:39:23 of genomes.
0:39:27 So some people call it, we call this GP Right, or Genome Project Right, but it could just
0:39:28 be heavy editing.
0:39:34 So we’ve set the record of 62 edits in the pigs, and we now have, we have 10,000 edits
0:39:35 in a single cell.
0:39:36 That’s unbelievable.
0:39:38 So it goes from two to 62 to 10,000.
0:39:39 And we want uses for each of these things.
0:39:45 So each of these projects, we have a driving societal benefit for each, and we have a driving
0:39:50 technology where we say, we don’t just want a factor of 1.5, we want a factor of a million
0:39:51 or 10 million.
0:39:58 And so that’s what pushes each of these projects is that triple criteria, which is cool, basic
0:40:04 science, philosophically interesting, technological factors of a million, and societal benefit.
0:40:08 Last question, 10 years from now, or just looking forward into the future.
0:40:11 Do we get the Neanderthal baby, or do we get the mammoth calf first?
0:40:16 Well, we never really said that we were going to a Neanderthal baby.
0:40:20 I mean, there’s a response to a journalist, whether it was technically possible, but nobody
0:40:21 has articulated a reason to do it.
0:40:26 But for the mammoth, there are lots of reasons, both for the environment and for enriching
0:40:29 the diversity of a living endangered species.
0:40:34 So this is not about de-extinction, this is about making hybrids, and many of the species
0:40:40 are already hybrids of multiple species, but now we can have the benefit of synthetic genes
0:40:41 and ancient genes.
0:40:42 Great.
0:40:43 Well, thank you, George.
0:40:44 Thank you for making time.
0:40:45 It was a real pleasure.

0:00:06 Hi everyone, welcome to the A6NZ Podcast. I’m Sonal. Today we’re continuing our series
0:00:12 on consumer tech trends with an episode that pulse checks the state of autonomy in 2019.
0:00:17 Where are we with autonomous vehicles right now? We also share some clarity on what levels
0:00:22 of autonomy means there, including touching on regulatory aspects, and also discuss quite
0:00:27 frankly capitalism, what cars mean nationalistically, and what it will take to bridge the worlds
0:00:33 of Silicon Valley and Detroit, which is why our special guests are Kazer Yunus, former
0:00:39 COO at Y Combinator and co-founder and CEO of Applied Intuition, and Peter Ludwig, CTO
0:00:44 at Applied Intuition, which builds software for the autonomous vehicle industry. Throughout
0:00:49 the discussion, we thread the analogy of mobile to autonomous vehicles, where it applies and
0:00:55 where it breaks down. Speaking of, be sure to also check out a6nz.com/autonomy for posts,
0:01:01 decks, and videos from Benedict Evans, Frank Chen, and others. But this conversation begins
0:01:07 by cutting through the hype on whether autonomous vehicles are coming soon or not.
0:01:12 It’s interesting. You can kind of read publications and within a three-month period you’ll hear.
0:01:17 We’re in the early days, the hype site we’re in, the trough of despair, disillusionment,
0:01:20 I think it’s called, right, the Gardner hype cycle.
0:01:22 Yeah, you have pessimism. It’s kind of all over the board.
0:01:24 And then you have people saying it’s here tomorrow.
0:01:29 Yeah, exactly. And I think probably a good analogy to think about where we are specifically
0:01:35 is that I like to use is where mobile was in kind of roughly 2005. What we consider
0:01:38 to be the modern smartphone isn’t really there because you’re like, oh, look at this
0:01:42 Motorola Razer. It’s not that powerful. And I can kind of extrapolate that maybe this
0:01:47 Blackberry will be cheaper, but it’s very hard to really extrapolate. And so being even
0:01:53 more specific, if you look at kind of 2010, 2011, 2012, mobile engineers highly coveted
0:01:58 in Silicon Valley. I had a mobile company. My last startup was a messaging company,
0:02:02 and that was kind of the bleeding heart of Silicon Valley, and that was the next wave.
0:02:05 You kind of have a lot of that right now in autonomy where autonomous vehicle engineers,
0:02:09 roboticists are highly valued, highly coveted.
0:02:14 I mean, didn’t Uber or Waymo suck up the entire CMU robotics department at one point?
0:02:18 Yeah, exactly. And that’s happened multiple times. These companies like May Mobility who
0:02:24 basically are the U of M lab. You have a voyage who came out of Udacity. And so, yeah, that’s
0:02:28 happening left and right. If you take that mobile analogy though, and then you think,
0:02:34 well, 2010, there’s 11, there’s excitement. Just a few years later, 2015, 2016, nobody’s
0:02:40 writing Objective C. These waves go really, really fast. And I think a good kind of adage,
0:02:44 I think it was Bill Gates who said this, “In two years, nothing looks different, but every
0:02:48 10 years, things are dramatically different.” So if we look back at 2017, autonomy doesn’t
0:02:53 look that much different. The players are generally the same. But I think 10 years from
0:02:57 now, autonomy would be very, very different. Insofar as it might even be a commodity.
0:02:59 Interesting. That’s kind of controversial.
0:03:03 I think there will be a lot of parallels as well on the hardware front as well as the
0:03:08 software front, looking back at mobile. A modern mobile phone has GPS and an inertial
0:03:12 measurement unit has all these advanced sensors that, prior to mobile becoming big, were very
0:03:16 expensive electronics that were only present in potentially military systems.
0:03:22 Right. Chris Anderson calls these components the peace dividends of the smartphone wars.
0:03:26 This idea that essentially all the supply chains and all that competition and the commodification
0:03:30 actually created this rich and thriving ecosystem of all these commoditized parts that can now
0:03:32 be recombined and deployed in many ways.
0:03:35 Exactly. And today, we’re just at the beginning of that for automotive sensors and autonomy.
0:03:39 Yeah. The real miracle of, I don’t want to get too philosophical with capitalism, frankly,
0:03:46 philosophy, go for it. This is the, not the engineer in me, this is the MBA in me speaking.
0:03:52 The real miracle of capitalism is, I mean, there are all of these things that are made
0:03:58 much, much better, much, much cheaper, that almost are the rails in which these industries
0:04:04 lie on. And I think, even if you’re sitting in 2012, I don’t think anybody had the, “Aha,
0:04:09 I can’t believe it. The mobile era has arrived,” exclamation mark. And people, when we talk
0:04:15 about autonomy, they almost want a declarative event, a Eureka moment, where autonomy is
0:04:19 suddenly unloaded into the masses. And then you look at mobile, that didn’t ever happen.
0:04:23 You just remember, one day you decided, “I’m going to find and get that iPhone.” In 2007,
0:04:28 I was at Harvard at the time, and I remember one of my buddies getting a phone, and I said,
0:04:31 “Oh, what do you, what do you think of this iPhone thing?” And he goes, “Ah, it’s kind
0:04:34 of like a toy.” Well, you know, Chris Dixon says this at ACC a lot.
