0:00:03 The content here is for informational purposes only,
0:00:05 should not be taken as legal business tax
0:00:06 or investment advice,
0:00:09 or be used to evaluate any investment or security
0:00:11 and is not directed at any investors
0:00:14 or potential investors in any A16Z fund.
0:00:18 For more details, please see a16z.com/disclosures.
0:00:22 – Hi everyone, welcome to the A6NZ podcast, I’m Sonal.
0:00:25 So this week to continue our 10-year anniversary series
0:00:27 since the founding of A6NZ,
0:00:29 we’re actually resurfacing some of our previous episodes
0:00:32 featuring founders Mark Andreessen and Ben Horwitz.
0:00:34 If you haven’t heard our latest episode
0:00:35 with Stuart Butterfield turning the tables
0:00:37 as the entrepreneur interviewing them,
0:00:40 please do check that out and other episodes in this series
0:00:41 that we’ve been running all week
0:00:45 on our website at a6nz.com/10.
0:00:48 But this episode was actually recorded in 2014
0:00:50 on the five-year anniversary of the firm
0:00:52 and features Michael Copeland interviewing Ben and Mark
0:00:54 about disruption theory,
0:00:57 as well as key traits of entrepreneurs.
0:00:59 – Disruption theory has been in the news of late
0:01:01 as it relates to Clayton Christiansen,
0:01:02 you know, the master of this.
0:01:04 And I just wanna ask you guys not so much
0:01:07 about the criticism of him,
0:01:10 but from where you set that theory
0:01:13 and his thinking kind of galvanized itself into a book
0:01:17 in 1997, you know, do you build companies differently today?
0:01:22 Does, do those theories still hold water or what’s changed?
0:01:26 – Yeah, so I think his book was actually quite brilliant.
0:01:28 It’s funny that it’s coming under criticism now
0:01:30 after he’s been proving like completely right
0:01:32 and the general idea that he had.
0:01:35 It’s kind of, it actually reminds me of the creationist attacks
0:01:39 on evolution where like, yes, from a,
0:01:41 it’s like intellectualism at its worst, right?
0:01:43 It’s like, oh, here’s something wrong
0:01:45 with Darwin’s original theory.
0:01:48 And it’s like, okay, now we’ve based all of biology on it.
0:01:49 We’ve made tremendous progress.
0:01:50 Like how about that?
0:01:52 And this is kind of like, you know,
0:01:55 I don’t believe in electricity, you know?
0:01:59 And you know, this is kind of the kind of business version
0:02:03 of that where, you know, he developed the theory,
0:02:05 all of us in high tech.
0:02:08 And it was an amazing business book at the time
0:02:11 because it explained a phenomenon that, you know,
0:02:12 and now is kind of obvious,
0:02:16 but in 1997 was tricky, which is why does there,
0:02:18 really why do there need to be new companies?
0:02:19 – Right.
0:02:22 – And what’s happened when we just got through talking about
0:02:24 like there’s an explosion of new companies
0:02:26 and these companies aren’t trivial,
0:02:28 they’re becoming very, very important company,
0:02:33 you know, companies like Google and Facebook and so forth.
0:02:37 And so he’s kind of been proven right.
0:02:39 And then not only has he been proven right
0:02:40 on kind of the large level,
0:02:45 but the mechanics that prevent the kind of incumbents
0:02:49 from innovating at the same rate as the new company
0:02:51 are still completely in effect.
0:02:56 And we, you know, use his models all the time
0:02:58 in our thinking and our analysis.
0:03:03 And no doubt there are probably some minor problems
0:03:07 with examples he’s used or like the way he worded it
0:03:10 or what have you, but like basically he was right.
0:03:11 – Yeah, I would also say two things.
0:03:14 Let’s say one is we actually use this theory basically
0:03:15 to tell us what not to invest in.
0:03:16 – Yes, right.
0:03:17 – As well as what to invest in.
0:03:18 – So how so?
0:03:20 – Well, so for example, we have this basically
0:03:22 this theory that basically it’s very, very dangerous.
0:03:24 So one of the great things about our industry
0:03:26 about venture capital is you get to do these things
0:03:28 that basically disrupt sort of the big
0:03:30 establishing incumbent companies.
0:03:32 Conversely, a very dangerous thing to do is to attack
0:03:35 companies that we, our internal term is the new incumbents.
0:03:37 And so it’s one thing to like go attack, you know,
0:03:39 a tech company that’s been in business for 50 years
0:03:41 that’s on its six C or something like that.
0:03:42 It’s another thing to go attack Google
0:03:43 being run by Larry Page.
0:03:44 – Right.
0:03:46 – Because Google being, you know, Larry Page
0:03:48 is like fully aware of the theory of disruption
0:03:49 and in full command of his company.
0:03:51 And if you like, he sees a disruptive threat coming.
0:03:53 He is quite capable of doing the things to head it off
0:03:55 that a, you know, fourth generation professional CEO
0:03:56 might not be able to do.
0:03:59 So anyway, so that was one thing I want to say.
0:04:01 The other thing I want to say is disruption.
0:04:02 It’s a, I agree with Ben.
0:04:04 It’s funny that this is a topic now,
0:04:05 but since it is, it’s worth talking about
0:04:09 which is the term disruption by its very nature,
0:04:12 the term itself has negative connotations, right?
0:04:13 It’s disruption seems like
0:04:15 it’s one step away from destruction.
0:04:17 And so it gets, it’s got this kind of,
0:04:19 you see it in this kind of popular kind of conception
0:04:21 that there’s something bad about it.
0:04:23 The actual way that Christensen used the term
0:04:25 was actually in a very sort of applied way
0:04:28 in a very specific circumstance in business.
0:04:31 And, and actually in a very positive way,
0:04:33 which is basically he described as a way
0:04:34 that progress happens, right?
0:04:37 So progress doesn’t happen by basically old companies
0:04:38 like deciding to do new things.
0:04:39 Company, progress happens
0:04:41 because new companies decided to do new things.
0:04:43 And then disruption is the process
0:04:44 by which the new things are able to take over
0:04:45 from the old things.
0:04:47 If you decide you don’t like disruption,
0:04:49 what you’re basically saying is you don’t like new things,
0:04:49 right?
0:04:51 It’s basically to be against disruption
0:04:53 is to basically be pro the status quo.
0:04:54 And pro the status quo means the way,
0:04:55 however the world is today,
0:04:58 like that’s it, like that’s all we’re gonna have.
0:04:59 Like the way things work today,
0:05:01 this is as good as it’s ever gonna get.
0:05:02 The disruption argument is no, no, no, no, no, no,
0:05:04 things can become much better.
0:05:06 Products can become much better.
0:05:07 Businesses can become much better.
0:05:09 Opportunities for people can become much better.
0:05:11 And so it’s a, it’s a negatively connotated term
0:05:13 that has very positive implications.
0:05:15 And I think that that’s really at least in the last couple
0:05:17 of years that’s been lost in a lot of the commentary.
0:05:19 – You mentioned Google and one of the things
0:05:20 that we’ve seen, you know,
0:05:22 through the technology industry’s history
0:05:25 is that it’s very, very hard to disrupt yourself
0:05:27 and kind of make a transition from one thing to another.
0:05:31 IBM, maybe the only company that’s done it.
0:05:34 Google, you know, they’re trying everything.
0:05:36 You know, and Facebook is trying everything.
0:05:39 And do these companies somehow change the rules?
0:05:42 Or is it the same rules applying and, you know,
0:05:45 disruption theory catches up with them in 50 years maybe.
0:05:48 – So I think you’re, I think kind of have to break
0:05:51 that back apart and go back to what Mark said.
0:05:53 I think that people often think of big companies
0:05:55 can’t innovate, little companies can,
0:05:59 but the real truth is new companies can innovate
0:06:03 and companies that are so old that the original inventors
0:06:06 are gone, have a lot of trouble doing it.
0:06:10 And so if you go back to HP or IBM
0:06:13 or any of these companies, when the founder,
0:06:15 when Thomas Watson was running the company,
0:06:17 when Dave Packard was running the company,
0:06:20 they didn’t have any trouble doing new things.
0:06:22 And they did a phenomenal, I mean, HP in particular
0:06:24 did like a crazy number of new things,
0:06:29 just amazing and in retrospect, really phenomenal.
0:06:30 And if Mark Zuckerberg’s running the company
0:06:32 or Larry Page is running the company,
0:06:33 you know, that’s not an old company.
0:06:34 That’s a new company.
0:06:39 And as innovators, they, you know, we believe
0:06:43 and this gets back to why we don’t attack them
0:06:46 because they’ll attack right back and very effectively.
0:06:48 You know, they’re going to be able to do new things.
0:06:50 And like sometimes that will mean
0:06:52 bringing in new talent through acquisition
0:06:55 or new technologies through acquisition,
0:06:59 but they’re going to be able to think about the problem
0:07:02 through a lens that is not the business they’re in.
0:07:05 And that’s kind of, this is the amazing thing
0:07:08 that Clayton Christiansen laid out was that, you know,
0:07:10 if you’re like, if you’re an old company
0:07:12 run by professional managers,
0:07:13 you’re really good at studying
0:07:15 and optimizing the business you’re in.
0:07:17 And so if there’s a new business that comes along
0:07:22 that doesn’t, is inconsistent with that, you get stuck.
0:07:23 But if you’re Mark Zuckerberg
0:07:26 who created a business from nothing,
0:07:27 then you have a very different view of the world.
0:07:31 And it’s not like, okay, how do I optimize the business
0:07:32 that I’m in?
0:07:33 It’s like, well, how do I get another business
0:07:34 that’s like Facebook?
0:07:36 That’s more the way you think about it.
0:07:37 – The other thing is the fact that Christiansen
0:07:38 was able to articulate this in a theory
0:07:40 that’s so clear and put in the book is,
0:07:43 I think that like the best professional CEOs
0:07:45 in the tech industry today, like now understand this
0:07:46 in a way that maybe their predecessors
0:07:48 10 or 20 years ago didn’t understand it.
0:07:49 So I’ll just give you two examples
0:07:51 of people I work with, John Donahoe at eBay.
0:07:53 Like when mobile came along, you know,
0:07:55 sort of classical professional CEOs,
0:07:56 when mobile comes along, you know, would look at it
0:07:58 and say, well, I’ve got this great business on the web.
0:08:01 If I move to mobile, it may or may not work as well.
0:08:03 And so maybe I don’t want to try to make the move.
0:08:05 Maybe I want to stay on the web and reinforce the web
0:08:07 and like not take the risk of quote disrupting myself
0:08:09 by making the jump to mobile.
0:08:11 But since John understands disruption theory
0:08:13 and it’s been like articulated and explained
0:08:15 in a way that makes sense, you know,
0:08:17 he was able to be based on a phenomenally successful job.
0:08:19 He went full throttle into mobile
0:08:21 and they made the jump and they’ve done it very well.
0:08:23 Meg Whitby doing the same thing with this.
0:08:25 I just, one example is this project Moonshot
0:08:27 which is these cartridge based servers.
0:08:28 – At HP, right?
0:08:31 – At HP that are a direct attack
0:08:32 on the existing blade server business.
0:08:33 And the blade server business at HP
0:08:35 is a very, very big and profitable business.
0:08:37 And HP is basically self disrupting
0:08:39 with this new kind of cartridge based server.
0:08:41 And so again, and when you have the discussion,
0:08:43 you know, HP board meeting and you have the discussion,
0:08:45 you’re like, okay, why are we taking the risk
0:08:47 of damaging this big existing profitable business
0:08:48 by doing this new thing?
0:08:50 The answer is because it’s the right thing to do
0:08:51 according to disruption theory.
0:08:54 Like it is, like there is a logical framework.
0:08:56 And again, think about what’s happening
0:08:58 which is something new is happening.
0:08:59 Progress is happening, right?
0:09:01 This is now the reason and the motivation
0:09:02 and the explanation and the justification
0:09:03 to be able to make progress.
0:09:06 So it’s an incredibly powerful positive thing.
0:09:09 – Let’s get to entrepreneurs and entrepreneurship.
0:09:11 You guys founded the firm.
0:09:15 In part, I’ve been told because you wished
0:09:18 you’d been told or helped in certain ways.
0:09:21 What’s one thing both of you wish you knew
0:09:25 or someone had told you as entrepreneurs?
0:09:28 – Well, that presumes we would have listened.
0:09:34 You know, there’s just so much that we did not know
0:09:36 going through it the first time.
0:09:37 And one of the great things
0:09:39 about the entrepreneurial experience
0:09:42 is it’s just an amazing learning curve
0:09:46 about everything from markets to organizational structures
0:09:48 to compensation to everything.
0:09:54 But probably one of the most challenging things
0:09:55 to learn while you’re out there
0:10:02 is kind of how macroeconomics impact markets
0:10:06 and particularly how they,
0:10:11 how private funding can change very, very rapidly.
0:10:14 You know, when we were, you know, particularly,
0:10:17 and this wasn’t as a harsh lesson at Netscape,
0:10:20 but at Opsware and LoudCloud,
0:10:23 it was like incredibly difficult for us
0:10:24 to go from the funding environment
0:10:28 where basically had the highest multiples
0:10:30 in the history of anything
0:10:32 to there was no money available, period.
0:10:33 I mean, like that was,
0:10:36 it was the most dramatic fall imaginable
0:10:38 from the highest of highs to the lowest of lows.
0:10:42 And, you know, to have the NASDAQ fall over 80%
0:10:46 and that not being, you know, that’s NASDAQ,
0:10:48 that’s not tech, tech fell 95%.
0:10:50 It’s just like not something you could even imagine
0:10:51 or get your head around.
0:10:55 So I wish, you know, like I wish we would have known that.
0:10:58 I wish, I don’t know if we would have believed anybody
0:10:59 if they had told us that,
0:11:03 but that would have probably made it a little less painful
0:11:05 if we had any idea how bad it could be.
0:11:07 – It would have made a worse book.
0:11:09 I’ll tell you that, that you wrote, but still.
0:11:10 – Yeah, yeah.
0:11:13 – On the other side of the table,
0:11:17 what do you want more or less from entrepreneurs,
0:11:19 more of or less of from entrepreneurs?
0:11:23 – Yeah, well, you know, it’s very different
0:11:25 across different businesses,
0:11:27 but like the one thing
0:11:30 that would probably be nice if there was less of
0:11:33 that’s pretty consistent is it’d be nice
0:11:36 if it wasn’t so important to entrepreneurs
0:11:40 what their peers valuations were.
0:11:44 Like that, that is probably the most meaningless thing
0:11:48 to focus your mind on as an entrepreneur imaginable.
0:11:50 It’s just like irrelevant.
0:11:53 – You don’t have anything else to base your value on, do you?
0:11:56 – No, no, no, that’s, you go ahead.
0:11:59 – Yeah, so it’s not actually,
0:12:02 you know, your company is your company,
0:12:03 their company is their company.
0:12:06 You’re looking at the price they got,
0:12:09 not any of the business metrics that they have
0:12:11 or like how the company is going.
0:12:13 So you’re not actually basing your valuation
0:12:15 on anything in that sense.
0:12:19 And there’s better data to be gotten for sure.
0:12:21 Like, you know, we have better data.
0:12:24 We can talk to them about all the kind of valuations
0:12:26 based on actual revenue and so forth,
0:12:28 as opposed to the person they went to school with
0:12:31 or the person they worked at their last company with.
0:12:33 And, but people get very wrapped around the axle on that
0:12:37 because there’s, you know, it’s kind of the thing
0:12:38 that Peter Thiel talks about,
0:12:40 whereas competition is actually like really destructive.
0:12:42 And that’s like the worst kind of competition
0:12:44 ’cause it’s competition that’s irrelevant
0:12:47 to anything in life other than, you know,
0:12:50 you can go tell your friend what valuation you got.
0:12:55 And I think that it causes bad, you know,
0:12:57 errors in judgment and delays in decisions
0:13:00 that need to be made quickly and things like that.
0:13:02 So, you know, it’s just,
0:13:05 it’s one of those things where humanity gets a better view.
0:13:10 And I wouldn’t like, less of that would be good.
0:13:13 – Mark, any, anything you would offer on that?
0:13:14 – Well, the thing that the great entrepreneurs
0:13:16 all have in common, we talk about this a lot,
0:13:18 but you just see it every day is the great entrepreneurs
0:13:20 all have amazing courage.
0:13:21 And so I would say we’re blessed
0:13:23 in that the entrepreneurs we work,
0:13:24 and we select for it.
0:13:25 I mean, we try very hard to select for it,
0:13:27 but the entrepreneurs we work with that are amazing.
0:13:28 One of the things they all have in common
0:13:29 is they’re incredibly courageous,
0:13:30 but which I mean, they don’t give up.
0:13:31 They don’t, they don’t quit.
0:13:33 Like they don’t, they don’t quit.
0:13:33 They don’t flinch.
0:13:34 They don’t get demoralized.
0:13:36 They don’t get, I mean, well, actually,
0:13:37 they may get demoralized or depressed,
0:13:39 but they show up to work the next day
0:13:41 and they work their way out of whatever problem they’re in.
0:13:43 And they just keep pounding and pounding
0:13:44 and pounding and pounding.
0:13:45 And I think there’s a little bit too much
0:13:48 in the Valley right now of the pivot
0:13:50 and the lean start, you know, the lean startup
0:13:51 and the, you know, the everything’s an experiment
0:13:54 and minimum viable product and failure is good
0:13:57 and kind of all of these excuses to be able to give up
0:13:59 when things aren’t going well.
0:14:02 And I think that the great entrepreneurs through history
0:14:04 have always been the opposite kind of personality
0:14:05 and all that they’ve always been.
0:14:06 “I’m gonna make this thing work
0:14:07 hell or high water no matter what.
0:14:09 I am going to knock my way, you know, headfirst
0:14:12 through any, you know, barrier that I run into.
0:14:13 I don’t care what people say about me.
0:14:15 I don’t care what kinds of problems I have.
0:14:18 I’m gonna figure this out and I’m not gonna give up.”
0:14:20 And so I would just say we love working with people
0:14:22 who have that personality type
0:14:23 and you can never have enough of them.
0:14:25 – Elon Musk comes to mind.
0:14:26 I mean, cars and space.
0:14:28 – Yeah, so to start, think about this,
0:14:30 to start a new electric car company.
0:14:32 And by the way, think about the last car company
0:14:33 started in the United States.
0:14:34 They literally made a movie about the catastrophe
0:14:36 that resulted, which is this movie, Tucker.
0:14:39 And so if you want like a story of like a horrible business.
0:14:40 – Which went better than DeLorean.
0:14:42 – Yes, well, actually, yeah, DeLorean.
0:14:43 Well, he had the added,
0:14:44 he had the cocaine smuggling business on the side,
0:14:47 which helped cover the, to free the expenses.
0:14:49 But, you know, car companies,
0:14:52 like all the car companies in the US that are successful
0:14:53 are like, you know, from the 1910s and 1920s.
0:14:55 And so to start a new car company
0:14:56 in the electric car category,
0:14:58 when all the electric cars had failed,
0:15:01 simultaneously to start the first new private rocketry company
0:15:04 in the United States in probably 40 years
0:15:05 to go straight up against the big boys,
0:15:07 to do those at the same time.
0:15:09 And then to go through the 2008 crash.
0:15:12 And he has actually recently opened up on this of like,
0:15:13 he almost lost both companies in 2008.
0:15:15 Like they’ve almost both vaporized.
0:15:17 And then a gut through both of those
0:15:18 and have both come out the other side,
0:15:21 like just like an excreaming successes is just a,
0:15:23 it’s a spectacular performance.
0:15:26 And a huge part of it is he didn’t give up.
0:15:28 – Ben, let’s touch on your book a little bit.
0:15:29 The hard thing about hard things.
0:15:31 One thing it got great reception,
0:15:34 but you’re like, well, yeah, that sounds good for you, Ben,
0:15:36 but that was your story.
0:15:38 How can I embrace that and make that my story?
0:15:41 But, you know, was there anything in the response
0:15:45 that you wish people had pushed you harder on?
0:15:48 – Well, the things that people pushed me on
0:15:49 actually annoyed me.
0:15:52 So I, it’s hard to say that I wish about that.
0:15:55 I mean, I think that to your point though,
0:16:00 it was my story and the reason for that,
0:16:02 I mean, there was a really specific reason for that,
0:16:04 which is building these companies
0:16:07 tends to be very dynamic and very situational.
0:16:10 And so a very frustrating thing
0:16:12 about management and advice in general,
0:16:15 and particularly, you know, both in books
0:16:18 and then things that you often get from board members
0:16:22 or kind of pattern matchers as it were,
0:16:23 is that they’re giving you advice
0:16:25 and it’s based on something.
0:16:28 And that advice and what it’s based on
0:16:30 may or may not be relevant to you.
0:16:31 And if you don’t know what it is,
0:16:33 it’s very difficult to interpret it.
0:16:35 And I always found that, you know,
0:16:37 management books would give like guidance
0:16:39 and you’d be like, well, okay,
0:16:41 is that what I should be doing?
0:16:44 But I have no idea where it came from.
0:16:45 And so it’s hard to say.
0:16:48 So a lot of putting my story in was just to say,
0:16:50 look, this is why I’m telling you this.
0:16:54 And like, if your situation is completely different
0:16:56 than this, then that might be the part of the book
0:16:59 that you ignore or like, at least,
0:17:01 or maybe you can map it on to what you’re doing.
0:17:03 But I think that without knowing
0:17:05 why somebody is telling you something,
0:17:09 it’s pretty difficult to get value out of it.
0:17:10 – And on the topic of the entrepreneurial journey,
0:17:11 we have to go see a pitch.
0:17:13 – All right, that’s what you guys get paid to do.
0:17:15 So Ben and Mark, thanks so much.
0:17:16 We will do this.
0:17:18 Well, it won’t do it in five years.
0:17:19 We’ll do it much sooner than that.
0:17:20 Thank you very much.
0:17:21 – Okay, thanks, Michael. – Thanks, Michael.

0:00:03 – Hi, and welcome to the A16Z podcast.
0:00:05 I’m Amelia Salyers.
0:00:07 Today’s episode is a special one,
0:00:09 since it’s the 10th anniversary of Andreessen Horowitz,
0:00:12 which was founded in late June, 2009.
0:00:14 So we decided to turn the tables
0:00:17 by asking Stuart Butterfield, founder and CEO of Slack,
0:00:20 to interview our founders, Ben Horowitz and Mark Andreessen.
0:00:22 The three of them discussed the differences
0:00:24 between founders in 2009 and today,
0:00:28 the business model of VC in A16Z’s history,
0:00:31 and technology trends then, now, and into the future.
0:00:32 And they also throw in a few good
0:00:35 summer book and TV recommendations at the end.
0:00:36 Please note that the content here
0:00:38 is for informational purposes only,
0:00:40 should not be taken as legal, business,
0:00:41 tax, or investment advice,
0:00:44 or be used to evaluate any investment or security,
0:00:45 and is not directed at any investors
0:00:48 or potential investors in any A16Z fund.
0:00:52 Any investments or portfolio companies mentioned, referred to,
0:00:53 or described in this podcast
0:00:56 are not representative of all A16Z investments,
0:00:57 and there can be no assurance
0:00:59 that the investments will be profitable,
0:01:01 or that other investments made in the future
0:01:03 will have similar characteristics or results.
0:01:05 A list of investments made by A16Z
0:01:08 is available at a16z.com/investments.
0:01:13 For more details, please see a16z.com/disclosures.
0:01:15 – One question I wanna ask is,
0:01:18 has the nature of the entrepreneurs or the founders changed?
0:01:22 So 2009, you have raised some money, you have some LPs,
0:01:24 now you’re out there looking for ways to invest.
