The Peter Attia Drive - #06 - D.A. Wallach： music, medicine, cancer screening, and disruptive technologies

00:00:00.000 Hey everyone, welcome to the Peter Atiyah Drive. I'm your host, Peter Atiyah.

00:00:10.160 The Drive is a result of my hunger for optimizing performance, health, longevity, critical thinking,

00:00:15.600 along with a few other obsessions along the way. I've spent the last several years working with

00:00:19.840 some of the most successful, top-performing individuals in the world, and this podcast

00:00:23.620 is my attempt to synthesize what I've learned along the way to help you live a higher quality,

00:00:28.360 more fulfilling life. If you enjoy this podcast, you can find more information on today's episode

00:00:33.000 and other topics at peteratiyahmd.com.

00:00:41.180 This podcast, I'll be speaking with my good friend, D.A. Wallach. I've known D.A. for maybe five years

00:00:48.440 now, maybe four, I can't recall, but he is truly a renaissance man. I have been accused of being a

00:00:54.740 renaissance man on occasion, but I am not. D.A., however, is. And if you look up renaissance man

00:01:01.240 in the dictionary, I think you'll just see his picture with his curly hair sitting there. He is

00:01:05.600 a recording artist, a songwriter, an investor, an essayist. He was discovered while an undergrad at

00:01:11.840 Harvard by Pharrell, among others, who signed him to a deal. He went on to become one half of the band

00:01:20.120 Chester French. They released three full-length albums, and he also has a solo album called Time

00:01:26.280 Machine, which was released in 2016. We'll link to all of that stuff. While with Chester French,

00:01:31.920 they toured with a number of legendary bands and artists such as Lady Gaga, Weezer, and perhaps my

00:01:37.140 favorite of them all, Blink-182. Beyond music, however, D.A. is sort of in a class of his own in terms of his

00:01:44.160 intellectual curiosity and his ability to assimilate information that seems so far outside of his area

00:01:51.100 of expertise. In fact, some of the most interesting discussions I remember ever having with D.A. is

00:01:55.700 sort of what prompted this podcast. I remember one day he came over, he was passing through San Diego

00:02:00.420 on his way down from L.A., came by, and we were sitting out at a park on the swings having a discussion

00:02:06.460 about liquid biopsies. And I was thinking to myself, how is it that I'm sitting here with this guy,

00:02:12.020 my buddy, who's a musician and a very good investor, having this discussion about liquid

00:02:17.520 biopsies at a level of detail that I don't get to have with pretty much anybody else outside of people

00:02:22.640 who are knee-deep in this field. And that was sort of when it clicked in my mind. I was like, you know,

00:02:27.300 D.A. would be a great guy to have on the podcast. He's advised a number of startup companies,

00:02:32.080 including SpaceX, Doctor On Demand, Ripple, Emulate. And of course, he was an artist in residence

00:02:37.620 at Spotify. And we talked actually quite a bit about Spotify on this episode for anyone who's

00:02:41.660 kind of interested in how it came to be. The other things that we talk about, of course,

00:02:44.880 is his background in music and his start. My daughter is a great drummer for a little kid, 0.92

00:02:50.700 and I've always been interested in how one can continue to encourage kids to be involved in

00:02:56.240 music. And we talk about some fun times that we've had when he's been over and has jammed with her.

00:03:01.340 We talk a lot about cancer screening, which anybody who's kind of ever heard me talk about this stuff

00:03:06.360 privately. I really think that when it comes to the major metabolic diseases, cardiovascular disease

00:03:12.220 and the other atherosclerotic diseases, cancer and neurodegenerative diseases, the big tool that is

00:03:17.600 really missing is these liquid biopsies. By the time cancer becomes visible on an imaging study,

00:03:23.700 you can make the case you've lost the war. I don't know that that's true, but I do believe 1.00

00:03:29.180 that if we can catch these things when they are not yet fully determined to be cancers based on

00:03:36.040 either looking at a DNA signature, an RNA signature, or even a protein signature, that we might have a

00:03:40.860 shot. We also get into some really kind of nerdy stuff that I think is very important for anybody

00:03:45.740 thinking about screening, such as positive predictive value, negative predictive value, sensitivity

00:03:49.380 and specificity. And we'll link to some information here that we use internally in our practice to

00:03:56.360 help patients navigate that. So if you're interested in music, if you're interested in liquid biopsies,

00:04:03.160 cancer prevention, general cancer screening, and just interested in listening to a really smart dude

00:04:09.140 who seems to know a lot about a lot of things and can speak very articulately about them, I think

00:04:15.060 you'll really enjoy this podcast. You'll be able to find the show notes for this at peteratiamd.com

00:04:21.360 forward slash podcast. And we'll link to a lot of the stuff that we talk about that will hopefully

00:04:26.500 allow those of you who are interested to follow up and learn a little bit more. So without further

00:04:30.220 delay, here is my conversation with the amazing DA Wallet. DA. Peter. Thanks for having me over to

00:04:40.780 your lovely place. You're welcome. You're welcome here anytime. I like how you saw that I was in the

00:04:46.720 driveway before I got here and I was kind of just hanging out. Well, we have this thing called the

00:04:50.080 doorbird, which is kind of the evolved version of a ring. And you, since you're interested in all

00:04:58.500 esoteric technical things, would be interested to know that this is the only web-enabled cloud

00:05:06.600 recording doorbell system that can hook into a electric strike. A strike being the thing that opens a

00:05:16.360 gate remotely. And so I'm able to see who rings the doorbell. And then if I want to let them in,

00:05:23.700 press a button from the same app that opens the gate.

00:05:27.320 Remarkable technology.

00:05:28.580 Yes.

00:05:29.420 Technology, as Ali G would say.

00:05:31.160 Yes.

00:05:32.720 So we have known each other. I don't even remember how we met. Actually, I think we met through Gary

00:05:37.120 Taubes, didn't we?

00:05:37.680 We met through Gary Taubes and I met Gary Taubes because I cold emailed him, which is how most things

00:05:43.040 started my life. And I cold emailed him because I became sort of obsessed with obesity and nutrition

00:05:50.100 research. This was maybe six years ago or thereabouts, six, seven years ago. And then I had

00:05:57.640 gotten coffee with Gary up in Berkeley and he thought it would be worth our meeting. And now I

00:06:04.520 talk to you more than I talk to Gary.

00:06:06.380 Yeah. No, it became a, it was a love at first sight actually. And you know, one thing that's really

00:06:11.040 funny, we're going to talk so much about sort of your musical career and things like that, but

00:06:14.940 I will forever be grateful to that one night that you and Adam were over for dinner. This was just

00:06:20.660 after my daughter, Olivia started to play the drums and you guys got up. Adam's played the piano.

00:06:27.100 You played the drums. Olivia then played the drums. And it was really exciting to see. She got to see

00:06:32.320 in action what like improvised music can look like. And I really think that that's part of the reason she

00:06:37.680 still loves the drums. Well, that's good. I mean, part of the drums that's fun is that

00:06:41.720 you don't necessarily need to know the musical material as well as other instrumentalists do

00:06:47.280 in order to play along with people. You know, you don't need to sort of learn the song. You can kind

00:06:52.240 of, as long as you can learn the beat of the song or figure it out quickly, you can play, which is

00:06:58.740 as someone who doesn't necessarily love practicing something that's always drawn me to drums.

00:07:02.920 And I remember her teacher when she was five and I started saying, you know, it's going to be really

00:07:07.960 hard because the music's really hard to read. So the only shot she'll have it starting this young

00:07:12.900 is if she sort of has an intuitive feel for the music, in which case she can get by on that until

00:07:18.360 she learns to actually figure out that, you know, two sixteenths is actually an eighth and that kind

00:07:22.640 of stuff. Right. Well, that's a good point. And I think the best way to learn instruments or to

00:07:29.020 learn music in general is kind of to start without any framework, play around and explore yourself

00:07:36.280 and then learn a little theory because what you don't want to do is become imprisoned by theory,

00:07:42.580 but it does answer some important questions that you'll arrive at if you allow yourself to get lost

00:07:49.060 in the first place. And so, uh, I've always said that, you know, when we have kids, my vision for

00:07:55.700 piano training would be, you just have to sit there for half an hour every day and you can do

00:08:01.480 whatever you want. You don't have to touch the piano if you don't want, but of course anyone's

00:08:04.740 sitting at a piano for 30 minutes will, and you can figure out how it works. And then theory,

00:08:10.680 if you have spent, you know, tens of hours messing around is like an amazing gift because it goes,

00:08:19.140 oh, okay, well that makes sense. That's how it works. It's just like, if you were trying to

00:08:22.960 reinvent mathematics with no orientation. Like Ramanujan. Yes. Like Ramanujan, which,

00:08:30.700 you know, I'm not. So you're almost a Ramanujan of music. I wish. So speaking of which, how did you,

00:08:37.560 what was your, how did you get started in music? Were you always musical as a child? Were you playing

00:08:41.240 instruments when you were young? I don't think I was particularly musical. I always liked listening

00:08:46.880 to music. And my dad would take me to jazz shows occasionally when I was young, which sort of

00:08:52.560 infected me with an interest in jazz. The first instrument I tried playing was the saxophone

00:08:57.600 because I thought it looked cool. And I remember we rented one. This was in the middle school band.

00:09:02.580 So it might've been fifth grade. This would have been an alto sax or a tenor. I don't even remember.

00:09:07.860 And in any event, we rented it. I brought it home and I tried to learn how to just make a sound with it,

00:09:13.220 which is not trivial because with a reed, you have to purse your lips in a particular way and

00:09:18.560 all this. And I was so frustrated in the first hour of trying to play the saxophone that I gave up

00:09:25.340 and became a drummer and then played drums throughout middle school and high school.

00:09:32.400 And I had, had sort of bands with high school friends and that sort of thing. And then in college

00:09:37.120 became a singer, which was something I had never done. Oh, so I didn't realize that you had never

00:09:41.320 sung until you got to college. I had never sung. And I met some cool guys in the dining room at my

00:09:49.180 college. And then they said, do you want to try out for this band we're thinking about starting? And I

00:09:54.140 said, sure. And I tried out as the drummer, but I got beaten by my friend Damien, who then became our

00:10:01.860 drummer. And as a consolation prize, they asked me if I wanted to be the singer. And I said, well,

00:10:08.040 I've never sung before, but I'll try. And I've been trying ever since.

00:10:13.160 So I didn't realize that. So you went into what became Chester French as the drummer.

00:10:17.840 Trying to be the drummer. And then Damien, who became the drummer, later quit and became a

00:10:23.880 filmmaker and is now won like 10 Academy Awards. He did Whiplash and then he did La La Land, which I have a

00:10:31.940 small cameo in. But Whiplash, if you've seen it, is about a drummer and it's somewhat autobiographical

00:10:38.640 about Damien. Oh, I didn't that I didn't realize. I mean, Whiplash is there's only probably five

00:10:45.200 movies I have stored on my iPad because, you know, it's just they take up a lot of room. Right. So

00:10:50.560 but the five that I have are like such that if I'm on an airplane and everything goes to hell in a

00:10:54.920 handbasket and the Wi-Fi is broken and I don't feel like doing work. Boom, boom, boom. And Whiplash is one of

00:10:59.420 those five. It's great, which means it's a movie I've seen more times than I can count. But I

00:11:03.600 especially like the last scene. Well, it's a high octane movie. I mean, it's about human

00:11:07.700 performance, basically. So I get why you like it. It is unbelievable. Yeah. But I had no idea about

00:11:14.460 this notion that it was not just purely fictional and that there was some autobiographical component

00:11:19.360 to it. A little bit. I mean, I don't think there was anyone as sinister in Damien's life as the teacher

00:11:23.980 in the movie. But Damien, both as a drummer and now as a filmmaker, is incredibly

00:11:29.020 self-critical and hardworking and perfectionistic. And so I think those elements are definitely a

00:11:36.040 reflection of his personality. I'm always amazed when you look back at sort of the annals of rock

00:11:41.320 and roll, how many musicians didn't come into it as singers. So for example, I remember hearing about

00:11:46.940 Bob Dylan and Jimi Hendrix and people who really never thought of it as their voice was what was going

00:11:51.800 to do things. And yet we still think of them as completely iconic. Can anybody learn to sing?

00:11:57.620 I don't know if anyone can learn to sing. I mean, you need a certain amount of physical

00:12:01.900 instrumentation that you just can't escape. I mean, so there are things I wish I could do with

00:12:07.540 my voice that I'll never be able to do, just like I wish I could dunk a basketball. That being said,

00:12:13.060 I think there's probably an enormous, I know there's an enormous range of refinement that can be

00:12:19.760 pursued because I started as a pretty bad singer and I think I've become an okay singer. And a lot of that,

00:12:26.800 just like any physical activity is about learning how to mentally control a part of your body to get

00:12:33.160 it to do what you want it to do. It's just that controlling your vocal folds and the way that you

00:12:38.440 express air is a relatively fine-tuned set of physical processes. And so the sort of detailed

00:12:47.440 control that you need to physically command over your anatomy is kind of difficult to obtain.

00:12:54.780 But, you know, I became a much, much better singer. I think the thing that is probably more natural

00:13:00.780 that you either have or you don't have is an ability to know when you're making a noise,

00:13:06.280 whether it is on pitch. And some people clearly don't have this, but I think most people have a

00:13:13.060 decent sense of pitch. And if they didn't, then, you know, they wouldn't be pleased by harmonious

00:13:20.220 music. I mean, we have a natural ability to hear whether someone's hitting a sour note in a chord

00:13:27.660 or something, or whether someone's singing off tune that, that bothers most of us. So if you can be

00:13:33.500 bothered by that, then chances are you can hear the difference between that and the right thing.

00:13:39.980 And that's what a lot of it comes down to when you're singing. It's very important that you hear

00:13:44.260 yourself because that's the feedback loop. Yeah. So do you remember the first time you

00:13:50.260 sang on stage in front of people besides your bandmates?

00:13:54.700 Well, it would have been freshman year in college. And we began by doing sort of weekly performances

00:14:01.260 in the student commons at Harvard. And, you know, the audience was 20 or 30 of our friends.

00:14:07.600 And the band always had a sense of humor to the music, although that trailed off in our final

00:14:15.820 album, which was not really funny. But earlier, we had always kind of been humorous at some level.

00:14:22.960 And I think that masked how poor my musicianship was for a long time, because we could kind of play

00:14:31.660 it off as a bit of a joke band. So sort of like Barenaked Ladies, like what was it?

00:14:36.320 Not that they were a joke band, but you know what I mean? Like they were always having fun.

00:14:39.720 They were silly. We were irreverent and somewhat subversive in the notion that what we did was 0.97

00:14:48.820 meant to almost be ironic. So we would do songs about medieval nights. We would do totally absurdist

00:14:58.520 music. And I think totally possible to hear it and either think that it was self-consciously funny or

00:15:07.700 to think that it was totally clueless.

00:15:10.780 And so that made it easier to sort of cut the tension a little bit and break the ice or so to

00:15:14.860 speak, you know, that it wasn't like you were scared to death getting up there. For me, like if

00:15:19.340 you said to me, Peter, you have to either go and climb Mount Everest or K2 and there's, you know,

00:15:26.340 like a 30% chance you're going to die or you have to sing in front of a thousand people. I'm

00:15:31.240 trekking, like I'm going to the mountain. Like the thought of actually singing in front of any human

00:15:36.920 being, like including people like my kids, like there's no way I'm singing.