0:04:37 The next big thing we’ll start out as a toy, which he’s modeling that off-crate Christians
0:04:41 and disruption theory. Yeah, who’s an HBS professor, bringing a full circle, right?
0:04:45 Exactly, which is essentially that the innovations happen at the lower end, or the underserved
0:04:49 end of the market, before they hit the mainstream, and the kind of tips enabled by some kind
0:04:57 of enabling technology underneath it. Yeah. And so this type of incremental kind of revolution
0:05:02 or incremental changes that one day bring to you a product, you know, even if you take
0:05:07 the iPhone, the iPhone rests on hundreds of companies and thousands and thousands of
0:05:12 innovations, and they’re not just in, you know, the screen. No, it’s an entire ecosystem.
0:05:17 It’s the store, the payment processing, all the way down to, you know, the analytics for
0:05:20 apps. The broadband and the connectivity. I mean, that’s actually the missing piece
0:05:23 for a lot of continuing installation in the mobile phase.
0:05:29 There’s so much, exactly. I think there are these statements made that, you know, the
0:05:36 path to autonomy is far, far away. It is better to start autonomy company today in 2019 than
0:05:37 ever before.
0:05:40 So I want to ask a few questions on this. So the first thing is the kind of theme of
0:05:46 what you’re saying, and I buy this, is that, you know, innovations, they seem incremental
0:05:51 at the time, and then they kind of tip to where they accelerate very fast. And there’s
0:05:57 some kind of combo of the two, where the iPhone was like 20, 30 years in the making. But then
0:06:02 I would also argue that there is, well, there may not be a discrete specific single event.
0:06:07 It is an accumulation. There is still a quote, iPhone moment in every industry where that
0:06:13 industry sort of mainstreams, and you really then begin to see and experience the potential
0:06:18 even if it does start off as a toy. So my first question is for autonomy, how far away
0:06:22 do you think we are? Not just in terms of time, but steps towards that quote, iPhone
0:06:23 moment.
0:06:28 So let’s define the iPhone moment first. You know, I think it was Steve Jobs who said,
0:06:32 you know, the ’60s really happened in the ’70s, and the iPhone moment really happened
0:06:40 in like 2012, 2013. Right? The iPhone moment in my mind is when you have, you know, frankly,
0:06:47 the 12 or 24-month period where Instagram, Snapchat, Uber, and WhatsApp are all created,
0:06:51 and they all are created in a roughly pretty tight time bound. That’s the iPhone moment
0:06:52 in my mind.
0:06:56 So you’re really saying the app layer where people are really using things.
0:07:01 I think that’s what generally the public thinks about it. Now, I think probably the more specific
0:07:06 moment is the announcement of the iPhone. But if you look back, the announcement of
0:07:09 the iPhone is met with skepticism.
0:07:10 We forget that now in hindsight.
0:07:14 There’s that famous video, I think, of Steve Ballmer talking about how great Windows Mobile
0:07:17 is in comparison to iPhone because it has so many more features.
0:07:18 Oh, I forgot that.
0:07:23 Blackberry was like, well, no way, this is going to be actually a real thing. Again,
0:07:28 the gimmick, the toy. So if you take the iPhone moment as a 2007, we’ve already had that.
0:07:32 That’s the Waymo shuttles. People are like, wow, these things can only go during the day.
0:07:36 That’s not very useful. I don’t live in Arizona. So what the general public, though, will consider
0:07:41 the autonomy moment is when you meet somebody who doesn’t live in Silicon Valley, doesn’t
0:07:46 work in technology, and they’ve had an autonomous ride. Maybe it’s on college campuses. Maybe
0:07:50 it’s an airport shuttle. Maybe some goods appear at their house with an autonomous robot.
0:07:54 That’s when you’re seeing the penetration of the market into areas which are far beyond
0:07:56 the early adopters or something like that.
0:07:59 I mean, one could argue that that, to your point, in terms of defining what the iPhone
0:08:06 moment is, is not the moment. It’s actually the experience of the iPhone. It’s the applications,
0:08:11 the iPhone phenomenon even. So that’s really what we’re talking about here. So then on
0:08:16 that front, where do you guys think we are? How far are we from there?
0:08:23 So we have shuttles already. This is another, I think, mischaracterization or classification
0:08:28 of autonomy. It’s almost always, excuse me, thought as robo-taxis. And autonomy is actually
0:08:34 much more the adage that anything that moves will one day be autonomous. We blew that very
0:08:40 deeply. And so the point being is they’ll come in these little waves. And each of those
0:08:46 are different. The robo-taxi wave is kind of a bit orthogonal to the shuttles wave,
0:08:52 which is a real thing, which is campus shuttles, retirement communities. So those are different,
0:08:56 which is orthogonal to the self-driving truck wave, which is orthogonal to the, I would
0:08:59 say, the warehouse robots.
0:09:01 Why do you think all of these are orthogonal to each other? One would argue that they’re
0:09:07 the same underlying kind of roboticization, automation. So why are they orthogonal in your
0:09:08 taxonomy and worldview?
0:09:12 There are many similar technologies that are shared across the different verticals, but
0:09:16 there is a lot of domain-specific work that’s still done to make the system actually production
0:09:23 worthy. For example, John Deere has had a production semi-autonomous tractor/trailer system for harvesting
0:09:27 crops for more than a decade. As these systems become more and more sophisticated and more
0:09:31 autonomous to the point where there’s no human in the loop, there is a lot of engineering
0:09:34 effort that sort of goes in that last 10% to get to that production quality.
0:09:38 Yeah, that’s what people always talk about is that sort of last 10%, that last mile, that
0:09:43 you get the 80%, the 99%, but then you have this percentage left, which is quote all the
0:09:47 edge cases and all the things that people are trying to tackle. There are levels out
0:09:51 there for how people describe these things. And so Elon Musk will make a claim about Teslas
0:09:55 and people will say, “Well, they can’t handle all these edge cases,” etc. So in this state
0:10:01 of autonomy 2019, where are we on the levels of autonomy? Can you quickly break down that
0:10:02 taxonomy for our listeners?
0:10:06 Sure, so going through the levels just one by one. So level zero is where most production
0:10:11 vehicles are today. And so this would be a car that perhaps has anti-lock brakes and
0:10:16 traction control, some version of electronic stability control. But the systems are all
0:10:22 fairly done in a sense that they’re not necessarily seeing the world in any way. Level one system
0:10:27 will mean that there’s some level of automation. So adaptive cruise control is an example of
0:10:31 a level one system where typically there’s a radar that’s seeing the vehicles in front
0:10:35 of you on the road and then the vehicle is able to accelerate and apply the brakes automatically.
0:10:39 Level two is where things get pretty interesting. That’s where you typically have a combination
0:10:44 of a lane-keep system with an adaptive cruise system. So for example, the Tesla autopilot
0:10:49 system is a level two system. It’s able to maintain its own lane safely on the highway.