0:01:26 Who are you meeting, where are they coming from,
0:01:27 and what are they like?
0:01:31 – Yeah, well, you know, I really feel like in retrospect,
0:01:35 that class of 2009 entrepreneurs were some of the most
0:01:38 special ones that we’ve met in the entire history
0:01:40 of the company, and when we see that again,
0:01:42 we always say like yourself,
0:01:45 Todd McKinnon, Martine, Brian Chesky.
0:01:47 You know, the thing all of you had in common
0:01:51 is you had gone through something like just unbelievable
0:01:53 to get to the position you were to start the company.
0:01:55 You know, earned your stripes.
0:01:59 Like everybody in 2009 seemed like they pay their dues,
0:02:03 like in a pretty serious way to get into position.
0:02:05 And I think, you know, one of the great things
0:02:07 that’s happened over the last 10 years
0:02:09 is it’s just become easier to start a company.
0:02:13 But as a result, you don’t have, you know,
0:02:16 people who know exactly what it is and what that means
0:02:21 and what they’re about to face, and still wanna do it,
0:02:25 which is, you know, that’s a very special thing,
0:02:26 but that’s an unusual person.
0:02:28 And then I had two categories of entrepreneurs
0:02:30 who have really risen in the last decade.
0:02:32 So one is, you know, we, Alex, our partner,
0:02:34 Alex kind of coined this term, O2O, you know,
0:02:39 kind of B2B, B2C, now O2O, and O2O is online to offline.
0:02:40 So it’s this whole broad category.
0:02:43 It’s, you know, Airbnb and Lyft and Uber
0:02:44 and many of these postmates in DoorDash and all these,
0:02:46 it’s sort of these companies that you have an online
0:02:48 experience that culminates in something happening
0:02:49 in the real world.
0:02:53 And so those founders are much more,
0:02:55 I would say, operationally focused maybe
0:02:56 than the previous generation, right?
0:02:57 Which is, it’s not just software,
0:02:59 but even beyond that, just being software, you know,
0:03:01 those companies have a big real world, you know,
0:03:04 logistical infrastructure operations component.
0:03:06 And so that has turned out to be different kind of founder,
0:03:08 which is really interesting.
0:03:09 Almost actually even a little bit of a throwback model.
0:03:12 There, those people arguably are a little bit more like
0:03:14 the semiconductor founders from like 30 years ago.
0:03:16 They’re like, they’re harder core.
0:03:17 They’re grouchier.
0:03:18 They’re dirt under the fingernails.
0:03:21 Yeah, and they have to worry about more real world stuff.
0:03:23 Yeah, stuff can go wrong, you know, people can die.
0:03:24 Like, you know, all kinds of, you know,
0:03:25 all kinds of stuff can happen.
0:03:29 So that’s been a particularly interesting kind of rise
0:03:30 of a new kind of founder, which has been super interesting
0:03:31 to work with those people.
0:03:33 And then the other on the bio side is we’ve gotten more
0:03:35 involved in bio and in healthcare.
0:03:38 The other is the rise of the deep domain expert
0:03:40 in a science like biology.
0:03:41 So somebody might come, you know,
0:03:43 biology PhD or chemistry PhD or something
0:03:45 where, you know, 10 years ago,
0:03:48 if you met a newly minted biology PhD out of Stanford or MIT,
0:03:50 they really wouldn’t know that much about computers.
0:03:51 It was kind of an afterthought
0:03:53 and they wouldn’t really know, you know,
0:03:54 they know a little bit how to code, but not much.
0:03:57 And now you meet a newly minted biology PhD out of Stanford.
0:04:00 They basically have a dual PhD in computer science.
0:04:02 Like they basically, you know, it’s kind of the story.
0:04:04 They’ve been programming since they were little kids
0:04:06 in most cases and then they kept current.
0:04:08 And in fact, a lot of the research that they did
0:04:10 to get their bio PhD often had to do with computer science
0:04:13 and math and algorithms and machine learning, right?
0:04:14 A lot of that stuff’s happening now.
0:04:17 And so you’ve got these kind of dual discipline founders
0:04:18 for the deep science stuff.
0:04:22 So dual discipline, biology CS, mechanical engineering CS,
0:04:24 physics CS, chemistry CS.
0:04:27 And that’s a real, like that those,
0:04:28 those people are like super enticing.
0:04:30 ‘Cause those are, you know, it’s like two superpowers.
0:04:32 Like, you know, he can, you know, he or she can fly
0:04:33 and they’re vulnerable.
0:04:35 Like is a really good combination.
0:04:37 And so a lot of the companies we’re seeing on the bio side
0:04:39 and in the harder sciences is that kind of founder.
0:04:40 I think that’s new.
0:04:41 – Yeah.
0:04:44 – Or like Satoshi economics and computer science.
0:04:45 – Actually, it’s funny.
0:04:46 There’s actually a revolution happening
0:04:48 even apart from crypto in the field of economics.
0:04:51 There’s a wrote in the, in the actual academic field
0:04:52 of economics, there’s a revolution happening
0:04:54 towards what’s called empirical economics,
0:04:56 quantitative economics in which a lot of the new economists
0:04:58 who are kind of 40 and below are very focused on data
0:04:59 and machine learning.
0:05:01 And, you know, the economists that are in their fifties
0:05:02 and sixties were more inspired by physics
0:05:04 and they’re, they’re much more into formulas
0:05:05 and they’re, they’re much more abstract about what happens
0:05:06 in the real world.
0:05:08 And so there are actually, there are actually,
0:05:10 there are companies increasingly driven by, started by,
0:05:12 or, you know, new invent, new inventions being created
0:05:15 by economists with a very strong CS background.
0:05:16 – That’s true. That’s true.
0:05:18 So from the beginning, I’m not sure if this was intentional
0:05:21 or not, but certainly the positioning in the press was,
0:05:24 this was different, like you’re going to blaze a new trail,
0:05:25 kind of a different model.
0:05:28 And right out of the gate, there was the investment in Skype,
0:05:30 which was not a thing that VCs normally did.
0:05:32 How, how intentional is that?
0:05:33 – Yeah, a lot of it was intentional.
0:05:35 So there, there was a big kind of throwback element
0:05:37 to what we were doing, which was basically,
0:05:38 we wanted to work with the best founders
0:05:39 to build the most important companies.
0:05:41 And there’s, we could talk a lot about kind of the history
0:05:43 of venture and the art and science of the whole thing,
0:05:45 but there is a rich tradition there
0:05:47 that we definitely drew from.
0:05:48 And we actually, we spent a lot of,
0:05:50 we, we, to get Ben and I leading up to this,
0:05:52 spent a lot of time really digging into the history
0:05:54 of kind of where all this stuff comes from
0:05:55 and kind of what made it work across various areas.
0:05:57 And so, so we were, you know, very inspired
0:05:59 by a lot of the people who came before us.
0:06:01 And then there were a bunch of new ideas.
0:06:03 And so, you know, maybe list several of the new ideas.
0:06:05 And so one of the new ideas was just that we thought
0:06:07 that a lot of the venture firms had lost their way
0:06:09 in the sense that they had been started by founders.
0:06:10 They’d been started by founders and operators
0:06:11 who had built businesses.
0:06:13 And so you would, you know, raise money from, you know,
0:06:15 a firm and you would get, you know,
0:06:17 somebody who’d been a CEO or general manager
0:06:18 of an important business on your board
0:06:20 and they could really help you figure things out.
0:06:21 And then over the years, the, a lot of the venture firms
0:06:23 just quote unquote, professionalized
0:06:24 and they ended up with a lot of GPS
0:06:25 who didn’t have that experience.
0:06:27 And so it started to get,
0:06:28 the advice started to get more abstract
0:06:30 and maybe less helpful.
0:06:31 So that was one difference.
0:06:33 Another difference was, you know, we, we took seriously
0:06:35 the, the, the idea of building the institution.
0:06:37 And then in particular around that building a network
0:06:41 and an ecosystem and making a real long-term systematic
0:06:42 and actually very costly investment in building a,
0:06:44 you know, building a network and that had to do
0:06:46 with the fundamental staffing model of the firm.
0:06:47 And then that, you know,
0:06:49 that’s why we have all these operating functions.
0:06:51 We have all these, all these, all these professionals here.
0:06:53 And then that rippled over even into things like compensation,
0:06:56 like we get paid very differently than most, than most VCs.
0:06:57 And so that was a big difference.
0:06:58 The Skype deal you alluded to,
0:07:00 we did start at the very beginning
0:07:01 with the idea of being stage agnostic.
0:07:03 And that was probably a pretty new idea at the time.
0:07:05 And the idea there basically was
0:07:07 if the priority is to work with the best founders
0:07:09 to help build the most important companies,
0:07:11 it shouldn’t matter that much what stage the company’s at.
0:07:14 I mean, ideally you’d like to start working with people early,
0:07:15 but like, you know, you make mistakes.
0:07:17 And so you miss things.
0:07:18 And we were starting from scratch.
0:07:19 And so there were a bunch of companies
0:07:20 we thought were very impressive
0:07:21 that we wanted to get involved in,
0:07:23 you know, even though they already existed.
0:07:25 And so we kind of had this idea
0:07:26 that there should be multiple entry points
0:07:27 from an investment standpoint.
0:07:28 And then there would be things
0:07:30 that we could actually do to be able to help
0:07:32 and work with the entrepreneur.
0:07:33 And so we kind of put ourselves in business
0:07:36 from the beginning to operate across all the stages.
0:07:37 You know, now that took time to get,
0:07:39 I would say it took time for us to get good at all the stages
0:07:41 and maybe we’re still, we’re still working on it.
0:07:44 But that core fundamental idea was in there.
0:07:45 – You referenced looking back at history
0:07:50 for some context and thinking about the investment theses.
0:07:52 For us, 10 years doesn’t seem like that long ago,
0:07:55 but then we hire people who have like three or four years
0:07:56 of experience, which means that they were born
0:07:59 even in college, sometimes 10 years ago.
0:08:00 – Yeah, we experienced that a lot.
0:08:02 – Increasingly frequently, yeah.
0:08:04 – But that’s interesting ’cause that was a very
0:08:08 particular moment in history that I have vivid memories
0:08:13 of the 2008 crisis that then like almost immediately
0:08:15 after that, the idea that we were in a tech bubble
0:08:17 and kind of the whiplashing back and forth.
0:08:18 What was that like?
0:08:20 – Well, you know, it was interesting when we started
0:08:23 and we had, we of course had been through
0:08:25 the actual tech bubble.
0:08:26 – The 2000.
0:08:27 – 2001 tech bubble, yeah.
0:08:31 The, you know, the 2008 crisis was a banking crisis,
0:08:33 not a tech crisis, it was a debt crisis,
0:08:34 not an equity crisis.
0:08:37 So it was very different in nature for startups.
0:08:39 So a big advantage we had was,
0:08:43 it really did not bother us at all.
0:08:46 And so walking in, one, like everybody said,
0:08:49 you can’t possibly raise a fund in 2009,
0:08:50 you can’t raise a venture capital fund.
0:08:53 So the only two new funds were raised that year,
0:08:56 us and COSLA, and because we, you know,
0:08:57 we didn’t know any better.
0:08:59 We were like, come on, like this isn’t bad.
0:09:02 This is like not bad at all.
0:09:04 And then like the, the other thing that really helped us
0:09:07 was the just grouchiness of a lot of the other VCs.
0:09:10 I remember I’ll go unnamed venture capitalists
0:09:12 ’cause I remember it so well
0:09:14 ’cause I ran back to tell Mark, you know,
0:09:15 he asked me, he’s like, well, like,
0:09:16 what are you interested in?
0:09:17 And I was telling him about OCTA
0:09:19 and how like important it was gonna be
0:09:20 in a SaaS world and so forth.
0:09:22 And he gave me like a 30 minute lecture
0:09:24 that SaaS was a bunch of BS.
0:09:26 It was never gonna happen
0:09:29 that the only successful SaaS company was Salesforce.
0:09:32 And that was ever gonna be the only SaaS company
0:09:34 was gonna be Salesforce and all this stuff.
0:09:35 And so it was just like that kind of,
0:09:38 like everybody was just like pretty grouchy
0:09:41 and like crusty and we were, you know, new.
0:09:44 So we were excited to invest in all that stuff.
0:09:47 – So thinking back to that point,
0:09:48 I don’t remember the exact year that everyone started.
0:09:51 Obviously, my company started in 2009.
0:09:54 I think Airbnb was like maybe that same year or year later.
0:09:56 Something’s happened a couple of years earlier
0:09:58 like AWS came out in 2006.
0:10:01 iPhone and nominally in 2007 or really 2008.
0:10:03 But a lot of those things hadn’t actually picked up.
0:10:05 So certainly my recollection is
0:10:08 none of that was really visible at that time.
0:10:09 But we were at this, in retrospect,
0:10:11 was an incredible inflection point.
0:10:13 To what degree do you think that worked to your advantage?
0:10:14 ‘Cause you hear you are starting to fund,
0:10:16 people are skeptical.
0:10:18 But meanwhile there’s these massive secular trends
0:10:21 which were pretty much invisible to everyone at that time.
0:10:22 – I think the two things that are true.
0:10:24 So one is the big secular trends do drive
0:10:25 a lot of what happens in this industry.
0:10:27 And so like it is absolutely the case in the last decade
0:10:29 that mobile and cloud in particular
0:10:31 like drove just giant growth and social as well.
0:10:35 So they made a lot of people look like geniuses,
0:10:36 including a few actual geniuses.
0:10:37 So that’s also true.
0:10:39 But I think it’s also true what Ben mentioned
0:10:40 is like super important underline,
0:10:42 which is like people don’t think these things
0:10:43 are obvious in the beginning.
0:10:45 Like they’re only obvious after the fact.
0:10:46 They really don’t look that obvious in the beginning.
0:10:48 And so you like you mentioned the iPhone came out.
0:10:50 Like I remember the iPhone in 2009,
0:10:51 it was like a cool gadget,
0:10:53 but like it couldn’t hold a phone call.
0:10:55 Like that was the era in which like it wasn’t even,
0:10:57 was it on 3G at that point?
0:10:59 It was like a big- – No it wasn’t, it was pre-3G.
0:11:01 – The original iPhone was actually pre-3G, right?
0:11:02 It had like edge data connection
0:11:04 and then the 3G iPhone was a big upgrade,
0:11:05 but then you couldn’t hold a phone call on the thing.
0:11:07 And that was in the era when Steve was telling people
0:11:08 that you were holding the phone wrong.
0:11:10 – Yeah. – If it was dropping
0:11:11 phone calls, you need that.
0:11:12 – And they built that crazy room
0:11:14 to prove it to the journalists.
0:11:16 – Exactly, and then they shipped everybody the bumper, right?
0:11:19 And so it’s like, okay, to squint from that to like,
0:11:20 okay, now it’s the mobile boom of all time
0:11:22 and we’re gonna be sitting here 10 years later
0:11:23 and they’re gonna have, you know, whatever it is now,
0:11:25 a billion and a half of these things, you know,
0:11:26 in the field and it’s gonna be kind of the defining,
0:11:28 you know, device and interface for a generation.
0:11:30 Like that wasn’t super obvious.
0:11:32 And then cloud, you know, cloud, I just, you know,
0:11:36 there were lots and lots and lots of companies of that era
0:11:37 that were incredibly powerful,
0:11:40 that were in the server business or networking business
0:11:41 or storage business or software business
0:11:44 where this whole cloud thing like AWS, like it’s a toy,
0:11:46 it’s a gimmick, like it’s never gonna make any money.
0:11:47 It is really interesting.
0:11:49 There is a big leap that has to happen
0:11:51 even when they are the really big megatrends.
0:11:54 Like it’s not, I had this in the early, the internet,
0:11:56 like a lot of people in the early 90s,
0:11:57 a lot of people in the press,
0:11:58 a lot of people in the investment community,
0:11:59 a lot of entrepreneurs.
0:12:01 – Well, the entire, actually you should tell the story
0:12:02 about like when you went to raise money,
0:12:04 you and Jim went to raise money
0:12:06 from all the magazines and the newspapers.
0:12:08 – Yeah, so we started in Escaping early ’94
0:12:10 and we went out and pitched all the media companies
0:12:11 to become customers, partners, investors
0:12:13 and every single big media company.
0:12:14 And in fact, at that point, they said,
0:12:16 no, no, the future is AOL
0:12:18 because AOL pays us for our content.
0:12:20 And on the internet, we have to spend money
0:12:20 to put our per content.
0:12:21 So that’s never gonna work.
0:12:22 And of course they all knew
0:12:24 that normal people wouldn’t use the internet
0:12:25 if it didn’t have Time Magazine on it.
0:12:26 Right, ’cause Time Magazine
0:12:28 would obviously be the killer for the internet.
0:12:29 You know, and by the way, it’s like,
0:12:31 a friend of mine says that this thing is happening.
0:12:32 It’s going to fundamentally change the world
0:12:33 and people poo poo it.
0:12:35 Like that might actually be the logical response
0:12:37 because there are many new things to come along
0:12:39 where people claim it’s going to change the world.
0:12:41 And then most of those things don’t change the world.
0:12:43 And so maybe, you know, on average,
0:12:46 the correct response is no, this thing is stupid.
0:12:48 And then maybe our lot in life as founders and VCs
0:12:50 is to, you know, be the fringe element
0:12:52 that like, that bucks that.
0:12:53 – Who’s wrong 97% of the time.
0:12:54 – Right, right, exactly.
0:12:55 And by the way, you know,
0:12:56 we looked dumb a lot of the time except, you know,
0:12:59 during the times we looked, you know, really, really smart.
0:13:01 And you know, and the reality is we’re probably neither, right?
0:13:02 We’re probably neither super dumb or super smart.
0:13:04 We’re probably just willing to take the risk
0:13:05 at a time when other people aren’t.
0:13:07 – So let’s fast forward a couple of years,
0:13:10 kind of like the middle era, 2012, Facebook went public.
0:13:12 But I think that, I don’t remember what year that happened,
0:13:14 but we started talking about unicorns
0:13:16 and there was another round of this is a bubble.
0:13:17 What was that like?
0:13:20 And what do you remember about the investment?
0:13:21 Like basically the partner meetings
0:13:24 after I would leave the room
0:13:26 and the debate was happening.
0:13:29 How much did, is this too expensive factor
0:13:30 into the conversations?
0:13:33 – Yeah, you know, we have in the entire time.
0:13:35 And I can say we’re totally consistent on this.
0:13:37 And I think it’s because of our history,
0:13:41 never thought it was a bubble in our entire time
0:13:42 doing the job.
0:13:44 And I think a lot of it has to look,
0:13:48 prices of companies are always incorrect,
0:13:49 like always, always, always incorrect
0:13:52 because they’re valued on like future performance,
0:13:53 which nobody knows what that is.
0:13:55 So most people are optimistic,
0:13:56 the prices go a little higher.
0:13:58 Most people are pessimistic.
0:14:00 At that time, the prices go a little lower.
0:14:01 But to get to a bubble,
0:14:03 everybody’s got to be optimistic.
0:14:07 And that’s what happened kind of in the 99, 2000 era.
0:14:08 And like in our whole time,
0:14:10 we never signed anything close to that.
0:14:13 Like they never went anywhere,
0:14:15 anywhere like within an order of magnitude
0:14:18 to what it did in 99, 2000
0:14:20 for similar kinds of companies.
0:14:22 And so we were always, no, there’s no bubble.
0:14:23 Like, what are you talking about?
0:14:24 There’s no bubble.
0:14:27 But people want to believe there’s a bubble so badly.
0:14:30 I think 2011, I was in a debate
0:14:32 with Steve Blank and the economist,
0:14:34 and I argued it wasn’t a bubble.
0:14:38 And he argued it was in 2011 tech bubble, right?
0:14:41 And at the end of the debate, he called me
0:14:44 and he said, Ben, like, I voted for you.
0:14:45 You won.
0:14:49 The economist readers voted 78%, 22% for him.
0:14:51 ‘Cause like that’s how much people wanted to believe
0:14:52 we were in a bubble.
0:14:53 – That still do.
0:14:55 – Yeah, there’s a, I’m not sure if it’s quite
0:14:58 a cognitive bias, but I feel like there is a predisposition
0:15:00 that a lot of people have to take the cynical bet.
0:15:03 So how that seems smarter, ’cause either way,
0:15:04 there’s a payoff.
0:15:06 And the payoff, if you said that’s bullshit,
0:15:08 and then it turns out you are right,
0:15:11 seems greater to people than the opposite.
0:15:11 And also there’s a little bit
0:15:13 that you can’t prove a negative,
0:15:16 popper, the valid hypotheses and stuff like that.
0:15:19 – Plus you get the victory as the most smug person
0:15:20 in the room too.
0:15:23 – Yeah, so you’re generally betting against the cynicism
0:15:24 in your business.
0:15:26 Is that like something you can actually take advantage of?
0:15:27 Or is that something that works in your favor,
0:15:29 that cynicism?
0:15:30 – Oh, 100%.
0:15:33 In fact, like we have this thing that our friend came up
0:15:35 with, which is the East Coast, West Coast arbitrage,
0:15:38 which is anything that the people in the East Coast
0:15:41 think is ridiculous in a toy and people in the West Coast
0:15:44 think is the next big thing, that’s the thing to bet.
0:15:46 – And we said, you just keep flying back and forth.
0:15:48 – Yeah, and you find out what those things are,
0:15:50 and then you just invest all your money in that.
0:15:53 – That really suggests a question for today.
0:15:56 There are some big things happening today
0:15:57 that aren’t obvious.
0:16:00 What kind of energy do you put in to find that?
0:16:02 Are there like specialist researchers?
0:16:06 Is it just every partner’s kind of contribution?
0:16:07 How much time do you spend looking
0:16:09 for what isn’t obvious today,
0:16:11 but will be obvious in retrospect?
0:16:14 – Yeah, I think that ends up being like half the job
0:16:18 is trying to understand like,
0:16:21 what is the next big platform?
0:16:22 Where are things going?
0:16:25 What’s going to be the user interaction model
0:16:27 after the iPhone?
0:16:29 That’s a big open question right now.
0:16:31 Like what’s the next platform?
0:16:33 What is AI really going to mean?
0:16:37 Is this whole crypto thing real?
0:16:41 These are all like VR and AR, like at what point?
0:16:42 It’s hard to imagine.
0:16:44 20 years from now, it’s not working.
0:16:47 So like how many years from now will that take?
0:16:51 – Yeah, and these are the kind of fundamental questions
0:16:54 always in the venture capital business, I think.
0:16:56 – And you might add genomics, you might add CRISPR,
0:16:58 you might add synthetic biology.
0:16:59 – Absolutely.
0:17:00 – It’s three of the big new frontiers
0:17:01 on the biological front.
0:17:02 They all have that characteristic.
0:17:04 – CRISPR seems like one to me
0:17:06 that is going to have really dramatic impact.
0:17:09 Obviously there’s big moral debate to be had
0:17:11 and policy debate to be had.
0:17:13 How do you take something like that
0:17:15 and try to look for the opportunities?
0:17:17 – Yeah, so the big thing is we default into thinking,
0:17:18 okay, this is going to happen.
0:17:21 We don’t spend a lot of time on, okay, will this happen?
0:17:23 Like is this going to be a thing?
0:17:25 We try, in fact, I tried one of my things,
0:17:26 I tried it at the firm, it started very hard
0:17:28 to actually kind of prevent us from having the discussion
0:17:29 of like, okay, is this going to happen?
0:17:30 It’s more a question of like, okay,
0:17:32 let’s assume it does happen, right?
0:17:34 And so then there’s kind of two really critical questions
0:17:35 that follow from that, which is like, okay,
0:17:37 if it does happen, then where does it go?