00:15:42.920 Yeah. I don't know why there is some sense of shame that attends singing. I mean, I know what you're

00:15:50.720 saying. I don't feel it, but I know what you're saying. I mean, like part, part of it is I think

00:15:56.220 realizing that singing is just a musical speaking. So we're, we're really singing all the time when

00:16:03.100 we're talking, we're just not using tonality and as deliberate away. I mean, there's a pitch to

00:16:10.100 every time you speak. So singing is just controlling that pitch, which is actually a helpful thing that

00:16:17.820 I learned from vocal coaches who, when I went to go touring, I went to see literally because it

00:16:23.960 became more of a sport for me. It was now something that I was going to have to do for an hour or two

00:16:29.860 every day. And I didn't have the physical endurance and the muscle control to endure that without

00:16:37.440 hurting myself. So I would go to some coaches to sort of get ready to tour and they would make this

00:16:44.920 point that, that singing is just musical speaking. And it really changed the way I thought about

00:16:49.820 singing and made it quite a bit easier. So let's talk through the transition. So you guys start this

00:16:55.280 band. Tell me about the name Chester French. Where did it come from? Chester French was the name of a

00:16:59.800 sculptor, Daniel Chester French, who did the Lincoln Memorial. And then he also did the John Harvard

00:17:05.760 statue, which is in the center of Harvard's main quad. And we didn't know that when we chose the name,

00:17:12.400 but we had a dining hall there that all freshmen ate in that sort of looks like the Hogwarts dining

00:17:16.920 hall in Harry Potter, big wood paneled room with soaring ceilings and busts all around the walls.

00:17:24.900 So we were going into lunch and we saw one of these busts and it had a little plaque under it that was,

00:17:31.100 we thought the name of the person who had been sculpted. And so we said, oh, Chester French,

00:17:36.760 that's kind of a cool name. And we picked it then and there. And then we later, of course,

00:17:41.540 learned that that was the sculptor of not only that bust, but also these other great American

00:17:46.780 sculptures. Got it. So when did you guys start to get some traction? And I mean, because most college

00:17:52.080 bands don't end up going on tour. So what was that transition like? We basically got no traction

00:17:57.340 doing what we initially set out to do, which was to build a live performance following.

00:18:02.860 So the conventional wisdom, if you started a band on the East coast in college, which a lot of people

00:18:10.000 did actually, if they were musicians, was start playing on your campus, start playing at surrounding

00:18:16.980 campus parties and sorority keggers and whatever, and then ultimately get a van and start driving around

00:18:23.160 the East coast and going to other colleges. And you can build a following among college students.

00:18:28.100 And we tried to do this, but we just totally ate shit. I mean, we could barely get people to come 0.97

00:18:34.720 to our shows on campus. And then we, I remember booked a gig at a, at a club called Great Scott

00:18:40.660 in Boston, which was kind of the, or Harper's Ferry was the, was the specific one that I'm thinking of

00:18:45.540 here. Great Scott was another club. We booked a show at this Harper's Ferry and we both tried to

00:18:52.520 recruit our friends from campus to, to make a 10 minute journey to this club. I think three of them

00:18:59.520 came. And then we also, for the three nights prior to the show, went and stood on the street in front

00:19:06.860 of the club, figuring that people who lived in the area probably walked by the club and, or people who

00:19:12.420 went to the club might come and see multiple shows, handing out flyers, trying to promote ourselves.

00:19:17.040 And when we got there to do the gig on a Thursday night or something, I think there was an audience

00:19:22.500 of like five or seven people. And you know, that's particularly depressing when you've spent two and

00:19:28.180 a half hours setting up and sound checking and hauling all your gear. So anyways, that strategy

00:19:33.920 failed. I'm not sure why, maybe we just weren't good live or something, but we took a turn at the end

00:19:40.940 of sophomore year in strategy and our focus became just making recordings. And the thought was ultimately

00:19:48.480 the product is a recorded music product. That's what we're really making. We're songwriters and we're

00:19:55.260 producers. And so our focus shifted to essentially just living in the recording studio, making stuff.

00:20:02.860 And the idea became, let's try and make an album that we think is a great complete representation

00:20:10.940 of our musical ideas. And if we do that and it's really good, it will speak for itself. And no one

00:20:18.660 in the record industry will care whether we have this big live following or not. Luckily that turned

00:20:25.100 out to work. I mean, it was not certain by any means, but we made this record over the subsequent two

00:20:32.080 and a half years. And then in the middle of senior year started sending out that album to as many

00:20:37.740 people as we could send it out to. And ultimately it got to Kanye West who gave us our big break

00:20:45.040 and flew us out to Los Angeles and offered us a record deal. And then that spurred a number of

00:20:51.620 other people attempting to sign us to record deals. We ultimately chose to work with Pharrell and did a

00:20:58.720 deal with him and a guy named Jimmy Iovine who ran Interscope Records. And as soon as we graduated,

00:21:04.400 we moved out to LA and that became our career. I mean, just put that in context for a moment.

00:21:08.660 That strikes me as like a fairy tale, right? I mean, what's the probability that Kanye West,

00:21:14.140 I mean, how many times is he getting something pitched and having to pick a needle out of a

00:21:19.940 haystack? Is that how that works? I think it was a little atypical that other artists supported us.

00:21:26.060 And now that I'm sort of not doing music full-time in retrospect, when I look at the band's career,

00:21:31.920 it's kind of clear to me that we were an artist's artist. In other words, our fans tended to be people

00:21:40.180 who were quite musical, whether they were famous musicians or whether they were just random people

00:21:44.520 who played guitar at home. But we were making a style of music that I think spoke to musicians

00:21:50.560 specifically. Part of that was the result of trying to make music that we wanted to listen to.

00:21:57.020 And so we were our own customer in a kind of weird way. So it certainly wasn't typical that other

00:22:04.020 artists would be the ones to jumpstart our career. On the other hand, it was at a time when the music

00:22:09.660 industry was changing. And so all of the paradigms by which artists got discovered were being

00:22:15.520 destabilized. There used to be these armies of A&R people who worked at record companies and they

00:22:21.540 went to concerts and scouted. And what was changing when we came onto the scene was that the internet

00:22:28.080 was becoming the primary distribution channel for music and also the primary place where people

00:22:33.100 discovered new artists.

00:22:34.520 This is what, like 2003?

00:22:36.560 2007.

00:22:37.240 2007, okay.

00:22:38.360 And so Facebook was a couple years old. MySpace was kind of at its height. And we were one of the

00:22:46.000 first artists to primarily build our audience online. So that was unique.

00:22:52.720 And was Facebook a better channel for that than, say, having your own site that you're hosting?

00:22:57.860 Facebook was and remains a bad channel for that. But MySpace was an amazing channel for that.

00:23:07.400 And MySpace had all of these features that allowed culture to kind of virally permeate

00:23:14.040 civilization. So-

00:23:16.020 Tell me what MySpace did better than Facebook with respect to that or what it enabled.

00:23:20.060 Well, there were, let's say, two or three distinctive things about it. The first was that it

00:23:24.280 essentially was a place where spam was totally legal and standardized. And so it was kind of like

00:23:31.180 MySpace felt almost like Times Square in the 80s or something. You know, it was like a little unsavory.

00:23:39.780 There were all these weird people on there. People had avatars. So you didn't know if it was really

00:23:45.080 them or not. Identity had not been formalized in the way that Facebook ultimately made it. So on the one

00:23:52.780 hand, you could spam people. So for us, we could go on there and just randomly send our music to

00:23:57.740 random people who we thought might like it. That was highly useful. The second thing was that they

00:24:03.680 had this thing top eight. So on everyone's MySpace page, you could pick your top eight friends.

00:24:10.720 And that became kind of like a prize to be one of the top eight of a famous person or one of the top

00:24:17.700 eight of a famous band or something. And so you could kind of go on MySpace and try and find artists

00:24:25.100 who you thought would share an audience or whose audience you wanted to steal, essentially. It's not

00:24:32.800 stealing because it's not a zero-sum thing. People can like a lot of artists. But we would identify

00:24:37.000 these artists who we wanted to, you know, steal the fans of. And then we'd try and get our music to

00:24:43.180 them to get in their top eight because then tons of people would discover you. And then the third

00:24:48.840 thing was that MySpace allowed each user an incredible amount of freedom in designing their

00:24:56.180 page. So you could go in and change the HTML or the CSS on your MySpace page. And this, from a user

00:25:04.480 experience standpoint, made it a very difficult place to navigate. But I think what it brought out was

00:25:11.160 that everybody, when you give them a creative canvas, likes to paint on it. And what you saw was

00:25:20.160 that ordinary people who were just MySpace users, consumers of content, themselves would really

00:25:28.120 wear their identity on their page. And so you could kind of navigate this universe of MySpace

00:25:34.880 and understand the cultural orientation of every person on it. And it made it really easy to figure

00:25:43.240 out who were your people. You know, I mean, if you were like a goth band or something, or a metal band,

00:25:50.900 it was very clear who the metal people were. And so for us, we could really take advantage of this

00:25:57.880 and kind of hack the system to find our people quickly. I don't want to go too down the rabbit

00:26:02.940 hole on this, but it's so interesting because I've never, I haven't thought about MySpace in

00:26:06.200 probably 10 years or something like that. When the history book's written, and it can probably be

00:26:10.820 written now, I'm assuming, why did Facebook win, MySpace lose at the risk of oversimplifying things?

00:26:17.400 I think there are a few reasons I perceive, and then there are probably many I have no idea about.

00:26:22.380 You just strike me as having a better sense than the average person based on a pretty robust

00:26:27.440 understanding of how both of them worked. MySpace was, even if they maybe thought they were being

00:26:32.480 scientific, it was not a sort of science project. In other words, the people building it were in LA,

00:26:39.020 they were kind of more inclined to understanding culture and the thinking about personal freedom

00:26:45.060 and expression. Whereas what Facebook was built around was the idea that if you created

00:26:52.260 an accurate map of people's real world relationships, then the behaviors that they

00:27:00.480 exhibit in the real world would end up mirrored in this digital environment. So social values like

00:27:09.120 trust were going to be essential in the Facebook universe. Whereas in MySpace, like I said, it was

00:27:15.640 like Times Square in the 80s. It was like there were seedy people on there. You didn't know if people

00:27:19.740 were who they said they were. It was a fake place. And the early internet was much more like that.

00:27:28.760 You know, everyone was in chat rooms with aliases and stuff. You never really knew who was who.

00:27:34.740 Facebook became a digital version of the real world. And so it was kind of more valuable from the

00:27:43.460 outset because things that you only previously could do in reality, like talk to someone or share

00:27:49.620 with them photos or something became something that you could do in a highly managed way through

00:27:56.200 Facebook. That's really interesting. I have never, I would have assumed just as a guy who knows

00:28:02.140 nothing that, you know, Facebook won because they figured out how to monetize this stuff better than

00:28:06.660 MySpace. But it sounds like there's much more cultural, emotional, sort of philosophical differences

00:28:12.640 that may have decided which direction was going to have greater mass appeal.

00:28:16.860 I think that's right. And you sort of see this resuscitated in the rise of Snapchat. Because

00:28:26.960 Snapchat is also kind of back to that LA mindset. It's much more expressive and irreverent and kind of

00:28:33.960 culturally contextualized. It's got a lot of character and personality. Its logo is a funny ghost

00:28:40.380 drawing and this sort of thing. And so I wouldn't give up yet on the idea that the future of the

00:28:46.640 internet may allow people to be much more expressive than you see them being on Facebook. Facebook is

00:28:54.860 a little bit totalitarian, not necessarily in a bad way. I mean, that's, that's a word that is

00:29:00.960 pretty much loaded, but yeah, loaded, but I'm looking for a antiseptic or anodyne, or I don't know what the

00:29:07.680 right word is, but it's, it doesn't have a lot of flavor. It's a reflection of the engineering

00:29:13.540 mindset behind it, which itself is really ingenious, but it as a digital place lacks culture. Mark

00:29:25.300 Zuckerberg, who I think very highly of, and who's a friend, is a genius business person and technologist,

00:29:34.020 but he's never, I don't think been especially interested in culture. And when he talks about

00:29:41.640 Facebook being a community and a global community, I think he is potentially always, you know, I don't

00:29:49.100 want to put words in his mouth, but I think he's maybe always thought that Facebook shouldn't have

00:29:55.360 too much of a culture that it imposes. It should basically be a blank canvas for all of the diversity

00:30:02.860 that is in the world. And when you allow people to express themselves more, it's unclear whether what

00:30:11.700 you're doing is loading a place with some cultural precepts of your own, even by dint of the way that

00:30:19.380 you design what those expressive tools are, or if you're actually giving people more freedom.

00:30:25.720 And so I think Facebook has always taken the approach of limiting people's expressive capacity,

00:30:32.640 forcing them to make comments or post things in these very structured, linear formats,

00:30:39.720 whereas some of these other tools have been a lot more liberal in the attitude that they take to

00:30:48.660 personal expression. When you went to college, remind me what you studied.

00:30:52.500 I did African-American studies and I have a minor in Kikuyu, which is a Kenyan language

00:31:01.000 that I no longer speak.

00:31:03.580 Were you thinking about going to grad school after? I mean, were you just pursuing your bliss? Like what

00:31:07.480 were you thinking you would end up doing before you ended up in LA, of course?

00:31:11.760 I wasn't thinking much about it at all. I mean, I think I'd been very interested in public policy

00:31:16.480 and so politics interested me and I think academia maybe interested me. The arts certainly did. And I

00:31:24.400 think halfway through my freshman year, I decided that the goal I wanted to shoot for was becoming

00:31:28.420 a musician. But part of that was a response to finding when I got to college that Harvard was

00:31:35.960 essentially a vocational school for investment banking and management consulting. I didn't know that

00:31:43.640 those were jobs in the world. I'd never heard of either of those professions. And when it became

00:31:49.360 clear to me that 40 or 50% of my classmates would end up in those jobs, I was kind of horrified. And I

00:31:56.540 thought, oh my gosh, is that the adult life that awaits me? And maybe as a bit of a rebellion against

00:32:02.820 that, I went full bore into, I want to be a musician.

00:32:06.060 That is hilarious. The vocational school for investment banking and management consulting.

00:32:13.640 I didn't realize it was that high. So roughly, but whatever, call it even if a third of people

00:32:18.600 end up as investment bankers and management consultants, that's a pretty interesting

00:32:21.860 concentration.

00:32:22.260 And not that there's anything wrong. I mean, I know you were a management consultant at one point,

00:32:25.440 but there was a sense in which I felt that my college and elite universities in general,

00:32:32.380 I've mentioned it a couple of times, but I'm not someone who's a big booster. I'm not involved in

00:32:37.180 alumni stuff. I don't take a lot of pride particularly in having gone to Harvard or anything. And part of

00:32:42.480 why is because I'm very suspicious of these institutions, both because they essentially serve a function

00:32:50.260 in the society of reproducing inequality and social hierarchy. And they've been doing that for

00:32:56.220 hundreds of years. And second, because they homogenize all of these talented young people.

00:33:02.940 So they do an extraordinarily good job of finding interesting, unique people in high school,

00:33:08.400 and then turning them into boring, high achieving technocrats. And that's kind of a tragedy for the

00:33:16.200 world, because it's a waste of the formidable resources that those universities do have,

00:33:21.000 that what they choose to do with it is essentially pump out functionaries in the financial services.