0:10:54 And right now the trend in production systems is automakers are trying to go to what they’re
0:10:59 calling level two plus, which is taking these level two sort of lane-keep plus adaptive
0:11:05 cruise systems, and they’re adding on functionality for automatically taking freeway interchanges.
0:11:09 And so if you can automatically take an exit and then perhaps automatically merge into freeway,
0:11:12 while the human is still behind the wheel and paying attention, that’s called a level
0:11:18 two plus system. That’s a level two plus. Exactly. And so major vendors, for example,
0:11:23 Mobileye, they are now marketing their level two plus systems to OEMs. Level three is sort
0:11:30 of a bit of a dubious classification where it’s essentially saying that the user should
0:11:36 be able to not pay attention, and the system should be able to alert them when they need
0:11:39 to take over. So it’s kind of like a passive driver, a passive human in the loop, not an
0:11:44 active human in the loop. Exactly. The problem with that classification though is it sort
0:11:49 of breaks down at the technical detail level. There are lots of situations where dangerous
0:11:53 things can occur where the system wouldn’t necessarily be able to warn the driver ahead
0:11:58 of time. Within the industry, there’s been hesitation to use that actual classification
0:12:02 of level three. And that’s where really the level two plus classification comes from.
0:12:06 Right. It’s a funny little distinction, but I get it. It’s almost like it’s like one,
0:12:11 two, three, four, five, and then you have like three in the middle of this weird blurry pivot
0:12:16 to quote true autonomy. Exactly. And I think I’ve seen some demos of systems that were
0:12:22 purported to be level three, but actually then in the demos, there were events that
0:12:25 required the driver to take over immediately. So that’s not really a level three. That’s
0:12:29 really a level two system. And then when you get to level four, that’s really where we’re
0:12:35 talking about these fully autonomous robotaxies that have some geographic fence. So for example,
0:12:39 the Waymo pilot in Arizona, that’s a level four system where there’s fully autonomous
0:12:43 vehicles, but only within a certain geographic region. So the geofencing is just like the
0:12:48 physical location of how far it can operate in. So generally, there’s what’s called an
0:12:54 ODD and operational design domain. And that’s the set of capabilities that the car has.
0:13:02 And so as long as the car is within the region of the world where it knows based on the engineers
0:13:05 who worked on the system, where they have good confidence that it’s able to handle all the
0:13:09 situations that can occur, that’s considered within the ODD. And oftentimes that also has
0:13:14 to do with the mapping system that’s on the vehicle. And the weather and time of day.
0:13:18 Is this by the way also where like a lot of these cart robots and delivery robots sit
0:13:22 because they’re only delivering on campuses and constrained spaces? Does that count as
0:13:23 level four?
0:13:28 That’s absolutely a level four system. Because for those, there’s no human operator typically.
0:13:31 And so it is a level four system within the ODD of the robot.
0:13:34 And so level four is fully autonomous in that there is no human in the loop. Or at least
0:13:39 is a human offsite, like not in the car, but maybe monitoring feeds.
0:13:46 So technically, you can have a human in a loop, but the system needs to be able to safely
0:13:50 handle any situation that it can be in for it to be considered level four. And so that
0:13:54 might entail the vehicle pulling off to the side of the road, waiting for a human to do
0:13:57 something. But typically, most of the systems that are considered level four operate the
0:14:02 vast majority of the time fully autonomously. And then the very last level is level five,
0:14:08 which is more of an idea than a reality. It’s the notion that there could be a vehicle that
0:14:12 is able to drive autonomously in all conditions where a human would be able to operate that
0:14:18 vehicle. And the truth is in the industry, no one is even close to that particular goal.
0:14:19 So that’s further off.
0:14:20 That’s quite a bit further off.
0:14:21 Yeah.
0:14:23 Okay. So what we’re talking about here when we’re talking about autonomy in this context
0:14:27 of this podcast, you guys are actually focusing more on level four.
0:14:31 We fundamentally believe that the tools that you’re using to develop your level two systems
0:14:35 should be actually the same tools that you use for three and four. And if you look at
0:14:40 kind of the tooling used to develop these systems, historically the tooling for level
0:14:46 two system, what Peter mentioned earlier, LCC and ACC lane keep and adaptive cruise control,
0:14:51 those were more hardware focused tools. And so they would, the quote unquote simulators
0:14:56 were trying to spoof the hardware that is actually controlling the system. So the radar
0:15:03 or the camera system, they’re literally tools where you actually point the camera that would
0:15:08 be sitting in a car in front of like a monitor. That’s the quote unquote simulation. Now the
0:15:13 fundamental differences you go up the levels is there’s a proliferation of scenarios. There’s
0:15:16 a finite number of scenarios when you’re just going down the highway trying to keep a lane
0:15:21 and a certain distance when you’re in an intersection with, you know, four lane intersection with
0:15:25 multiple agents, all those agents can behave in many, many different ways. And the vehicle
0:15:31 needs to be able to understand and then navigate in that environment. And so we build tools
0:15:35 that not only start with a level two but then take, you know, take development all the way
0:15:41 to level four. Interesting. So you drew the analogy earlier about the mobile analogy,
0:15:43 but where does that apply? And where does that fall apart? Because the one, a couple
0:15:48 of differences, all right, I would argue here are one that with mobile, we knew there would
0:15:52 be some application, but there’s been a lot of second and third order, you know, applications
0:15:56 that no one could have predicted or maybe would not have known that selfies would be
0:16:00 such a big deal or social would be so, you know, powerful. They might have thought it
0:16:06 might, transactions, commerce, I think people predicted. So that’s one thing. So with autonomy,
0:16:09 it feels like it’s the other way around where actually I think people do know what a lot
0:16:12 of these things could be. Of course, there’ll be second and third order effects. One of
0:16:16 our partners Frank Chen did a whole series on the, you know, second order effects of
0:16:17 autonomy.
0:16:18 Real estate. Yeah, exactly.
0:16:22 Insurance. How does it change? You know, infrastructure. I didn’t op-ed when I was at Wired on folks
0:16:25 from Autodesk that we’re thinking about the future of infrastructure because you have
0:16:26 cars out there today.
0:16:28 Exactly. So we have an analog.
0:16:32 You have an analog, which mobile, you had computers, but they’re kind of fundamentally
0:16:33 different.
0:16:35 People didn’t even believe that they could even handle the constraints in this way because
0:16:39 that completely changes the design. And here we are talking about cars still look like
0:16:44 cars for the most part. And yeah, Google’s cars could be a little Kauai-like and cutesy
0:16:47 and Waymo’s and all the other ones have different looks and feels, but overall, they look like
0:16:48 cars.