0:17:39 And so the financial version of that question
0:17:41 is kind of how, we call how high is up,
0:17:43 which is like, okay, how big could it get, right?
0:17:45 Which is sort of a very interesting question
0:17:46 for venture capitalists, because it’s like,
0:17:47 if you make an investment in something,
0:17:49 even if it happens, it turns out to be small,
0:17:50 then it’s still not worthwhile.
0:17:51 And so you’re looking for the things
0:17:52 that could get really, really big.
0:17:53 And then you’re obviously looking for,
0:17:55 you know, you’re then looking for the founders
0:17:56 and looking for the specific ideas
0:17:57 and applications that you spend a lot of time on that.
0:17:59 The other thing you think a lot
0:18:00 about in this business is timing.
0:18:02 And so like my observation is basically,
0:18:04 basically everything happens,
0:18:05 like my entire history in this industry,
0:18:07 and at least for 25 years is basically everything
0:18:09 that people said was going to happen happened at some point,
0:18:11 up to an including online pet food delivery,
0:18:13 like it all actually happens.
0:18:15 – All the things they made jokes about
0:18:19 in all the early 2000s movies are all actually.
0:18:19 – They’re all actually happening,
0:18:21 but it’s like, okay, when is it going to happen?
0:18:23 And like, is it ready now?
0:18:24 You know, is it going to be on this cycle?
0:18:25 And then how do you bet this?
0:18:28 Like sometimes these things take three, four, five cycles,
0:18:30 right, for the founders to really figure things out
0:18:32 for the technology to kind of fall into place.
0:18:33 So it’s kind of like, what are the exploratory bets?
0:18:36 How are you kind of vetting whether the stuff is real?
0:18:37 And then there’s this kind of multi-dimensional question
0:18:39 of like, you know, kind of to your point,
0:18:40 there’s this multi-dimensional question of like,
0:18:42 okay, is the technology ready?
0:18:44 And then you got to kind of cross that with like,
0:18:45 do you think the market’s ready?
0:18:47 Like, do you think people are going to want this?
0:18:48 And then you might have to cross like,
0:18:49 in CRISPR, you might have to cross other issues
0:18:51 like regulatory issues,
0:18:52 like are the regulators going to buy into this?
0:18:55 And so, and that’s where I think you have to kind of explore
0:18:56 as you go, right?
0:18:58 Which is you have to kind of feel your way through it.
0:19:00 Like it’s often not the first company in a category
0:19:01 that ends up being the winner, right?
0:19:03 Well, this is a Peter Thielism that we quote a lot.
0:19:05 It’s not the first company that gets all the money.
0:19:06 It’s the last company in the market
0:19:07 that gets all the money, right?
0:19:08 In other words, it’s the company
0:19:09 that actually takes the market, right?
0:19:11 It ends up actually being the dominant company
0:19:13 and forecloses the opportunity for there
0:19:14 to be new startups behind it.
0:19:17 And so sometimes that’s a pioneer, sometimes it’s not.
0:19:19 And so you’re kind of having this constant discussion
0:19:21 about timing.
0:19:23 And then, at the end of all that,
0:19:25 we kind of try to park that to a certain extent
0:19:26 and just start talking to entrepreneurs
0:19:28 and figure out who is the person
0:19:30 who’s got this the most decoded,
0:19:32 how much time and effort has that person put into it?
0:19:34 How qualified are they to pursue this?
0:19:35 What’s their personality?
0:19:36 And can they build a company around it?
0:19:39 Yeah, that’s what I was gonna actually go right there
0:19:42 because knowing you, as I do, I can’t imagine it’s ever,
0:19:44 we have a thesis that this thing is going to work out,
0:19:46 now we’re going to look for the entrepreneurs
0:19:47 who are doing this thing.
0:19:50 It’s much more gonna be the confluence of who you’re meeting,
0:19:53 who you get to talk to, what people are up to,
0:19:56 and these background theses that give you the opportunities.
0:19:57 Yeah, that’s right.
0:19:58 Yeah, I think that’s right.
0:20:00 And actually, I mean, I think for a while,
0:20:04 it was always like started with the entrepreneur
0:20:06 in the early days of it, just ’cause they were allowing
0:20:08 the two of us and we couldn’t cover all the spaces
0:20:10 and enough depth to do it any other way.
0:20:14 But the other thing kind of related to that
0:20:18 is the platforms that kind of we think
0:20:19 are getting proximate from a timing standpoint
0:20:22 are the ones where like the smartest entrepreneurs
0:20:24 are all working.
0:20:27 So if we see 20 genius entrepreneurs all working on crypto,
0:20:30 that makes us pay attention, for example.
0:20:30 Right.
0:20:33 So Mark, you’ve talked about five-year cycles in tech.
0:20:36 Is that something that you think is a good way
0:20:40 to imagine what’s going on or to picture it in context?
0:20:42 So I would say there’s two big sites.
0:20:43 It’s hard to, you know, the stuff is just,
0:20:44 these are just general frameworks.
0:20:46 And so they vary a lot in practice,
0:20:47 but they’re two big general concepts.
0:20:51 And so like I would say, one is the big technology changes,
0:20:52 like the ones we’ve been talking about,
0:20:54 like they’re generational changes,
0:20:54 like they’re quite literally,
0:20:56 they’re human generational changes.
0:20:59 So it’s like the typical cycle in those is like 25 years.
0:21:02 And the reason literally is because a lot of the time,
0:21:04 you just, the people who are in positions of power
0:21:06 and decision and influence when the new thing comes out,
0:21:08 they just will not accept it.
0:21:10 They won’t accept it, they won’t adapt to it.
0:21:11 They won’t recalibrate to it.
0:21:15 And fundamentally they need to age out of the cohort
0:21:16 that has the power, right?
0:21:18 Purchasing authority and all the decision-making authority
0:21:19 and all these other things.
0:21:21 And so they, and then you need a new generation
0:21:22 that like takes the stuff seriously
0:21:23 ’cause they grew up with it, right?
0:21:25 And you need them to age into the cohort, right?
0:21:27 So it’s like, it’s literally a generational turnover.
0:21:29 And so you see these, and that’s why you get these things.
0:21:31 You’ll see these, you know, some of these new things
0:21:32 that’ll grow for 25 years.
0:21:33 And I’m convinced like a big part of it
0:21:35 is just simply that generational effect.
0:21:37 So that’s the good news is like when they work,
0:21:39 you can have literally decades of growth
0:21:43 off of enormous skepticism from day one.
0:21:46 You know, the bad news is, as you’re well aware as a founder,
0:21:48 no individual company gets 25 years, right?
0:21:49 To prove something, right?
0:21:50 (laughing)
0:21:53 Nope, in fact, something well short of that, let’s say.
0:21:57 And so like our basic mental model is a company on average
0:21:59 gets maybe five years to prove something,
0:22:00 to prove the hypothesis.
0:22:01 Like, and you can kind of like,
0:22:03 if you’re a top end founder and you’re super credible,
0:22:04 you could probably raise the seed around,
0:22:06 you know, series A, series B,
0:22:07 you can get yourself five years of runway,
0:22:09 you can get engineers to some product people
0:22:11 to sign up for that and you can prove it or not.
0:22:12 But after five years, if it’s not working,
0:22:14 like you start to have a problem and,
0:22:16 or I should say you have two problems.
0:22:18 One is you start to have a morale issue
0:22:19 where people start to lose faith
0:22:21 and they spin off and go to other things.
0:22:23 You can also end up with an architecture issue, right?
0:22:25 Which is like, even if you’re right,
0:22:26 even if it starts to happen,
0:22:28 you’re built on the prior architecture, right?
0:22:30 And so, you know, imagine being a mobile developer
0:22:33 that started in, you know, 2002, right?
0:22:36 And even if they were still around when the iPhone came out,
0:22:37 you know, they’d built their entire, you know,
0:22:38 system on brew and, you know, Java
0:22:40 and all these technologies that were now archaic.
0:22:42 And so, so you kind of have this,
0:22:45 this kind of aging in place thing that happens.
0:22:47 And so each company kind of has a five year shot.
0:22:48 So then what happens is-
0:22:49 – There are exceptions.
0:22:50 – Yes, there are particular founders
0:22:51 who can, who can, who can get through this,
0:22:53 but it does tend to be the exception.
0:22:55 And so, but, but then you think about it,
0:22:56 then there’s sort of the psychological thing
0:22:57 that happens as a consequence,
0:22:59 which is if the founder starts the company
0:23:00 in the first cycle, runs for five years
0:23:02 and it doesn’t prove the hypothesis,
0:23:03 that founder usually ends up,
0:23:06 so bitter about the whole experience
0:23:07 that they become cynical about that category
0:23:09 for the rest of their lives, right?
0:23:11 And then, and then somebody else in sort of, you know,
0:23:13 cycle to generation two, three, four does figure it out.
0:23:15 And like, you know, that, by the way,
0:23:17 I’m speaking out of also myself out of experience,
0:23:18 like talk about upset,
0:23:19 the fact that you couldn’t get it to work
0:23:22 in this other person did, it’s just like absolutely maddening.
0:23:23 And that’s just human nature.
0:23:25 The part of it that really bites the VCs
0:23:27 is the VCs do that too.
0:23:29 If you as a VC make a bet and go on a board
0:23:31 and you’re in the board meetings for five years
0:23:32 and it doesn’t work and the company shuts down
0:23:36 and then a new kid shows up, you know, three weeks later
0:23:38 and says, hey, I’ve got an idea.
0:23:40 Why don’t we do that, right?
0:23:41 And of course, what really makes you frustrated as a VC
0:23:43 is that kid half the time isn’t even aware
0:23:44 of the previous failed experiments
0:23:47 ’cause like they literally weren’t paying attention, right?
0:23:49 And so what happens is actually the VCs will free,
0:23:52 it’s actually, the VCs will actually freeze themselves out.
0:23:54 And so, and it’ll be VCs who are much more naive
0:23:56 and much less aware of the previous failures
0:23:57 that will actually make the bet.
0:23:59 And so that puts you in this very weird spot.
0:24:00 If you think about being a VC
0:24:01 or running a venture capital firm,
0:24:03 which is you would like to say that the person
0:24:04 who knows the most about the domain
0:24:06 is the person who should make the investment decision.
0:24:07 But it may also be the case,
0:24:09 the person who knows the most about the domain
0:24:10 has the most scar tissue
0:24:11 and has the most followed up psychology.
0:24:14 And so a big part, exactly to your comment,
0:24:16 like a big part of this job, just like being a founder,
0:24:18 is like you have to suppress your natural instincts
0:24:20 to get bitter and resentful and envious and upset.
0:24:22 And it goes to even a more fundamental question is,
0:24:24 like, can you learn lessons, right?
0:24:26 Like what do you learn, like in this business,
0:24:27 what do you learn from a failure?
0:24:29 And maybe the answer is you should learn a lot
0:24:31 from a failure ’cause like it’s those are all hard-won lessons
0:24:33 and maybe the answer is you should learn absolutely nothing.
0:24:35 Maybe all the lessons are wrong, so.
0:24:37 – Yes, it is a lesson to just, that didn’t work.
0:24:38 – Yeah.
0:24:41 – That’s something that I spent a lot of time trying to,
0:24:44 to convince people on the team of that,
0:24:46 it’s the right brothers or Thomas Edison.
0:24:49 It’s just every day, what’s the best idea we got?
0:24:50 That didn’t work.
0:24:51 All right, what’s the next best idea we got?
0:24:55 And the characterization of celebrating failure
0:24:58 sometimes misleads people to characterize that as a failure.
0:24:59 ‘Cause if this plane didn’t fly
0:25:00 or this light bulb didn’t light up
0:25:02 or it blew up or whatever, is that a failure?
0:25:04 Or is that just the process of getting to the light bulb
0:25:06 that works, that they are playing that flies?
0:25:09 – Yeah, Edison tried 3000 compounds, I think,
0:25:10 for the light bulb.
0:25:11 – Yeah.
0:25:11 – Before he figured out the filament.
0:25:15 – That is a level of persistence I would like to hire.
0:25:18 So Mark, you just mentioned a little bit of like control
0:25:20 over your own emotions or your reactions, the cynicism.
0:25:22 Ben, that’s something that you talked a lot about
0:25:23 in the hard thing about hard things,
0:25:24 just like the power to overcome.
0:25:26 To what, I mean, obviously you’ve seen a lot of this,
0:25:29 you’ve experienced it yourself, to what degree
0:25:32 do you think you’ve been helpful and let me just say,
0:25:34 you have been helpful to me personally
0:25:36 in helping entrepreneurs through some of that.
0:25:38 Whether it’s the overwhelming emotional reaction
0:25:40 to a bunch of good stuff happening
0:25:42 or a bunch of bad stuff happening.
0:25:44 – Yeah, so I think, I mean, that’s probably
0:25:47 that the number one kind of consistent thing
0:25:49 that I get back on that book is like,
0:25:52 you know, what I’m feeling is so intense
0:25:55 and there’s nobody to talk to about it.
0:25:58 And then the book kind of goes like,
0:26:01 this is what it feels like, this is what it looks like.
0:26:05 And I think that just knowing that you’re not
0:26:07 the stupidest entrepreneur of all times
0:26:10 is like really valuable.
0:26:11 And it’s something that I always wish that I had.
0:26:13 It’s a lot of the reason I wrote the book
0:26:17 ’cause I used to go around, you know, and I, you know,
0:26:21 and I was like in the kind of horrible period of 2001
0:26:24 and I would talk to other founders and I’d be like,
0:26:25 you know, how’s it going?
0:26:27 And they’d be like, it’s amazing.
0:26:30 I can’t, this is the greatest experience of my life.
0:26:32 And I’d just be like, wow, I am like
0:26:34 the stupidest motherfucker of all times.
0:26:37 Like, ’cause my business is in horrible trouble.
0:26:38 And it just seems so bad.
0:26:40 But then like, you know, as because I’ve lived long enough
0:26:43 to see like most of those guys went bankrupt.
0:26:44 So they were all going through it.
0:26:45 I was going through.
0:26:46 They just like nobody would tell each other
0:26:48 ’cause it’s so embarrassing.
0:26:50 So it was, you know, one of those, those kinds of things.
0:26:53 But it’s been, you know, it’s been great help to me
0:26:57 in the work because when I sit down with an entrepreneur,
0:27:00 they go, okay, yeah, I know, you know what I’m talking about.
0:27:03 So we can talk about the real like horrible shit
0:27:06 and not just, you know, the happy stuff.
0:27:08 – I feel like there’s an obligatory question here
0:27:10 to talk about the things that you guys tried
0:27:12 that didn’t work and the failures.
0:27:14 But before we get there, just on the way,
0:27:15 there’s a bunch of things that obviously did work.
0:27:20 So building up a bunch of capabilities in the firm
0:27:21 and in the partnership with something
0:27:23 that is now pretty widely emulated,
0:27:26 like having those services that the companies can call on,
0:27:28 being a little bit more stage agnostic
0:27:32 and even like industry technology vertical agnostic
0:27:35 has been something that’s worked out really well.
0:27:38 When you look back before we get to the failure question,
0:27:40 what do you think the best decisions you made are
0:27:42 in the way that you set up the firm?
0:27:44 – Yeah, so it’s a great question.
0:27:49 I think, you know, at the core, I think just this belief
0:27:54 that a venture capital firm has got to be able
0:27:57 to help the technical founder grow into a CEO.
0:28:00 It’s just so, you know, in retrospect,
0:28:02 that turned out to be profound
0:28:04 and just differentiating and important.
0:28:06 And it really is the work.
0:28:08 So when we think about what do we do
0:28:11 and why are we here, that’s it.
0:28:13 And then kind of backing that up,
0:28:15 the thing that helped us the most
0:28:19 is just taking what Michael Ovitz had done at CAA.
0:28:22 And that jump started us, I mean,
0:28:24 we probably saved five years by copying his model.
0:28:27 So that, and I can’t even believe how well it worked,
0:28:29 like every aspect of it worked.
0:28:32 So, you know, those are probably the two things.
0:28:34 – What was it from Michael Ovitz that you were copying?
0:28:35 – He has censor it in the book.
0:28:36 And so there’s a great book.
0:28:38 – Yeah, he finally revealed some of it.
0:28:40 – Some of it’s in the book, not all of it’s in the book,
0:28:42 but it’s the book is called “Who is Michael Ovitz?”
0:28:45 And it’s a highly entertaining book and we recommend it.
0:28:46 I mean, it’s actually really funny.
0:28:48 He kind of sat down and described the whole thing.
0:28:50 And it basically was this idea of, you know,
0:28:51 we’re not just going to be a collection of individuals.
0:28:53 We’re going to be an actual true team.
0:28:55 And then it’s not just going to be the principles.
0:28:57 It’s going to be an entire system, right?
0:28:58 It’s going to be an entire operating platform,
0:28:59 an entire infrastructure.
0:29:00 It’s going to be professionals
0:29:02 across all these different domains.
0:29:04 And, you know, we’re going to build this
0:29:06 enduring long run network that’s going to, you know,
0:29:08 it’s just going to constantly compound year after year
0:29:09 and build more and more value.
0:29:11 And then the next client comes longer,
0:29:12 the next, you know, founder comes along
0:29:14 and they can plug into this entire system, you know,
0:29:15 that’s been built, you know,
0:29:17 and we’ve been building the system now for a decade.
0:29:19 And so, you know, a new founder who works for this today,
0:29:21 like they’re walking into a, basically a system
0:29:22 that’s been built for a decade.
0:29:23 So then he said, basically what happens is
0:29:24 then it’s compounding advantage,
0:29:25 which is every year that goes by,
0:29:27 you just get more and more differentiation
0:29:28 off of the status quo.
0:29:30 He was competing at the time with William Morris,
0:29:31 which was this huge talent agency.
0:29:33 And it’s like, well, why wouldn’t they just copy you?
0:29:34 And he’s like, well, they’d have to vote themselves
0:29:36 big salary cuts, right?
0:29:38 Like they’re paying themselves all the money right now, right?
0:29:40 And so they had to go hire, you know,
0:29:41 100 people to go do the stuff that we’re doing.
0:29:42 They’d have to like,
0:29:43 they’d have to free up that money from something.
0:29:45 So they’d have to vote themselves giant salary cuts.
0:29:47 And he’s like, they don’t like each other.
0:29:48 Like they don’t get along to start with.
0:29:50 And so imagine getting into the room.
0:29:50 And they’ve all got like, you know,
0:29:52 they’re like very successful people.
0:29:53 They’ve all got very high personal burn rates, right?
0:29:54 They’ve got all kinds of hobbies, you know,
0:29:56 they’ve got vineyards and yachts and all this stuff.
0:29:59 And so, you know, they’re going to now decide
0:30:01 to give themselves an 80% pay cut, right?
0:30:03 To compete with the startup, like no chance.
0:30:05 And so anyway, that was his explanation.
0:30:07 – Yeah, that’s been a long lasting advantage.
0:30:08 I think so.
0:30:10 – Yeah, but yeah, that continues.
0:30:11 As it actually turns out,
0:30:13 that didn’t just happen in the talent agency business.
0:30:16 What I discovered doing more research after that was that,
0:30:18 it was also exactly what happened for law firms.
0:30:20 It’s exactly what happened management consulting firms.
0:30:22 It’s what happened to ad agencies, accounting firms.
0:30:23 And then also investment banks,
0:30:25 private equity firms, hedge funds.
0:30:26 All these other industries have gone
0:30:27 through this transformation.
0:30:30 They basically professionalized and upleveled.
0:30:31 And it just happens that the venture industry
0:30:33 is doing that now.
0:30:34 And in fact, I think at this stage,
0:30:36 like it’s beyond just us.
0:30:36 – There’s something else there
0:30:38 that I actually hadn’t really realized,
0:30:40 but maybe it was implicit the whole time,
0:30:43 is CA’s investment was to make the people
0:30:44 that are representing more successful.
0:30:47 Smart idea, given that you got points
0:30:49 on their success is a little bit of the same thing.
0:30:51 ‘Cause you can make an investment decision
0:30:54 that it’s just like I’m gonna buy some copper futures
0:30:56 or oil or something like that.
0:30:59 And I can’t do anything about to make oil more valuable
0:31:01 for people or copper more valuable.
0:31:02 I invest in the startup
0:31:03 and there’s a ton of stuff I can do.
0:31:05 I have my connections and I know that personally
0:31:08 I benefited from being able to call on both of you
0:31:11 from John O’Farrell who joined our board,
0:31:14 from Margaret, from Jeff Stump on recruiting,
0:31:16 from the whole team running the EBCs.
0:31:18 There’s more than I can mention here.
0:31:20 And I think that has,
0:31:24 I don’t know what the ROI for you is on that,
0:31:26 on top of the dollars that you put in,
0:31:29 but I would say it’s probably 70% of the value to us
0:31:31 and 70% of the value created came
0:31:33 from that additional support beyond just the money.
0:31:36 – Yeah, and there’s also this little knock-on effect
0:31:37 which you appreciate, I’m sure,
0:31:42 which is part of the trouble with an inventor becoming a CEO
0:31:46 is you just don’t feel like a CEO.
0:31:48 You don’t know the people who CEOs know,
0:31:50 you don’t know how to do the things CEOs know how to do.
0:31:54 And so a lot of what you get out of the firm is,
0:31:55 no, I’m a CEO.
0:31:57 Like if I need to know how to do it, I’ll call Margaret.
0:32:00 Like, you know, I can do that, I can step up.
0:32:03 And so like that’s a lot of what it conveys
0:32:06 at the end of the day is just like that confidence
0:32:08 which is often that little difference
0:32:10 between being able to stay in the job
0:32:13 and having to raise your hand and tap out.
0:32:15 – All right, so I promised that I was gonna ask
0:32:20 about any dumb decisions, mistakes, failures along the way.
0:32:21 What do you got?
0:32:22 – We haven’t made any.
0:32:24 I don’t really know why you would even ask that question.
0:32:26 There’s obviously nothing to talk about.
0:32:29 So we have a very specific philosophy on that
0:32:30 and the book I’d really recommend.
0:32:32 We were lucky enough to have her in the podcast a while ago.
0:32:34 So Annie Duke wrote a book called “Thinking in Bets”
0:32:36 where she talks about basically what is the nature
0:32:39 of a mistake in a probabilistic domain,
0:32:41 you know, with uncertainty of outcome.
0:32:44 And she uses the term in the book “resulting.”
0:32:46 It’s basically the process of looking at a bet
0:32:48 that was made in a probabilistic domain
0:32:49 that did not pan out.
0:32:51 And then concluding that that was a mistake
0:32:53 as compared to a bet that didn’t pan out.
0:32:54 And so basically what she says in the book,
0:32:57 she says the book is basically resulting
0:32:59 is the root of all evil if you’re in a probabilistic business
0:33:00 ’cause you will learn the wrong lessons
0:33:03 and you’ll torture yourself mentally to death by doing that.
0:33:06 And so she says the thing to do is basically
0:33:08 to very clearly separate in your own mind process
0:33:09 and outcome, right?
0:33:11 So you’re in a probabilistic domain,
0:33:11 you don’t know the outcome
0:33:13 of any particular bet ahead of time.
0:33:15 And so you need to design the best possible process
0:33:18 to generate the best possible set of outcomes over time.
0:33:21 And then basically when things go quote unquote wrong,
0:33:23 you don’t second guess the outcome.
0:33:24 You go back and you just make sure the process
0:33:26 was as good as it could have been.
0:33:28 And so, you know, from the outside,
0:33:30 the mistakes of venture firm makes are always like,
0:33:31 well, what’s the investment that you didn’t make
0:33:33 that worked or the investment that you made
0:33:34 that didn’t work inside the firm?