00:33:28.600 But somewhere along the way, you learned to think. And is it safe to then assume that you

00:33:33.340 learned that before college and that college, if anything, didn't diminish that ability rather than

00:33:39.140 augmented it? I'll tell you where I'm going with this. And we're going to talk a lot about biology

00:33:43.760 and our mutual interest in hacking and lifespan and all that sort of stuff. But I remember after

00:33:49.820 even the first day we met, maybe it was like after our second meeting or something, I remember

00:33:54.240 thinking, how does this guy know so much about this stuff? Like you had more than just a superficial

00:33:59.480 understanding of stuff that suggested like you'd read a couple of things about it, which obviously

00:34:05.020 just suggested to me that you learned how to learn. And as a result of that, when you decided,

00:34:10.000 hey, I really want to understand this area and I want to understand the science of this stuff,

00:34:14.380 you at least had a toolkit that you could use to do that. Where do you think you got that?

00:34:18.700 I think I've maybe just always been curious and that is a personality trait or disposition.

00:34:25.380 Just wanting to understand things has always driven me to some degree. When I went to college,

00:34:33.280 I was interested in race and the history of race in America and how this set of ideas about who we are

00:34:42.140 influences what the world is and who we become as people. And I was essentially interested primarily in these

00:34:50.680 very complicated phenomena in the social world. So I wanted to, you know, I remember the first time I read

00:34:59.560 Marx and whatever, it was so engaging to discover a set of these totalizing ideas about how history works or

00:35:09.120 something, you know, whether I agree with them or not, it's a separate issue. But the idea that one could

00:35:16.280 theoretically describe reality was always compelling to me. And what has happened intellectually as I've

00:35:26.980 gotten older is that I've realized that science is essentially the best methodology that we have for

00:35:34.060 describing reality. And I've realized that I think the things that most interested me, like social

00:35:41.200 phenomena, government, public policy, history, are very difficult to address with the science currently

00:35:48.620 at our disposal. So ultimately, I believe we will be able to have an empirical understanding of some of

00:35:55.620 these sorts of things. But I was going after them, back then naively thinking that we already had those

00:36:04.600 tools at our disposal, or that I could by reading a lot about history, develop a accurate picture of how it

00:36:12.440 works. And one learning how to think is about understanding the limitations of our own biology or our own

00:36:21.040 cognition. But then it's also about figuring out what's knowable, and how you can understand things

00:36:29.280 and what processes you can use to understand things. And so when I encounter a subject, I'm instantly

00:36:37.340 tracking on what's the theoretical framework behind it. The details don't really interest me. It's what's the

00:36:45.840 skeleton behind it. What's the structure of it. And for that reason, I don't think, you know, I've never

00:36:52.720 been good at facts. I'm not good at dates. I'm not good at memorizing names of biomolecules. But what

00:36:59.420 sticks in my brain are these kind of theoretical constructs. Yeah, somebody asked me on Twitter a

00:37:05.700 while ago, and I decided it would be a good question for an AMA at some point, which was if you could change

00:37:11.120 something about the medical school curriculum, what would it be in it? I wouldn't describe it probably as

00:37:15.220 eloquently as you, but it's basically, I remember the first day of medical school, they said, you know,

00:37:20.580 the average college graduate has a vocabulary of X words. And I don't actually remember what X is,

00:37:25.120 but I feel like it was about 10,000 or something like that. And they said, you know, in the next two

00:37:29.640 years, which is before you go into the clinical stuff, you're going to learn. And it was a little

00:37:33.080 more than that. So if it was 10,000 was the college grad, they said like, you'll learn 12,000 new words.

00:37:38.080 And obviously you do need to learn a new vocabulary in medicine. But when I look back at medicine

00:37:42.980 in medical school, rather, I think the biggest deficit was no attention paid deliberately

00:37:48.700 towards how to think. And that of course becomes relevant because, you know, all facts have a half

00:37:53.800 life and some facts like anatomical facts have very long half lives. So in an area like that,

00:38:00.080 it's probably reasonable to know stuff. But in other things, like by the time you're done medical

00:38:04.660 school, the information is wrong or irrelevant. And yet nothing's really put in place to teach

00:38:09.940 that skill of how do you go about, you're now out there in the real world, you're taking care of

00:38:17.260 patients. And yet maybe what you learned about cholesterol, or maybe what you learned about

00:38:21.860 subject XYZ is kind of not right, or at least should be revisited or the probability of it being

00:38:27.480 correct is lower. So do you and Liz think you'll have kids? Yeah. And so how do you, now I'm asking

00:38:34.060 this as a curious parent, but how do you think about creating an environment to produce that type of

00:38:39.140 curiosity and that skillset? What do you think a parent can do deliberately? I'm asking this based

00:38:45.440 on you having been the kid, not the parent so far. Well, you certainly have more experience with this

00:38:51.080 because you've got kids. What I've heard people say, and what I sense may be real is that you

00:38:58.220 actually don't need to teach kids this. They're born with a lot of it. And what the world tends to do

00:39:03.300 is squeeze it out of them. And so I guess I would invert the question and frame it as how do you not

00:39:10.600 kill a child's curiosity and how do you feed it? And certainly one of the things I think my parents

00:39:18.620 did was not sanction in a negative way, asking questions or asking why questions. And by contrast,

00:39:27.560 you can inculcate a child in religious ideologies or strictures about how to live that are so

00:39:34.700 imprisoning that becomes a fundamental part of their brain. And so I think just not making those

00:39:43.400 mistakes is probably the most important thing. I like the example you gave earlier of your piano

00:39:49.220 lesson is 30 minutes of sitting in front of a piano for some period of time before you start to

00:39:54.120 read sheet music and start doing drills and things like that. I like that idea. That's.

00:39:59.340 Absolutely. I mean, look, there's a balance between the sort of rote discipline that is required in

00:40:06.480 anything and then the creative and exploratory potential within that area. In my mind, I tend to

00:40:16.340 wait the latter. And so what's difficult for me is the brutal, repetitive skill building. And I've had

00:40:25.980 to figure out ways in life of not making my success depend on that aptitude and figure, you know, that's

00:40:33.240 stuff that I look to other people to help me with. You know, in my band, as an example, my partner,

00:40:38.560 Max is an, I mean, he's good at everything. I mean, he's a musically brilliant person at a

00:40:44.180 compositional level too. I mean, he can write great music and he's very theoretical, but he also has

00:40:50.480 that OCD capacity to sit there and practice guitar for 10 hours. And that's really fun. That's his

00:40:55.880 happy place. I'd go crazy doing that. And so I think to come back to your question about children,

00:41:02.960 I think the struggle for me will be figuring out how to impart to them a greater capacity to do the

00:41:09.720 part that I don't like doing. Because if you can really have both, that's the superpower.

00:41:16.020 But it sounds like to me, at least maybe part of the, the secret is knowing how much emphasis to

00:41:22.820 put on each of those two. Again, they're not mutually exclusive, but you understand that this

00:41:27.520 is a kid that if pushed too hard on the rote stuff will actually dampen the creativity versus

00:41:33.540 sounds like in Max's case, that's not the case. He can live in both of those worlds and be quite

00:41:37.720 successful. And I think that's right. You know, the one metaphor I use to think about this because

00:41:43.060 a debate that comes up in all contemporary art, be it painting or writing or music is essentially

00:41:50.520 whether amateur work can be just as valid as quote unquote professional work. So if you think about

00:41:57.200 punk rock, the whole idea is screw learning these instruments. I'm just going to pick up the guitar

00:42:03.020 and express myself. And so of course there's a level of vitality that can be produced through

00:42:11.720 that type of artwork. Both the absence of formal rules gives birth to a lot of expressive freedom

00:42:22.780 and you can get a kind of visceral version of someone, but there are only so many ways to smash a guitar

00:42:30.020 on stage. And so if a hundred people- None of them are as good as Jimi Hendrix at Monterey.

00:42:35.340 None of them are. And Jimi Hendrix can, he can still smash a guitar on stage, but he can also do a

00:42:41.780 thousand other things that the Ramones can't do. And so the metaphor that I've been thinking of lately

00:42:47.880 is almost like granularity or resolution. So, you know, if you imagine a photograph of something

00:42:56.340 in low resolution versus high resolution, it's the same thing that's being expressed,

00:43:02.860 but the greater the resolution, the more detail there is. And when someone is making music or

00:43:10.220 singing or writing a poem or doing anything, what you're essentially, they're communicating.

00:43:16.580 And what you're getting is you're getting sort of a photo of their brain.

00:43:20.160 And the more technical skill they have in this metaphor, the greater the resolution of that

00:43:29.720 picture they can produce. And so I once remember being in the studio with Pharrell and Herbie Hancock

00:43:37.860 came by and Pharrell is a kind of folk musician. I mean, he's incredibly successful and he's very,

00:43:45.360 very talented, but he's not a trained, he didn't go to music conservatory. He can't play you a Bach

00:43:51.940 sonata. He can write amazing stuff. He's a theoretical musician, right? He's a great songwriter

00:43:57.980 and he's got great taste. And Herbie Hancock came and he was showing us a bunch of piano exercises that

00:44:05.440 he and his friends used to do in Chicago in the sixties. And Pharrell kind of expressed that he was scared

00:44:12.540 to learn that stuff because he was worried it would eat into his roughness or his authentic

00:44:22.240 expression. And I remember Herbie Hancock saying, it's just going to give you more colors to paint

00:44:30.180 with. It doesn't cost you anything. It just, it just gives you more things that you can say

00:44:35.260 with those ideas you've got. And I thought that was a great argument. You know, I think it was

00:44:40.720 convincing to Pharrell too. When you were in high school, who do you in retrospect, look back at

00:44:46.220 and say, you know, those were really, those musicians shaped my, either my taste or my philosophy

00:44:52.580 or my way of writing or performing, like who just broadly speaking, who were your influences in that

00:44:58.520 regard? They're a little bit mundane, but they're sort of the obvious suspects. Uh, the Beatles,

00:45:04.420 Bob Marley, Curtis Mayfield. Those are the big ones. Stevie Wonder. Maybe, maybe those four

00:45:13.340 Motown, you know, uh, all the Holland Dozier Holland were the three great songwriters who did a lot of

00:45:20.340 the Motown stuff, a lot of the four top songs, a lot of the temptation songs. Any James Brown?

00:45:25.360 Yeah. I like James Brown, but I would put, I'm, I'm more of a songwriter and James Brown,

00:45:31.220 I think is most important as both a performer and as a groove guy. So, you know, different music is

00:45:40.480 good at different things. And there's a tradition in black American music that I think ultimately

00:45:46.160 begins in Africa and in the African diaspora of music that is not really linear. It's not about a

00:45:53.780 narrative story being told through something like a song. It's much more about setting a musical mood

00:46:01.200 and almost being a place that you hang out. Indian music is also like this. If you think about

00:46:06.200 Indian ragas, for example, it's sort of a tonal place and the piece of music may be half an hour

00:46:15.380 long. It may be two hours long, but you kind of go there and you're just hanging out there in that

00:46:21.220 vibe. Whereas the sort of music that influenced me much more that I sort of work in is, is more

00:46:28.940 about, it's again, it's kind of more theoretical. It's more about a story. It's a linear story.

00:46:34.940 It's a three minute song. There's a beginning, a middle and an end. It has some conventions about

00:46:41.020 the way that it moves through those places. And people I cited, I think are emblematic of that style.

00:46:49.940 Now you're a drummer. So what drummer would you have identified with most?

00:46:54.920 I'm sort of now going to double back on what I just said, because the key function of the drums

00:47:01.780 is most highly expressed in those types of music that are essentially groove musics. So

00:47:08.660 Clyde Stubblefield, who played with James Brown is one of my favorite drummers of all time.

00:47:14.860 That style of music that James Brown made is all about the drummer, you know, give the drummer some

00:47:19.900 and other people who are, you know, giants of that format are Bernard Purdy. Who's just, he's still

00:47:30.000 still around and he's incredible. I think he claims he's the most recorded drummer of all time, or has

00:47:35.520 maybe played on the most hits of any drummer of all time. And, you know, then there are great rock

00:47:41.980 drummers like Keith Moon or John Bonham. In jazz, the drums are much more of a instrument. You can sing

00:47:50.780 through the drums in jazz music because the drummer is not just hitting hard and fast and loud. The

00:47:58.780 drummer has an incredible dynamic range of tonalities and expressive options. So in jazz drumming, you know,

00:48:09.860 Elvin Jones is probably my favorite who played a lot with John Coltrane, but there are an almost

00:48:17.140 unlimited number of incredible jazz drummers who I look up to. I think in the show notes for this

00:48:23.680 podcast, we should link to some performances that you find particularly notable. Oh, that'd be cool.

00:48:29.620 I remember once you, you gave me a nice list of stuff for Olivia, which was like, here's some great

00:48:35.980 Michael Jackson songs. Here are some great James Brown songs. Here are some great boom, boom, boom.

00:48:39.640 And it was like, this is a great way for a kid to start thinking about music. And I thought that's,

00:48:44.500 uh, that's such a helpful thing to provide. So, so, okay, let's go back to LA. So you get to LA,

00:48:49.880 you're living the dream, right? Like it's, you've hit a home run here. It's hard to imagine

00:48:53.360 given what the denominator is on the number of people that are trying to accomplish what you guys

00:48:57.680 have accomplished. And are you primarily now still a studio band or are you, once you've got this

00:49:04.300 break, are you becoming kind of more performance based as well? Well, by that time, the band was

00:49:10.300 just me and my bandmate max. So it was a duo really. And we, as we made music, we're a studio

00:49:16.440 thing still, but we quickly realized you were playing. So you were doing vocals, you were doing

00:49:22.400 drums and were you keyboard as well? You know, I was primarily writing the songs with max and singing

00:49:29.660 them. And max in the band was the primary instrumentalist. He can play pretty much

00:49:34.860 everything and he can learn new instruments in a week. So max on the recordings was playing almost

00:49:43.220 all the instruments. Occasionally we would have studio musicians come in and do stuff that he

00:49:47.060 couldn't do. You know, if we needed a viola player or something, but my function in the band was largely

00:49:54.120 to write the songs with him. I was writing all the lyrics. I was typically writing the vocal melodies of

00:49:58.980 the songs and then singing them. So how did Spotify come into the picture? Was that true,

00:50:05.140 true and unrelated? Was that just a, did Sean start that or how did that? No. So let's fast forward a

00:50:10.840 few years. So we basically, because there's, there's not a lot, I think that is super interesting about

00:50:15.240 this period in my life, although you may disagree, but I don't know a period of your life. It's not

00:50:20.080 interesting. This is 2007. We get our record deal. We move out here and then we spent three years

00:50:24.380 making records and touring. And a lot of stuff happened. And you toured with Blink at one point,

00:50:30.060 right? We toured with all sorts of people, Lady Gaga, Blink 182, Weezer, huge number of really cool

00:50:36.480 artists. And then I concluded that I didn't like touring. And it was around that time that I discovered

00:50:45.800 Spotify. I think at the time I already had an interest in investing, but I didn't really know

00:50:52.400 which direction to go. I had met Ashton Kutcher who became a really good friend and he and I got

00:50:58.620 along, I think right away because we were both from the Midwest. We were both artists and we both had

00:51:03.380 an interest in business and technology, but he was a decade ahead of me in terms of being a famous

00:51:09.060 entertainer. And he was three or four years ahead of me in terms of thinking that Silicon Valley

00:51:16.020 venture capital could be an interesting canvas for his artwork. And so he gave me a little bit of a

00:51:23.880 template for an artist who was going to get into the technology business. And the question became for

00:51:33.440 me, you know, how do I go from being a musician who neither has capital nor investing experience

00:51:40.760 and become a professional investor. And Spotify ended up being in retrospect, the pivot that allowed

00:51:49.360 me to do that. You know, historically, the music industry has not birthed many good businesses. And so

00:51:57.020 it was a pretty happy accident that right at this time that I wanted to make this transition,

00:52:02.440 this company that would become now a $28 billion public market behemoth was starting

00:52:10.560 to become ubiquitous in Sweden. And I had friends who knew about it. And when they told me about it,

00:52:19.280 the appeal of that product of this idea that you could install a thing on your computer and have

00:52:24.840 all of the world's songs ever at your instant recall was an incredibly cool thing to me. And I was just

00:52:34.900 thinking of it as the customer, basically, you know, I was someone who collected tons of music,

00:52:38.860 had hard drives full of it, was always struggling to manage the data. You know, I'd get a new computer,

00:52:44.840 now I have to copy all the songs, and I have to put them in folders. And I, it was such a pain in

00:52:49.440 the ass that when someone showed me this, I thought, whoa, this is what I've always wanted. 0.99

00:52:53.920 And was Pandora on the scene at that time? Pandora was really popular. 0.99

00:52:57.440 And it was just streaming. Pandora was streaming, but it was streaming algorithmic radio.