0:16:50 I think that’s where we’re getting to the edge. The interesting stuff happens beyond
0:16:58 that we can draw the Instagram analogy now because we’re in 2019 and not 2005. In 2029
0:17:02 or 2039, we’ll be able to say, well, it was actually in hindsight, it was so obvious that
0:17:07 there’d be people who are living maybe in autonomy or some more unique and crazy things or implications
0:17:11 that we just don’t have right now. I think sci-fi, I’m a big fan of sci-fi. And I think
0:17:17 our imagination only goes so far and there will without a doubt be autonomy applications,
0:17:18 which we’re just not thinking of.
0:17:19 I agree.
0:17:22 I think it was in the sci-fi world, I think it was William Gibson, I can’t remember who
0:17:25 said that quote about, you know, don’t predict the car of the future, predict the traffic
0:17:27 jam of the future or whatever that is.
0:17:32 Or the iRobots scene, right, where I think Will Smith jumps in and he says, I want to
0:17:35 drive this manually. So are you crazy? Are you going to drive this manually? Like that
0:17:44 will become a norm. We always get caught up of will that be 2025, 2029, 2035, 2045? I’m
0:17:48 less concerned about the preciseness of when that date will come, but that will happen.
0:17:49 Yeah. You’re just saying it’s inevitable.
0:17:54 It’s inevitable because of the kind of the three prongs of, you know, of new products,
0:17:58 which is cost convenience and safety. And guess what autonomy gives you all three of
0:18:02 those things. It’s cheaper, it’s safer, and it’s more convenient.
0:18:06 And safer in the sense of accidents overall, right now, focus on the outlier incidents,
0:18:09 which are real and we have to worry about them, but we’re not there yet.
0:18:14 Exactly. I mean, again, the mobile analogy is relevant here. In 2005, you’d see those
0:18:18 bumper stickers, you know, get off your phone. And now if you get in the car and somebody
0:18:22 is in your phone, there’s like, what are you crazy? How are you? It’s like the opposite
0:18:26 because it’s mapping and all these other things that you wouldn’t have thought of when you
0:18:27 had the Motorola Razer.
0:18:30 Well, since we’re talking about right now, and I agree that we don’t know what we don’t
0:18:35 know, who are the players in the ecosystem right now? Like I can guess some of the obvious
0:18:41 ones like the manufacturers of cars, the mapping companies, the mobilize that, you know, supply
0:18:45 components and sensors, like how would you break down the taxonomy of the players?
0:18:50 I think the automotive industry is a good analog to some degree of what I think the
0:18:55 autonomy industry will be. You’ll have end consumer facing companies who will have brands
0:19:00 that interface with the consumer. Now, whether those are ride sharing companies, AV providers
0:19:05 or the continue to be the BMW or the Tesla’s, I think that’s up for debate. Then you’ll
0:19:10 have folks who are supplying services. Right now in the automotive business services
0:19:15 quote unquote are the dealer services. But in the autonomy world, we always talk about
0:19:19 and is the emergence of the software car. And so in the software car, those services
0:19:24 are much more, they look like kind of your phone. I think that seems fairly obvious because
0:19:29 you see some of those already. CarPlay and Android Auto are early indications of that.
0:19:33 And then you have the thing that you can call the infrastructure companies. Just like you
0:19:38 have in phones and in the web, there’s this, you know, every time you go to San Jose, you
0:19:43 see these office parks of companies you’ve never heard of. And you wonder, why do they
0:19:48 have 10 glass buildings? Yeah, totally. And they’ll be those companies and they’ll exist
0:19:53 in automotive as they exist right now. They’re, I mean, people, you know, Forci and Magna,
0:19:57 these are becoming more known in the valley. When I worked at Bosch before, Bosch was unknown
0:20:02 just a few years ago. And it’s finally now because of autonomy, Bosch is like a relevant
0:20:07 name. And so I think the ecosystem will be like that. Each of the things that you have
0:20:12 in mobile and web or more accurately automotive will continue to exist just in different shapes
0:20:16 and shapes and forms because the change is pretty significant.
0:20:19 What we’re talking about autonomy, the human driver becoming a software product, but you
0:20:23 also have the internal combustion engine becoming an electric drivetrain.
0:20:27 And so an electric drivetrain, for example, that doesn’t just impact the propulsion system,
0:20:31 but it actually impacts every other component in the vehicle. For example, the cooling system,
0:20:35 an air conditioner, that’s on a typical gas car is going to be different from an air conditioner
0:20:37 system that’s on, it’s on an electric car.
0:20:43 Yeah. I mean, I have a Prius, which is nowhere near autonomous, but it is electronic partially.
0:20:47 And I have to say, it was like a huge like mindset chef for me to even realize like,
0:20:51 oh, all those tips about how to check your coolant and open your hood in case of an emergency
0:20:56 before triple A comes like, they don’t apply anymore. The mere fact of pushing a button
0:20:59 to turn it on instead of using a key, like, I mean, those are really mundane examples,
0:21:03 but it’s an example of what you’re talking about, which is like a changes everything.
0:21:04 Things you don’t even think about.
0:21:06 All these great revolutions are very mundane.
0:21:09 Yes. I like that concept, actually, because I think about that even in terms of self-improvement
0:21:13 in your life. Like it’s always like at the mundane level that the real shit happens.
0:21:18 Yeah. Day to day, nothing looks different, but when you reflect, you know, five, 10 years,
0:21:20 it’s pretty significant.
0:21:26 I think in front of the old and new and zooming into just autonomy, you have this rich universe
0:21:31 of companies now that are either form forming or are quite mature, that are doing individual
0:21:35 components. So you have sensor companies that Peter mentioned earlier, you have mapping
0:21:39 companies, you have companies like us, infrastructure companies.
0:21:42 Do you guys would categorize yourself as infrastructure?
0:21:46 Yeah. I think that’s probably the most accurate term. What’s different about simulation in
0:21:51 the past versus simulation today, is simulation in the past was usually used to build hardware
0:21:55 products and we’re using simulation to build a software product.
0:21:57 That’s actually really interesting. Let’s pause on that for a moment. I love talking
0:22:00 about simulation on this podcast, actually, in general, because to me, to your earlier
0:22:06 point about virtual worlds, it reminds me of one of my old edities concepts of mirror
0:22:10 worlds, David Galerinter. And this idea that you can essentially turn everything into something
0:22:15 that can be in a virtual system. And that is, I think, what you mean by virtual world
0:22:19 as opposed to quote, you know, VR virtual world like only immersive. And so this idea
0:22:25 that you can essentially softwareify everything, that’s pretty significant. So that swap that
0:22:28 you’re talking about that before we would use simulation to build hardware, now we’re
0:22:32 using simulation to build software. Let’s talk a little bit more about that.
0:22:37 Yeah. Simulation is not new to automotive or aerospace. These methodologies that existed
0:22:44 for decades and even longer than that, you would develop a product, let’s say a turbine,
0:22:49 and then you would manufacture, you develop a bridge, and then you would build it. Using
0:22:53 software simulation is different because you have these products that are out there in
0:22:57 the real world and they’re going to continue to inform the thing that you’re developing
0:23:01 in the simulator world. And so this connection of, it’s almost like reality in the loop.