0:33:35 What we try to do is say, okay,
0:33:37 what have we done well in our process
0:33:39 and how can we improve our process?
0:33:41 You know, that’s a much more boring topic to talk about
0:33:42 ’cause it has to do with things like meeting structure
0:33:44 and memo documents and, you know, research
0:33:46 and due diligence and all these topics.
0:33:49 But that is the actual answer to the question, which is,
0:33:51 you know, when we started with a, as Ben said,
0:33:53 we started with a relatively lightweight process
0:33:54 on investments because it was just Ben and me
0:33:57 and then over the time we’ve evolved
0:34:00 to a much more rigorous process.
0:34:01 Today we do a lot more work on the investments
0:34:03 than we used to.
0:34:05 And then the other side of that is to try to keep all that
0:34:06 work from preventing us from making
0:34:07 the controversial investments.
0:34:08 – Getting people to that position.
0:34:11 And I wish I could remember who gave me this analogy,
0:34:14 but if you got super drunk and then you drove home
0:34:16 and you didn’t have a crash,
0:34:18 it’s not that that was a good decision.
0:34:18 The result was good.
0:34:21 And we really, it’s a tough thing to build structures
0:34:23 inside the company to celebrate.
0:34:26 That was a great idea to change the homepage
0:34:28 so that whatever, and it turned out that it was wrong,
0:34:30 but you still got a bonus.
0:34:32 You get a bonus, you get a promotion,
0:34:35 you get recognition, even though it didn’t have the result
0:34:37 ’cause people do get super result fixated.
0:34:40 – Yeah, Bisa says real good thing where he says,
0:34:42 “We rate people on the inputs, not the outputs.”
0:34:45 – One of the last questions I wanted to get to
0:34:47 is the nature of the entrepreneurs.
0:34:48 Now it’s been long enough.
0:34:51 You’ve seen some two-time, three-time entrepreneurs
0:34:53 and you’ve backed some of them.
0:34:55 What kind of difference do you see between that,
0:34:56 the first time and the second time?
0:34:59 – So for top-end venture, basically the rule is
0:35:00 if you’re a first-time fund,
0:35:01 to get funded by a top-end venture firm
0:35:03 as a first-time founder, you have to have something working.
0:35:04 You have to have a product.
0:35:06 You have to have some of a product market fit.
0:35:08 Google.com already existed.
0:35:09 Facebook already existed.
0:35:11 Airbnb already existed when they raised money.
0:35:13 So that’s the general pattern.
0:35:15 And then the question of the first-time founders is,
0:35:16 do they know what they’re doing, right?
0:35:19 Can they then do the job of being a founder CEO
0:35:22 of a scaling company and some can and some can’t?
0:35:24 The second-time founders are like a huge relief
0:35:25 to deal with on the one hand
0:35:27 because it’s like, okay, they’ve been through it before.
0:35:28 Now they know what they’re doing, right?
0:35:30 They’ve got some experience and some gray hair
0:35:32 and some operational experience.
0:35:34 The problem with the second-time founders,
0:35:37 they can raise money before they have something working.
0:35:40 And then there’s this question of like, okay, what’s the idea?
0:35:41 And then we talk a lot about like,
0:35:44 is it an organic idea or is it like a synthetic idea?
0:35:45 Was the process, I have a great idea there
0:35:46 for I’m going to start a second company
0:35:48 or was the process, I want to start a second company
0:35:50 and therefore I have to come up with an idea.
0:35:52 And what you often find is they want to start
0:35:55 a second company so badly that they come up
0:35:58 with basically a fragmentary idea, partial idea.
0:36:00 It’s a conceptually interesting idea,
0:36:02 but with nothing underneath it.
0:36:03 And we have this other concept
0:36:06 we use called the idea maze, which basically is the process
0:36:08 that a founder uses to figure out what the actual idea is,
0:36:09 which is like a hundred-step process
0:36:11 to work your way through all the different permutations
0:36:12 of the idea before you actually finally figure out
0:36:13 the real thing.
0:36:15 And the second-time founders often just haven’t gone
0:36:17 through the idea maze, but it’s really bizarre as a VC
0:36:19 because it’s like, here’s this founder who you love
0:36:21 and like they’ve showed incredible persistence
0:36:22 in the last company and like you so badly
0:36:24 want to work with them again and you can just tell like,
0:36:26 for the idea has almost become interchangeable.
0:36:28 Right, and it’s just like, that’s a super bad sign.
0:36:30 And so that’s what tortures you on those.
0:36:32 – Yeah, from the entrepreneur side, I can tell you,
0:36:33 having a whole bunch of money does take away
0:36:36 a very critical forcing function, which is like,
0:36:38 I’m about to find out of money, I better figure this out.
0:36:41 So we added overcorrect for that.
0:36:42 – What’s the job of the founder?
0:36:45 Is the job of the founder to figure out the product,
0:36:47 figure out the market and get the idea nailed?
0:36:49 Or is the job of the founder to staff
0:36:51 an executive team and an employee base, right?
0:36:54 And those are two like, they’re overlapping responsibilities,
0:36:55 but there’s a lot of-
0:36:57 – The first one turns out to be a lot more important.
0:36:58 – Yeah, it’s like the startups where it’s like,
0:37:01 they’re doing all the outward facing things
0:37:03 involved with being a startup, but like there’s nothing there.
0:37:05 – Well, you know, and you always kind of know it
0:37:09 because the founding team is all vice presidents
0:37:10 and no engineers or something like that.
0:37:13 It’s just like, okay, what are you doing?
0:37:16 You can’t execute your way through like no ideas.
0:37:18 – 10 people in the company, they’ll have a chairman,
0:37:20 they’ll have a CEO, they’ll have a CEO, they’ll have a president,
0:37:21 they’ll have a VP of sales, a VP of marketing
0:37:23 and a VP of engineering.
0:37:25 That’s not a good, yep.
0:37:27 – A little bit of a pet peeve for me too.
0:37:30 So the first time that I met Ben, it was with Mark
0:37:32 and it was at the Creamery in Palo Alto.
0:37:34 And I think you were about probably about six months away
0:37:36 from starting the fund.
0:37:40 I never would have predicted how things would have turned out.
0:37:42 Did you predict how things would have turned out?
0:37:47 – You know, no, I think we dreamed that we would kind of
0:37:48 get to where we got to,
0:37:51 but it was a much longer timeframe on the dream.
0:37:54 I mean, things worked out way, way better
0:37:58 than I think either of us set out and expected.
0:38:01 And, you know, we had a lot of good luck along the way.
0:38:03 And then a lot of, you know, great help
0:38:07 from a lot of people, you know, people like actually starting
0:38:08 with like people like Jim Breyer,
0:38:13 who kind of taught us what it meant to like create an LP base
0:38:16 and how to think about investors and those kinds of things.
0:38:19 And then, you know, we ended up getting like very lucky
0:38:21 on the hiring, I think our first hire was Scott Cooper,
0:38:25 who we probably kind of built the firm with that.
0:38:28 And then, you know, our first consultant was Margaret.
0:38:31 And, you know, there’s no way we would have like
0:38:32 pulled off the marketing thing without her.
0:38:35 So a lot of, a lot of really great luck
0:38:36 and a lot of really great help.
0:38:38 Oh, and Andy Rackliffe, yeah, he helped us understand
0:38:39 what venture capital was.
0:38:41 – It turns out to be.
0:38:43 – Neither of us had any experience.
0:38:44 – Then obviously all the partners who have joined us
0:38:47 and so about 150, 150 people now,
0:38:49 by definition numerically, they get almost all the credit.
0:38:52 – So one thing I definitely want to get to is the transition
0:38:57 from a VC firm to a financial advisor for whatever that means.
0:38:59 And you were a little bit iconoclastic and different
0:39:00 from the beginning.
0:39:04 This also seems, we’ll see looking back at iconoclastic,
0:39:05 but definitely different.
0:39:06 What was the idea there?
0:39:08 – Yeah, so, you know, kind of the thing that catalyzed it
0:39:10 was actually crypto.
0:39:14 There’s a rule that exempts VCs from having to do
0:39:17 a bunch of kind of regulatory compliance stuff.
0:39:21 And part of the thing that keeps you as a VC
0:39:26 is you can’t invest more than 20% of your funds
0:39:30 and things that aren’t like primary equity investments.
0:39:33 So crypto would fall into that category secondary
0:39:34 and so forth.
0:39:37 And look, we believe crypto is going to be important.
0:39:39 Now there’s a lot of VCs who do,
0:39:41 who won’t take the step that we did to become regulated
0:39:43 in the way that we have.
0:39:46 But, you know, like this is kind of another advantage
0:39:49 from our background is we’re not afraid of governance
0:39:50 or regulation or these kinds of things.
0:39:52 And that, you know, it’s something that we understand
0:39:54 pretty well from being a public company.
0:39:55 We’ve done it before.
0:39:57 And it opens up a lot of opportunities
0:39:59 that we can now think about
0:40:02 because, you know, we’re in another category.
0:40:05 – Yeah, so this changed the categorization
0:40:06 and then the regulatory environment,
0:40:07 but you’re still a V-serfer.
0:40:08 – Yes, yeah.
0:40:11 Now we still are in the exact same business we always were.
0:40:13 – Mark, I gotta also do TV shows that you recommend
0:40:17 ’cause you are a source of excellent viewing.
0:40:18 – There we go, good.
0:40:19 All right, well, I’ll struggle.
0:40:20 So I can’t help myself.
0:40:23 Deadwood is the best TV show of all time.
0:40:24 And it’s actually very relevant for founders.
0:40:27 It’s basically, it’s the story of the American frontier
0:40:29 through kind of a modern lens.
0:40:31 And it’s just astonishingly high quality.
0:40:32 And it’s basically the creation of a city.
0:40:33 It’s basically the creation of a city
0:40:34 and ultimately the creation of a state,
0:40:36 the state of North Dakota.
0:40:39 And it is, you know, in the face of just like, you know,
0:40:41 horrifying obstacles.
0:40:42 You know, and by the way, you know,
0:40:44 many ethical issues along the way and everything else.
0:40:45 If you think starting a tech company is hard,
0:40:47 you wanna watch a couple of seasons of Deadwood.
0:40:48 It’ll put you in the right frame of mind.
0:40:50 – And then the movie that it got canceled,
0:40:51 it gave us a decade ago
0:40:52 and it got canceled after three seasons
0:40:54 and it really should not have been.
0:40:56 And so they did a very rare thing.
0:40:56 They went back 10 years.
0:40:59 And all these other people in the movie became huge stars
0:41:00 afterwards in the show.
0:41:01 So they got them all back
0:41:03 and made the fourth season into a movie.
0:41:04 – Wow.
0:41:05 All right, can’t wait to see it.
0:41:06 – Yep.
0:41:07 You can give me two books if you want, two books.
0:41:09 – Yeah, give me two books, give me five books.
0:41:11 – Favorite two books of the year.
0:41:14 Book number one, “Why History is Always Wrong?”
0:41:18 is not written by a historian and it is basically,
0:41:19 if you’ve read Nassim Taleb,
0:41:21 he talks about something called the narrative fallacy,
0:41:24 which basically is, okay, why did something happen?
0:41:25 And then there’s some story as to why it happened.
0:41:27 And then it usually turns out,
0:41:28 like if you talk to the principles involved,
0:41:29 it wasn’t that story at all.
0:41:31 It was something much more complex.
0:41:34 And so this is like the next level of that theory
0:41:36 that basically says all of recorded history
0:41:37 is the narrative fallacy.
0:41:39 And so everything that we think we understand
0:41:41 about why the American revolution happened
0:41:44 or why Rome fell, right, or why Christianity emerged
0:41:45 or like any of these stories that we,
0:41:47 all the, any of these things that we teach you,
0:41:50 we’ll take 12 years of history class in school
0:41:51 and all this stuff, like it’s all wrong.
0:41:53 Like it’s worse than wrong
0:41:55 because it’s not like it could be corrected.
0:41:57 You couldn’t actually make a bunch of edits
0:41:59 to the book and make it correct.
0:42:00 You can’t do that at all.
0:42:02 And the reason why it’s worse than wrong
0:42:03 and you can’t ever get it right
0:42:06 is because reality is so complicated, right?
0:42:07 Reality is a complex adaptive system
0:42:09 when you’ve got human agents involved in everything.
0:42:12 And so you’ve got, and anything big that happens,
0:42:13 you’ve got thousands or millions of people
0:42:15 who are making all kinds of random decisions every day
0:42:16 for all kinds of random reasons
0:42:19 and it just happens that things result in a certain way.
0:42:21 – Is this wrong in the same way to think that
0:42:22 it didn’t rain because God was mad
0:42:24 or it did rain because we performed the ritual
0:42:27 in the right way, like puts some agency into a system
0:42:29 that there isn’t anyone making decisions.
0:42:30 It’s just the emergence.
0:42:31 – Yeah, exactly, in reality, the weather system.
0:42:33 I mean, you know, this is after 100 years
0:42:33 of meteorological science
0:42:35 and they still can’t predict it’s gonna rain tomorrow.
0:42:37 And it’s ’cause the atmospheric system
0:42:38 is a complex adaptive system
0:42:40 is too complicated to model.
0:42:42 And just like, and you can keep throwing supercomputers at it
0:42:44 and it’s still too complicated to model.
0:42:46 So, by the way, it’s the same thing, the human body.
0:42:47 Like we don’t, it’s actually,
0:42:48 we think about this a lot in the bio fund.
0:42:49 It’s like, we don’t understand.
0:42:50 Like we don’t even, there’s not even settled
0:42:51 in traditional science.
0:42:53 Like we still don’t know, like there’s still,
0:42:55 and there’s no, a whole new category revision of science
0:42:58 now questioning this whole, the whole protein fat,
0:42:59 you know, the whole protein fat thesis.
0:43:02 And so like it’s, the example he uses in the book
0:43:04 at the fall of Rome is like the, you know,
0:43:06 the single most studied kind of historical story
0:43:07 is like the rise and fall of Rome.
0:43:09 And it’s like, and basically what happens is like,
0:43:10 if like a science is working properly,
0:43:12 you converge on the correct answer.
0:43:13 Like you converge on Newton’s law.
0:43:15 So you converge on quantum mechanics
0:43:16 or something like that.
0:43:18 He’s like, the problem in history is that
0:43:19 the more time goes, Pat,
0:43:21 the more explanations they come up with,
0:43:23 the more new explanations they come up with.
0:43:24 And there’s like historians have documented,
0:43:26 there’s like 250 now different causes
0:43:28 for the fall of Rome, right?
0:43:30 And so like, it just, it leaves you with nothing.
0:43:33 And so it’s, it’s, it’s a, it’s a disconcerting theory
0:43:34 ’cause it basically says getting a handle
0:43:37 on cause and effect in the world is impossible.
0:43:38 It’s a very convenient theory
0:43:41 ’cause it means you can just ignore history.
0:43:42 That saves you a lot of time.
0:43:43 It saves you a lot of time.
0:43:45 And then it’s an inspiring story.
0:43:48 Our theory, I find, ’cause it’s like, okay,
0:43:49 things can change.
0:43:51 Like nothing is actually carved in stone,
0:43:53 like not even a little bit.
0:43:55 And who knows what’s going to be the next person
0:43:56 who’s going to make the decisions
0:43:56 that’s going to cause everything
0:43:57 to go one way or the other.
0:43:58 And that could be you.
0:43:59 And so I find that inspiring.
0:44:03 And then the other book I love to tell people about
0:44:06 is David Goggins, who’s a, the only guy in history,
0:44:08 he’s a triple qualified as an ABCL and army ranger
0:44:11 and what’s called an Air Force tactical air controller,
0:44:13 which is a special forces unit of the Air Force.
0:44:15 So triple qualified special forces.
0:44:18 He wrote a book called “Can’t Hurt Me.”
0:44:19 And it is one of the most amazing stories
0:44:20 that anybody has ever written.
0:44:23 His story is really amazing.
0:44:25 He’s one of the only African-American Navy SEALs
0:44:29 in history and just manages incredible accomplishments
0:44:30 both inside and outside the military.
0:44:32 And it’s the book.
0:44:34 Like if Ben’s book is about like the struggle in business,
0:44:36 like David’s book is about the struggle in life.
0:44:39 And so anytime anybody feels mopey.
0:44:43 About what’s happening in their startup or in their life.
0:44:44 This is the book to read,
0:44:46 to kind of reset all the expectations.
0:44:46 All right, great.
0:44:48 Well, thank you so much.
0:44:51 It was a pleasure and an honor to be able to do this with you.
0:44:52 All right, Stuart, thank you so much.
0:44:53 Thank you, Stuart.

0:00:02 Hi and welcome to the inaugural edition
0:00:04 of the A16Z BioJournal Club.
0:00:05 I’m Hannah.
0:00:08 Our goal here is to take an interesting new research paper
0:00:10 in the field and talk about why it’s cool,
0:00:12 break down a little of the science involved,
0:00:14 and consider what the implications of this research
0:00:15 for industry might be.
0:00:17 So in our first take, A16Z General Partner
0:00:21 on the Bio Fund Jorge Conde and Deal Team Partner Andy Tran
0:00:24 chat with me about two papers recently published.
0:00:27 The first transposon encoded CRISPR/CAS systems
0:00:30 direct RNA guided DNA integration
0:00:33 was published by a group under Samuel H. Sternberg
0:00:36 at Columbia in nature of June 2019.
0:00:39 The second is RNA guided DNA insertion
0:00:42 with CRISPR associated transposases
0:00:45 with a team under Feng Zhang from the Broad Institute
0:00:48 published in science also in June 2019.
0:00:50 We talk about what these papers are all about
0:00:53 in the field of CRISPR development and beyond.
0:00:54 This mini podcast is available
0:00:57 as part of our new A16Z Bio Newsletter.
0:01:00 So if you like it and you want to hear more or read more,
0:01:05 please sign up for the newsletter at a16z.com/subscribe.
0:01:07 Let’s talk about what specifically is happening here.
0:01:10 What does, what a transposon encoded CRISPR/CAS system
0:01:13 direct RNA guided DNA integration?
0:01:14 Like, what does that actually mean?
0:01:16 Can you help me understand what was the interesting science
0:01:17 that was going on here?
0:01:18 – Yeah, so what’s really interesting
0:01:19 in the field of CRISPR lately,
0:01:22 and actually a few papers that came out in the recent times
0:01:26 developed a way to basically use these transposon machinery
0:01:28 inside the cell and use CRISPR to direct it
0:01:30 into a specific place in the genome
0:01:34 and really edit the genome without cutting it open at all.
0:01:36 So you can think of it as this scarless type
0:01:37 of genetic modification.
0:01:39 – Essentially what a transposon is,
0:01:42 is this sort of phenomenon that’s been observed
0:01:44 of where you have sort of these genes
0:01:46 that jump into the genome.
0:01:47 – It’s kind of mysterious what the jumping
0:01:48 is really used for.
0:01:50 People surmise that, you know, potentially, you know,
0:01:54 it helps, you know, the cells and organism evolved in general.
0:01:56 So what this paper shows is to utilize
0:01:58 this CRISPR machinery known as Cascade,
0:02:02 and it’s formed by, you know, CASS6, CASS7, CASS8 proteins
0:02:04 to really insert genes into the genome.
0:02:08 And this Cascade protein is directed to the chromosome
0:02:11 by using guided RNA, you know, CRISPR machinery,
0:02:15 and it then binds to this transposase-associated protein,
0:02:18 TNIQ, and it allows them to recruit
0:02:20 these transposable elements and then effectively integrate
0:02:22 the gene into the genome.
0:02:25 And this is super powerful because now we’re able to,
0:02:27 you know, add genes in the genome directly
0:02:29 and precisely without cutting it open.
0:02:32 It’s a scarless way to modify the genome.
0:02:35 – So in some ways, this is almost like if CRISPR
0:02:38 is about surgically inserting or editing DNA,
0:02:41 this is almost in many ways like plastic surgery, right?
0:02:42 It’s scarless. – Ah, right, right.
0:02:45 – It doesn’t leave a mark and therefore in some ways
0:02:48 it’s less risky from an intervention standpoint
0:02:49 on the genome side.
0:02:51 – What it really boils down to is that
0:02:53 this machinery utilizes a way
0:02:56 to really efficiently integrate genes into the cell,
0:02:58 into the genome specifically,
0:02:59 without having to cut it open at all.
0:03:02 And to the metaphor of the surgery,
0:03:04 this is really important in the whole context
0:03:06 of gene therapy as a whole, right?
0:03:08 ‘Cause when we started off in gene therapy,
0:03:09 we had these random integration
0:03:12 of these trans genes into the cell.
0:03:15 And, you know, this was a more stochastic process.
0:03:17 So think of it almost as, you know,
0:03:20 the arrival of this paper and papers like it
0:03:24 are showing us that we’ve gotten to a point
0:03:28 where we have a fully programmable ability
0:03:32 to integrate new genes or new DNA into a genome
0:03:34 without having to first open up
0:03:36 or break apart the DNA to do that.
0:03:38 – So let’s go back and actually situate it
0:03:40 into the development of the science,
0:03:42 what it represents for where we’ve gotten
0:03:44 from where we began.
0:03:45 – Yeah, so if you want to do the really,
0:03:47 the fast forward montage version of this.
0:03:49 – Yeah, the 80s montage.
0:03:51 – Yeah, exactly, so I’m sure you’ll put
0:03:52 in the pop music behind it.
0:03:55 – Right, yeah, and the overalls in the paintbrushes.
0:03:56 – He always started in the 60s.
0:03:57 – In the time motion.
0:03:58 – Yeah, let’s go.
0:04:00 – But look, I mean, if you go all the way back,
0:04:02 you know, one of the earliest sort of technologies
0:04:04 that came to the fore was the discovery
0:04:05 of something called the restriction enzyme.
0:04:07 And the restriction enzyme was this ability
0:04:10 to take this protein that could cut DNA
0:04:12 at these predetermined sites,
0:04:15 essentially open up the DNA,
0:04:19 and then you could introduce new genetic material,
0:04:20 and it would eventually integrate randomly,
0:04:23 but it would integrate into that DNA.
0:04:25 That’s what gave us the ability to do
0:04:27 or recombinant DNA technology,
0:04:29 which gives rise to the entire biotech field.
0:04:32 One of the earliest applications of biotechnology is
0:04:35 getting bacterial systems to integrate human insulin
0:04:38 so that we could coax bacteria to make that drug
0:04:41 or human insulin on our behalf, of course, to treat diabetics.
0:04:43 So that starts the whole field.
0:04:45 Now, you know, if you sort of move forward,
0:04:46 the objective always has been,
0:04:51 can you cure disease by repairing
0:04:55 or replacing what’s broken in DNA?
0:04:58 And so the whole field of gene therapy arises
0:04:59 with this idea that, you know,
0:05:02 can we introduce a corrected version
0:05:05 of a non-functional gene in a patient
0:05:06 and have it do the job
0:05:08 that the non-functional gene cannot do?
0:05:12 The first version was just put that gene
0:05:16 into a viral vector, into almost like a delivery vehicle,
0:05:18 introduce that into patient cells,
0:05:20 that gets taken in by the cells,
0:05:22 and the gene just starts to sort of do its job.
0:05:23 It integrates randomly in the genome,
0:05:24 but it does the job it needs to do.
0:05:26 Hopefully it integrates, or maybe it doesn’t,
0:05:28 but it just does the job to compensate
0:05:30 for some mutated gene.
0:05:33 Yeah, so it’s basically almost a parallel support system
0:05:35 that the cell has.
0:05:37 We just had the first gene therapies approved
0:05:39 over the course of the last couple of years.