00:53:01.720 In other words, you couldn't choose the song. What was so cool about Spotify was that it recreated

00:53:08.480 the experience of having bought every song ever. You had MP3s that you'd buy through the iTunes store,

00:53:16.800 for example, and they'd be on your computer, but maybe you'd bought 200 of them. That was your

00:53:21.780 library of music. Whereas when I first saw Spotify, I thought, wow, for $10 a month,

00:53:27.220 it's as if I bought everything ever. I mean, that was such a shocking value proposition

00:53:33.260 to me that I thought this is almost certainly what everyone should have if the powers that be

00:53:43.840 allow it. And the business question, which was much more difficult than the product design question,

00:53:51.500 was can this company get everyone who owns music to be a part of delivering this experience to the

00:54:01.020 consumer? And so was the hardest sell with the artists or with the studios?

00:54:05.640 Everything was a hard sell. Spotify was, I'm not going to take much credit for it because there

00:54:10.800 were thousands of people working on this from the beginning almost, but everything about building

00:54:16.000 that business was difficult. The first thing that was difficult was that record companies and music

00:54:21.480 publishing companies owned the rights to the music. And they had seen their business destroyed in

00:54:28.540 the nineties. The record industry went to a quarter of its former size in the nineties because of

00:54:34.680 piracy. And so you'll remember the record labels were suing individual customers.

00:54:40.320 Is that because CDs had become so ubiquitous that it was easier to copy a CD than say a vinyl or a cassette?

00:54:46.320 I mean, not that it's not hard to copy a cassette, but the quality is not as good. Is that?

00:54:49.620 What was coincidental with CD-ROMs? So two things happened. One sites like Napster arose that allowed

00:54:58.800 people to trade music freely without paying for it. And then what they did with that music after

00:55:04.980 they traded it on the internet, which was so difficult to control or put a lid on was then

00:55:09.960 they would burn CDs and copy the CDs and whatever. So not only did the format of CDs get unlocked with

00:55:16.840 the ability to write CDs. I mean, you remember getting your first CD-R that was really cool,

00:55:22.600 but also you had this unfettered digital environment in which people could share music without anyone

00:55:29.440 getting paid for it. What preceded this in the eighties and nineties or really in the nineties

00:55:35.420 was the CD revolution. And so not only was piracy painful in its own right, it succeeded a decade

00:55:46.800 during which the music companies resold the entire history of music in a 10 year period.

00:55:54.920 In other words, they not only sold in the nineties on CDs, the new music that was being made in the

00:56:01.960 90s, but everything that had been made today. So they sold you an entire library of the world's

00:56:07.720 history of music in a 10 year period. They were rolling in cash. And when they made new music,

00:56:13.820 they would sign an artist. They'd make one hit song that they'd play on the radio. And then you had to

00:56:20.380 go buy an 1899 album with 14 other songs you didn't want just to get the one you liked. So they were

00:56:30.160 raking in the cash by completely controlling the consumer's experience. Piracy blew the lid off that

00:56:37.860 and the consumers rebelled and everyone stopped paying for music. And so the industry that Spotify

00:56:44.260 came into was one that had been eviscerated by piracy. And then to add insult to injury,

00:56:52.480 the legal paid download model that Steve jobs pioneered with iTunes unilaterally recaptured

00:57:03.460 the distribution system of what before then had been fragmented. So the record labels used to in

00:57:10.080 the good old days of the nineties, both control the radio stations so they could make things get

00:57:15.720 popular. But then they had a variety of physical retailers, Sam Goody, Virgin Megastore, all these

00:57:23.660 different CD retailers that they could play off each other. They had all the leverage. They could

00:57:31.080 make the thing big and then they could tell you at the record store that you had to give them this

00:57:35.920 promotion or else your artist wasn't going to show up there and do a CD signing. Well, when Steve jobs

00:57:41.480 invented the iTunes store, he became the only retailer that mattered. And so the record labels

00:57:47.840 had this one, two punch they'd been subjected to of, of the industry disappearing and going to a quarter

00:57:53.600 of its size. And then this one guy who didn't need them, Steve jobs, who made all his money selling

00:57:59.660 hardware, gaining total unilateral control over them because he became the only place they could sell

00:58:06.660 music. And so they were desperate never to let either of these sorts of things happen to them

00:58:12.760 again. They were incredibly paranoid about a new model that they viewed as threatening the paid

00:58:21.600 download business. So Spotify was showing up telling consumers for 10 bucks a month, you get all the music

00:58:27.520 ever. The labels at that point were still embedded in a model where what you did was you sold people

00:58:33.740 downloads for a dollar a piece. They were used to selling you 10 songs a month for $10 and Spotify

00:58:40.160 came along and said, let's give them 22 million songs for $10. So selling the record labels was

00:58:46.300 incredibly difficult. And then the artists were ardently against this as well, because for them,

00:58:53.620 their lifeblood was selling downloads. They too were in an environment that was severely nutrition

00:59:01.160 restricted from the past decade. And they were struggling to make money. Most of them were only

00:59:06.980 making money touring and recovering a little bit of income selling their recordings. Spotify came along

00:59:14.600 and what we said was, we're going to fix this problem. We're going to convince most people in the world

00:59:21.220 that they should pay for music. And the way we're going to do it is we're going to give them an experience

00:59:26.240 that is every bit as good as the free illegal experience. And we're going to get them hooked on

00:59:32.940 it. And then we're going to charge them 10 bucks a month. And if we can get a huge percentage of them

00:59:38.780 paying 10 bucks a month, that is far more than the average consumer at that time was currently spending

00:59:45.260 on music. I still don't understand the economics of Spotify. So if you're spending that 10, and by the

00:59:52.240 way, I couldn't just sign up for 10 bucks a month today, could I? Aren't those sweet?

00:59:55.260 You can. 10 bucks a month.

00:59:56.680 I thought that was like a student deal or some.

00:59:58.840 You can sign up for $3 a month for the first three months today, and then it'll go to $10 a month.

01:00:03.760 I'm ashamed to say I still don't use Spotify. I'm still wed to iTunes.

01:00:07.180 You got to get on there.

01:00:07.880 And I don't know if it's just like the hurdle rate of switching out of iTunes,

01:00:11.100 but it strikes me as ridiculous because I pay a buck 29 per song. I probably buy an average of

01:00:16.680 10 to 20 songs a month. So I'm obviously spending more money than I would.

01:00:21.380 But let's just talk about the remuneration for the artist. So when I spend a buck 29

01:00:26.440 on iTunes, where does the rent go?

01:00:30.920 30% goes to Apple. 70% goes to the record label. The record label pays out roughly 10% of their 70%.

01:00:42.100 So 7% to the music publisher, which owns the song. Note that the song, which is called publishing,

01:00:50.440 is different from the recording of the song. So if you and I write happy birthday and people sing

01:00:56.100 it all over the world, we own the song no matter who's singing it. But if Toni Braxton records a

01:01:01.940 version of happy birthday, she owns the recording. The song owner gets about 10% of the 70%.

01:01:07.080 And the artist will typically get somewhere between 15% and 20% of what's left after their 15% or 20%

01:01:19.460 cut has fully paid back the record label for all the money that label has spent promoting them.

01:01:26.660 So Apple's getting a straight 30. That's pretty straightforward. Of the 70, the label's getting

01:01:31.380 the majority of it. The label is getting the majority of it. And the artist is getting anywhere

01:01:36.120 from zero to maybe 10%. 15%, something like that. Now that sounds grossly unfair. What makes it less

01:01:45.740 obviously unfair is that the record label is the one taking all of the financial risk

01:01:51.260 in financing the music and very often paying for the artist's life.

01:01:55.780 So in this economy, they are still the only companies that take a financial risk making

01:02:04.400 original artwork. And they certainly deserve to be paid for taking that risk.

01:02:10.460 So how does it work at Spotify? So now you paid your 10 bucks a month.

01:02:13.420 How is that rent spread out?

01:02:16.220 Sure. So what happens is you listen to music all month. At the end of the month, Spotify looks at the

01:02:22.160 amount of time that you spent listening to each of those songs and your $10 gets split pro rata

01:02:30.300 to the originators of that music.

01:02:33.760 In roughly the same distribution as the 70% out of iTunes got split?

01:02:37.520 Yes.

01:02:38.180 How much does Spotify get to keep of the 10 bucks?

01:02:40.400 30%.

01:02:40.800 Okay. So Spotify keeps 30 and the other 70 goes the same way, but now it's pro rata.

01:02:44.720 Yeah, that's correct. And so for a consumer like you, who maybe before Spotify was spending

01:02:51.180 more than $10 a month, it results in less financial benefit than it might have before. On the other

01:02:59.520 hand, it depends how much you listen to those songs. So if on Spotify, you know, if you buy the song

01:03:05.740 on iTunes and you listen to it one time, you're paying $1.29 for one listen. But a lot of people,

01:03:11.380 when they fall in love with a song, listen to it thousands of times and they listen to it for years.

01:03:16.540 And if you take $1.29 and split it among thousands, effectively, they're paying very little each

01:03:23.180 listen. So the paradigm shift is towards a pay for listen model. It's a consumption-based payment

01:03:30.840 model.

01:03:31.160 Now, back before any of this stuff came along, if you had a radio station and you played a song,

01:03:36.580 not satellite radio, just straight FM radio, for example, did you have to pay a royalty for that?

01:03:42.260 But the radio station had an agreement with an organization called a Performing Rights Society.

01:03:50.840 These are societies that manage a particular type of right in music. So if you and I write a song,

01:03:57.460 we, and record it, we've just generated a whole basket of different types of legal rights,

01:04:03.020 all of which we begin by owning. And then we can sell those rights off to different third parties

01:04:08.740 that will help us collect the income that attaches to the exploitation of those rights.

01:04:14.260 In the case of songs that get played on the radio or performed at a basketball game or these sorts of

01:04:19.780 things, that right is called a public performance right. And the radio stations do deals with these

01:04:26.920 rights societies. In the US, the two most prominent ones are called ASCAP and BMI. And those

01:04:33.320 organizations are responsible for monetizing that right on behalf of artists. So the radio station

01:04:41.120 at the end of the year will give ASCAP a million dollars. And then ASCAP will do something like

01:04:47.220 Spotify does, which is tabulate what that million dollars corresponded to, and they'll pay out the

01:04:54.380 royalties to artists pro rata in accordance with that.

01:04:57.600 And Sean Parker was involved with Spotify and with Napster, right?

01:05:01.140 That's right.

01:05:01.880 So he had kind of a, this was, this was the evolution of what he and Sean Fanning had done.

01:05:07.080 Yeah, it was, they had intended Napster to become a legal service. And what they wanted,

01:05:11.980 you know, 20 years ago was to go do deals with the record labels to legalize it. But the industry was so

01:05:19.240 defensive towards them and so aggressive that it couldn't happen. It took 20 years and it took the

01:05:28.160 industry being decimated for everybody to be desperate enough that they were willing to entertain

01:05:35.820 solutions to the internet. And that's what Spotify was.

01:05:39.680 You know, it's interesting when you look at all of the businesses that have been disrupted by

01:05:43.800 the internet and contrast it with those that have not, right? So you've just given us a very eloquent

01:05:49.040 description of how the music industry was disrupted, but you also pointed out something

01:05:53.240 interesting, which is it wasn't just the internet that disrupted. It was the desperation that followed

01:05:58.860 the decimation that permitted the restructuring. Of course, then you can rattle off all the glib and

01:06:04.780 obvious examples, right? What Amazon has done to retail, what Netflix has done to, you know,

01:06:09.740 Blockbuster and, you know, stores on the side of the street, what Uber has done to taxis, what,

01:06:14.900 you know, all of the travel sites have done to travel agents. I mean, it's been incredibly disruptive.

01:06:18.980 And yet when I think about medicine, it's like one of the least disrupted industries by technology.

01:06:27.440 And, you know, you basically still have hospitals, eat all the rent payers. Although most people

01:06:33.280 generally perceive them as bad guys generally don't get paid that much providers get less and less on a

01:06:38.840 per encounter basis. And it's, you know, overall a pretty much similar system to what it was 20 years

01:06:45.340 ago. And that's, I don't know. I mean, I guess I'd have to think about it, but I can't think of too

01:06:49.860 many industries that are as unchanged in the presence of this revolution as healthcare.

01:06:55.460 No, I think you're right. The appeal of investing in healthcare is exactly what you just expressed.

01:07:02.220 And that's what drew me into it several years ago. And what I've learned since then has been

01:07:08.260 slightly disillusioning, but I'll give you my take on it. The first thing I would say is that

01:07:14.400 healthcare is still an industry that largely gets paid through labor. So most of the money

01:07:23.520 ultimately goes to labor, goes to doctors, it goes to healthcare personnel. The hospitals don't make

01:07:32.220 that much money. They're not that great a business. You know, if you look at Kaiser, for example,

01:07:36.700 which is an integrated payer provider, you know, in California, I think they have a three or 4%

01:07:41.880 gross margin. It's not a phenomenal business. The payers to your point have very low margins.

01:07:49.240 They in fact have legally imposed limits on the amount of gross margin that they can generate,

01:07:54.560 which, you know, you can argue about the ethics of the pharmaceutical companies who everyone hates

01:08:00.780 seemingly extract enormous profits. On the other hand, it's a very competitive industry.

01:08:08.560 And apart from the legalized monopoly given to them via patents on new drugs,

01:08:14.460 they are the only companies, sort of as I said about record labels, they're the only ones who are

01:08:21.160 out here designing new solutions. And taking an enormous risk.

01:08:25.640 Taking an enormous risk. And their shareholders are taking that risk. Now, whether they're being

01:08:29.240 overpaid or underpaid to take that risk is totally debatable.

01:08:33.000 The other thing is that it's unclear where technology can deliver dividends in terms of human

01:08:42.580 health. And so we're thinking about medicine at a moment in which we as a species have already made

01:08:49.780 enormous progress. I mean, whatever we've doubled or tripled lifespan. And, you know, I mean, you know

01:08:55.080 these numbers better than I do, but even if you get rid of the infant mortality factor and these other

01:09:01.240 things that distort the numbers, people are living quite a bit longer on average than they were even

01:09:06.620 seemingly 30 years ago. And that's especially true of affluent societies until they get afflicted by the

01:09:13.080 diseases of affluence. So where can we get more life from or where can we reduce suffering and disease

01:09:22.320 by way of technology is a little unclear. There are all sorts of things that we could, of course,

01:09:28.880 do to make hospitals more efficient and to spend less than we do on things like end of life care

01:09:36.060 that maybe aren't a good use of capital. But there's not so obviously a Spotify of healthcare

01:09:44.520 waiting to happen. It's more like there are a million little IT operating efficiencies that McKinsey

01:09:53.120 consultants will over the next several decades implement in large health organizations. And then

01:09:59.080 the big promise is around new types of medicine that I think will arise in the biotechnology industry.