0:23:04 It’s a little feedback loop, but you’re right. Reality in the loop is a more significant
0:23:10 wave. It’s less linear. X creates Y, Y creates Z, Z goes and influences X and it’s a nonlinear
0:23:16 circle. And because of that, we’re more infrastructure than purely simulation because, like for instance,
0:23:19 if you’re managing large amounts of data, is that really simulation? Technically, it’s
0:23:25 not. But you need to do that in order to make your simulations useful. Connecting to the
0:23:29 car, or is that part of simulation? No. So that’s why I think where the larger umbrella
0:23:35 is infrastructure. You could say HD mapping is really an infrastructure play. Those are
0:23:40 the rails of which the train rides. It’s also infrastructure in the sense that it’s used
0:23:45 continuously on an ongoing basis, whereas the traditional forms of simulation were typically
0:23:51 used sort of for this big, big moment, which is the creation of this final hardware specification
0:23:55 which is then going to be made. It’s shipped. It is delivered. It is done. This is never
0:23:58 done. You guys, I mean, it’s a terrible analogy, but it’s a little bit like a Kanye album.
0:24:03 It’s continuing to evolve in the wild. And after it’s dropped, it’s going to keep getting
0:24:10 modified. It’s a real life goal I had of comparing my simulation company to… Well, we’re all
0:24:15 fans of music. So using your analogy of trains, because you mentioned the train tracks. So
0:24:19 this is interesting because what we’re really describing here is laying down the tracks while
0:24:22 also inventing the train itself. And the two things are kind of like moving targets against
0:24:27 each other, et cetera. So what does that mean for the evolution of the ecosystem? I think
0:24:31 overall, there is a co-evolution of sorts that happens between each of the different
0:24:36 components involved. And so, for example, sensor companies and mapping companies, ensuring
0:24:40 that the latest advancements that they have in their own products are then accurately
0:24:44 represented inside of simulation. It’s like the phone supply chain we’re talking about
0:24:50 earlier. So all these little revolutions and miracles happening. And the untold story that
0:24:57 hasn’t been discussed is if you’re building autonomy yourself, what’s the right path?
0:25:01 Is the path to go vertical and build everything yourself? Or is the path to buy things off
0:25:08 the shelf? Where in this ecosystem, where do you draw that line of what is critical for
0:25:17 autonomy, quote-unquote, and what is not? And so my rough view, and I would say largely,
0:25:22 what is not differentiated between the companies, the mapping companies included, you should
0:25:27 basically buy off the shelf. That’s commoditized. And you should be differentiating elsewhere.
0:25:30 Because mapping companies are sensor companies, which are all kind of have the same role in
0:25:37 different ways. We’re spreading our R&D costs across 10, 20, 30, 50 players. And therefore,
0:25:42 each individual player gets a more advanced product for a cheaper cost. And that’s capitalism,
0:25:47 right? You’re driving market efficiency. When people talk about a new industry, we’ll drive
0:25:51 market efficiencies like tactically. How does it happen? It happens where you have individual
0:25:57 players who are now unbundling the cost onto a bunch of people, a bunch of different companies,
0:26:02 and then those people who are participating almost in that consortium are getting the benefit
0:26:05 of it. Now, that doesn’t mean you can go ahead and you can definitely go and do that
0:26:08 at a vertical company. There will always be an apple in every ecosystem.
0:26:15 Yeah, exactly. But you better be Steve Jobs. Right, exactly. And what I love about what
0:26:18 you’re describing, and this is capitalism, it’s funny because we might as well say cloud
0:26:24 is capitalism at this point to make that syllogism. But it is the AWS moment in this ecosystem.
0:26:29 And it’s talking about the fact that you can actually then free a whole new wave of companies
0:26:34 to do things. I do find that very fascinating because until now, I would have thought that
0:26:38 autonomy is only for like the big, the big, big five car companies.
0:26:42 So the AWS example is exactly right. You can roll your own server. Some people have pride
0:26:47 in running their website off their local, but guess what? Your consumers actually don’t
0:26:53 care if you’re running on-prem, AWS, GCP, Azure or whatever.
0:26:57 They just want the service. They just want the service. And so by the way, this happened
0:27:01 in automotive. Automotive started, you know, Alfred P. Sloan.
0:27:03 Wait, who’s Alfred P. Sloan? I don’t even know who that is.
0:27:05 Oh, okay. So they are early automotive pioneers.
0:27:07 Oh, I would have thought it was Henry Ford.
0:27:12 Sloan was, so he’s not technically the founder of General Motors, but Sloan and Kettering
0:27:17 were essentially the leaders of General Motors. GM was founded by a person named William Durant,
0:27:21 who had started another car company. The amazing thing about the AV industry today is it’s
0:27:25 almost copy and paste of the automotive industry a hundred years ago because you have these
0:27:30 individual personalities who are shaping companies in their own way. Some get fired, they start
0:27:34 competing companies. There’s all this drama between the existing players who are coming
0:27:38 in. There’s a lot of M&A activity happening.
0:27:42 Oh, this is one of my favorite things when we talk about how software and tech evolution
0:27:46 is taking you back to an earlier era. That’s one of my favorite themes ever.
0:27:52 The Sloans of the world and the Henry Fords of the world, they wrote and they tried to
0:27:58 build vertical companies. I mean, Ford used to do everything. They used to get rubber
0:28:04 from plants. They would forge steel. I think they even owned the farms where things were
0:28:09 grown. So guess what? We don’t do that. Why? Because it’s actually more efficient to have
0:28:12 a supplier ecosystem. Well, that’s like capitalism to the T. I mean,
0:28:17 that’s like the classic, you don’t want to, someone did an experiment where they tried
0:28:21 making their own sandwich from scratch. If they grew the vegetables, I think they had
0:28:26 to outsource the cheat. They take the cows and the cheese and I think they estimated it
0:28:31 to be like over almost $2,000 and capitalism makes that sandwich $7.
0:28:37 So try doing that for like a computer, something that’s more manageable. You can go on YouTube
0:28:40 and watch videos about people trying to build their own phones. They end up just going to
0:28:43 China and buying for a bunch of suppliers because that’s actually the faster way to
0:28:49 do it. And the ecosystem conversation that’s happening every single day in these autonomy
0:28:55 teams is, oh, wow, we don’t have that many engineers. Oh, wow, there’s another huge pilot
0:29:01 that somebody has announced and how can we move faster? One of the easy rules of thumbs
0:29:07 of you can see how sophisticated and AV leadership is just asking them, where’s that line? And
0:29:12 that line, that circle of competence should be as small as possible. That small circle
0:29:16 in autonomy is algorithms. That’s the coveted golden nugget.
0:29:19 Takes confidence to focus, narrow laser focus like that.