0:05:42 If you look at the spark therapy, gene therapy drug for,
0:05:44 this rare inherited form of blindness.
0:05:47 And so that was one big advance forward.
0:05:49 Now, that is introducing a full gene
0:05:51 and just hoping it gets taken up by the cell.
0:05:52 Just kind of throw it in the mix.
0:05:54 Yeah, throw it in the mix and it has its own risk.
0:05:56 The original discovery of CRISPR-Cas9,
0:05:59 that system, the way that system works
0:06:02 is by making what’s called a double-stranded break
0:06:03 in the DNA molecule.
0:06:04 So if you think of, you know,
0:06:07 we all remember DNA from high school biology
0:06:10 as sort of the beautiful double helix.
0:06:12 So imagine sort of cutting that double helix in two
0:06:14 and making an end of it, right?
0:06:17 And then, you know, and then putting it back together.
0:06:18 That’s not riskless, right?
0:06:20 And by the way, it’s very well known
0:06:22 and documented that one of the big setbacks
0:06:25 that the gene therapy field had a couple of decades ago
0:06:27 was when they ran a clinical trial
0:06:28 at the University of Pennsylvania,
0:06:31 there was a patient named Jesse Gelsinger
0:06:32 who died after receiving the therapy
0:06:34 just because there’s a lot of risk associated
0:06:37 with introducing, you know, a viral capsid with a gene
0:06:39 into a cell system, into human being
0:06:41 where you could have a catastrophic result
0:06:41 and that patient died.
0:06:44 And that actually put a big pause on how we thought
0:06:47 about developing gene therapies for humans.
0:06:49 But that approach will have
0:06:51 and does have therapeutic potential.
0:06:52 And there are several companies
0:06:56 that are pursuing developing CRISPR-Cas9-based therapeutics.
0:06:58 As that advanced in parallel,
0:07:02 we started to see other gene or genome editing technologies
0:07:03 come to the fore.
0:07:06 The first one was one known as zinc finger nucleases,
0:07:08 which really didn’t have as much uptake
0:07:10 as one would expect
0:07:13 because it’s just very hard to deal with these proteins.
0:07:16 – How zinc fingers work is that you actually use these proteins
0:07:18 to bind onto DNA sequences.
0:07:21 And every time you want to iterate to find a new target,
0:07:22 cost tens of thousands of dollars
0:07:24 in a few months to develop one protein.
0:07:26 – So it’s enormously expensive and labor intensive.
0:07:27 – Exactly.
0:07:28 So it never really caught on like wildfire
0:07:30 in the community, right?
0:07:31 – It’s a very bespoke process.
0:07:34 It’s one of the type of thing.
0:07:36 And then there were other technologies.
0:07:38 There was Talens,
0:07:40 which was a bit better than zinc finger nucleases,
0:07:45 but really the big sort of shift in the field
0:07:47 was the arrival of CRISPR.
0:07:49 – And so when scientists discovered CRISPR,
0:07:52 they noticed that it was just always constant evolutionary
0:07:55 warfare between bacteriophages and bacteria
0:07:56 for billions of years.
0:07:58 – Bacteriophages are viruses.
0:07:58 – Yes.
0:08:03 – And so when these viruses inject their viral genome
0:08:04 into the bacteria,
0:08:06 you can think of it as the bacterial immune system,
0:08:10 whereas they’re able to sense these little snippets
0:08:11 of viral genome,
0:08:13 they would create a vaccination card from it,
0:08:14 and then they’ll put it back into this,
0:08:16 what is known as this CRISPR array.
0:08:18 And this would be like a vaccine card
0:08:20 or a vaccine database, if you will.
0:08:22 So every time it recognizes that foreign viral sequence.
0:08:23 – Right, then it knows what to do.
0:08:25 – It would know how to snip it away, right?
0:08:27 And then people have hijacked this machinery.
0:08:31 So what if we can program this outside of bacteria
0:08:32 and use it in human cells
0:08:35 and program it to sense not viral DNA,
0:08:38 but specific targets in the human genome.
0:08:42 And then then we can program and effectively target
0:08:44 anywhere we want in the cell, right?
0:08:46 – So now bring us forward to today.
0:08:48 So is this new development,
0:08:50 does it mean that the therapeutic potentials
0:08:52 are essentially less risky?
0:08:54 Or are there new possibilities
0:08:56 that we haven’t been able to do before?
0:08:56 Or both?
0:08:59 – Basically, you know, it’s really powerful
0:09:01 because when we talk about the first gen of gene therapy
0:09:03 to the second gen of CRISPR,
0:09:05 this paper really represents this third wave
0:09:07 in this scarless genomic editing.
0:09:10 This is basically genomic surgery at its finest, right?
0:09:13 You know, when we think about laparoscopic surgery
0:09:15 and all these advanced surgery tools,
0:09:16 we want to have, you know,
0:09:19 basically a scarless methodology of doing surgery.
0:09:21 This is a way to really cleanly integrate,
0:09:23 you know, genomic segments into the cell
0:09:25 without even touching, right?
0:09:28 And then this also represents an even broader tool
0:09:31 of, you know, the entire, you know, genomic toolkit landscape.
0:09:33 – The real promise on the near term side
0:09:37 is that we get to a point where you can make
0:09:40 scarless integrations of genetic material into the DNA.
0:09:42 So it’s less traumatic in that regard.
0:09:44 So if you can do that more precisely,
0:09:46 there’s less scar tissue that hopefully is, you know,
0:09:48 a better intervention altogether.
0:09:50 So that holds great promise from a potential
0:09:53 for future therapeutics based on this type of technology.
0:09:55 I think the other thing that’s worth noting
0:09:57 is that in a relatively short period of time,
0:10:01 the programmability of these kinds of systems
0:10:04 has improved dramatically.
0:10:06 So the kinds of things that we could do,
0:10:09 that we can do now with, you know, with CRISPR
0:10:11 based on these kinds of advances
0:10:15 gives us a very broad repertoire and toolkit to work with,
0:10:17 whether it’s for therapeutic applications
0:10:18 or for diagnostic applications
0:10:20 or for any other number of things.
0:10:25 When most people think about CRISPR developing CRISPR
0:10:26 for human health,
0:10:28 they’re thinking about therapeutic applications.
0:10:31 But the reality is there’s also a lot of potential
0:10:32 for diagnostic applications.
0:10:34 So going back to the original discovery
0:10:35 of the CRISPR-Cas system,
0:10:38 this was essentially the memory banker immune system
0:10:40 for bacteria to remember what viruses
0:10:41 had attacked it before
0:10:43 so they could protect themselves going forward.
0:10:47 And the way it does that is by essentially cutting
0:10:49 that viral genetic material.
0:10:52 So it’s ineffective essentially basically, you know,
0:10:56 cutting it off at its Achilles heel, so to speak.
0:10:57 And so as you can imagine,
0:11:00 if you’re hijacking that capability
0:11:01 for therapeutic purposes,
0:11:03 you could also hijack that capability
0:11:05 for diagnostic purposes.
0:11:07 And a way that that could potentially work, for example,
0:11:12 is if you know what bacteria or what virus
0:11:15 or even what, you know, mutation in DNA you’re looking for
0:11:18 in say a human sample, a blood sample
0:11:20 or urine sample or anything like that.
0:11:24 If it’s present, it will get cut by the right CAS system.
0:11:27 You program the CRISPR-Cas system to say,
0:11:31 these are the sequences of genetic material
0:11:33 that I am looking for in the sample.
0:11:34 This is basically the search engine.
0:11:36 Like you’re doing essentially a Google search.
0:11:38 – And you basically just look for the kill switch
0:11:40 to have been activated.
0:11:41 – You just look for the identifier.
0:11:42 So you basically say,
0:11:44 if I want to look in this patient’s sample,
0:11:47 let’s say I want to look for a specific bacteria
0:11:48 or a specific virus
0:11:51 or specific genetic mutation associated with disease,
0:11:53 you can say, if you find the presence
0:11:55 of any one of these sequences,
0:11:56 those are almost like the search terms,
0:11:58 cut them and when you cut them,
0:12:00 you can essentially engineer the system
0:12:02 to send out some sort of a reporter,
0:12:04 a reporter, usually it’s a visual marker.
0:12:06 And so basically if it lights up,
0:12:08 it’s because the CRISPR-Cas system cut the DNA
0:12:10 you told it to look for.
0:12:12 So that has a potential diagnostic application
0:12:13 and you could run that diagnostic
0:12:15 essentially without a lab, right?
0:12:16 ‘Cause it just happens with the biology.
0:12:19 So a whole other kind of new tool, essentially.
0:12:21 – Yeah, and I think the diagnostic applications
0:12:23 for these kinds of technologies
0:12:26 are actually pretty intriguing
0:12:28 because most of the way we do diagnostics
0:12:33 is based on developing some very specific biological
0:12:34 or chemical assays.
0:12:36 So look for something and if so,
0:12:38 have the reaction take place
0:12:40 and one that I can visualize and quantify
0:12:42 or quantitate in some way.
0:12:44 But here you’re just basically letting the biology
0:12:45 do the work for you.
0:12:47 In the CRISPR toolkit,
0:12:51 a lot of the initial applications was using
0:12:53 CRISPR-Cas9, this one nucleus.
0:12:56 And actually the diagnostic application
0:12:57 that Jorge was talking about
0:12:59 was actually using these other nucleas
0:13:01 known as Cas13 and Cas12.
0:13:04 And so all these different CRISPR proteins,
0:13:08 Cas9, 12, 13, X, Y, that we’re continuing to discover
0:13:11 has all different fundamental applications, right?
0:13:12 Even fundamentally changing
0:13:14 what these CRISPR nucleases even do.
0:13:15 It doesn’t even cut anymore.
0:13:17 It can do scarless editing.
0:13:20 And then we can even add different applications on it.
0:13:22 There’s new applications adding,
0:13:24 deaminase is just to do base editing.
0:13:28 So we can really do single base pair resolution editing.
0:13:30 There’s really the final frontier of precision
0:13:32 in terms of genomic modification.
0:13:34 And I think what’s really important is that
0:13:36 we’ve really seen this shift from
0:13:38 when it became this random bespoke science
0:13:40 and really turning into full-fledged,
0:13:42 modified engineering tool.
0:13:44 And this paper that we talk about here
0:13:47 is not only a great representation
0:13:49 of this engineering biology thesis,
0:13:53 but also a pretty huge potential step change
0:13:54 for the field as a whole.
0:13:56 – You can program this to turn genes up and down
0:13:58 as opposed to just editing them.
0:14:01 There’s even work that’s ongoing to use this technology
0:14:03 to image DNA directly,
0:14:04 which is a pretty remarkable thing
0:14:07 because since this is acting locally on DNA,
0:14:10 you can add all kinds of agents to make it imageable.
0:14:12 So therefore, you can observe chromatin
0:14:14 or genome structure directly.
0:14:16 So there’s a lot that can be done with this technology,
0:14:20 and you hear the old adage about the pickaxes for the gold.
0:14:22 And with these toolkits,
0:14:25 I mean, we are quite literally panning for gold here.
0:14:28 I mean, where these things get found,
0:14:31 they’re found in soil and in ocean vents.
0:14:32 – New York City subways.
0:14:34 – The New York City subway.
0:14:37 So these people are quite literally looking in nature
0:14:39 because nature has ingenious ways
0:14:41 to do a lot of the things that we’re trying to do
0:14:45 from an engineering biology or programmable biology standpoint.
0:14:50 And so I think it’s a remarkable moment to take pause
0:14:53 and see how far this technology has come
0:14:55 in a relatively short period of time.
0:14:58 This generation’s recombinant DNA or restriction enzymes
0:15:01 that really gave rise to the biotechnology industry.
0:15:03 I think this tool kit, this CRISPR tool kit,
0:15:06 as we’re describing it and discussing it,
0:15:07 represents sort of the next frontier
0:15:10 for what will happen in biology.
0:15:13 So an incredible development of precision
0:15:15 in what the tools can do and at the same time,
0:15:17 a huge expansion of what that will enable us
0:15:18 to do going forward.
0:15:20 Thank you guys so much for joining us.
0:15:23 Should I say on the A16Z Journal Club?
0:15:23 – No, no, please don’t say that.
0:15:24 – Yeah, okay.
0:15:34 [BLANK_AUDIO]

0:00:05 The content here is for informational purposes only, should not be taken as legal business
0:00:10 tax or investment advice or be used to evaluate any investment or security and is not directed
0:00:15 at any investors or potential investors in any A16Z fund. For more details, please see
0:00:22 a16z.com/disclosures. Hi, and welcome to the A16Z podcast. I’m
0:00:27 Hannah, and this episode is all about synthetic fraud, a new evolution of consumer fraud that’s
0:00:32 emerging in financial services to the tune of $1-2 billion a year.
0:00:36 In this episode, Naftali Harris, co-founder and CEO of Centelink, which builds technology
0:00:41 to detect and stop synthetic fraud, talks to me and A16Z operating partner for information
0:00:46 security Joel de la Garza, all about what this new kind of fraud is, including the life
0:00:52 cycle of this long con, how these synthetic identities get made, incubated, and finally
0:00:57 busted out, and some of the wildest stories behind the strange fraud rings he’s seen.
0:01:02 We also touch on why this new fraud is on the rise, who the true victims are, and at
0:01:06 the end of the day, what the foundational security issue at the heart of it all truly
0:01:11 is. We’re here to talk about synthetic fraud, which I have to confess I didn’t even know
0:01:15 what that really meant when we first started talking about it. What does synthetic fraud
0:01:18 even mean? Almost no one hears about it in the public
0:01:22 outside of financial services industry. I hope that neither of you have been the victim
0:01:23 of identity theft. I have.
0:01:29 Okay, I’m sorry to hear that. And if you haven’t, you probably know someone that has been.
0:01:32 And so the general public is very aware of identity theft because there’s that consumer
0:01:38 victim. With identity theft, you’re stealing a real person’s identity. With synthetic fraud,
0:01:41 you’re saying, forget the real person, I’m going to make up a totally fake one.
0:01:44 And that means like fake from the very ground up.
0:01:49 Yeah, from the ground up. So a fraudster will use a synthetic identity, so a made up, named
0:01:54 date of birth, an SSN combination in order to open up an account with a bank or get a
0:01:58 loan from a bank. The key thing here is that there’s no one record, there’s no one one
0:02:02 actual person that it all belongs to. And then what they’ll be able to do is actually
0:02:07 acquire quite a bit of credit, take out a lot of loans, usually a few tens of thousand
0:02:13 dollars from every major bank and lender, and then use that to get a lot of money and
0:02:14 not repay any bit.
0:02:19 How prevalent is this kind of fraud in the industry? I mean, how much is this happening
0:02:21 versus like we all hear about identity theft all the time?
0:02:25 So this is actually one of the super interesting things. So we’ve added up the losses across
0:02:29 the industry and within lending, it’s somewhere from one to two billion dollars a year of losses
0:02:30 annually.
0:02:34 Wow. And how aware of it are the banks? At what point do they catch on?
0:02:38 That’s also one of the really interesting things because there is no consumer victim.
0:02:42 The banks have a really hard time figuring out which of their losses are attributable
0:02:46 to synthetic fraud as opposed to somebody that had a hardship or lost their job.
0:02:48 Oh, right. Same pattern of behavior.
0:02:52 Exactly. With identity theft, what happens is somebody opens up an account, they get
0:02:56 a new credit card, they steal a lot of money from the bank, and the way the bank finds
0:03:01 out about it is eventually the victim contacts them and says, “Hey, I didn’t take out this
0:03:04 credit card. This wasn’t me.” And they’ll sign an affidavit and then the bank will realize
0:03:09 this was an actual victim of identity theft as opposed to someone that just had a hardship
0:03:13 and took out more money than they should have. And with synthetic fraud, all the bank sees
0:03:19 is a large set of people that haven’t been making payments for the loans, and they have
0:03:23 a really hard time of figuring out which of these are people that have had some toward
0:03:26 of local economic challenges.
0:03:29 Yeah. Legitimate need for the loan, basically.
0:03:32 Exactly. And which of them are people that were actually defrauding them?
0:03:37 Synthetic fraud is a relatively new-ish phenomenon. So I think it’s something that’s kind of
0:03:41 grown up. As banks have gotten better at spotting identity theft and credit freezes and those
0:03:46 sorts of things, it seems that that correlated to the rise in synthetic fraud. Identity theft
0:03:52 used to be ridiculously simple. If you think back 10 to 15 years ago, as bank fraud teams
0:03:55 got better, they got better tools to catch this kind of thing you had credit freezes
0:03:59 come to effect, it seems like the fraudsters pivoted in this direction.
0:04:02 Yeah, that’s exactly right. I mean, another big one actually is the rise of the EMB chip.
0:04:04 Oh, that is a factor in this?
0:04:09 Absolutely. Fraudsters are committing fraud as a business. And what they do is they gravitate
0:04:14 towards channels, so to speak, that are profitable for them. And it used to be you can make a
0:04:17 lot of money doing card skimming. The EMB chip made that a lot harder. So you saw a lot
0:04:23 of fraud move online to card not present fraud. So people stealing credit cards online. There’s
0:04:27 been a lot of great technology that’s arisen there recently, which has made that harder
0:04:32 to do. Still certainly happens as we all know. And then a lot of progress towards identity
0:04:36 theft and that’s gotten harder. And so they’re moving on to synthetic fraud, which is very
0:04:40 challenging for banks and lenders to detect and quite lucrative for the fraudster.
0:04:44 But can we just go back to like that moment of opening the account? Why is it so hard
0:04:50 to verify like an actual birthday against an actual name against an actual SSN? Like
0:04:56 if those things are not matching, why is that initial moment not the place to catch it?
0:05:02 So what most people don’t realize is that financial institutions, so banks and lenders
0:05:08 do not have a list of all name, date of birth, and SSN combinations in the United States.
0:05:11 A lot of people think that the credit bureaus have this list, you know, experiencing Equifax
0:05:17 and TransUnion, and they don’t have it either. Essentially, the banks and lenders believe,
0:05:22 certainly until recently, had believed that the three credit bureaus had lists of all
0:05:25 name, date of birth, and SSN combinations. So they, everybody’s thought somebody else
0:05:26 was doing it.
0:05:27 Exactly.
0:05:28 That’s absurd.
0:05:31 It’s quite funny, actually. And this is the way that fraudsters actually create these
0:05:37 synthetic identities. If you apply for credit with a name, date of birth, and SSN repeatedly,
0:05:41 the credit bureaus will believe that it’s a real person, and they’ll create a record
0:05:43 for this totally fake person.
0:05:48 Because they’re only tracking the applications, they’re not backing it up to reality.
0:05:50 Yeah, and they have no way of doing so.
0:05:56 I feel like we’re giving tips to everybody in the world. I don’t like how to create this.
0:05:57 Do not do this.
0:05:58 Exactly.
0:06:03 But that is such a gaping hole in the information flow, a weird blind spot that everybody else
0:06:05 just kind of assumes that…
0:06:09 Yeah, it’s pretty interesting. I mean, so banks and lenders believe that the bureaus have
0:06:13 records on everybody, and mostly general public believes that as well. The logic on the bureau
0:06:18 side is essentially banks and lenders have strong know-your-customer procedures, they’re
0:06:21 doing a great job of risk. And so consequently they say, “Oh, you know, everyone’s talking
0:06:26 about John Smith, that must be a real person.” But actually, nobody really knows here. And
0:06:28 so everyone’s pointing fingers at everybody else.
0:06:34 It seems like, actually, it was this gaping hole for quite a while, right? So why… Was
0:06:37 there always some level of this, and then it just spiked?
0:06:44 I think the interesting point is sort of the actual genesis of this whole situation, which
0:06:48 is that there is no source of truth for proofing identity. And that really lies at the center
0:06:53 of kind of a lot of these issues. There’s sort of a coordination and a collaboration
0:07:00 that has to happen in between entities that, while wanting to minimize fraud, these entities
0:07:04 are also competing with one another in a number of different product categories. And so there
0:07:09 isn’t always necessarily a line financial incentive for them to collaborate.
0:07:12 It’s always been possible, but the thing that’s really challenging about synthetic fraud is
0:07:15 it is such a long con. It’s challenging.
0:07:16 What do you mean by that?
0:07:22 It’s not sufficient to just make a fake identity. You can do that, and it’s pretty easy. But
0:07:28 when you do that, all you have is a person who exists on one of the bureaus, or all three
0:07:33 of them, but doesn’t actually have real credit to their name. No bank is going to give them
0:07:35 $100,000 or even $10,000.
0:07:38 Right. So it’s like me when I first got out of college or whatever.
0:07:42 Exactly. It’s like when you first entered the credit space. And so there are some fraudsters
0:07:46 that will just try to churn through $300 cards, but there’s not a ton of money in that. The
0:07:52 real money that the fraudsters are pursuing is getting access to all the prime credit
0:07:59 cards, to big auto loans, to huge unsecured personal loans. And that requires building
0:08:03 up their credit over a period of one to two years. Get some low limit credit cards, start
0:08:08 making a little bit of payment, build their credit. They do it quite aggressively because
0:08:12 they’re optimizing to when can they get to that 700 plus credit score or better, but
0:08:13 it does take a long time.
0:08:18 And I think that’s the answer as to why we hadn’t seen it in the past because in the
0:08:23 old days, you could go steal someone’s identity, open a line of credit, have access to that
0:08:27 credit within a week, maybe even a couple of days, depending on how you did the disbursement
0:08:33 of funds. But then sort of as people got better about reporting those things as consumers
0:08:37 actually started to notice when lines of credit were open for them, or they had credit
0:08:42 monitoring capabilities, the response time was a lot quicker. So you couldn’t necessarily
0:08:46 get those funds out in the amount of time. And so this is kind of the new process that
0:08:51 they’ve moved on to. And to the earlier point, like this does take some amount of time and
0:08:55 preparation. So creating lots of identities, going through the process of establishing
0:08:59 credit for them over a period of one to two years, and then getting to a cash out that
0:09:03 in the old days, you could have done in five days to maybe a month.
0:09:06 So a lot more work for that same size hit.
0:09:10 So the hit actually can be even bigger than for identity theft. So with identity theft,
0:09:15 you’re racing against the clock, because the victim will actually notice this at some point
0:09:20 and they will, they will say, this wasn’t me. And so they go back to the, to the bank,
0:09:24 they go to the lender and they say, stop doing this. And they’ll put a freeze on their cutter
0:09:29 report and so forth. But with synthetic fraud, there’s, there’s no race for the clock. There’s
0:09:35 no one who’s watching for this. There’s no one, there’s no one that, that is going to
0:09:37 notice this until they stop making payments.
0:09:40 Yeah. Are you seeing the synthetic fraudsters actually make payments?
0:09:41 Oh, absolutely.
0:09:42 Absolutely.
0:09:43 Wow.
0:09:44 So they’re, they’re taking out loans. They’re making the payments except for the initial
0:09:50 fraud of the identity. They’re not, the behavior is not caught, is not at that point doing
0:09:51 anything wrong.
0:09:56 So there are three phases in the lifetime of a synthetic identity. The first part is
0:10:02 the creation phase. So this is where a synthetic identity starts applying for credit a couple
0:10:06 of times. Oftentimes, we’ll actually start with any lender that does a poll from all
0:10:12 three credit bureaus. So most lenders only pull from one of the three bureaus. So TransUnion
0:10:17 Experian and Recofacts. But when you first create a synthetic identity, you want to get
0:10:22 that synthetic ID to have credit records on all three of the major bureaus. So one of
0:10:27 the things that we see synthetic identities doing is initially the first place that they’ll
0:10:32 apply for credit is anywhere that does a tri-bureau poll that pulls from all three of the major
0:10:33 bureaus.