01:10:09.100 And that's where I'm now spending a lot of my time and attention.

01:10:12.380 Yeah, it's sort of interesting because I think one of the challenges of healthcare, and I think this was

01:10:16.380 most evident during the debates many years ago, not many, but what I call it, eight, 10 years ago

01:10:21.320 around the ACA and around complete revisiting a restructuring of healthcare. And one of the things

01:10:26.960 I found a little painful in listening to politicians talk about healthcare was they spoke about it like

01:10:32.220 there was just one problem to be solved. But the way I sort of think about it is there are at least

01:10:37.640 three completely separate but highly related problems, which is, you know, quality of care. And that can be

01:10:43.980 broken down into two levels. You can sort of have it as just

01:10:47.340 keeping people from dying versus like generating alpha. So, you know, really improving quality. There's cost,

01:10:54.540 which is just the cost to the system. And

01:10:57.120 about 90% of the dollars are basically spent by three entities, the government, the employer, and the payer,

01:11:05.880 depending on who's at risk. The consumer is on the hook for about 10% of that cost. So how could you reduce

01:11:11.720 the cost? And then the third leg of that stool is access. How do you ensure access?

01:11:18.340 And so when we talk about technology disrupting healthcare, it's like, where do you start, right?

01:11:23.240 Because, you know, for me, I can see things where technology can be quite helpful on quality of care.

01:11:30.000 Not entirely obvious to me how to leverage technology to lower cost. People much smarter than me, I'm sure,

01:11:35.300 think about that. And not entirely clear, at least at a practical level, how it can drive access.

01:11:41.380 Although in theory, that should be the first place you'd want to go after leveraging technology. But I

01:11:46.920 think I'm trying to think about what you said about how Spotify went about it and trying to see,

01:11:51.340 is there a parallel there? Because the thing with Spotify that was hard was you had multiple

01:11:55.620 constituents, right? You have artists, you have labels, you have consumers, and they all kind of want

01:12:01.160 something a little different. And that speaks of an optimization problem, which is clearly what

01:12:05.240 healthcare is. But going on what you said a moment ago, it sounds like today you're a little less

01:12:11.760 encouraged than you would have been four years ago when you're thinking on this problem.

01:12:16.420 Well, let me try and recapitulate your three legs of the stool in one simple framework,

01:12:22.560 which is something that we should all be aiming for is to generate universal access

01:12:30.620 access to the current best available healthcare that anyone in the world receives. You know,

01:12:37.260 one of my favorite quotes, I always forget whose quote it is, but venture capitalists talk about it

01:12:42.680 a lot, which is that the future is already here. It's just not evenly distributed. And I think that's

01:12:47.240 almost always true. Look at what rich people are doing and assume that in 20 or 30 years,

01:12:53.920 everybody will have that. And so what are rich people doing right now? Well, they're paying

01:12:59.000 doctors like you a lot of money to do very bespoke types of medicine that leverage all sorts of deep

01:13:08.560 biometric analyses of their bodies, a lot of thinking by you and your team, and a lot of spend on the usage

01:13:19.060 of what are currently expensive machines, MRI machines to do preventative screening for tumors every

01:13:27.720 year, things like this, that if you're really rich, you know, you might as well go get an MRI every

01:13:32.420 year, get a full body MRI every year. I mean, just see if there's anything going on. Why not?

01:13:37.080 And so what stands between today and orphans in Somalia receiving that level of care and what stands in

01:13:48.340 the way I think is to some degree, a level of technological leverage. What is happening in your head

01:13:56.640 when you look at a patient's blood tests and try to gel that with your biophysiological

01:14:05.780 mental models is something that I believe computers will be able to do in the future.

01:14:11.300 And so right now we use physicians for thinking and for bedside manner and for guidance. And then we use

01:14:25.240 some physicians like surgeons for physical procedures that they're essentially athletes at,

01:14:32.820 you know, you want high performing athletes who don't mess up, who are very precise, who've done

01:14:36.820 it a million times. Well, the roboticization of physical medicine like surgery is something that I believe

01:14:47.100 will just continue to get better and better. It may take longer, it may not be replaceable immediately,

01:14:52.220 but that will, you know, we will be primarily doing robotic surgeries a hundred years from now.

01:14:57.260 The cognitive work that physicians do will, I believe, largely be replaced by computational systems

01:15:05.700 that can leverage the same types of theoretical frameworks that our minds are utilizing to make

01:15:12.600 good decisions. And what you're left with is all the touchy feely personal stuff that you as a physician

01:15:19.420 and any other physician knows really matters. It's not superficial. The ability to help someone

01:15:27.580 make good decisions about their own health, to make someone care about their physical health,

01:15:32.400 to help them navigate difficult decisions that aren't straightforward, that involve trade-offs,

01:15:38.880 are the sort of things that physicians right now don't have the time to do that they did when they

01:15:44.420 were taking their medicine bag around to people's houses and hanging out for a couple hours.

01:15:48.040 And so in a certain way, we may go back to an older style of medicine, but the people who do that may

01:15:55.180 look more like highly skilled nurses than like physicians. So I believe that what you'll see is

01:16:00.680 globally a lot more technology involved in the deployment of medicine and a shift in the composition

01:16:08.400 of the labor market around medicine towards something that looks more like a large population of highly

01:16:15.360 skilled nurses, fewer overpaid or appropriately paid, but maybe over-trained physicians that right 0.74

01:16:24.580 now are very scarce that only rich people can access. And then most excitingly, you're going to see a lot

01:16:31.000 of new medicine. That's the stuff I'm focused on now as an investor. And that's the stuff that I think

01:16:37.340 is going to be the most game-changing. That's the stuff that's going to buy us years of additional life.

01:16:42.460 So when did you make that transition where you sort of your, your big foray into this was obviously

01:16:49.300 Spotify as far as like transition one. So DA 2.0 to DA 3.0 and then DA 4.0 doing the investing. So

01:16:58.500 the biotech stuff specifically, when did you make that transition?

01:17:01.740 Well, I'm still making it because it's a hard transition to make. And there's been so much that

01:17:06.920 I've already had to learn and so much more that I have yet to learn. I think the starting point was

01:17:13.460 this realization that what mattered most in healthcare as a capital allocator, meaning the

01:17:20.400 most useful place that I could help direct money was to what goes through the pipes of the healthcare 0.97

01:17:28.760 system, as opposed to the healthcare system itself. The healthcare system in America, for the reasons

01:17:34.520 we've just discussed, is a creaky old infrastructure. And it's important to fix it, just like it's

01:17:45.080 important to fix the government or the broken bridges that we have all over the country. But it's not the

01:17:51.180 most interesting thing to do. And it also doesn't have a global impact because fixing the US healthcare 0.81

01:17:57.260 system doesn't do a lot for folks in Sudan or in Saudi Arabia or in Japan or anywhere else. You're

01:18:04.560 just dealing with a human system where what's broken are the human structures that we've constructed

01:18:11.060 to deliver the thing. So that was my starting point was realizing that the medicine itself was the thing

01:18:18.220 that mattered more. And ultimately, despite all of the valuable things that physicians do,

01:18:25.500 the biggest game changers in human health in the past 50 years have been pills. They've been pills

01:18:32.160 that very brilliant people invent, and that we can give to people through any number of channels,

01:18:39.380 and that actually fix the problem sometimes. And we're moving into an era of medicine that people

01:18:47.080 call preventative medicine or precision medicine or whatever you want, that really boils down to

01:18:52.720 addressing illness at the level of mechanism. Understanding that diseases have a functional,

01:19:04.400 concrete explanation, and that we can develop interventions that actually fix what's broken

01:19:12.760 in a very precise way and without a lot of collateral damage. And our ability to do that better and better

01:19:19.120 is what's going to give people much more health in the future. And we're in inning two or one of that

01:19:28.180 journey. So as an investor, there couldn't be a more exciting place to hang out. And the advances

01:19:35.800 that we're making on a monthly or yearly basis are staggering. So when I noticed that that was

01:19:41.980 happening in the world, it just became clear to me that this was not only an exciting and interesting

01:19:46.760 place to be, but also one that could be enormously profitable, both financially and in terms of the

01:19:53.580 benefit to humanity that it delivered.

01:19:55.980 What got you interested in the problem of living longer personally?

01:20:02.160 Well, I guess I am not primarily preoccupied with living longer or with longevity as I am with

01:20:10.880 the reduction of suffering that already exists. And so people who are really obsessed with longevity,

01:20:17.980 maybe you are one of them, will make the argument that death is the great tragedy of human life and

01:20:25.040 that we basically don't realize it's so bad because we take it for granted and can't see any alternative.

01:20:31.780 That may be right. But I also think there's a lot of evidence in nature that death and the turnover of

01:20:39.040 populations has some utility. Steve Jobs talked about famously that this is nature's way of clearing

01:20:45.660 out the old ideas and bringing in new ones. And creating urgency. And creating urgency. But I don't

01:20:51.360 want to downplay the credibility of the argument for longevity. I just think that there are enough

01:21:00.960 people dying obviously prematurely from things that should be preventable. And there are lots of

01:21:07.260 people dying in the second half of their life or suffering from things that should be preventable that

01:21:14.080 we should start by focusing on those things. Now, it may be the case that longevity is the skeleton key

01:21:24.360 to all of those problems. That the number one risk factor for all diseases is age.

01:21:30.960 And that if we could just figure out what's happening with aging, then we could get out in front of all

01:21:37.300 these other diseases as opposed to playing whack-a-mole with all of them. That may be the case. And I think

01:21:43.280 it's a legitimate hypothesis to explore, but it's not the one that I'm starting from or else all of my

01:21:50.760 investing would be in longevity. Whereas most of my investing at this point is preoccupied with

01:21:55.700 individual diseases and whether we can arrest them and detect them earlier.

01:21:59.860 So what are you most passionate about right now? Not necessarily from the investment thesis

01:22:04.920 perspective, but from the technology or promise that the idea holds?

01:22:09.440 Well, the holy grail for me is approaching what I think of as a singularity-like moment in

01:22:21.020 biomedicine. People throw around the term singularity mean all sorts of things, but I

01:22:25.680 actually believe it's useful in this context. Who coined it?

01:22:28.220 You know, it may have been Kurzweil, Ray Kurzweil, or I'm not sure, but, or it may have been some of

01:22:34.300 those old cybernetics dudes. I first heard it from Ray's work. I don't know enough. Why don't you

01:22:41.420 explain briefly what, how he's referring to it? Well, I think Ray Kurzweil is referring to the

01:22:46.280 singularity as this moment when human minds merge with machines. And so we, for example,

01:22:56.220 download our brains into computers and then are effectively immortalized because our consciousness

01:23:02.740 exists now in a digital format. That's not the singularity I'm talking about. But the concept of a

01:23:11.900 singularity is that this is a moment at which sort of everything changes and there's an almost infinite

01:23:17.480 acceleration in our capacity to do something. The singularity that I envision in our understanding

01:23:25.420 of biomedicine is the moment at which we can digitally represent complex biology and therefore

01:23:36.840 study it at zero marginal cost. So let me explain this right now, what we essentially do in biology

01:23:46.060 and what medicine has always tried to do is understand how the machine of a living system works.

01:23:52.340 And in the work you do on metabolism, for example, the ways that we try to represent how the machine

01:23:58.820 works are incredibly complicated. There are these flow charts with all these arrows and pictures, and

01:24:03.900 this connects to this, and this connects to these 10 things. And if you pull this lever, these four

01:24:08.320 things happen. And how do we even make sense of that? A way station to the singularity I'm talking

01:24:14.980 about is the field of so-called systems biology, in which what people do is they observe these biological

01:24:21.500 systems, and then they try to translate the mechanistic connections that they have uncovered

01:24:29.540 into code, into formal representations, into a formal language that captures how something works in a way

01:24:40.480 that a computer can understand, and in a way that is unambiguous, and not sort of stifled by the

01:24:48.100 inherent ambiguity of human language, by which we describe a lot of these things in everyday life.

01:24:53.620 Ultimately, systems biology, I think, will allow us to simulate biological systems in their full

01:25:04.380 glorious complexity. And when we can do that, which we might be 20 years away from, or we might be a

01:25:10.640 thousand years away from, I don't know. But once we can do that, then we can effectively run

01:25:17.080 intervention-based experiments on digital systems at zero cost. And that would be the moment at which

01:25:26.000 we will start to understand biology and the ways that we can intervene in it at an extraordinary

01:25:32.960 rate. Until then, unfortunately, we have to do tests on humans, or mice, or dogs, or monkeys.

01:25:42.440 And unfortunately, we can only learn one thing at a time that way. We do controlled experiments where

01:25:49.220 we hold everything constant except one variable, and in so doing, we try to understand how that

01:25:54.300 variable works. And this is a very slow way of learning, and it's a very expensive way of learning.

01:26:01.300 And although the people who participate in clinical trials are heroes to whom we owe most of our

01:26:07.800 medicine, clinical trials are a brutal form of learning that cause many people to die from

01:26:14.900 experimental medicines, or that cause many people to endure enormous suffering without any benefit.

01:26:22.320 And so that's the thing I'm most excited about, getting to that singularity of digitally representing

01:26:29.520 complex biology. Most of the things that I'm focused on as an investor, I view as somehow being

01:26:35.960 stepping stones towards that singularity. If you look at other fields that could benefit from the

01:26:41.740 ability to do digital experiments, consider macroeconomics. Wouldn't macroeconomics be even

01:26:48.500 more complicated than biology? In other words, this sounds like the hardest problem I've ever heard of.

01:26:54.240 All of these systems, people put in the same bucket of complex adaptive systems.

01:27:01.260 Kind of ground zero for studying these things is the Santa Fe Institute, an organization in which

01:27:05.860 I'm involved and I'm very passionate about. And part of the reason I'm passionate about all of these

01:27:11.220 types of systems is because they share certain characteristics. So in a kind of vague way,

01:27:17.540 they're all complex in just the colloquial sense. Like they're really complicated. There are a lot of moving

01:27:23.060 parts, but it may turn out that we can come up with theories that actually describe all of these types of

01:27:32.940 systems. So there's some sense in which these problems of economies and the weather and physiological

01:27:42.560 systems are all kind of similar. And we may be able to come up with theories that are predictive

01:27:50.040 that help us make predictions about these different types of systems. So that's a kind of intellectual

01:27:58.620 promise land that I think is worth chasing, whether or not it's going to happen or not. I think even

01:28:05.760 trying for those sorts of theories is going to generate useful information. Whether we'll be able

01:28:12.480 to simulate these things or not is sort of an open theoretical question, but I'm optimistic about it

01:28:19.040 because it seems like it's the logical endpoint of all of our digital technology is to create digital

01:28:24.960 representations of reality. And on the one hand, reality is really complex, but on the other,

01:28:32.180 it seems to be driven at the bottom from relatively simple theories. And so to the extent that the

01:28:40.420 laws of physics are ubiquitous and are driving everything from the bottom up, and that everything

01:28:47.880 that happens subsequent to the laws of physics are these emergent phenomena, it's not clear to me that

01:28:55.180 we won't come up with simple parsimonious descriptions of these types of systems. And the physics of the

01:29:03.800 universe before Newton probably also appeared intractably chaotic. So the way that anything looks to us

01:29:12.640 before we understand it is, I think, pretty imposing, which is why when we have breakthroughs,

01:29:19.920 they feel almost like this enormous relief because something that seemed really complicated becomes

01:29:24.820 really simple. Yeah. I mean, the obvious example, right, would be the coding of the human genome,

01:29:30.240 which is now approaching 20 years in its anniversary. Do you feel that that discovery has underperformed,

01:29:37.900 overperformed, or met expectations on what we would have thought prior to that codification?