0:29:24 So you can go to like a completely different industry, go to consumer CPG or you can go
0:29:30 to consulting. If McKinsey or a, you know, Unilever or whoever it is will very clearly
0:29:35 say, hey, you know what, this is the thing, this is the hill we die on. This hill, we
0:29:38 have to be better than everybody else. The only way we win this hill is we abandon every
0:29:39 other hill.
0:29:43 Right. Well, this begs the question and Benedict often asked a similar question in his post
0:29:48 on autonomy a lot, which is, you know, will Tesla become more like Detroit? Is Detroit
0:29:53 more likely to acquire the Silicon Valley mindset faster or is Silicon Valley going
0:29:56 to move faster in sort of learning the skills of Detroit?
0:30:00 I think there’s no path to autonomy that doesn’t go through Silicon Valley and Detroit.
0:30:01 So it’s an and not a word.
0:30:04 And when you say Detroit, we mean roughly the automotive centers. Jit guard included.
0:30:08 I mean, the Japan and Korea and China included in that.
0:30:12 Right. You don’t mean Detroit geographically, you mean the entire category of automotive.
0:30:13 Detroit is the concept of automotive.
0:30:14 Right.
0:30:20 Yeah. As a second hand for the automotive industry because Detroit has the delivery mechanisms,
0:30:27 which are the brands and the factories which build these vehicles and the channel for lack
0:30:28 of better words.
0:30:33 This is not an internet product. The channel is not a website. The channel is the traditional
0:30:40 OEM business. But the thing that you’re distributing through this channel is almost ideally built
0:30:43 in Silicon Valley. Again, we’re talking about that circle of competence and how small you
0:30:44 can make it.
0:30:50 So we’re Silicon Valley, I think strays is when we start doing things, which frankly
0:30:56 speaking, are outside this very small circle of software. And I get a little nervous when
0:31:01 companies are doing a lot of hardware because there are other hardware centers in the world
0:31:06 which are arguably better or when even broadly like on podcast, people start talking about
0:31:08 like these other things.
0:31:13 And it’s like, if we went to some group of factory owners who I don’t know are specialists
0:31:20 and they don’t get on podcasts and then start advocating about things outside of their little
0:31:26 circle of competence, they talk about a leather price and how are you getting cheaper electricity?
0:31:30 I hear you. It’s both. It’s both arrogant and charming at the same time.
0:31:31 Exactly.
0:31:37 But it’s good because it pushes you to go into trying new things, agree. And so the magic
0:31:42 happens where you’re pushing trying new things in your area of competence. I can go and try
0:31:47 to be an NBA basketball player, but guess what? It’s probably not going to work no matter
0:31:52 how much effort and, you know, I try, I put into, but so I think there’s a similar, you
0:31:58 know, relationship between Detroit and Silicon Valley. There is a real merger. And my background,
0:32:01 you know, both Peter and I, we grew up in a group in Detroit area.
0:32:02 I had no idea. You guys grew up in Detroit?
0:32:06 Yeah. Of all the random coincidence, not only we grew up in the same town, we grew up in
0:32:11 the same subdivision. We’re literally at the same crossroads for people who are in Detroit
0:32:16 as 22 and Shayner and Shelby Township. I went to GMI or now Kettering University, which
0:32:18 is the General Motors Institute. Peter went to U of M.
0:32:22 So I actually started my career at a small engineering tool company in Michigan, but
0:32:23 really my entire family works on a motive.
0:32:25 So you guys are like Detroit, born and bred?
0:32:29 Yeah. I worked five years at General Motors, two years at Bosch. And then we’re in the
0:32:33 same team on Google Maps. This is five, eight years, seven years ago, a long time ago. And
0:32:39 we saw chauffeur, which was, which became Waymo. And I remember saying to Peter, man,
0:32:42 this is going to hit Detroit like a ton of bricks.
0:32:46 Kettering is located in Flint, Flint, Michigan. So I spent five years at Flint. And when you
0:32:52 look at places like Flint, you really start thinking long and hard about like, well, where
0:32:56 do people get in these new jobs? That was the theme in the 90s and the early 2000s when
0:33:00 I was growing up was, oh, there’s going to be this revolution and all these people in
0:33:03 Michigan are so suddenly going to have these great new jobs. And guess what? My family
0:33:09 included those jobs didn’t come. My dad never became a software engineer in his late 50s.
0:33:13 That doesn’t happen. And then also I think any business in the human experience is emotional
0:33:18 to some degree. I mean, we very much like practice that belief that there is this connection
0:33:19 between these two.
0:33:23 You guys are really, we’re really long on the Detroit Silicon Valley and not the ore.
0:33:28 So what do you think then that the winning company, maybe it’s not a winning company,
0:33:32 there’s plenty of room for many, but where is it going to sit? And how is it going to
0:33:33 look?
0:33:39 Well, it’s like, you know, where does the winning automotive player today sit? I think
0:33:40 it’s very hard to answer that question.
0:33:43 There’s at least a few in every major geography.
0:33:48 And the supply chain, which is really what the auto business is, is everywhere. These
0:33:53 are such massive industries. They have epicenters. So I think the autonomy software stack will
0:33:59 probably for a long time be in Silicon Valley, but even you can look at like TRIAD, the Toyota
0:34:05 Research Institute’s Autonomous Division based in Tokyo. You have other companies, BMW,
0:34:10 even TRIAD actually has presence here in the valleys as well in the Detroit area.
0:34:15 So I think this concept of like, there’s a company that wins it for a town, I think
0:34:19 that’s different. I think we sometimes get that analog because of the internet where
0:34:22 you have Google, which is basically home home, home team, which is Mountain View.
0:34:23 Yeah.
0:34:27 A lot of the companies are like Silicon Valley and Seattle and there’s like a few centers
0:34:28 that are very focused.
0:34:33 I think these large industries that are very intertwined with each other, it’s a lot less
0:34:39 concentrated like that. I think the real fundamental issue we have, and this is getting more philosophical
0:34:44 again is what the internet has done and what software has done is it’s concentrated wealth.
0:34:48 We talk about wealth concentration as like somehow blaming sometimes, you know, a certain
0:34:53 political viewpoint, but really they’re so efficient software companies that does bring
0:35:00 a disproportionate amount of money to where the epicenter is. And so how can we make sure
0:35:09 that concentration, you know, that the next wave, which is autonomy, doesn’t keep just
0:35:12 kind of underlying that. One of the other things that is not talked a lot about autonomy
0:35:18 but should be talked about autonomy is that these are national questions. The German government
0:35:24 won’t just let Waymo come take over Germany and let Daimler and BMW go under business.
0:35:29 And the same thing is true for Hyundai in Korea, Hyundai and Toyota in Japan and the Chinese
0:35:32 companies because there’s a recognition that if all of these cash flows end up going to
0:35:36 these little neighborhoods in the suburbs of San Francisco, maybe that’s not good for
0:35:43 our national interest. In the internet, because it was a new market, it wasn’t very visceral.