0:10:34 They want immediately to disperse that information.
0:10:35 Exactly.
0:10:36 Okay.
0:10:42 So in this creation phase of the synthetic identity’s life, they will apply for credit
0:10:46 at places that do tri-bureau polls. They’ll sign the synthetic identities up for an email
0:10:49 address and for a phone number.
0:10:53 So it’s really, it’s becoming like a real identity almost in a lot of dimensions.
0:10:54 Yeah.
0:10:58 They’ll sign them up for social media accounts. So get them a Facebook or even better as a
0:11:03 LinkedIn or a Twitter. The reason being that later on, a fraud investigator is going to
0:11:08 be looking for this person and this gives them a little bit more legitimacy.
0:11:12 That is so much, that’s so much attention paid at that early phase.
0:11:13 Absolutely.
0:11:17 So one of the things that we’ve, we noticed with a lot of the fraud rings, the traditional
0:11:23 fraud rings was a tremendous amount of technical sophistication. So highly automated, really
0:11:28 well, a really deep understanding of not just the fraud controls, but the entire technical
0:11:29 stack.
0:11:34 With this kind of fraud, it seems very manual. It seems very kind of almost like an artisanal
0:11:35 form of fraud.
0:11:37 Yeah. It’s like a bespoke, like you literally create these lives.
0:11:38 Absolutely.
0:11:47 The whole cloth. So, okay. So that’s phase one and then the birth of the fake person.
0:11:48 Yeah.
0:11:49 Exactly.
0:11:53 So then in phase two, that’s the buildup phase. This is where it takes one to two years. And
0:11:57 in this phase, the synthetic identity is acquiring credit as quickly as they can. So often this
0:12:03 means getting small credit cards, introductory credit cards, and actually making oftentimes
0:12:09 the minimum payments, but anything that shows this person has a good repayment history.
0:12:13 Now when eventually down the road, this is discovered and people are presumably going
0:12:18 back to figure out, can you start tracing those payments when you look back and start
0:12:23 understanding where that money comes from and have like understanding into the fraud from
0:12:24 that route?
0:12:30 Well, those payments often come from bank accounts in the names of the synthetic identities.
0:12:34 Isn’t there a point when you open the bank account where you need more than those three
0:12:35 pieces of information?
0:12:39 You’re supposed to collect four. It’s technically name, date of birth, SSN, and address.
0:12:40 Okay.
0:12:45 It’s called the customer identification program. And you’re supposed to verify these things
0:12:49 in a number of different ways, but because there’s simply no way of doing it, a lot of
0:12:54 times people say, “Oh, they have a credit record that’s sort of sufficient.”
0:13:00 Most of the account opening anti-fraud stuff people do is focused on identity theft, which
0:13:03 has traditionally been the big account opening from a fraud.
0:13:09 But for account opening, if you want to prevent identity theft, what you’re doing is trying
0:13:14 to see whether the person submitting the application is the same as the identity that they’re using
0:13:15 to apply for credit.
0:13:21 So as an example, if you see John Smith apply for credit using naftaliharris@gmail.com as
0:13:22 their email address.
0:13:23 Problem.
0:13:28 Yeah, problem. Exactly. It’s probably not John Smith doing it. It’s probably naftaliharris.
0:13:35 But if you see John Smith applying for credit with johnsmith@gmail.com, then it looks fine.
0:13:40 But what if it’s actually naftaliharris that made John Smith and made John Smith at gmail.com?
0:13:44 Let’s go back to the life cycle. So we talked about the birth, then we talked about the
0:13:45 development.
0:13:46 Incubation.
0:13:52 The incubation, where is the moment where they die, where you generally get killed?
0:13:56 So that’s every fraudster’s favorite part of the life cycle. It’s the bust out. Once
0:14:03 you have a synthetic identity that has been making payments, which has gotten access to
0:14:07 higher credit lines. So at the end of that incubation period, the synthetic ID has a
0:14:11 credit score over 700 or 750 plus, or even less than 800.
0:14:12 Pretty good.
0:14:13 Yeah.
0:14:14 Yeah, they look great.
0:14:15 Yeah.
0:14:19 And they make every bank and lender, especially in today’s low rate environment, wants to
0:14:23 throw as much money at them as they can.
0:14:27 And so in the bust out phase, fraudsters acquire as much credit as they possibly can. They max
0:14:32 out any credit card they’ve had, and all of a sudden they just stop making payments. They
0:14:35 go from your model customer to your worst one.
0:14:39 They stop paying their loans, and then what happens next?
0:14:43 So someone stopped making payments, and so the bank starts pushing them through their
0:14:44 collections process.
0:14:46 So somebody starts calling.
0:14:52 Yeah, it’s usually, it’s a polite email, “Hey, John Smith. Notice you missed your payment.
0:14:54 Could you please do that as soon as you can?”
0:14:58 And then that becomes a little bit more stringent, and then it starts paying phone calls. In
0:15:02 some cases, the fraudsters will ignore it completely and vanish from the face of the
0:15:08 earth. And in that case, it’s uncollectable. In other cases, they’ll pick up the phone
0:15:14 and they’ll say, “Oh, I’m really sorry. I couldn’t make payments. I lost my job. I had
0:15:19 a hardship. Someone in my family got ill. I can’t make payments right now.”
0:15:20 And they buy some time.
0:15:24 And they buy some time, and eventually the loan gets charged off.
0:15:29 Why does this not, at that moment, trigger when you suddenly, your behavior suddenly
0:15:35 changes and you take a big loan? There are all sorts of legitimate reasons for that kind
0:15:41 of sudden big loan. But why is that not automatically getting flagged just for a little check at
0:15:42 that point?
0:15:46 To the earlier point, right? There’s a very big interest to grow your creditor base, to
0:15:51 grow the base of people you’re loaning money to. And in that process, friction is generally
0:15:57 found upon, right? It’s a risk determination. Some of these organizations, they’ve built
0:16:00 risk models that feel comfortable enough about the validity of this identity, and they make
0:16:06 kind of the business decision to take a risk on extending credit to them. And it’s probably
0:16:09 one of those things where they need to make some adjustments to that risk model.
0:16:14 So that’s, I would say that there’s probably some perfectly rational process-driven reason
0:16:19 why this is happening. Fraud, like most of these kinds of criminal enterprises, are very
0:16:24 much games of cat and mouse. And this is just sort of the mouse finding a way around the
0:16:25 cat in this instance.
0:16:29 So where in the life cycle do you guys try and intervene? Like, how do you look at this
0:16:33 life cycle? And where do you think is the weak point? And with what kind of tools?
0:16:37 The places where they really are experts are on the U.S. credit system. They understand
0:16:44 that very deeply, honestly better than probably a lot of people who have that as their careers.
0:16:50 They know who does a tribunal credit poll. They know how to get through the KYC processes
0:16:56 at different organizations. They know who is weak at the beginning. And so, at a high
0:17:01 level, the way we actually solve this problem is we have a team of risk analysts that manually
0:17:06 review transactions, looking for fraud, investigating cases, deeply trying to understand individual
0:17:11 fraud transactions, and understanding what is new in the fraud world.
0:17:16 And then on the other hand, we have a sister team of technologists, so engineers, machine
0:17:20 learning engineers, data scientists, who are taking the insights and the labels from the
0:17:25 risk operations team and using those to build production-alized machine learning models
0:17:28 that actually can detect this sort of fraud in real time.
0:17:32 It almost sounds like a detective agency on one side, and then building the tech on top
0:17:33 of the knowledge.
0:17:38 So, I mean, a lot of the tech is based on the fact that we understand synthetic fraud
0:17:42 extremely well. Different kinds of products naturally fall in one or different parts,
0:17:48 so like a high limit rewards credit card from a top 10 card issuer. Those will tend to get
0:17:53 hit towards the end of that process a little bit before the bust out. And so, in that case,
0:17:58 we have more history through which to actually identify an application as synthetic.
0:17:59 Right.
0:18:06 But we also work with card issuers that are trying to give cards to immigrants or to young
0:18:11 people even as early as in college. And there, we’re really playing at sort of the very beginning
0:18:14 during phase one or the very beginning of phase two to differentiate between those real
0:18:16 people and those fake people.
0:18:21 A big thing that we do is around clustering, connecting together applications that come
0:18:26 from the same fraud ring. So, for this form of synthetic fraud, most of it comes from
0:18:33 organized crime rings, and $100,000 per identity is great, but if you want to make a business
0:18:40 out of it, the fraudsters are a lot more ambitious. And so, they make a number of these different
0:18:43 synthetic identities and incubate all of them at the same time.
0:18:45 Oh my gosh, it sounds like the Matrix that way.
0:18:50 Yeah, it’s a lot of fake people. We’ve seen them be so ambitious as to actually make
0:18:51 families.
0:18:53 So, they’ll have like a mother…
0:18:56 But only a families of lendable ages.
0:19:02 Exactly. So, they’ll be like a mother and father. So, they’ll have the same last name
0:19:07 with birthdays that are a couple years apart. And they’ll have like five kids, all of them
0:19:12 are in their early 20s or something like that, address history that’s shared at different
0:19:15 points and they tried to make the ages staggered and stuff like that.
0:19:20 It’s like scripting a story. So, you’ve seen that more than once?
0:19:24 We’ve seen a number of such families, quote-unquote, created. Internally, we call it the Keeping
0:19:29 Up with the Joneses approach because the first time we saw this, the last name was
0:19:30 Jones.
0:19:32 You know, a family that commits fraud together, stays together.
0:19:36 We need like a symbol, put on, but shh.
0:19:40 Thank you. I’ll be here all week. I was going to suggest we call this a fraud cast.
0:19:41 Yeah.
0:19:42 There you go.
0:19:49 Another good one. So, what are some of the other types of fraud rings that you guys see?
0:19:54 We oftentimes see alleged people that have no relationship with each other who are sharing
0:20:00 address history at some point. And it’s really interesting what causes that. So, one reason
0:20:05 this happens is that a fraudster will oftentimes reuse the same address or for that matter
0:20:08 the same phone number or email address if they’re lazy.
0:20:13 But during the incubation period, one of the ways in which fraudsters boost up someone’s
0:20:19 credit quite a bit is by purchasing authorized user trade lines. That’s when you give a credit
0:20:25 card to your spouse or one of your kids. So, like when you’re younger, sometimes your parents
0:20:29 will give you a credit card. The credit card, it actually is in the name of your parents
0:20:32 and they’re the ones that are actually responsible for making the payments.
0:20:36 But what a lot of people don’t realize is that that credit card will oftentimes show
0:20:41 up on the recipient’s credit report. So, if you’re a kid and your parent gives you a credit
0:20:45 card, which they’re responsible for, it’ll end up on your credit report. And that’s sort
0:20:51 of what all the major card issuers had historically thought was the point of having an authorized
0:20:57 user card. It’s to usually within a family or at most friends or maybe employees or something
0:21:04 like that. But actually, you’ll find hundreds of these, hundreds of these marketplaces that
0:21:08 let you purchase or sell a high-limit credit card that you have.
0:21:10 And that’s legitimate?
0:21:13 It’s not, but it is as far as I know, legal.
0:21:18 Whoa. So, you sell your ability to borrow to somebody else? I mean, it sounds like such
0:21:19 a bad idea.
0:21:25 The recipient won’t actually get the card. The card will show up on their credit report,
0:21:31 but the card actually won’t get sent to them. And the purpose of it actually is essentially
0:21:36 credit score arbitrage. If you have a high-limit $20,000 credit card that you’ve had since
0:21:41 2005, it looks really good when it shows up on somebody else’s credit report and they’re
0:21:42 willing to pay for it.
0:21:49 So fraudsters who are very prolific about buying and selling these authorized user cards will
0:21:53 oftentimes have shared addresses. And the reason the addresses are shared is that multiple
0:21:57 of these synthetic identities at one point or another bought the same authorized user
0:22:03 credit card. Our technology can detect this and realize that these people, 50 of them
0:22:07 throughout the United States, who should have no relationship to each other nonetheless
0:22:08 have shared history.
0:22:12 What’s the weirdest thing you’ve seen besides the Joneses?
0:22:19 So we saw one case where the fraudster actually had taken two totally different people and
0:22:23 mashed their identities together. And one of the identities that was mashed together was
0:22:29 someone that was actually in prison for murder. So that person, if they ever get out, might
0:22:30 be pretty upset about this.
0:22:36 So it’s like half identity fraud, half synthetic, like a kind of weird Hollywood mashup. Like
0:22:40 you take two movies and splice them together with lazy storytelling, basically.
0:22:46 One that I thought was just really amusing. And we saw a fraud ring that had so many identities
0:22:53 in it that the way they kept track of which identity had which SSN is actually included
0:23:00 the last four of the SSN in the email addresses of the synthetic identities. So lots of people
0:23:07 have, you know, Naftali Harris and then monthday@gmail.com or a lot of people have, you know, Naftali
0:23:14 Harris, year of birth@gmail.com. These fraudsters actually use Naftali Harris last four of SSN
0:23:15 at gmail.com.
0:23:16 Wow.
0:23:19 And they did this for all several hundred of their identities.
0:23:21 So that was an immediate first signal.
0:23:26 Yeah. And essentially the identities all looked very cookie cutter to us as though somebody
0:23:31 was following directions for how to create a synthetic identity. They had something that
0:23:38 worked. They all used the same original institution as their first inquiry. They all were structured
0:23:44 the same way. They all had first name, last name, last four of SSN@gmail.com was the one
0:23:49 that they used. Everything about them was sort of similar even though none of the information
0:23:51 was overlapping in that case.
0:23:57 So, you know, when we looked at this, people use the SSN4 in their email address. Almost
0:24:01 everyone who did that was fraudulent, but there were some that were not. And some people
0:24:05 just didn’t realize that you’re not supposed to put the last four of your SSN in your email
0:24:10 address. I think most of us realize that, but, you know, some people don’t.
0:24:13 Yeah. That’s another tip for our listeners. If you’re doing that change of email right
0:24:14 now.
0:24:18 So you look for patterns. You look for clustering. Are there other hallmarks that you look for
0:24:21 that you guys are paying attention to?
0:24:26 It’s a lot around the consistency of the history. Synthetic identities have histories
0:24:31 that are not really cohesive. So we’ll do things like look at state-by-state migration
0:24:37 patterns. So it’s pretty common for people to move from Florida to Georgia. It’s a lot
0:24:42 less common for people to move from Florida to Alaska. Obviously it does happen. And apologies
0:24:47 to whoever’s listening and did just that. But statistically, it’s there are certain
0:24:52 patterns that are more or less likely. So we’ll look at when SSNs were issued and then
0:24:56 when and where those were issued and see if they match up with someone’s actual credit
0:25:01 history. We’ll look at where they’ve been moving, how fast that’s happening. It’s pretty
0:25:07 rare for someone to be in a, have a residential home in a new state every, you know, one or
0:25:12 two months. It’s just not very frequent. So we’ll look for a lot of things around cohesiveness
0:25:13 of the identity.
0:25:14 And weird outliers.
0:25:20 I think there’s a really interesting salient point here that’s being made, which is that
0:25:27 kind of the first two generations of large-scale consumer fraud were mostly about technical
0:25:31 weaknesses, underlying technology weaknesses, lack of two-factor authentication, inability
0:25:37 to secure endpoints, right? It was very kind of software-driven or computer breach-driven.
0:25:42 This is actually a business process hack or a hack of sort of existing broken business
0:25:46 process. Yeah. You know, essentially it’s social engineering and scale.
0:25:50 So in some ways, it sounds terrible to say, but it kind of feels a little bit like a
0:25:55 victimless crime because you’re not stealing money from another person. You’re stealing
0:25:56 it from this like institution, right?
0:25:57 The funny thing about that is…
0:25:58 I know that’s not true.
0:26:04 Having worked at institutions that had lots of things attempted to be stolen from them.
0:26:07 Yeah. Like can you talk about how that impacts the whole?
0:26:12 Yeah, absolutely, right? Losses of these nature, of this nature go directly against
0:26:17 the bottom line of the corporation, right? Losses like these translate directly into
0:26:21 the financial performance of the stock, and these are the kinds of things that shareholders
0:26:26 and board members and anyone with the fiduciary responsibility that they want to tackle as
0:26:30 quickly as possible because reducing losses in these kinds of categories can translate
0:26:35 into meaningful movement of stock, especially if you’re talking about a billion to $2 billion.
0:26:41 That’s not trivial. So usually the way that these start to materialize is that this will
0:26:46 translate into higher costs associated with borrowing for legitimate customers. So these
0:26:50 expenses, they’re not going to get eaten by the corporation. They’re going to get probably
0:26:55 pushed out in the forms of new fees or higher interest rates to people opening new accounts.
0:27:00 It’s going to translate probably into more internal controls, more expense on the back
0:27:05 end to start validating some of these transactions to do more verification, and we’re going to
0:27:10 pay for it. And it’s going to be maybe a tenth of a percent, maybe a fifth of a percent,
0:27:13 but it’s going to start to drive up costs of borrowing for consumers, and that’s usually
0:27:14 where it turns out.
0:27:19 There’s actually two other sorts of ways in which certain groups are victims. So one
0:27:24 of them is that synthetic identities look like people that are new to credit, and those
0:27:29 populations, the legitimate populations there are often young people and immigrants.
0:27:33 Oh, so it’s making it harder for all the people who need credit the most.
0:27:38 Exactly. So it makes banks a lot less comfortable lending to immigrants and a lot less comfortable
0:27:42 lending to young people, or even just people that decided they didn’t need credit for
0:27:46 a long time. A lot of money will say, “I have no reason I should get a credit card and get
0:27:51 trapped in debt,” until they decide they might want a mortgage, and it makes it harder
0:27:54 for those kinds of people to acquire credit because they look like they might be not a
0:27:55 real person.
0:28:00 We did a podcast before about sort of different areas of cybercrime and different geographic
0:28:03 concentrations of different kinds of fraud. Is there a geographic concentration or is
0:28:07 there a type of fraudster that tends to gravitate towards this kind of fraud?
0:28:13 Yeah, a lot of this form of fraud is geographically concentrated. So we see a lot from Southern
0:28:20 California, a lot nowadays from the Atlanta region, a lot from South Florida.
0:28:24 And is that just because people get good at it and then the organization gets bigger?
0:28:26 Or they’re telling their friends, it’s like Amway or something.
0:28:27 Yeah.
0:28:28 Yeah.
0:28:29 A lot of it is organized crime.
0:28:33 Typically, the way we’ve seen these illicit criminal industries develop is that they start
0:28:38 off as sort of what you could think of as sort of familial clusters, right? Small groups
0:28:43 of individuals that figure out a neat trick, share it among a couple friends, perhaps locally,
0:28:46 which is why you’re seeing geographic concentration.
0:28:51 And then that information gets distributed more broadly, and other more professionalized
0:28:54 career type criminals start to move in.
0:28:59 And in industry develops, you’ll get sort of a one-stop shop, right? A group of individuals
0:29:03 that do soup to nuts, this kind of fraud. Specific tasks now will start to get broken
0:29:08 up. So you’ll be able to probably go buy these identities in the dark web. There’s probable
0:29:12 places that are actually farming them, developing them, and then selling them to other parts
0:29:16 of the organization. And then you’ll get specific groups that are focusing on kind of the bust
0:29:18 outrings and those sorts of things.
0:29:20 The industrialization of synthetic fraud.
0:29:24 I would suspect that we’re either in that phase or we’re moving towards it. We’re seeing
0:29:28 sort of that hockey stick growth of a new industry, right? And it’s just kind of the criminal
0:29:33 variant of it. And so as that starts to ramp up, it’s going to be interesting. So I am
0:29:38 not aware of any large scale arrests of people involved in this kind of activity. I’m interested
0:29:43 if you know of like any of the regulators have said anything about synthetic fraud or interested
0:29:44 in looking at it.
0:29:48 You know, that’s one of the really interesting things. Synthetic fraud right now is a huge
0:29:54 money laundering issue, but a totally underappreciated one. If you look at the regulations around
0:30:02 KYC, so specifically the laws that require this, they really contemplated identity thefts
0:30:08 and did not contemplate synthetic fraud almost at all. Everyone’s assumption for a really
0:30:14 long time has been that identities that are used to apply for credit are real. And as we’ve
0:30:17 discovered over the last couple of years, that’s really not the case.
0:30:21 So the banks are starting to understand it and noticing it and getting new tools to try
0:30:26 and notice it. When the banks catch this and they stop it, do they then alert the authorities?
0:30:28 Do people try and pursue this at all?
0:30:34 No, the first instinct of banks is to try to have it not happen again. And they’re not
0:30:42 quite as focused on having law enforcement step in and apprehend the people doing it.
0:30:46 What would be the tipping point for that to have to happen if it becomes this big industrial
0:30:49 loss? So it’s dollars, right? Arrests typically
0:30:55 happen towards the end of the life cycle of something like this. And so as it gets professionalized,
0:31:00 as you see kind of the industrialization of this sort of activity, regulators will start
0:31:05 to notice, law enforcement will start to notice. They may already be active investigations,
0:31:10 we don’t know, but they’ll start to kind of move against these sorts of organizations
0:31:16 as large scale criminal organizations that are engaged in things that may be drugs, may
0:31:21 be terrorism, could be things that are life-threatening. They’re always looking for new conduits for
0:31:25 money laundering. So sometimes what happens is that money or some of that activity will
0:31:31 find its way into some of these channels as a way to clean and rinse some of these funds.
0:31:33 And that’ll also draw the attention of law enforcement.
0:31:35 And then you really have to pay attention to where interesting.
0:31:39 Absolutely. If you don’t necessarily see criminals from other forms of crime moving into this
0:31:44 sort of crime. So you won’t see racketeers or you won’t see narcotics traffickers like
0:31:46 quitting their day jobs and deciding to do synthetic fraud.
0:31:47 It’s the specialist.
0:31:52 Exactly. But they will sort of give money to people to run it through these systems
0:31:57 to clean it for a fee, right? And that’s usually where you start to see the real professionalization.
0:32:00 That’s where it starts spreading through the criminal system.
0:32:04 And then you start to see the cases come and you’ll see arrests made. And that’s usually
0:32:05 how these things start to get rolled up.
0:32:09 Are there sort of fundamentally new human behaviors that you’re noticing? Or is it the
0:32:15 same fundamental criminal behavior, but just manifesting itself in new and different ways?
0:32:20 I think that’s actually a really interesting point here about all of this. And I think
0:32:26 most of the fraud discussions and just broadly a lot of security issues we have in general,
0:32:29 it all comes back to that kind of earlier discussion about like the social security
0:32:33 number that, you know, if you look at your social security card, it says this is not
0:32:37 to be used for identification, right? Like this is this is this number should mean nothing
0:32:41 to you. I mean, it’s almost like money Python, right? Like we’ve built all these things on
0:32:46 something that said, don’t make me the Messiah. And we kind of did that. And then as a country,
0:32:50 we’ve sort of refused to meaningfully consider any kind of national level identity or identity
0:32:55 management. And so you have the proliferation of a lot of these issues. And that’s that’s
0:32:59 sort of the really fascinating thing about almost all the fraud discussions.
0:33:05 So if there is this huge kind of foundational crack in all these systems that we’ve built
0:33:09 up, but that make it feels like a house of cards almost with this missing kind of giant
0:33:14 verification piece at the bottom, how do you get at the heart of that problem?
0:33:18 So I think one thing that Joel mentioned earlier was the sort of cat and mouse nature of a
0:33:22 lot of fraud. We want to go a step beyond that. There are many organizations out there
0:33:28 or even beyond financial services that are verifying identities as part of their business.
0:33:35 So every major bank in lender does this, but so do online marketplaces like Lyft or Airbnb.