01:29:45.680 The pithy aphorism that I think applies to genomics and a lot of things in technology is that it

01:29:51.780 is probably short-term overestimated and long-term underestimated. So certainly when the Human Genome

01:30:00.500 Project reached its initial milestones in Craig Venter's genome being sequenced, there was a

01:30:07.500 extreme optimism that was unjustified that now that we've cracked this code, we're going to figure out

01:30:14.100 everything in five or 10 years. Clearly that has not happened. And so to that extent, those predictions

01:30:21.700 have failed. On the other hand, the sequencing technology that underwrote that is now being used

01:30:31.760 used to sequence mRNA, to sequence methylation of the genome, to sequence neoantigens in cancer patients'

01:30:42.960 tumors, this has become an extraordinarily useful set of tools for reading all kinds of biology.

01:30:52.540 And so long-term, I believe the impact of the human genome and our understanding of it is almost

01:31:02.820 impossible to overstate because it represents a basic part of a toolkit that is now integral to

01:31:12.440 everything at the cutting edge of medicine. Yeah. So in other words, the tool might actually be more

01:31:17.540 valuable than the initial application. Absolutely. But the only way that people were able to justify

01:31:23.120 paying for that tool was in the hope. That's what I say about Bitcoin and stuff, by the way, as well,

01:31:33.940 because this is now part of my theory of technology, which is that you take something like Bitcoin and

01:31:40.100 all these blockchains and all this. I was an early investor in this company, Ripple, about six years ago

01:31:45.540 now. And so I've been interested in all of this for years, but I don't take the view that there's any

01:31:51.680 end point I predict and know with any level of confidence. All I can tell you is that having

01:31:59.320 the smartest computer science, 19 year olds in the world in mass shift their attention to an area of

01:32:08.960 technology is going to produce something. So whether Bitcoin becomes a replacement for the United States

01:32:16.280 dollar is almost irrelevant. What I'm interested in is what's going to happen now that millions of

01:32:21.760 people are working on this thing. Similarly, the human genome project was this kind of catalyzing

01:32:27.840 event is very vivid. Bill Clinton had a press conference with Craig Venter and it got everyone

01:32:33.660 excited. And that sort of moment is what inspires a lot of 18 year olds or a lot of 15 year olds

01:32:41.300 to go into molecular biology undergraduate programs and then to become genetics researchers. And, you

01:32:48.120 know, so this is how history happens. And a lot of things have already come out of genomics and will

01:32:53.880 continue to. Have you been paying a lot of attention? I mean, you alluded to it a moment ago, but have you

01:32:58.260 just on a personal level been paying a lot of attention to the liquid biopsy space? A lot. Yeah. I've been

01:33:03.140 following the liquid biopsy space now for about four years. Yeah. You and I spoke about this a couple

01:33:08.400 of years ago and I wasn't sure how much you'd still been into it. This is something that's very

01:33:11.980 interesting to me. You know, one of the things I talk about with patients is even with my most best

01:33:17.820 attempts at providing the most bespoke insights into how to prevent diseases. When you take a big step

01:33:23.960 back and say, let's look at the three main diseases that are going to kill most people in a civilized

01:33:30.000 society where you're basically taking care of the blocking and tackling that you've alluded to

01:33:34.840 earlier, it's atherosclerotic diseases, it's cancers and neurodegenerative diseases. And my personal

01:33:41.700 viewpoint is that the atherosclerotic diseases and the neurodegenerative diseases, we have much more

01:33:45.660 insight into how to prevent. Cancer is tough. And I sort of explained to people when you look at a blood

01:33:51.600 test and let's talk about the best blood test money can buy, it's probably offering you 70 to 80%

01:33:58.040 of your predictive value on the atherosclerotic side, probably offering you 60 to 70% of your

01:34:04.960 predictive value on the neurodegenerative side. It probably isn't offering you even 30% of insight

01:34:12.000 on cancer because of course the mutations that kill are somatic, not germline. So we're not measuring

01:34:18.920 those. We don't have great assays for measuring adoptive immune cell function. We can measure innate

01:34:25.040 immunity, but that's so crude and it's irrelevant in cancer. And so we talk about something else you

01:34:30.760 alluded to, which is how aggressive can we be in screening whilst solving for optimizing around

01:34:37.420 minimizing physical harm and emotional harm, physical harm from the actual screening tool,

01:34:42.380 emotional harm from risk of false positives, because the harder you look, the more you'll find.

01:34:45.780 And the open gaping, open hole in this is we can't do a liquid biopsy. So we got interested in this

01:34:55.620 together, you and I mutually around this ENOX2 protein, which was quite interesting, but unfortunately

01:35:02.180 that company, that technology doesn't really seem to exist right now. It also had a number of issues

01:35:06.960 with it, but basically it was a Western blot. It was a protein assay. They were looking at a protein,

01:35:11.660 this ENOX2 protein. And the belief was that that protein was found exclusively on malignant cells and

01:35:20.380 could only be shed off, not just cancerous cells, but cancerous cells that had the potential to spread

01:35:25.080 and attach to other cells. Now a company that we both know pretty well, Grail, is taking a different

01:35:31.100 approach that seems slightly more logical. Tell us a little bit about Grail.

01:35:35.320 Grail is an interesting company. It's a spin out of Illumina, which is the behemoth

01:35:40.460 genome sequencing business. They make all the machines that do sequencing that most people use.

01:35:47.800 And the premise of Grail is that by sequencing peripheral blood, you can find all sorts of

01:35:54.500 cellular refuse from somatic tissues. And that this refuse, if you are able to amplify it,

01:36:04.500 will tell you about the cells that it's coming from. And so one of the proofs of principle of this

01:36:11.940 concept in general was non-invasive prenatal screening. Pregnant women have in their blood,

01:36:17.900 I think, potentially 10% of their peripheral blood is actually coming from the fetus.

01:36:23.800 And so you can detect a lot of things about the fetus by looking at the mother's peripheral blood.

01:36:30.760 And in that case, maybe what you need to effectively do is separate the 10% from the 90% 0.98

01:36:36.880 in order to look at it. When you're talking about early cancers, you might be dealing with

01:36:43.840 sub half percent or sub 10th of a percent concentrations of tumor DNA that are showing

01:36:52.140 up in the blood. And so the premise of Grail is that if you sequence the blood with enough depth,

01:36:58.780 meaning you sequence it over and over and over again, you can detect just that DNA that's coming

01:37:06.460 from the tumor. And by doing so, you can identify things about it that give you a guide to where the

01:37:14.900 tumor is and what its genetic characteristics are. And this is much harder than what you just said a

01:37:22.740 moment ago, because the problem you described earlier has two things going for it, which is one,

01:37:27.380 you've got much more of it there. Let's just assume the number is 10%. But the other thing is half of

01:37:32.480 that DNA is foreign. That's right. 50% of that DNA is from the father, 50% is from the mother. So you 0.63

01:37:38.160 have half foreign greater quantity. If you have colon cancer, and we are lucky enough to get the RNA or 0.54

01:37:44.980 DNA of that colon cancer, it might have 20 mutated genes. It's effectively self-DNA.

01:37:50.560 Well, it is self-DNA. Exactly. But it may be tumor DNA. It's tumor DNA.

01:37:55.660 It's tumor self or tumor. Yeah, yeah, exactly. So I mean, that's a double whammy. It's not just a needle in a

01:38:01.060 haystack. It's like a needle of hay in a haystack. That's correct. And it's also difficult because- I just came up with

01:38:06.880 that, by the way. I'm really proud. That's the first smart thing I think I've ever said. That's a good metaphor.

01:38:09.880 The needle of hay in the haystack. I like that. You can use that as much as you want. I will. I'm

01:38:14.340 going to start right now. When you're looking for that needle of hay in the haystack,

01:38:18.620 the premise of your search has to be that you know what you're looking for. In other words,

01:38:25.300 you have to identify a series of potential oncogenes in which you're seeking aberrant genetic

01:38:33.480 variants that would indicate the gene has essentially been co-opted by the disease and is now

01:38:39.760 working for the cancer as opposed to working for you who want to defeat the cancer. And there are

01:38:48.480 so many different ways that this technology is going to be difficult. That is matched by billions

01:38:56.580 of dollars that people are investing in this because solving this problem is, as the company's name

01:39:02.440 suggests, sort of a grail. I'm not an investor in grail, but I've looked at them and I've also looked at

01:39:08.280 many of their competitors. There's another approach that's really cool that I'm investing in right

01:39:13.700 now. A company out of Boston called Glimpse is a business that was founded by Sangeeta Bhatia,

01:39:20.260 who's an extraordinary researcher at MIT. Sangeeta is a hepatologist, but a bioengineer whose work has

01:39:28.320 uncovered two really interesting ideas that I think are useful in cancer screening.

01:39:32.440 One is that when cancer is setting up shop in a tissue, it remodels the microenvironment so as to

01:39:39.520 build defenses for itself against the immune system and against the other physiological processes that

01:39:45.260 would stand in its way. And it turns out that there are in the human genome 550 different endoproteases

01:39:53.760 that are in some combination utilized in the remodeling of tissues in different processes.

01:40:02.040 So they're used primarily in healthy processes, but when cancers are beginning, these proteins get

01:40:09.840 recruited or these enzymes get recruited rather to do this remodeling work. They're sort of the

01:40:14.820 construction workers of the disease as it gets set up. And Sangeeta's work and the work of several other

01:40:21.780 researchers that she's leveraging has identified what proteases tend to be implicated in the early

01:40:29.520 formation of different diseases, ranging from fatty liver disease to liver cancer to lung cancer and a

01:40:36.840 set of others. And if you can, it turns out, identify 10 or 15 enzymes that you basically only see

01:40:46.140 upregulated enormously when a particular disease is beginning. Then there's another way of detecting it

01:40:52.760 that Sangeeta is working on, which is to engineer nanoparticles that you send into the body. She calls them

01:41:00.440 a synthetic biomarker. You let them circulate and you design them such that they break apart if they

01:41:08.660 encounter those enzymes. And when they break apart, they fragment into smaller nanoparticles that can

01:41:16.900 be detected in the urine. So Sangeeta's approach is rather than look for these trace amounts of

01:41:24.020 refuse in the blood, why not send essentially a team, a SWAT team into the body to circulate and hunt for a

01:41:33.960 thing. And if that SWAT team finds the thing, get a much larger signal in the urine. I think that's

01:41:42.700 a very elegant approach. Will that have tissue specificity? It'll have tissue specificity to the

01:41:48.720 extent that the disease you're looking for has a different enzymatic signature when it's in different

01:41:56.300 tissues. Got it. So in the context of, let's say you get the signal and you would have to believe

01:42:07.160 then that you're going to have different enzymatic pathways in the mesenchymal system of colon cancer

01:42:13.040 versus lung cancer. I see. So it's not, you know, when you were saying this, I wasn't, I'm not familiar

01:42:18.120 with lymphs. So I was, I was wondering if what you were going to say was it got taken up in residence

01:42:22.260 and then you basically would do, you know, it was labeled with gadolinium or something and you do

01:42:26.320 an MRI and you'd see, well, Hey, this stuff is being broken apart in the lung. Therefore that's

01:42:30.700 where we want to look, but that's not what you're saying. This is all happening in the periphery in

01:42:35.260 the blood itself. The breakup of the nanoparticles is occurring at the site of the disease, but the

01:42:41.860 refuse from that interaction is going through the kidneys and into the urine. It would be interesting

01:42:48.400 if there was a way to see if there was a way to capture the local signature at the tissue site of

01:42:53.640 origin. There may in fact be ways to do this. And there are several nanoparticle based approaches

01:42:58.140 that follow the lines that you described, where it's an imaging agent that is meant to bind

01:43:03.560 specifically to the tissue. What's cool about this though, is you don't need to target. You just use

01:43:09.380 the circulatory system to get these things everywhere. And if they detect the thing that you're probing,

01:43:15.880 they change. And that's what you measure. I think this is a really clever workaround for some of

01:43:23.600 the problems that I think have been bedaviling liquid biopsy. But I do think liquid biopsy will

01:43:28.240 probably work. And the other reason it may work, and there's a company called Freenome, for instance,

01:43:33.200 that's taking a different angle on it. Freenome's approach is don't look for the tumor DNA in the

01:43:40.400 blood. Look at the entire genome and see if there is any sort of smoke from the fire. If the cancer's

01:43:51.080 the fire, try to detect the body's systemic response to the cancer and recognize that as

01:43:59.420 aberrant. In other words, even if the cancer itself is sending off a very small signal, maybe the systemic

01:44:06.020 response to it is a much larger signal. And you can detect that. So people are going to try and get

01:44:13.140 after this in many different ways. My prediction is that within a decade, we'll have one or more of

01:44:19.000 these that work quite reliably. And each of us will, at our annual physical, have a routine blood test

01:44:25.100 that is pretty good at detecting cancer. Now, that begets another problem that people have raised, which

01:44:30.160 is, you know, what if we're all getting cancer all the time? And most of the time, the immune system

01:44:34.660 is disposing of it in short order. If that's happening, then we'll have another problem to

01:44:40.600 solve, which is how do we know when we should treat people who are detected early versus when

01:44:45.840 shouldn't we? But that's fine. Problems create solutions which create more problems. And that's

01:44:51.960 what we do. You know, that's the problem that I probably worry about the most. I mean, I guess the

01:44:56.880 second most. The problem I worry about the most is, can we crack this? But that problem, I got this

01:45:02.200 glimpse into with all the deep dive I did into Oncoblot three years ago, which was, I was asked

01:45:09.620 by a company that was interested in acquiring them to help with the due diligence. I actually

01:45:14.620 thought it was complete buffoonery, total quackery, complete and utter nonsense. But after about a year 0.93

01:45:21.340 and a half, I thought there really was something there that unfortunately had been sort of bastardized

01:45:27.980 by unsavory characters that were involved in basically creating kind of a profit center around

01:45:33.580 selling alternative products that were complete nonsense. But when we did our own analyses on raw

01:45:40.780 data that they gave us, we saw something pretty interesting. There's two questions you have to

01:45:46.480 ask, right? When you're doing this is the first is, and rather than describe them in terms of

01:45:51.400 sensitivity, specificity, negative predictive value, and positive value, just describing this stuff in

01:45:55.660 English, you want to understand something, which is if you do a blood test on a person, and it comes

01:46:01.300 back and says nothing, how confident are you that they have nothing? That's the concept of negative

01:46:08.460 predictive value. And then conversely, if the test comes back and says you have something, how confident

01:46:15.720 can you be that you have something? That's the idea of positive predictive value. Now, if you look at how

01:46:21.360 Oncoblot was created and you look at how the FDA gave it a type of approval, which is not an FDA

01:46:27.900 approval, but there's, you know, and I don't want to get into the weeds on that stuff, but basically

01:46:33.200 it was approved, quote unquote, only for patients presenting with metastatic cancer of a, with an

01:46:39.780 unknown primary because of its remarkable ability to create a differentiation around 27 different types

01:46:46.080 of tissue. So someone shows up with a lung nodule that's not lung cancer. It's actually important to

01:46:51.200 know, is that breast? Is it thyroid? Is it what? Because they could have a primary that's occult,

01:46:55.240 but it's still metastatic cancer. And this is a life and death decision as to how to treat it.