0:35:48 Daimler is a visceral, Bosch is a German thing. Peugeot is a French thing.
0:35:53 It’s like the classic discussions around manufacturing, like, you know, this idea that like it’s a
0:35:58 physical product that is made in India, made in China, made in Japan, made in Italy, made
0:36:04 in Italy. You know, it’s very specific and you’re right, there is a very national sentiment.
0:36:07 But what I love about what you’re describing too, though, is it is true capitalism because
0:36:12 I think capitalism gets a bad rap for the inequality, which is a fair complaint and
0:36:17 a fair criticism. But to me, true capitalism is something that raises the all boats in
0:36:18 the ocean.
0:36:23 Yeah. It’s on Pakistani, by birth on Pakistan. My family were from a small farming village
0:36:27 for the first, you know, seven, rough seven years of my life. I was in this, you know,
0:36:31 in this remote farming village in the, roughly in this valley. And you know what’s a real
0:36:35 luxury? Hot showers. Yeah. And so, you know, when I, when I had that hot shower in the
0:36:37 morning and I drank that cold water.
0:36:38 You don’t take it for granted.
0:36:39 Capitalism.
0:36:43 I know. I go the same way about electricity. I mean, my dad, I was born and raised here,
0:36:47 but my dad’s from India, small village, and he grew up without electricity. And then he
0:36:51 later got electricity and I was just marveling just very recently in our families at the
0:36:58 valley that we had electric BIAS. That’s insane. Like before electricity was not even available
0:37:03 to people. And now you have mass produced little tiny LED lights and like little bases
0:37:04 as candles.
0:37:05 That’s fricking amazing.
0:37:06 Exactly.
0:37:07 No, I agree.
0:37:08 So on that front.
0:37:13 I think where autonomy is different is I think it has the potential. And I think whether
0:37:17 we like it or not, there is a regulatory aspect to this entire conversation.
0:37:20 So I have a question about this because you brought up the point about there being a national
0:37:25 interest. There’s also a local city and state level of interest. Mark wrote an op-ed a few
0:37:30 years ago in Politico, arguing that you can use it for a regulatory arbitrage where like
0:37:34 say, Detroit could actually loosen some of the barriers. Just like, you know, I think
0:37:38 Governor Ducey is doing in Arizona where you have different cities offering different
0:37:42 incentives and doing more experiments so that they can ensure the ecosystem kind of grows
0:37:47 up locally. How is that really happening? Given your thesis, it sounds like you’re saying
0:37:52 that everything can happen everywhere and there’s room for all kinds of players and
0:37:57 B, what do you see as sort of the regulatory and policy issues in the autonomy ecosystem?
0:38:00 Yeah, I think everything can happen everywhere is more of this concept of there are so many
0:38:04 components and these components will come from everywhere. One of my friends who’s Indian
0:38:09 who was wished that states themselves in India would have more of a control over their own
0:38:14 laws because he believes that within the US, the states creating their own regulatory, many
0:38:20 regulatory environments is almost like a mini form of capitalism. It’s a grand laboratory
0:38:25 of capitalism actually. States are laboratories of innovation. There’s cities are to that
0:38:31 sort of federalist style. I think it was an enabling condition for success, not a bug.
0:38:36 And it’s a great feature because the state like Indiana says, Hey, listen, maybe this
0:38:40 is in our best interest because we’re a state that trucks go through and want to make sure
0:38:44 we make that toll, toll income. But if you’re a state like Arizona, and maybe you don’t
0:38:49 have that and you have this great testing ground, historically Arizona belong before
0:38:54 autonomy has been approving grounds for the auto business, that, Hey, we see that, you
0:38:58 know, the Arizona approving grounds for General Motors brought all of this, you know, business
0:38:59 over the last 20 years.
0:39:02 Right. Why can’t we do the same for this? It takes a lot of courage, by the way, because
0:39:06 they did have, I think, the first instance of a fatality through autonomy. So when we
0:39:11 do talk about these states sort of taking the leap, there is sort of a cost you pay.
0:39:15 Because in the case of Arizona, I think they were the first to have the first fatality
0:39:19 related to autonomy. And of course that’s going to happen. I’m not trying to minimize
0:39:24 it. That’s a really big deal. But that is, I think one of the trade offs is that cost.
0:39:29 I think the states that are making those decisions are opening some of them and their citizens
0:39:34 to that risk. And so the citizens then elect those, you know, those representatives who
0:39:39 then say, Hey, this is or is not the trade off that I want to have.
0:39:44 That matches their needs. And I think, I think though, probably what again doesn’t get covered
0:39:49 is I think that night, there were also 10 other pedestrian accidents in America where
0:39:50 people died.
0:39:54 You’re right. There is a statistical thing, which is hard to think about when you’re
0:39:56 talking at a personal level.
0:40:01 It is tough because that’s a real family. And if you’re that person, you don’t care.
0:40:06 There’s 10 other 11 other people that died that night. Also, these are the guardrails
0:40:13 we roughly think are ones that can be employed. You don’t want to have complete laissez-faire
0:40:16 open. Everybody does whatever and
0:40:21 Pure permissionless innovation. We’re talking about moving, killing robots.
0:40:25 This is like a human being. Like they’re like one of the things we don’t think about in
0:40:30 Silicon Valley. A lot of the, a lot of times engineers in the auto business over the last
0:40:32 years have gone to prison.
0:40:33 I had no idea.
0:40:39 Just a Volkswagen diesel scandal that put employees of Volkswagen in prison. And so there is real
0:40:45 consequences when you’re dealing with a product, which an automotive product, which can harm
0:40:46 the public.
0:40:51 The other end though, is if you put in regulations and, and they’re onerous and they’re significant,
0:40:52 guess what?
0:40:53 Squelch is innovation.
0:40:59 100%. And we are, we’re talking this, this, this conversation has been very US centric.
0:41:03 In 1980 capitalism was a very, you know, regional thing. The real revolution that’s happened
0:41:07 in the, in the last, you know, 30, 40 years is a capitalist revolution. It’s just the
0:41:10 shade of capitalism. And so when you think about China, which is a different shade of
0:41:13 capitalism, you think about Europe, which is a different shade of capitalism, generally
0:41:18 different approaches. But, you know, some, some, some regulatory environments are very
0:41:24 open and we have to be aware of that for not only the Silicon Valley and Detroit companies,
0:41:29 but just in general as Americans of being in an economy which is healthy and productive
0:41:34 and at the cutting edge. But at the same time, you as citizens, you don’t want to be a laboratory
0:41:42 for private entities to make a profit. And so there is a very nuanced approach there.
0:41:47 At the end of the day, we’re really advocates of best practices for safe development. And
0:41:53 so that means really taking the steps necessary to ensure that the systems, the software are
0:41:58 safe before they actually go to the public. Yeah. You’re talking about simulation here.