0:33:41 So do also retailers. So one that you’re constantly having to do this. Yeah, you probably did
0:33:45 this a couple of times even today. Yeah. And one thing that we’ve observed is that these
0:33:52 organizations with respect to customer identification don’t really work together, despite the fact
0:33:54 that it’s fundamentally the same problem they’re solving, like figure out if someone
0:33:59 is who they say they are and if they actually exist. All these organizations are fighting
0:34:04 the same fraudsters and they’re verifying the same 300 million Americans. So the way
0:34:10 this really should work is the government should step in and make a sort of national
0:34:17 ID, I think to really solve this. One that does have printed on it. You should use this.
0:34:20 There’s web standards for how to do this and cryptography is advanced quite a bit and there
0:34:25 are ways of doing this. I don’t think we’re going to see the U.S. government step in and
0:34:29 do this. And so we’re building it. Thank you so much for joining us on the A16Z podcast.

0:00:05 The content here is for informational purposes only, should not be taken as legal business
0:00:10 tax or investment advice, or be used to evaluate any investment or security and is not directed
0:00:14 at any investors or potential investors in any A16Z fund.
0:00:17 For more details, please see a16z.com/disclosures.
0:00:21 Hi, and welcome to the A16Z podcast.
0:00:22 I’m Frank Chen.
0:00:28 Today, I’m here with Carnegie Mellon’s Professor Tom Mitchell, who has been involved with machine
0:00:30 learning basically his entire career.
0:00:34 So I’m super excited to have this conversation with Tom, where he can tell us a little bit
0:00:37 about the history and where all of our techniques came from.
0:00:40 And we’ll spend time talking about the future, where the field is going.
0:00:45 So Carnegie Mellon’s been involved in sort of standing up the fundamental teaching institutions
0:00:50 and research institutions of the big areas, computer science, artificial intelligence
0:00:52 and machine learning.
0:00:56 So take us back to the early days, the you and Newell and Jeff Hinton are teaching this
0:00:57 class.
0:00:58 What was your curriculum like?
0:00:59 Like what were you teaching?
0:01:02 Pretty different, I imagine, than what we teach undergrads today.
0:01:03 That’s right.
0:01:05 Well, at the time, this was the 1980s.
0:01:13 So artificial intelligence at that point was dominated by what we would call symbolic methods.
0:01:19 Where things like formal logic would be used to do inference.
0:01:24 And much of machine learning was really about learning symbolic structures, symbolic representations
0:01:25 of knowledge.
0:01:31 But there was this kind of young whipper snapper, Jeff Hinton, who had a different idea.
0:01:41 And so he was working on a book with Rommelhart McClellan that became a very well-known parallel
0:01:47 data processing book that kind of launched the field of neural nets.
0:01:49 And they were, if I remember, psychologists, right?
0:01:58 Yeah. Jay McClellan is a psychologist here at CMU, Rommelhart, kind of a neuroscientist.
0:02:01 And more, he was a very broad person.
0:02:07 And Jeff, so the three of them were kind of the rebels who were taking things off in a
0:02:09 different paradigm.
0:02:14 The empire wanted us to do research on knowledge representation and inference and first-order
0:02:15 logic.
0:02:20 I remember as an undergrad, I took this computer aided class that John H.M.M.D. wrote called
0:02:24 Tarski’s World, where we learned all about first-world logic.
0:02:25 What could you prove?
0:02:26 What could you not prove?
0:02:30 And so that’s what the establishment, quote, unquote, was teaching.
0:02:37 And then Jeff was the rebel off in neural network land, and he gets his reprise later.
0:02:40 So take us back to the world of knowledge representation, because I’m actually seeing
0:02:44 a lot of startups these days who are trying to bring back some of these techniques to
0:02:49 complement deep learning, because there are well-known challenges with deep learning.
0:02:52 Like, we’re not encoding any priors, we’re learning everything for the first time, we
0:02:56 need tons of labeled data sets to make progress.
0:03:00 And so take us back to the days of knowledge representation.
0:03:04 What were we trying to solve with those set of techniques, and how might we use them today?
0:03:13 So back in the ’80s and the ’90s, and I have to say that some of the really senior people
0:03:20 in the field were totally devoted to this paradigm of logical inference, logical representations.
0:03:27 People like John McCarthy, for example, were very strong proponents of this, and really
0:03:34 essentially just saw that reasoning is theorem-proving, and therefore if we’re going to get computers
0:03:38 to do it, that’s what we have to do.
0:03:42 There were some problems with that, and there still are.
0:03:48 One thing that I remember from back then that was an example was the banana in the tailpipe
0:03:50 problem.
0:03:55 These logical systems were used to reason to do things like, how would you plan a sequence
0:03:57 of actions to achieve a goal?
0:03:59 Like, how would you get from here to the airport?
0:04:05 Well, you’d walk to your car, you’d put the key in, turn the car on, you’d drive out of
0:04:10 the parking lot, get on the interstate, go to the airport exit, et cetera.
0:04:12 But what if there’s a banana in the tailpipe?
0:04:17 Even back then, before it became a meme in Beverly Hills Cop, we were worried about
0:04:19 the banana in the tailpipe.
0:04:20 That’s right.
0:04:25 And the point of the banana in the tailpipe is there are an infinite number of other things
0:04:31 that you don’t say when you spin out a plan like that.
0:04:39 And any proof, if it’s a proof, really, is going to have to cover all those conditions,
0:04:43 and that’s kind of an infinitely intractable problem.
0:04:48 You couldn’t encode enough to do the inference you needed for your plans to be successful.
0:04:49 Right.
0:04:56 And so one of the big changes between the ’80s and 2019 is that we no longer really think
0:05:03 in the field of AI that inference is proving things, instead it’s building a plausible
0:05:05 chain of argument.
0:05:10 And it might be wrong, and if it goes wrong, if there is a banana in the tailpipe, you’ll
0:05:12 deal with it when it happens and when you figure it out.
0:05:13 Right.
0:05:18 So we move from certainty and proof to sort of probabilistic reasoning, right, Bayesian
0:05:20 techniques started becoming popular.
0:05:21 Right.
0:05:25 And so around 19– in the late ’90s, in fact.
0:05:31 So if you look at the history of machine learning, there’s an interesting trajectory where in
0:05:38 maybe up to the mid ’80s, things were pretty much focused on symbolic representations.
0:05:43 Actually, if you go back to the ’60s, there were the perceptron, but then it got swallowed
0:05:50 up by the end of the ’60s by symbolic representations and trying to reason that way and trying to
0:05:52 learn those kind of symbolic structures.
0:05:59 Then when the neural net wave came in around the late ’80s, early ’90s, that started competing
0:06:02 with the idea of symbolic representations.
0:06:09 But then in the late ’90s, the statisticians moved in and probabilistic methods became
0:06:11 very popular.
0:06:16 And at the time, there was this– if you look at this history, you can’t help but realize
0:06:22 what a social phenomenon technology advances in sciences and technology.
0:06:24 People influencing each other at conferences.
0:06:25 Right.
0:06:29 They were shaming them into adopting a new paradigm.
0:06:36 And so one of the slogans, or one of the phrases you kept hearing when people started working
0:06:41 on probabilistic, statistical probabilistic methods, they would never call them that.
0:06:48 They would have called them, instead, principled probabilistic methods, just to kind of shine
0:06:53 the light on the distinction between neural nets, which are just somehow tuning a gazillion
0:06:57 parameters in the principled methods that were being used.
0:07:03 And so that became really the dominant paradigm in the late ’90s and kind of remained in charge
0:07:12 of the field up through till about 2009, 2010, when now, as everybody kind of knows, deep
0:07:18 networks made a very serious revolution showing that they could do all kinds of amazing things
0:07:20 that hadn’t been done before.
0:07:21 Yeah.
0:07:26 We really are living in a golden age here in deep learning in neural network land.
0:07:29 But let’s go back to the original sort of rebel group.
0:07:35 This is Jeff Hinton hanging out in the shadow of sort of first order logic and saying, “No,
0:07:36 this is going to work.”
0:07:40 I think they were loosely inspired by the architecture of the brain.
0:07:41 Is that–
0:07:42 Definitely.
0:07:43 Definitely.
0:07:49 The kinds of arguments Jerry Feldman was one of the people who gave some of these arguments.
0:07:55 She said, “Look, you recognize your mother in about 100 milliseconds.
0:08:03 Your neurons can’t switch state in faster than a few milliseconds.
0:08:10 And so it looks like at most the chain of inference that you’re doing to go from your
0:08:16 retina to recognize your mother can only be about 10 deep, just from the timing.”
0:08:17 Oh, fascinating.
0:08:21 So it was an argument of sort of how long it took to recognize your mother.
0:08:22 Right.
0:08:23 And then how slow your neurons are, right?
0:08:26 Because they’re basically, these are biochemical processes, right?
0:08:27 Right.
0:08:28 Fascinating.
0:08:31 So really a computational efficiency argument.
0:08:37 And therefore, Jerry would say, “There must be a lot of stuff happening in parallel.
0:08:41 It must be a very wide chain of inference if it’s only 10 layers deep.”
0:08:45 And then he says, “Look at the brain, look at visual cortex.”
0:08:46 Yeah.
0:08:47 You got it.
0:08:51 And so neuroscientists at this time were making progress in understanding the structure of
0:08:55 neurons and how they connected to each other and how they formed connections, and those
0:08:58 connections could change strength over time, right?
0:09:02 All mediated by chemical interactions in the computer science community was inspired by
0:09:03 this.
0:09:04 Definitely.
0:09:12 And the level of abstraction at which the computational neural nets met up with the
0:09:19 real biological neural nets was not a very detailed level, but where they kind of became
0:09:28 the same was this idea of distributed representations, that in fact it might be a collection of hundreds
0:09:35 or thousands or millions of neurons that simultaneously were firing that represent your mother instead
0:09:37 of a symbol.
0:09:38 Right.
0:09:39 Right.
0:09:46 It’s a completely different notion of what it even means to represent knowledge.
0:09:51 And really, one of the most exciting things that has come out of the last decade of research
0:10:00 in neural and deep networks is a better understanding, although we still don’t fully understand of
0:10:07 how these artificial neural networks can learn very, very useful representations.
0:10:14 And for me, a simple example of that, that in a sentence summarizes it, is we have neural
0:10:22 networks now that can take as input an image of photograph and output a text caption for
0:10:23 that photograph.
0:10:28 What kind of representation must be in the middle of that neural network in order to
0:10:35 actually capture the meaning well enough that you can go from a visual stimulus to the equivalent
0:10:43 textual content, that it’s really, it must be capturing a very basic core representation
0:10:45 of the meaning of that photograph.
0:10:46 Yeah.
0:10:50 And one of my favorite things about the brain, which is otherwise this very sort of slow
0:10:54 computer, right, if you just look at neuron speeds, is that not only can they do this,
0:10:59 but they can actually use this, the representation they’re deriving to actually inform our actions
0:11:01 and our plans and our goals, right?
0:11:05 So not only is it like this picture has a chair in it, but like I can sit in that chair.
0:11:06 I can simulate sitting in that chair.
0:11:10 I think like that chair is going to support my weight, and all of these things happen
0:11:14 in like milliseconds, despite the fact that the basic components of the brain are very
0:11:15 slow.
0:11:16 Yeah.
0:11:17 It’s an amazing thing.
0:11:21 In fact, now that you mentioned it, I have to tell you, a half of my research life these
0:11:29 days is in studying how the human brain represents meaning of language.
0:11:32 We use brain imaging methods to do this.
0:11:39 And in one set of studies, we put people in a fMRI scanner, and we showed them just common
0:11:49 nouns like automobile, airplane, a knife, a chair, and so forth.
0:11:56 And we would get a picture literally with about three millimeter resolution of the three-dimensional
0:12:02 neural activity in their brain as they think about these different words.
0:12:06 And we’re interested in the question of all kinds of fundamental questions like what do
0:12:08 these representations look like?
0:12:12 Are they the same in your brain and my brain?
0:12:17 Given that they don’t appear instantaneously, by the way, it takes you about 400 milliseconds
0:12:22 to understand a word, if I put it on the screen in front of you.
0:12:24 What happens during that 400 milliseconds?
0:12:30 How do these representations evolve and come to be?
0:12:36 And one of the most interesting things we found, we studied this question by training
0:12:46 a machine learning system to take as input an arbitrary noun and to predict the brain
0:12:49 image that we will see if a person reads that noun.
0:12:56 Now, we only had data for 60 nouns at that time, so we didn’t train it on every noun
0:12:57 in the world.
0:13:00 We trained it on 60.
0:13:05 In fact, what we did was we trained it only on 58, so we could hold out two nouns that
0:13:06 we hadn’t seen.
0:13:12 And then we would test how well it could extrapolate to new nouns that you had never seen by showing
0:13:18 it the two held out nouns and having it predict the images.
0:13:21 Then we’d show it two images and we’d say, “Well, which of those is strawberry and which
0:13:23 of those is airplane?”
0:13:26 And it was right 80% of the time.
0:13:29 So, you could actually predict essentially brain state, right?
0:13:31 I’m going to show you a strawberry.
0:13:35 Let me predict the configuration of your neurons and who’s lighting up and who’s not.
0:13:42 So then we had a model that we trained with machine learning that captured something about
0:13:44 representations in the brain.
0:13:50 We used that to discover that the representations are almost identical in your brain and mind.
0:13:56 We could train on one set of people and decode what other people were thinking about.
0:14:02 And we also found that the representations themselves are grounded in parts of the brain
0:14:05 that are associated with perception.
0:14:10 So, if I give you a word like “peach,” the parts of your brain that code the meaning
0:14:16 of that are the ones associated with the sense of taste and manipulation because sometimes
0:14:19 you pick up a peach in visual color.
0:14:21 Yeah, that is fascinating.
0:14:25 Well, it’s so exciting to think that the brain structures are identical across people
0:14:30 because what everybody wants is sort of that, remember that scene in the matrix where you
0:14:34 sort of like you’re jacked straight into your brain and you’re like, “Oh, now I know Kung-Fu.”
0:14:35 Right?
0:14:36 Like this is what we want, right?
0:14:42 We want to learn new skills and sort of new facts and new inferences just like loading
0:14:43 an SD card, right?
0:14:48 And so, the fact that we’re sort of converging to the same structures in the brain at least
0:14:50 makes that theoretically possible.
0:14:51 We’re a ways away from that.
0:14:52 We’re a ways away from that.
0:14:53 But I’m with you.
0:14:54 Yeah, awesome.
0:15:01 So, another area that interests you is finding biases and why don’t we start by distinguishing
0:15:06 sort of two types of biases because when you hear the word “bias” today in machine learning,
0:15:10 you’re mostly thinking about things like, “Gee, let me make sure my data set is representative
0:15:13 so I don’t draw the wrong conclusion from that,” right?
0:15:18 So the classic example being here that I don’t do good recognition on people with darker skin
0:15:23 because I didn’t have enough of those samples in my data set and so the bias here is you’ve
0:15:29 selected a very small subset of the target data set that you want to cover and make predictions
0:15:31 on and therefore your predictions are poor, right?
0:15:33 So, that’s one sense of bias.
0:15:37 But there’s another sense of bias, that statistical bias, which is kind of what you want out of
0:15:38 algorithms.
0:15:39 So, maybe talk about this notion.
0:15:40 Yeah, sure.
0:15:45 And this is really a very important issue right now because now that machine learning
0:15:52 is being used in practice in many different ways, the issue of bias really is very important
0:15:54 to deal with.
0:16:00 You gave an example, another example would be, for instance, you have some historical
0:16:06 loan applications in which ones were approved, but maybe there’s some bias that say people
0:16:12 of one gender receive fewer loan approvals just because of their gender and if that’s
0:16:18 inherent in the data and you train a machine learning system that’s successful, well, it’s
0:16:22 probably going to learn the patterns that are in that data.
0:16:29 So, the notion of what I’ll call social bias, socially unacceptable bias, is really this
0:16:36 idea that you want the data set to reflect the kind of decision making that you want
0:16:39 the program to make if you’re going to train the program.
0:16:44 And that’s kind of the common sense notion of bias that most people talk about.
0:16:50 But there’s a lot of confusion in the field right now because bias is also used in statistical
0:16:55 machine learning to really with a very different meaning.
0:17:02 We’ll say that an algorithm is unbiased if the patterns that it learns, the decision
0:17:10 rules that it learns for approving loans, for example, reflect correctly the patterns that
0:17:12 are in the data.
0:17:18 Also that notion of statistically unbiased just means the algorithm’s doing its job of
0:17:21 recapitulating the decisions that are in the data.
0:17:28 The notion of the data itself being biased is really an orthogonal notion.
0:17:32 And there’s some interesting research going on now.
0:17:39 So for example, typically when we train a machine learning system, say to do loan approval,
0:17:45 a typical thing would be you can think of these machine learning algorithms as optimization
0:17:46 algorithms.
0:17:54 They’re going to tune maybe the parameters of your deep network so that they maximize
0:18:00 the number of decisions that they make that agree with the training examples.
0:18:06 But if your training examples have this kind of bias that maybe females receive fewer loan
0:18:09 approvals than males.
0:18:13 There’s some new work where people say, well, let’s change that objective that we’re trying
0:18:22 to optimize in addition to fitting the decisions that are in the training data as well as possible.
0:18:29 Let’s put another constraint that the probability of a female being approved for a loan has
0:18:33 to be equal to the probability of a male being approved.
0:18:38 And then subject to that constraint, we’ll try to match as many decisions as possible.
0:18:44 So there’s a lot of work right now in really technical work trying to understand if there
0:18:51 are ways of thinking more creatively, more imaginatively about how to even frame the
0:18:58 machine learning problem so that we can take what might be biased datasets but impose constraints
0:19:01 on the decision rules that we want to learn from those.
0:19:03 Yeah, that’s super interesting.
0:19:08 We’re sort of envisioning the world we want rather than the data of the world that we
0:19:13 came from because we might not be happy with the representation of the representedness
0:19:16 I guess, of the data that we came from.
0:19:20 And it’s causing people to look a lot more carefully at even the very notion of what
0:19:25 it means to be biased and what it means to be fair.
0:19:29 Are there good measures for fairness that the community is driving towards or do we
0:19:34 not really have a sort of an objective measure of fairness?
0:19:35 We don’t.
0:19:41 We don’t have an objective measure and there’s a lot of activity right now to discussing
0:19:48 that, including people like our philosophy professor, David Danks, who is very much part
0:19:54 of this discussion and social scientists, technology people, all getting together.
0:20:01 And in fact, there are now a couple of conferences centered around how to introduce fairness and
0:20:07 explainability and trust in AI systems.
0:20:11 This is a very important issue, but it’s not only technical.
0:20:16 It’s partly getting our philosophical, social, trying to get our heads around what it is
0:20:17 that we really want.
0:20:21 That’s a beautiful thing about AI and about computers in general.
0:20:27 It forces you to be way more precise when you are getting a computer to do it about
0:20:29 what you want.
0:20:36 And so even if you just think about self-driving cars, we have, when I was 16, I took a test
0:20:39 and I was approved to be a human driver.
0:20:43 They never asked me questions about whether I would swerve to hit the old lady or swerve
0:20:44 to hit the baby carriage.
0:20:45 Right.
0:20:47 Charlie’s problem was not on the DMV test.
0:20:48 Exactly.
0:20:50 But it’s on the test for the computers.
0:20:51 Right.
0:20:53 Yeah, it’s really interesting that we sort of hold computers to a different standard
0:20:55 because we’re programming them, right?
0:20:59 We can be explicit and we can have them sort of hit goals or not, right?
0:21:04 And those are design decisions rather than sort of, you know, bundled into a brain, right,
0:21:05 of a person.
0:21:06 Yeah.
0:21:12 And so I think of, you know, look, banks historically have hired loan officers.
0:21:16 Those loan officers may or may not be fair, right, according to the definitions that we’re
0:21:21 sort of talking about now, but we kind of hold those humans, those human loan officers to
0:21:24 a different standard than we would hold the algorithms.
0:21:25 That’s true.
0:21:27 I mean, who knows which way it will go in the future.
0:21:33 If we continue to have human loan officers and some computer loan officers, will we
0:21:39 up the constraints on the humans so that they pass the same qualifications?
0:21:45 Or will we drop the constraints on the computers so that they’re no more titrated than the
0:21:46 people?
0:21:47 Yeah.
0:21:48 That’s fascinating, right?
0:21:49 Who is master?
0:21:50 Who is the student, right?
0:21:55 The intuitive thing is let’s make the humans the models that we train our systems to approach,
0:21:56 right?
0:21:58 Like human competence being the goal.
0:22:00 The other way to think about it is no, right?
0:22:04 We can actually introduce constraints like, you know, equal number of men and women or
0:22:07 equal number of this ethnicity versus another ethnicity.
0:22:12 And our algorithms as a result of those constraints could be more fair than humans.
0:22:14 And so we invert it, right?
0:22:16 Let’s get humans up to that level of impartiality.
0:22:17 Right.
0:22:21 Like maybe the algorithm can end up teaching the human how to make those decisions in a
0:22:26 different way so that the fairness outcome you want is really achieved.
0:22:27 Yeah.
0:22:28 That’s fascinating.
0:22:32 And it’s great that sort of not just computer scientists are involved in this conversation
0:22:36 about the ethicists and the social scientists who are weighing in.
0:22:40 So that gives me hope that, you know, sort of smart people across disciplines are really
0:22:41 grappling with this.
0:22:43 So we get the outcomes that we want.
0:22:49 Well, sort of a related topic to this, right, sort of social impact of AI.
0:22:56 You recently co-wrote a paper with MIT’s Eric Bernholzen about the workplace implications.
0:23:00 And I think you also testified on Capitol Hill about what AI is going to do to jobs.
0:23:04 So why don’t you talk a little bit about what you guys found in the paper?
0:23:10 Well, this actually started with Eric and I co-chairing a National Academy study on
0:23:18 automation in the workforce, which was a two-year affair with a committee of about 15 experts
0:23:24 from around the country who were economists, social scientists, labor experts, technologists.
0:23:30 And in that study, I think we learned so much.
0:23:35 It turns out when you really dig into the question of what’s going to be the impact
0:23:39 of AI and automation on jobs.
0:23:46 You can’t escape noticing that there are many different forces that automation and technology
0:23:48 is exerting on the workforce.
0:23:50 One of them, of course, is automation.
0:23:55 Toll booth operators are going away, do not sign up to be a toll booth operator.
0:24:02 But in other kinds of jobs, instead of the job going away, there will be a shift, a redistribution
0:24:05 of the tasks.
0:24:07 So take, for example, a doctor.
0:24:12 A doctor has multiple tasks, for instance, they have to diagnose the patient, they have
0:24:17 to generate some possible therapies, they have to have a heart-to-heart discussion with
0:24:25 the patient about which of those therapies the patient elects to follow, and they have
0:24:26 to bill the patient.
0:24:31 Now computers are getting, computers are pretty good at billing, but they’re getting better
0:24:35 at diagnosis and they’re getting better at suggesting therapies.
0:24:41 For example, in just in the last couple of years, we’ve seen computers that are at the
0:24:49 same level, if not a little better than doctors at things like diagnosing skin cancer and
0:24:50 other kinds of diseases.
0:24:55 The radiologists, the tissue biopsies, all of these things, we’re using these computer
0:24:59 vision techniques to get very good performance.
0:25:02 So what does this mean about the future of doctors?
0:25:07 Well, I think what it means is automation happens at the level of the individual tasks,
0:25:09 not at the job level.
0:25:16 If a job is a bundle of tasks, like diagnosis, therapy, heart-to-heart chat, what’s going
0:25:24 to happen is computers will provide future doctors with more assistance, in some degrees,
0:25:30 hopefully automating billing, but some amount of automation or advice giving.