01:46:58.960 And that's where Oncoblot was actually very useful. And so Oncoblot became really, really good. If

01:47:05.400 someone had cancer and you did a blood test on them, you had about 99.4% likelihood of correctly

01:47:11.740 guessing, not just that they had cancer, but what kind of cancer it was. But the flip side's really

01:47:16.940 important, which is if you take a whole bunch of people that don't have cancer and you can only do

01:47:21.000 this out of a blood bank prospectively, you can't even do this in the population. What's the likelihood

01:47:26.040 that you're not going to be over calling cancer? And when we did an analysis, and it was not a huge

01:47:31.400 analysis, we were limited by the amount of data we had. What we saw was by about 5X, it overestimated

01:47:40.260 the prevalence of cancer at a given age for a given histology. And our interpretation of that was

01:47:47.980 either this is categorically useless, or it's picking up a bunch of cancers that don't go on

01:47:54.580 to become cancers. They get winnowed out by the immune system, which is exactly what you just

01:47:59.320 described. And that raises a very difficult question as a physician, as a patient, which is,

01:48:06.020 what do you do if it's your annual physical? Now, sometimes you'll get lucky, right? When

01:48:10.820 what's getting lucky, getting lucky is, it says you have colon cancer, and you can go into a

01:48:15.960 colonoscopy, which is a great way to assess that. It's not a great way, though, if a colonoscopy

01:48:21.540 costs the system three or $4,000. And you now are giving people 1000 times more colonoscopies than

01:48:29.660 we're giving today, because we've got some screening test that suggests this. I'm reading

01:48:36.020 right now a phenomenal book that everyone listening to the podcast, and you should read the book of

01:48:40.160 why by Judea Pearl, or Judea Pearl, who's a professor at UCLA. And his area is causal inference. And he goes

01:48:49.600 through in the book, some of the fascinating math that corresponds with diagnostics in particular. And

01:48:57.660 frankly, I'm always confused by sensitivity and specificity measurements, and which is the inverse

01:49:03.220 of which. And so do you want me to send you my primer? Please send me it. I think we've put so

01:49:07.700 much work into this. It's our favorite. I believe I understand it now. But for your listeners, I'll just

01:49:14.440 from the book, give you one example of it. And the conditional probabilities that express it. People

01:49:22.300 will probably remember a few years ago, the guidelines on mammography changed to suggest that women, 0.99

01:49:27.160 I think what under 40 should not get mammograms as a screen. And I was always a little bit skeptical of

01:49:34.500 that conclusion, because there's there's certain intuitive way in which you think, well, if it gives

01:49:39.200 you a false positive, kind of who cares, because just go get a CT, whatever, you know, I mean, what's

01:49:44.800 obviously don't go get your breast cut off. But if all the cost is to a woman is that she goes and 0.99

01:49:50.840 gets another test, maybe that's not good for the system, from a cost standpoint. But if you're the

01:49:57.980 patient, you certainly don't want to not get screened if you do have breast cancer. And so Pearl, though,

01:50:05.860 in his book walks through the actual statistics of this. So if the incidence of breast cancer in the wild

01:50:11.280 population is one in 700, that means that out of 4000 women, say five of them would have breast cancer is 0.87

01:50:19.100 what you'd expect. If the sensitivity of the test is 73%, which sounds high, that means it's deplorable,

01:50:27.620 that still means, which is what the sensitivity is of a mammogram, 73%. It means that if you get a positive test

01:50:37.840 result, you still have a sub 1% chance of having breast cancer. Now, Bayesian statistics adds another

01:50:47.620 level of complication to this in that Bayesian statistics gives us a way of updating probability

01:50:55.420 estimates that we make. And so there's a difference between you being a member of the random population

01:51:02.060 and you being a woman who say has a BRCA mutation that radically increases your chances of getting breast

01:51:08.960 cancer. And so if you are someone who say has a 50% chance of getting breast cancer in your life, then a

01:51:16.880 positive mammogram gives you a much, much higher likelihood because it effectively updates your estimate that

01:51:24.760 you began with of your likelihood of getting the disease. And so what we probably end up with in any of these

01:51:30.960 tests is a way of stratifying populations based on their genetic risk of getting different diseases.

01:51:38.800 And it's only within each of those cohorts of relative risk that we can consider the importance or

01:51:49.720 usefulness of a screening test, like a liquid biopsy, because its utility to each person depends upon

01:51:58.720 the extent to which that person is already at risk of the disease you're surveilling.

01:52:04.860 Yeah. The other thing I add to that is, is the way we think about this problem clinically is if I

01:52:10.780 gave you a piece of Swiss cheese and I said, how many pencils could you drop through here? It's a lot.

01:52:15.880 But what if I put four pieces of Swiss cheese on top of each other and rearrange them in such a way

01:52:21.100 that one and only one hole could accept the pencil? And so where I see the real application of the

01:52:29.240 liquid biopsy is as follows. So let's use the mammography example. So, so mammograms are good

01:52:35.280 for some things and bad for some things. And let's ignore cost because I think once you layer cost into

01:52:41.900 this, it becomes way too complicated a problem. So it's highly relevant. But as a physician, I'm trying

01:52:47.900 to solve the problem of the patient, not society. That's my cop-out. So then it becomes a question

01:52:53.280 of, is the radiation significant? And the radiation mammography is trivial. MBI, of course, which is a

01:52:58.140 more elaborate type of mammogram, has very high radiation, but we almost never do that anymore.

01:53:02.440 So now we've taken away physical harm and we've decided to just discount cost. Well, mammography is

01:53:07.900 good for seeing calcified lesions, but it's not good for seeing non-calcified lesions. Conversely,

01:53:14.500 MRI won't pick up a calcified lesion, but if you're using something called diffusion-weighted imaging,

01:53:20.160 it's infinitely better for picking up pretty much everything else. So then the question becomes,

01:53:25.060 what happens if you layer the mammogram with the DWI MRI with the liquid biopsy? Those are three

01:53:31.700 pieces of Swiss cheese. And then the fourth layer would be the Bayesian piece, which is what is the

01:53:37.100 woman's probabilistic likelihood of breast cancer based on her family history, known mutations,

01:53:44.700 and known mutations of unknown clinical significance, which is basically the types of

01:53:49.060 mutations we have. And now all of a sudden you've got a much smarter way to think about it. So my only

01:53:53.720 take on this one is if and when the liquid biopsies become viable, I don't think they should ever be

01:53:59.400 leading candidates. I think they should only be used as confirmation. So once you do, you know,

01:54:06.400 X, Y, and Z, you then come back. So for example, if you do, you know, you go and get a whole body MRI

01:54:10.980 and there's a little shadow in the pancreas. Well, what's the next step there? That's a big step.

01:54:16.660 Are we going to submit that patient to an ERCP or a biopsy of that? That's the real deal.

01:54:22.360 Well, I think you're right to some degree, but I would push back against the a priori assumption

01:54:31.180 that we wouldn't lead with it because that does really hinge on sensitivity and specificity.

01:54:37.280 You know, I mean, if we have 99% specific, sensitive and a hundred percent specific

01:54:42.860 liquid biopsies. Over what timeframe though? That's see, I think that that's the challenge

01:54:48.520 with the liquid biopsy is you're right. I mean, technically there's no test that can have a hundred

01:54:53.800 percent specificity, a hundred percent sensitivity. That's sort of like having a rock receiver operating

01:54:57.660 characteristic curve. That's a square. No such thing exists. What I think is you take something

01:55:03.300 where you can, can you generate a liquid biopsy that has 100% sensitivity, even if specificity is

01:55:11.120 only 95%, which a lot of people think is great, but 95% is not.

01:55:15.860 I actually think you want the inverse. I think you never want false negatives. In other words,

01:55:22.280 you want a hundred percent specificity. You never want to tell people.

01:55:25.060 Yeah, yeah, yeah. Sorry. I misspoke. You want negative predictive value to be a hundred percent.

01:55:29.480 You never want someone who has cancer to be told they don't have cancer.

01:55:33.380 You can tolerate some people who don't have cancer being told that they do.

01:55:38.400 That's right.

01:55:39.100 The other dimension here, I mean, I know we put cost aside, but you can't really,

01:55:43.360 because the question is what are the relative costs of these different surveillance measures?

01:55:47.720 If liquid biopsies are trivially inexpensive, which is the goal, say this glimpse test, you know,

01:55:55.360 I mean, forget all the health economics of it. If just the cost of doing it might be five or $10

01:56:00.100 or $20 or something. If the sensitivity and specificity are both high enough,

01:56:06.080 then of course you would rather use them as a universal screen and then follow up with much

01:56:13.800 more expensive interventions.

01:56:17.080 Yeah. And with things like fecal occult blood testing or fecal DNA as a precursor to colonoscopy,

01:56:24.860 it makes tons of sense because we have a pretty well understood pathway for the tumor.

01:56:31.560 Again, not having an answer to this dilemma, my big concern with everybody shows up and willy-nilly

01:56:40.120 gets a liquid biopsy that shows you've got pancreatic cancer is what do you do next?

01:56:45.580 Well, that's a question of sensitivity.

01:56:48.040 Yeah. Yeah. So you go and you get the MRI and the MRI shows nothing. To me, that's a great outcome.

01:56:53.180 That's the best outcome is the MRI shows nothing. And if the test is vetted and you're positive,

01:56:58.800 and you can be confident that that was not a false positive, but rather it's a plea, a preclinical

01:57:04.260 tumor, then you're in a pathway of what's the righteous path of surveillance going forward.

01:57:10.160 And equally important, if not more important is what steps can the patient take in that moment

01:57:14.900 to ward this thing off, to aid their immune system? Is that something as quote unquote simple

01:57:21.240 as you got to sleep more, give your immune system a boost, figure out ways to de-stress yourself,

01:57:26.780 which of course is pretty hard given that I just told you you have pancreatic cancer.

01:57:30.140 But you know what I mean? Like, so anyway, that's, that's the art of, of how to think

01:57:34.760 about using these things, which I, I look forward to being able to think about these problems.

01:57:38.860 I mean, I think what we're getting at is a, a really interesting framework though,

01:57:42.360 that is almost certainly going to become standard, which is the utility of the genomics and family

01:57:49.440 history, which is, you know, genomics tells part of that story, but it doesn't tell you what

01:57:53.800 happened with your ancestors. So family history remains incredibly important, but you'll, you'll

01:57:59.580 start when a baby's born with a whole genome and a family history, and that will generate for you

01:58:08.460 the prior probabilities of different diseases being the diseases that afflict that person in the rest of

01:58:14.460 their life. Those prior probabilities will in a personalized way determine the utility of a wide

01:58:23.100 array of low cost screening tools that we're going to have at our disposal. And we'll also lead you to

01:58:30.720 make personalized estimates of the sensitivity or rather the diagnostic power of those tests for

01:58:40.000 that person, which to the Bayesian point is different for every person. This is a fundamentally

01:58:45.220 different way than how we think about using diagnostics today. There's one detail of it that I don't have

01:58:51.380 enough knowledge about, but you've made me interested in, in your, in your Swiss cheese metaphor,

01:58:55.440 which is what from a probabilities and statistics standpoint is the right way to think about that

01:59:02.920 Swiss cheese metaphor does adding tests on top of each other. How do you mathematically combine

01:59:10.880 the sensitivity and specificity of the different tests? And can you even do so without running

01:59:18.080 clinical trials on the combined usage of the tests to have some ground truth to reference them against?

01:59:25.960 My intuition is directionally you can do it, but you will not have a number. And in other words,

01:59:33.660 you can't say that mammography plus DWI MRI plus liquid biopsy will have this sensitivity and this

01:59:39.400 specificity calculated from the three pairs that you layered on. But I think philosophically,

01:59:47.420 what you're trying to do is arbitrage each strength and weakness of different screening tools.

01:59:56.340 I think that's right.

01:59:56.980 So it's a conceptual model more than it is a theoretical and precise.

02:00:00.900 A hundred percent where it gets interesting though is in the math because that's where it gets very

02:00:06.200 counterintuitive. So if you think about the example that I'm citing from Pearl's book,

02:00:11.140 the sensitivity of the 73% test says that if you have breast cancer, there is a 73% chance

02:00:21.820 this test is positive. This test is positive. The much harder probability is what is the chance that

02:00:31.200 you have breast cancer? If this test is positive, these are two different conditional probabilities.

02:00:38.220 And the shocking result is that if the test says you are positive, you have less than a 1% chance

02:00:47.800 of having breast cancer. So what's counterintuitive is if we thought, oh, we have a one test that's 73%

02:00:54.800 sensitive. We have one test that's 90% sensitive. We have one test that's 85% sensitive. You would assume

02:01:02.860 that if you layer the three pieces of Swiss cheese, the possibility that if you are positive on two of

02:01:11.280 those three and you have the disease is quite high. It may turn out though, that it's not very high.

02:01:17.840 You know, the thing I always caution people against is you never want to talk sensitivity without

02:01:21.920 remembering the specificity and vice versa. So in our little primer that we use with our patients,

02:01:27.000 we use two extreme and very glib examples. If I mail a letter at random to a thousand women and tell 0.58

02:01:35.660 them all that they have breast cancer, I have a hundred percent sensitivity. I have zero percent

02:01:42.180 specificity, but I have a hundred percent sensitivity because let's just assume three of those women have 1.00

02:01:47.940 breast cancer. We told them all they have breast cancer. Nevermind the fact that nevermind the fact

02:01:52.780 that 997 of them don't have breast cancer. And I incorrectly told them they do, but you can have

02:01:58.000 a hundred percent sensitivity and you're still, it's a dog shit test if the specificity is really 0.98

02:02:03.720 low. And similarly, you can send a thousand letters out to women at random and say, you absolutely don't 0.99

02:02:09.740 have breast cancer. And you will say that with 100% specificity. Now what's cool about, and this may be

02:02:16.460 true of incumbent diagnostics as well, but what's kind of cool in the liquid biopsy companies I'm looking at,

02:02:22.020 and I'm here exposing my lack of statistical knowledge, is that because the test is effectively

02:02:29.160 a computer analysis of data, you can manually trade off between sensitivity and specificity.

02:02:37.480 And that's like a PSA. Where do you decide the cutoff is?

02:02:40.720 So for the liquid biopsies, what everyone is doing is they're saying we have to have a hundred

02:02:45.640 percent specificity. We can't tell anyone who has the disease that they don't have it.

02:02:50.680 And that's the way it should be. That is how it should be.

02:02:52.660 I think that that's, so what, you know, if you picture one of those, and maybe we'll link to

02:02:56.560 this in show notes. So people know what we're talking about when we talk about a receiver

02:02:59.420 operating characteristic curve, but that's, you picture pulling to the upper left, tightening

02:03:05.880 that curve, getting it as close to an area under the curve of one. And that was the thing I could

02:03:11.760 never get those Oncoblock guys to understand was they couldn't understand. They thought of this

02:03:17.680 exclusively as a binary. Yes, no. And it's like, no, no, no. There's no such thing as yes. No,

02:03:22.240 you could say PSA is prostate cancer. If it's higher than one, and you will catch every single

02:03:28.620 person who had prostate cancer and a million people who don't conversely, you can make the cutoff 20

02:03:35.000 and you'll miss a million people with prostate cancer, but you'll be guaranteed that everyone

02:03:40.420 who you say has prostate cancer has prostate cancer. That's facetious and not even true,

02:03:45.020 but directionally that's the problem that you face. And I like the question you posed DA,

02:03:50.840 which, which I've never thought of, which is, could you take my sort of hand-waving Swiss cheese

02:03:56.480 approach and mathematically map it out without doing the clinical trial? And that also goes beyond my

02:04:03.780 pay grade. I would have to consult with a statistician. You could certainly do it retrospectively.