0:42:03 We’ve talked about simulation earlier in terms of the industry evolution, but simulation
0:42:06 itself got a bad rap for a while. You know, there’s a lot of companies that sort of felt
0:42:10 like, oh my God, simulation, it had a bad rap for a while. It’s like a, you know, I think
0:42:16 trying to do VR and AR to some degree in the 1980s. And so I think simulation, which has
0:42:21 been different than AR VR is there are no complex systems that are being developed without
0:42:28 simulation aircraft, military systems, automotive, internal combustion engines, microprocessors,
0:42:36 simulation is everywhere. And so I think that’s because the underlying kind of software industry
0:42:43 has become so much more advanced. It’s computationally more efficient. You can apply, you have things
0:42:48 like the cloud revolution, the ability to point lots and lots of resources at the problems.
0:42:52 What to say is shifted from constrained to abundant. And that essentially creates abundant
0:42:58 sensors, abundant data. You can waste bits. You can essentially simulate complex things
0:43:03 unbounded in a way that humans can’t even remotely conceive of. That does answer the
0:43:08 why now question. What are the limits of simulation? I mean, we are talking about complex systems
0:43:13 on a ton of edge cases here. Yeah. I mean, at the more technical level,
0:43:18 simulations are never perfect. There’s always going to be some difference between a simulation
0:43:22 and a real physical system. We like to get our simulations to the point where they are
0:43:26 plenty good enough for useful development. Good enough for development. Good enough for
0:43:30 development. Good enough for pushing things forward and to give a very high confidence
0:43:35 that the behaviors and simulation are representative of the real behaviors. But with that said,
0:43:40 there will always be situations and scenarios where there are differences in behavior between
0:43:43 the simulation environment and the real environment. Of course. Right. Okay. So what’s also interesting
0:43:47 about this is that it essentially lets you get the three C’s that you described earlier,
0:43:51 cost, convenience and safety in one system. And to the regulatory point that you brought
0:43:56 up, Casar, it is unless you kind of strike that just right balance in there. But the
0:44:00 big thing now, because you’ve been talking in this podcast about this importance of differentiation,
0:44:05 if this is a tool that everyone has, it sounds like they would differentiate on data. So
0:44:08 how do you in this ecosystem where you’re making this argument that there’s this horizontal
0:44:12 versus vertical layer, are all these players willing to share in the ecosystem, the mapping
0:44:17 companies, the sensor companies, the big vehicle companies? How do you navigate the data side?
0:44:22 Data means a lot of different things. It’s not like scenarios and, you know, the data
0:44:28 that you have for autonomy, but it is the autonomy engineer who themselves are understanding
0:44:34 how are the methodologies to best develop an autonomy system. There is some, what we
0:44:38 call light network effects there between companies. Well, I mean, if you go to Stanford, they
0:44:43 teach classes that help you learn ANSYS’s simulation tools. So there’s literally this
0:44:49 public company called ANSYS that does simulation tools and you can learn how to use it by taking
0:44:53 classes at Stanford. And that’s the same thing with AutoCAD. If you look back, if you look
0:44:58 there, there are many tools that kind of fall into the, into this group. I mean, when you
0:45:01 learn how to program, you’re actually just learning tools. Now, what’s happened with
0:45:06 software development is those tools, it becomes a really just a commodity and there, and there’s
0:45:11 many different ways. And so we’re still in a quite a nascent niche field with autonomy.
0:45:15 So the tools are not a commodity. These tools are so hard to build. These two simulations,
0:45:17 it’s not a trivial thing to build.
0:45:22 At the end of the day, the lowest cost solution will win, but of course it has to be a real
0:45:23 solution.
0:45:24 Yeah, it has to solve something.
0:45:25 And that’s what the industry is still working on.
0:45:29 Right. It’s actually kind of funny because the conundrum here is that software is bits
0:45:34 and it’s abundant and therefore it’s accessible to everybody. But the specialties and the algorithms
0:45:39 I’m clearly hearing and like the nuance of the art, and we used to call it know-how when
0:45:44 I used to be at park. It’s kind of the idea of the know-how and the differentiation. But
0:45:48 the point is, it’s basically going the way of mobile and you’ve been drawing the analogy
0:45:51 and we’ve talked a little bit about where the analogy breaks down and where it applies.
0:45:55 How do you think this plays out given that you are a horizontal player? There aren’t
0:46:01 really big horizontal huge like apples and Googles. There are vertical companies.
0:46:07 Well, they are, but actually, you know, each of the sub components, the phone manufacturers
0:46:12 themselves, companies that do analytics for mobile, companies that do ads for mobile,
0:46:19 a lot of horizontal players, anything that exists both on Android and iOS is in some
0:46:25 way a cross-platform horizontal play. And so I think where the commoditization has happened,
0:46:30 quote unquote, is in the apps themselves. I’m reading this book, The Five Ages of the
0:46:32 Universe, which is The Physics of Eternity. Fascinating.
0:46:40 So what happens at the end of the universe, right? All the stars have now died.
0:46:45 Oh my God, I really need to read this book. This is totally my jam. I’m like really obsessed
0:46:48 with space and evolution right now. Yeah, it can be dry. I find it very interesting.
0:46:55 But the point is, once you get into these outer edges, the strange things start happening.
0:47:02 And so we’re now in that mobile age where there are applications that are gaining users
0:47:07 very, very, very quickly and still not being valuable or some applications that might not
0:47:11 have as many users but can become super, super valuable because they’re catered towards
0:47:18 a very specific audience that needs that thing. And so I think with autonomy, I think the
0:47:23 arc here, you’ll see all these individual modules will be run by individual players
0:47:27 because there is this natural arc in capitalism, which says the independent providers can do
0:47:32 cheaper, better and faster than anybody doing it vertically. But the question is, will the
0:47:36 algorithms themselves ultimately commoditize? And I think that’s when you get into this
0:47:40 far edge of the universe. It’s like, could we be in a situation in 10 or 15 years that
0:47:46 like today, starting a mobile app is very easy. That starting an autonomy company is very
0:47:47 trivial. It’s just that easy.
0:47:52 I think it would be very hard in 2005 to think that Kim Kardashian or whoever would have
0:47:56 their own app and it would make millions and tens of millions of dollars. But that’s the
0:48:01 reality today because that’s so niche. It’s not just, hey, it’s a phone app. It’s a phone
0:48:05 app on a specific platform for a specific celebrity and just their fans because everybody
0:48:10 else can just consume Instagram or something else. And so that real edge, I think that
0:48:15 far off the world of autonomy, 10 or 15 years from now, imagine if you could build an autonomous
0:48:20 vehicle very quickly and very easily. If that could happen, how does that, what does that
0:48:21 make the industry?
0:48:24 Yeah, no, it’s like a theme we talk about actually is that the edge is where it’s at.
0:48:28 I mean, in computing and innovation, I mean, it’s basically the democratization of autonomy.
0:48:31 Well, you guys, thank you for joining the A6NZ podcast.
0:48:32 Thanks for having us.