02:04:10.300 The way that these tests are developed is that a researcher gets lots of tissue and blood samples

02:04:18.260 from patients and they know which of those are coming from cancer patients or not. And then they

02:04:23.820 build a set of assays and analytics pipeline that look at the samples and try to class them correctly.

02:04:31.520 And so you could certainly run multiple of those types of tests on the same samples from the same patients if

02:04:40.200 you had large enough sample volumes and you could develop these tests in a layered manner. I don't think that

02:04:49.360 the statistics would hold if you developed them separately from one another. But that effectively is a

02:04:55.820 controlled experiment that I'm describing.

02:04:57.300 Yeah. If you did it as a biased and unbiased sample, so you'd have to bifurcate the sample,

02:05:02.180 do all of your learning on the biased piece, and then only verified on the unbiased piece.

02:05:08.940 That's correct.

02:05:10.140 And hope that you had the data.

02:05:11.900 That's, you know, that's how they, that's how they develop these is they,

02:05:14.400 they teach an algorithm essentially what the differences are between healthy and disease.

02:05:20.420 And then they feed it new samples and they ask whether those new samples can be classed

02:05:26.140 algorithmically into either of the buckets.

02:05:29.760 I don't know. The liquid biopsy space is so interesting to me. I really, you know,

02:05:34.620 I tell my patients that I think this is the single area that will most influence our care

02:05:41.000 hopefully in the next five years is right now. Like, you know, we're learning a lot about

02:05:46.400 cardiovascular disease. We are still learning, especially in the inflammatory side of that

02:05:51.740 disease. We're not learning a hell of a lot around the lipoproteins that, you know, we're

02:05:54.980 just trying to educate people to actually know what's true and what's not true. But where we're

02:05:59.940 learning geometrically is around inflammation and potentially with these two clinical trials

02:06:05.240 that were published this year, or one that was published this year, one that was halted likely,

02:06:10.060 the, the methotrexate study was halted early, likely because of a positive effect.

02:06:14.000 And that won't be announced until the fall, but that's two really interesting proof of

02:06:19.200 concepts. So the anti-IL-1 study, and then the methotrexate study where making no change

02:06:25.100 in lipoproteins, you're improving cardiovascular outcomes. In my mind, the next frontier on

02:06:30.140 cardiovascular is what can we do to strengthen the endothelial resilience? Because if you have

02:06:34.720 a strong endothelium and muted inflammatory response, and you can control lipoproteins,

02:06:39.260 you're taking the only inevitable disease our species has ever faced and knocking it on its

02:06:44.420 heels. But cancer, man, like we're no better at curing cancer today than we were 40 years ago with

02:06:50.180 five exceptions.

02:06:51.400 Well, cancer, like aging, seems to be an enormously resilient set of processes. And, you know, biology

02:07:00.920 and evolution have been pretty thrifty. So the mechanisms, as you know, of many cancers are

02:07:08.300 mechanisms that are used in early fetal development by the body. So these are tools that our organism

02:07:14.880 has for developing in the first place. And unfortunately, they run some risk of getting

02:07:20.340 reactivated later in life and doing crazy stuff that's contrary to our interests. And so this trade-off

02:07:28.060 between the apparatus that we've been given to become healthy and mature and the processes that

02:07:34.460 ultimately are undoing seems quite difficult to overcome. Atherosclerosis is a less obvious one,

02:07:43.940 although the sort of endothelial damage that is at its core and the function of the immune system

02:07:52.420 in the processes that ultimately degrade that juncture are also used for all sorts of good

02:08:00.760 things. I mean, the immune system seems to be at the core of much of this, that, you know, if you do

02:08:05.980 view atherosclerosis ultimately as a sort of auto-inflammatory disease, then you can think of it as a trade-off

02:08:14.120 against the immune system. Yeah. Just like allergies. Right. Yeah, exactly. And, you know, we,

02:08:22.020 I'm getting allergy shots right now, which funny enough, I guess have always been referred to as

02:08:27.620 immunotherapy before it was hot. But, you know, I mean, I would still rather have to get allergy shots

02:08:34.040 than be susceptible to almost certain death if I didn't have this immune system.

02:08:40.140 Do you have any anaphylactic reactions or are your allergies pretty mild? Thankfully, no.

02:08:45.580 And thankfully, I haven't had any anaphylactic reactions to the immunotherapy.

02:08:51.000 Yeah. That's always bad. Which is still witch doctor science. One of the things Sean Parker's

02:08:56.740 working on is allergy. And it's one of the things that led him into immunotherapy because

02:09:01.280 he has horrible allergies and so has always been fascinated by the immune system. And when he

02:09:08.180 perceived that there was a crossover between his allergies and cancer biology, he got obsessed with

02:09:15.360 the idea that immunotherapy was going to be the root. But allergies themselves, as mundane as they

02:09:21.460 are, still can't be addressed in a particularly rigorous or precise way. So precision medicine for

02:09:28.460 tree pollen allergies is still beyond the frontiers of current medicine.

02:09:33.420 Although the work that Dr. Nadeau, who is the largest sort of recipient of the work that Sean's

02:09:39.580 doing up at Stanford, I mean, she is amazing. And the work that they are doing, I mean, I've sent

02:09:44.280 multiple patients there. Carrie's always very kind and takes my referrals. I mean, I have seen before my

02:09:50.460 own eyes people who once would have died from peanut dust and they can eat peanuts again through this

02:09:56.220 sensitization. So it's, yeah, I don't, I sometimes wonder if Sean gets enough credit for the amazing

02:10:04.420 stuff he's done, both in terms of funding cancer research, but also this incredible center at

02:10:08.860 Stanford that, you know, little by little we're seeing, you know, Mount Sinai is doing this in New

02:10:13.140 York. CHOP is doing it in Philly. I mean, other centers are still, are taking this research forward.

02:10:18.640 And I don't know. I mean, I think for people out there listening to this who have children or who

02:10:22.500 themselves have anaphylactic reactions, I am way more optimistic about this than I was four years

02:10:27.360 ago. Well, and this is, you know, it comes back to the earlier point about the future's already here.

02:10:31.840 Yeah. Now how do you make that accessible to everyone? How do you make it accessible to everybody? And

02:10:35.420 how do things that make sense and have been proven in a sense become standard of care? Unfortunately,

02:10:44.200 in the United States, and this is something that I discovered very slowly, it's not as if there's

02:10:50.260 one arbiter of standard of care. You've got all these different medical societies that govern

02:10:55.700 physicians in, and standards in clinical guidelines in every specialty area. They have their conferences

02:11:02.740 and their board of directors and the payers sometimes respect their clinical guidelines and

02:11:09.060 the government sometimes respect their clinical guidelines. So one of the things that in terms of

02:11:14.040 human systems, we should absolutely put more energy into is how once something like Dr. Nadeau's work

02:11:24.580 is sort of definitive science, do we rapidly make sure that it becomes standard for every patient

02:11:33.500 everywhere? And right now it has to go through this long cascade of human minds and bureaucratic

02:11:40.420 organizations, which is an enormous detriment to patients. I mean, this is something that has nothing

02:11:47.040 to do with science. It just has to do with people. And it's the sort of thing that physicians should

02:11:54.080 care much more about as a community. It's the sort of thing that people like, I think, Atul Gawande

02:12:00.820 rightly are always reminding us to think more about. It's very easy to get drawn into the science and

02:12:07.900 technology frontiers, but we forget that we could make such an enormous impact with just these simple

02:12:15.600 changes in our own behavior. DA, I know I could sit here and have this discussion for another two

02:12:20.900 hours, but I want to be respectful of your time. And you've been super gracious to open up your home

02:12:25.680 this morning. Where can people learn more about you? I know you've got a huge following on Twitter,

02:12:31.400 so they can obviously see you there. What's your Twitter handle?

02:12:33.700 It's DA Wallach. That's D-A-W-A-L-L-A-C-H. And I have a website, which is equally simple,

02:12:42.160 DAWallach.com. And I'm on Instagram, also DA Wallach. I tend not to do that much social media

02:12:50.580 these days, but I still occasionally post pictures. And I'm very easy to reach too. So I presume that in

02:12:58.340 your audience, there is a mixture of crazy people, geniuses, and smart, nice people who think they're 0.71

02:13:07.880 geniuses, but are not. And I'm particularly interested in meeting the geniuses. So any of you

02:13:14.660 who are listening- So right now, anyone listening, please put yourself in the bucket of crazy genius or 0.97

02:13:18.760 smart, not genius. Correct. And if you're in bucket two, can you reach out to DA? 0.93

02:13:23.340 If you're smart, not genius, but do have an interesting novel idea that itself could be

02:13:29.220 deemed genius, I'm also interested. That's great. Do you still perform?

02:13:34.280 Not regularly. The most recent performance I did was after a dinner party when I was embarrassingly

02:13:42.960 asked to play a song on someone's piano in their living room and got peer pressured into doing it.

02:13:51.080 But most of the time I don't perform because I don't have a great way of performing.

02:13:55.800 If someone was going to buy just one piece of music, would it be your most recent album? What

02:14:01.080 would you recommend? I think so. I would recommend my most recent album, which came out in October,

02:14:06.280 about three years ago. Yeah. Yeah. It's called Time Machine. Oh, I thought it was 2016. It came out.

02:14:10.700 Was it 15? It may have been 16. I don't even remember. You're not good with dates. You told us this.

02:14:14.440 I'm terrible with dates. Okay. I think it was 2016. I was thinking the other day

02:14:18.420 that it felt like it was a year ago. And then when I sort of realized, oh, it was three years ago,

02:14:23.440 I thought, oh shit, I should probably finish some of these songs I've been writing, 0.88

02:14:27.360 you know, because I can't go like a decade between it. I don't have the Axl Rose credibility to wait 0.97

02:14:35.360 that long between albums. All right. So Time Machine it is, which I believe came out in 2016,

02:14:40.440 but I could be wrong. Of course, like a fool, I bought all your stuff on iTunes. So I overpaid for it.

02:14:46.940 I could have just been on Spotify. Well, if you're not listening to it much,

02:14:49.420 I made out like a bandit on that. So thank you.

02:14:52.620 No, I tend to, when I listen to music, I just listen. I'm like a big repeater.

02:14:57.720 Right. It drives everyone around me nuts. See, you've got that OCD focus.

02:15:01.140 Yeah. Yeah. Yeah. Yeah. I can go, I can go five songs on repeat for an entire five hour flight.

02:15:06.880 Amazing. Yeah. Usually Zeppelin, but yeah. Zeppelin's good. I don't know if I mentioned Bonham before.

02:15:11.760 You did. Yeah. I wouldn't have let you get this far in the conversation if you hadn't at least

02:15:15.340 mentioned John Bonham once. What's cool about both Bonham and Elvin Jones is that they both play

02:15:20.300 behind the beat. They're always kind of catching up with the song. And when you play behind the beat,

02:15:26.880 it sounds cool. I don't know why, but just like you can recognize cool, if someone dresses cool or is

02:15:34.500 cool, playing behind the beat sounds cool. I, to this day, listen to good times, bad times,

02:15:41.640 at least once every couple of days. And I think to myself, how is it possible that this was the

02:15:47.000 first song on the first Zeppelin album and Bonham's opening riff? It's like incredible.

02:15:53.900 And I think to myself, what would it have been like to have heard that for the first time

02:15:58.120 on vinyl as like a new sound? I just, I mean, I would have worn that record out.

02:16:03.680 Well, and they had a level of mystery too. I don't know if you know this, but they were never

02:16:08.740 on the album covers, pictures of them. So people would go to the concert to figure out what they

02:16:15.060 looked like. So this, I do somewhat regret not having been around in a time where there was that

02:16:23.820 level of magic around who was making some music you heard. I mean, it must've been even crazier with

02:16:28.900 wax cylinders and stuff like this. People famously thought when they heard the phonograph that they

02:16:35.940 needed to pull the curtain back and see the source of the audio there because it sounded so realistic.

02:16:42.600 Now, if you hear a primitive wax cylinder recording, it's like shocking that anyone could have thought

02:16:47.900 that was in the room because it's so scratchy and simplified, but the magic's happening in people's

02:16:55.480 brains. So they just couldn't piece together how that was coming out of the speaker if it wasn't

02:17:00.580 there. Do you ever listen to vinyl anymore? Yeah. I love vinyl. I mean, vinyl is great. Vinyl is a

02:17:05.460 peak technology in music reproduction. So you get a level of accuracy with digital music that you don't

02:17:13.120 get with vinyl, but analog media are pretty cool. Yeah. Golly. And last question, just because we're

02:17:21.440 bringing it back to music, if you go back in time and see three or four concerts, actual pick the date,

02:17:28.500 pick the band, pick the venue that you just otherwise wouldn't have been able to have seen

02:17:32.520 because you weren't old enough or whatever, what do you think they'd be? Well, I would have liked to

02:17:38.080 see, I saw Prince one time and it was the most virtuosic live performance I've ever seen. So I might

02:17:44.460 devote two of those concerts to Prince or something. People say Chuck Berry was pretty awesome live and

02:17:51.520 pretty unhinged. I bet, you know, I don't love early rock and roll and I don't love blues, but some of

02:17:58.400 those early rock folks, you know, if you had never seen it before probably would have been incredibly

02:18:03.520 shocking. So I would have liked to see that. And watching the documentary on Nina Simone that I saw a few

02:18:11.840 years ago, I concluded that she must've been one of the greatest live performers ever. She's just a

02:18:18.020 virtuosic live performer. Uh, so I would have liked to see her. It's like one of my favorite games to

02:18:23.900 play in the time machine. Oh yeah. All right, BA. Well, thank you so much. It was a pleasure. Thanks

02:18:29.840 Peter. You can find all of this information and more at peteratiamd.com forward slash podcast.

02:18:37.480 There you'll find the show notes, readings, and links related to this episode. You can also find

02:18:42.720 my blog and the nerd safari at peteratiamd.com. What's a nerd safari you ask? Just click on the

02:18:48.740 link at the top of the site to learn more. Maybe the simplest thing to do is to sign up for my

02:18:52.740 subjectively non lame once a week email, where I'll update you on what I've been up to the most

02:18:57.440 interesting papers I've read and all things related to longevity, science, performance, sleep,

02:19:02.300 etc. On social, you can find me on Twitter, Instagram, and Facebook, all with the ID

02:19:07.420 peteratiamd.com. But usually Twitter is the best way to reach me to share your questions and comments.

02:19:12.600 Now for the obligatory disclaimer, this podcast is for general informational purposes only and does

02:19:17.180 not constitute the practice of medicine, nursing, or other professional healthcare services,

02:19:22.040 including the giving of medical advice. And note, no doctor patient relationship is formed.

02:19:27.460 The use of this information and the materials linked to the podcast is at the user's own

02:19:31.940 risk. The content of this podcast is not intended to be a substitute for professional medical advice,

02:19:37.040 diagnoses, or treatment. Users should not disregard or delay in obtaining medical advice for any medical

02:19:42.460 condition they have, and should seek the assistance of their healthcare professionals for any such

02:19:47.020 conditions. Lastly, and perhaps most importantly, I take conflicts of interest very seriously. For all of

02:19:53.040 my disclosures, the companies I invest in and or advise, please visit peteratiamd.com forward slash about.

02:20:01.940 Thank you.

02:20:02.940 Thank you.

The Peter Attia Drive - July 23, 2018

#06 - D.A. Wallach： music, medicine, cancer screening, and disruptive technologies

Episode Stats

Summary

Transcript