0:25:36 But for other tasks, like having that heart-to-heart chat, we’re very, very far from when computers
0:25:40 are going to be able to do anything close to that.
0:25:44 Good bedside manner is not going to be a future of your RoboDoc anytime soon.
0:25:53 And so what you find, if you look into this, and Eric and I recently had a paper in Science
0:25:59 with a more detailed study of this, but what you find is that the majority of jobs are
0:26:04 not like toll-booth operators, where there’s just one task, and if that gets automated,
0:26:05 that’s the end of the job.
0:26:15 The majority of jobs, like podcast interviewer, or computer, or professor, or doctor, really
0:26:16 are a bundle of tasks.
0:26:23 And so what’s going to happen is that, according to our study, the majority, more than half
0:26:29 of jobs, are going to be influenced, impacted by automation, but the impact won’t be a
0:26:30 elimination.
0:26:34 It’ll be a redistribution of the time that you spend on different tasks.
0:26:43 And we even conjecture that successful businesses in the future will, to some degree, be redefining
0:26:49 what the collection of jobs is that they’re hiring for.
0:26:53 Because they still have to cover the tasks through some combination of automation and
0:27:01 manual work, but the current bundles of tasks that form jobs today might shift dramatically.
0:27:06 So the key insight is to think of a job as a bundle of tasks, and that bundle might change
0:27:13 over time as AI enters and says, well, look, this specific task I’m very good at in algorithm
0:27:16 land, and so let’s get humans to focus on other things.
0:27:19 We just need to think of them as differently bundled.
0:27:24 Well, the last topic I wanted to talk with you about, Tom, is around whether this is
0:27:27 the best time ever for AI research.
0:27:29 So we started the grand campaign.
0:27:32 Some would argue summer of 1956 with the Dartmouth conference.
0:27:35 And we’ve had several winters and summers.
0:27:36 Where are we now?
0:27:39 And then what are you most excited about looking into the future?
0:27:45 I think we’re absolutely at the best time ever for the field of artificial intelligence.
0:27:48 And there have been, as you say, ups and downs over the year.
0:27:55 And for example, in the late ’80s, AI was very hot, and there was great expectation
0:27:57 of the things it would be able to do.
0:28:02 There was also great fear, by the way, of what Japan was going to do.
0:28:03 Yeah.
0:28:08 This is the fifth generation supercomputer and the entire national policy of Japan, right?
0:28:10 Focusing on this area.
0:28:16 And so in the U.S., there was great concern that this would have a big impact, Japan would
0:28:18 take over the economy.
0:28:21 So there are some parallels here.
0:28:24 Now again, AI, it’s very popular.
0:28:27 People have great expectations.
0:28:31 And there’s a great amount of fear, I have to say, about what China and other countries
0:28:34 might be doing in AI.
0:28:44 But one really, really important difference is that, unlike in the 1980s, right now, there’s
0:28:48 a huge record of accomplishment over the last 10 years.
0:28:57 We already have AI and machine learning being used across many, many different, really economically
0:29:00 valuable tasks.
0:29:06 And therefore, I think, really, there’s very little chance that we’ll have a crash, although
0:29:11 I completely agree with my friends who say, “But isn’t AI overhyped?”
0:29:13 Absolutely, it’s overhyped.
0:29:20 But there is enough reality there to keep the field progressing and to keep commercial
0:29:25 interest and to keep economic investment going for a long time to come.
0:29:27 So you would argue, this time it really is different.
0:29:28 It really is different.
0:29:31 Because we have real working stuff to point to.
0:29:36 And over the next 10 years, we’ll have a whole lot more real working stuff that influences
0:29:38 our lives daily.
0:29:44 So as a university researcher, I look at this and I say, “Where is this going and what should
0:29:47 we be doing in the university?”
0:29:51 If you want to think about that, you have to realize just how much progress there was
0:29:52 in the last 10 years.
0:29:59 When the iPhone came out, I guess that’s 11 years ago, computers were deaf and blind.
0:30:03 When the iPhone came out, you could not talk to your iPhone.
0:30:09 This is such a weird idea, but you could not talk to your iPhone because speech recognition
0:30:11 didn’t work.
0:30:19 And now computers can transcribe voice to text just as well as people.
0:30:26 Similarly, when you pointed your camera at a scene, it couldn’t recognize with any accuracy
0:30:29 the things that were on the table in the scene.
0:30:34 And now it can do that with about the same accuracy comparable to humans.
0:30:38 And in some visual tasks, like skin cancer detection, even better than you know.
0:30:40 Even better than trained doctors, yeah.
0:30:41 Better than trained.
0:30:44 So it’s hard to remember that it’s only been 10 years.
0:30:47 And that’s the thing about progress in AI.
0:30:53 You forget because it becomes so familiar, just how dramatic the improvement has been.
0:30:55 Now think about what that means.
0:31:00 That means we’re really in the first five years of having computers that are not deaf
0:31:01 and blind.
0:31:08 And now think about what are the kinds of intelligence that you could exhibit if you
0:31:13 were deaf and blind, well, you could do game playing and inventory control.
0:31:17 You could do things that don’t involve perception.
0:31:24 But once you can perceive the world and converse in the world, there’s an explosion of new
0:31:25 applications you can do.
0:31:30 So we’re going to have garage door openers that open for you because they recognize that
0:31:32 your car coming down the driveway.
0:31:37 We’re going to have many, many things that we haven’t even thought about that just leverage
0:31:43 off this very recent progress in perceptual AI.
0:31:49 So going forward, I think a lot about how I want to invest my own research time.
0:31:51 I’m interested still in machine learning.
0:31:53 I’m very proud of the field of machine learning.
0:31:55 It’s come a long way.
0:32:01 But I’m also somebody who thinks we’re only at the beginning.
0:32:05 I think if you want to know the future of machine learning, all you need to do is look
0:32:10 at how humans learn and computers don’t yet.
0:32:16 So we learn, for example, we do learn statistically like computers do.
0:32:24 My phone watches me over time and statistically, it eventually learns where it thinks my house
0:32:27 is and where it thinks my work is.
0:32:30 It statistically learns what my preferences are.
0:32:33 But I also have a human assistant.
0:32:39 And if she tried to figure out what I wanted her to do by statistically watching me do
0:32:43 things a thousand times, I would have fired her so long ago.
0:32:46 A lot of false positives and false negatives, right?
0:32:47 Right.
0:32:48 So she doesn’t learn that way.
0:32:51 She learns by having a conversation with me.
0:32:56 I go into the office and I say, “Hey, this semester I’m team teaching a course with
0:33:00 Katarina on deep reinforcement learning.
0:33:02 Here’s what I want you to do.
0:33:04 Whenever this happens, you do this.
0:33:10 Whenever we’re preparing to hand out a homework assignment, if it hasn’t been pretested by
0:33:16 the teaching assistants two days before handout, you send a note saying, “Get that thing pretested.”
0:33:22 So what I do is I teach her and we have a conversation, she clarifies.
0:33:32 So one of the new paradigms for machine learning that I predict we will see in the coming decade
0:33:35 is what I’ll call conversational learning.
0:33:42 Use the kind of conversational interfaces that we have, say, with our phones to allow
0:33:52 people to literally teach their devices what they want them to do instead of have the device
0:33:54 statistically learn it.
0:34:01 And if you go down that road, here’s a really interesting angle on it.
0:34:07 It becomes kind of like replacing computer programming with natural language instruction.
0:34:15 So I’ll give you an example of a prototype system that we’ve been working on together
0:34:19 with Brad Myers, one of our faculty in HCI.
0:34:25 It allows you to say to your phone something like, “Whenever it snows at night, I want
0:34:27 you to wake me up 30 minutes earlier.”
0:34:33 If you live in Pittsburgh, this is a useful app, and none of the California engineers
0:34:34 have created that app.
0:34:40 And today, I could create that app if I took the trouble of learning the computer language
0:34:43 of the phone, I could program it.
0:34:49 But only far less than 1% of phone users can actually have taken the time to learn the
0:34:51 language of the computer.
0:34:56 We’re giving the phone the chance to learn the language of the person.
0:35:03 So with our phone prototype, if you say, “Whenever it snows at night, wake me up 30 minutes
0:35:08 earlier,” it says, “I don’t understand, do you want to teach me?”
0:35:12 And you can say, “Yes, here’s how you find out if it’s snowing at night.”
0:35:17 You open up this weather app right here, and where it says current conditions, if that
0:35:20 says SNOW, it’s snowing.
0:35:25 Here’s how you wake me up 30 minutes earlier, you open up that alarm app, and this number
0:35:27 you subtract 30 from it.
0:35:36 So with a combination of showing, demonstrating, and telling voice, we’re trying to give users
0:35:44 the opportunity to create their own apps, their own programs with the same kind of instruction,
0:35:49 voice, and demonstration that you would use if you were trying to teach me how to do it.
0:35:50 I love that.
0:35:53 It’s sort of a natural language for an end to, we have an investment in a company called
0:35:57 ift, if this, then that, which is you can program those things, but you have to be a
0:35:59 little sophisticated.
0:36:03 You’d like to just be able to talk to your phone and have it figured out how do I fill
0:36:05 the slots into ift that ift wants.
0:36:07 Exactly, and if this, then that is a wonderful thing.
0:36:14 It has a huge library of these apps that you can download, but as you say, you still have
0:36:17 to learn the language of the computer to create those.
0:36:21 We’re trying to have the computer learn the language of the person.
0:36:27 If that line of research plays out, and I believe it will this decade, we’ll be in a very different
0:36:34 world because we’ll be in a world where instead of the elite few, less than 1% of phone users
0:36:39 being able to program, it’ll be 99% of phone users who can do this.
0:36:45 Now, think about what that does for the whole conception of how we think about human-computer
0:36:46 interaction.
0:36:47 Yeah.
0:36:49 That’s a profound shift in society, right?
0:36:55 Just like everybody became literate, not just the priests, and look what happened to society.
0:36:56 Exactly.
0:36:57 Yeah.
0:36:59 Think about what it means for the future of jobs.
0:37:04 Right now, if you have a computer introduced as your teammate, you, the human, and the
0:37:11 computer are a team, well, the computer is frozen, and the teammate who gets to do the
0:37:16 adapting is the human because the computer is a fixed functionality.
0:37:21 What if in that team, the human could just teach the computer how they want the computer
0:37:22 to help them do their job.
0:37:25 It would be a completely different dynamic.
0:37:27 It would change the future of work.
0:37:28 Yeah.
0:37:29 That’s fascinating.
0:37:33 And then I think another thread that you’re super interested in on the future of machine
0:37:37 learning is something around never-ending learning, so tell us about that.
0:37:38 Sure.
0:37:44 Again, I just go back to what do humans do that computers don’t yet, and computers are
0:37:49 very good at, say, learning to diagnose skin cancer.
0:37:54 You give it some very specific tasks and some data, but if you look at people, people learn
0:37:57 so many things.
0:37:59 We learn to all kinds of things.
0:38:00 You can tap the hands.
0:38:02 You can do double entry bookkeeping.
0:38:03 Right?
0:38:04 Right.
0:38:05 You can add numbers.
0:38:11 You can play music, all kinds of things, and a lot of those things we learn over time
0:38:17 in a kind of synergistic way, in a staged sequence.
0:38:21 First you learn to crawl, then you learn to walk, then you learn to run, then you learn
0:38:28 to ride a bike, and it wouldn’t make any sense to do them in the other sequence, because
0:38:31 you’re actually learning to learn.
0:38:37 When you acquire one skill, it puts you in a position that you now are capable of learning
0:38:38 the next skill.
0:38:44 So I’m very interested in what it would mean to give a computer that kind of capability
0:38:51 to do learning for days and weeks and years and decades.
0:38:59 We have a project we call our Never-Ending Language Learner, which started in 2010, running
0:39:03 24 hours a day trying to learn to read the web.
0:39:04 Fascinating.
0:39:07 And there’s something about sort of the longitudinal, right?
0:39:09 We started it in 2010 and it just keeps on going.
0:39:13 So it’s not just like transfer learning from one model to another.
0:39:15 It’s like long running.
0:39:20 It’s long running and it has many different learning tasks.
0:39:25 It’s building up a knowledge base of knowledge about the world.
0:39:33 But to keep it short, I’ll just say we’ve learned so much from that project about how
0:39:39 to organize the architecture of a system so that it can invent new learning tasks as it
0:39:46 goes so that it can get synergy once it learns one thing to become better at learning another
0:39:47 thing.
0:39:53 How, in fact, very importantly, it can use unlabeled data to train itself instead of
0:39:57 requiring an army of data labellers.
0:40:04 So I just think this is an area that’s relatively untouched in the machine learning field.
0:40:11 But looking forward, we’re already seeing an increasing number of embedded machine learning
0:40:14 systems in continuous use.
0:40:22 And as we see more and more of those in the internet of things and elsewhere, the opportunity
0:40:29 for learning continuously for days and weeks and months and years and decades is increasingly
0:40:30 there.
0:40:36 We ought to be developing the ideas, the concepts of how to organize those systems to take advantage
0:40:37 of that.
0:40:38 Yeah.
0:40:41 I love both of these design approaches in that they’re sort of inspired by humans and
0:40:50 sort of humans are mysteriously good at learning and adapting and they sort of shine a spotlight
0:40:53 on where machine learning algorithms are not yet.
0:40:56 So it’s such a fertile area to look for inspiration.
0:41:00 Well, Tom, it’s been a great delight having you on the podcast.
0:41:03 Thanks for sharing about the history and the future of machine learning.
0:41:06 We can tell you’re still fired up after all of these decades.
0:41:11 And so that’s a great delight just to see somebody who is committed, basically, their
0:41:17 life to understanding the mysteries of learning and wishing many good decades to come as you
0:41:19 continue working on it.
0:41:20 Thanks.
0:41:21 Thanks for doing this podcast.
0:41:25 I think it’s a great thing to get a conversation going and it’s a great contribution to do
0:41:25 that.

0:00:04 The content here is for informational purposes only, should not be taken as legal business
0:00:10 tax or investment advice or be used to evaluate any investment or security and is not directed
0:00:14 at any investors or potential investors in any A16Z fund.
0:00:18 For more details, please see a16z.com/disclosures.
0:00:21 Hi everyone, welcome to the A6NZ podcast.
0:00:22 I’m Sonal.
0:00:26 I’m here today with a very special guest visiting Silicon Valley, the former prime minister
0:00:32 of the United Kingdom, Mr. Tony Blair, who now runs an institute for global change working
0:00:35 with governments, policymakers and others all around the world.
0:00:39 Also joining us, we have Andreessen Horowitz, managing partner Scott Cooper, who has a new
0:00:44 book just out called Secrets of Sandhill Road, Venture Capital and How to Get It.
0:00:49 And given that startups are drivers of economic growth and innovation, Cooper also often weighs
0:00:54 in on various policy issues, especially those that affect the flow of capital, people and
0:00:55 ideas around the world.
0:00:58 And that’s the focus and theme of this episode.
0:01:02 It’s more of a hallway-style conversation where we invite our audience to sort of eavesdrop
0:01:04 on internal meetings and convos.
0:01:09 We discuss the intersection of governments and technology and where policy comes in,
0:01:13 focusing mainly on the mindsets that are required for all of this.
0:01:17 But then we do also suggest a few specific things we can do when it comes to supporting
0:01:21 tech change for the many, not just for the few.
0:01:22 Welcome Tony.
0:01:23 Thank you.
0:01:24 Did you ask me to call you that?
0:01:25 Thank you.
0:01:26 Can everyone know?
0:01:27 He said it was okay.
0:01:28 And Cooper, welcome.
0:01:29 Thank you.
0:01:30 So let’s just get started.
0:01:34 I think the context is that there’s so much discussion right now about tech in the context
0:01:35 of inequality.
0:01:40 One of the points of view that I have, particularly coming from a background where my family came
0:01:44 from India, et cetera, is that it’s also very democratizing.
0:01:49 And a lot of people can do new things in better ways because of technology.
0:01:53 But I think the big question, the question I think we care about today is how do we bring
0:01:58 more people into the system and make sure that tech benefits everyone?
0:02:03 The first thing I would say from the political perspective is that technology is essentially
0:02:04 an empowering and enabling thing.
0:02:08 So I regard it as benign, but it’s got vast consequence.
0:02:10 So the question is how do you deal with the consequence?
0:02:14 How do you access the opportunities and mitigate its risks and disbenefits?
0:02:17 So that is, I think, the right framework to look at it.
0:02:24 But because it’s accelerating in its pace of change and because the change is so deep,
0:02:28 and I look upon this technological revolution as like the 21st century equivalent of the
0:02:32 19th century industrial revolution, it’s going to transform everything.
0:02:37 So I think the fundamental challenge is that the policy makers and the change makers are
0:02:39 not in the right dialogue with each other.
0:02:44 And this is where misfortune will lie if you end up with bad regulation or bad policy and
0:02:49 where the tech people kind of go into their little huddle, because I say this with great
0:02:55 respect, but you come to this Silicon Valley and it’s like walking into another planet,
0:02:56 frankly.
0:02:57 Yes, that’s actually really interesting.
0:03:00 I’m personally offended by that comment.
0:03:03 Now I think the difficulty is that, yes, you’re right, it’s very empowering.
0:03:07 On the other hand, it’s actually quite frightening to people because you kind of all understand
0:03:11 it and the rest of the world doesn’t quite understand it.
0:03:16 And as far as they do understand it, they find it somewhat dystopian and it’s look.
0:03:19 I was actually sitting with some people from my old constituency in the north of England
0:03:24 a few months back and I said to them, I wonder what’s going to happen when we have driverless
0:03:28 cars and their attitude was, it’s never going to happen.
0:03:34 And the role of a politician is to be able to, in a sense, articulate to the people those
0:03:37 changes and then fit them into a policy framework that makes sense.
0:03:41 And that’s the worry because if the politicians don’t understand it, they’ll fear it.
0:03:43 If they fear it, they’ll try and stop it.
0:03:46 You articulated the vision that we’ve always had, which is we’ve always invested around
0:03:48 this theme called software is eating the world.
0:03:53 It’s exactly what you describe, which is technology no longer kind of sits in its own box.
0:03:57 It really is the case that technology will permeate almost every industry over time.
0:04:01 I think that’s where the big change is happening now is it used to be that technology was a
0:04:02 piece of the puzzle now.
0:04:04 Every company is a technology company.
0:04:05 Yeah, exactly.
0:04:07 So that is the kind of board on which people are playing.
0:04:11 So the issue, I think, is this, is how do you get the structured dialogue between the
0:04:13 change makers and the policy makers?
0:04:16 What would you say the number one thing if you could give advice to entrepreneurs in
0:04:21 the Valley that they should do differently to engage this kind of a framework that you’re
0:04:22 describing?
0:04:26 My advice would be stop looking at your own narrow interest in what you’re doing and
0:04:31 understand you’ve got a collective interest in making the world of policy and politics
0:04:33 understand the technology.
0:04:34 What’s going to happen?
0:04:41 How you get A, the right system of regulation and B, how you allow government to enable
0:04:43 these transformative changes.
0:04:44 Yes.
0:04:45 Well, I actually have a question for both of you.
0:04:48 Today is the 30th anniversary of the web, the World Wide Web.
0:04:52 And I just had Google Doodle this morning and a note from Tim Berners-Lee, his original
0:04:55 memo, information management, colon, a proposal.
0:04:59 The question I have is that a lot of people would argue that the best technologies can
0:05:04 develop when you don’t try to, A, priority predict the consequences because, A, you cannot
0:05:09 their complex adaptive systems and, B, there was an environment of quote permissionless
0:05:13 innovation that allowed the web to thrive because the original makers may have foreseen
0:05:17 some apps, but the whole point is that the innovation is what allowed it to thrive.
0:05:20 So I’d love to hear from both of you on how to balance that perspective.
0:05:21 So I agree with that.
0:05:25 I think though what’s different is we used to be able to compartmentalize technology.
0:05:29 It was a piece of software that you used at work to help you be more productive.
0:05:33 But if technology really is going to be part and parcel of everything, then I think it
0:05:37 changes the nature of how we think about that responsibility because it is regulated industries
0:05:42 in many cases that have been largely immune over time from technology in a way that appears
0:05:43 to be different today.
0:05:45 So I would say that then there are two questions that derive from that.
0:05:48 One is how do you make regulation intelligent?
0:05:53 How do you make it abide by the public interest or enhance the public interest, but at the
0:06:00 same time not dampen that creative and in a way entrepreneurial drive behind the development
0:06:01 of new ideas?
0:06:06 And then secondly, what are the ways that government should be working with those that
0:06:08 are going to be impacted by technology?
0:06:11 If you’re in the car industry, it’s going to be a huge change, right?
0:06:15 I mean, if you get these driverless cars, it’s going to change obviously jobs.
0:06:17 It’s going to change insurance.
0:06:22 It’s going to change the method of production, what you produce, probably change the concept
0:06:23 of car ownership in some way.
0:06:24 Absolutely.
0:06:25 It might even reshape entire cities.
0:06:26 Everything will be impacted by it.
0:06:27 Exactly.
0:06:32 I am fascinated by the potential of technology to allow African nations and governments to
0:06:36 circumvent some of the legacy problems we have within our systems.
0:06:41 And that goes for everything from basic healthcare and education through to how you help agricultural
0:06:45 small holders develop a better yield, cooperate better together, and link up better with the
0:06:46 market.
0:06:49 And in fact, one thing that’s happening in Africa today is there are applications of
0:06:52 tech that are growing up in interesting ways.
0:06:55 So my point is, you’ve got all these different facets.
0:07:00 And yet at the moment, the curious thing is, if you were to go to virtually any European
0:07:06 country or if you were to come here and say, okay, name the top four issues, where would
0:07:09 technology be in that list?
0:07:10 Would it be at the top of the list is what you’re saying?
0:07:12 No, I think it wouldn’t be in the list.
0:07:16 Cooper, you spent a lot of time actually in your role as a partner in front of Congress
0:07:20 and various entities giving testimonies about policy and curious for your take on this.
0:07:21 Yeah.
0:07:25 There is a concern I often hear from entrepreneurs, which is, how do we know if we go there?
0:07:29 How does that not just bring us into the fold of regulation and therefore have negative
0:07:32 consequences versus, you know, we talk about things out here.
0:07:35 Sometimes you do things you ask for permission later is a better strategy.
0:07:36 Right.
0:07:37 Ask for forgiveness.
0:07:38 Ask for permission.
0:07:39 Yeah.
0:07:40 I complete to get that.
0:07:44 That’s why I think it’s got to be a big, it’s got to be done in a big way from the collectivity
0:07:47 rather than individual people going because of course you’re absolutely right what will
0:07:50 happen is that the entrepreneurs think, okay, if I go and say, I’ve got the following five
0:07:53 problems that I can see in this technology I’m developing, they’re going to regulate
0:07:54 it away.
0:07:55 Yeah.
0:08:01 I think the hard question will be you’re getting people and companies that will enormous power.
0:08:06 I mean, not just the big tech players, but the others as well.
0:08:11 So I think one of the things that in a sense, it’s my question to you is, how do you manage
0:08:18 to get into that dialogue with policymakers where, you know, these very powerful people
0:08:23 recognize that in the end, you know, however powerful they are, they are not more important
0:08:25 than the public interest.
0:08:28 Part of we believe our role is to help provide, you know, visibility.
0:08:31 I wouldn’t, I don’t want to say education because I think politicians are very well
0:08:35 educated and certainly well meaning, but to connect the divide between, in our case, DC
0:08:36 and Silicon Valley.
0:08:40 And so we will often reach out to regulators, legislators and help them understand this is
0:08:43 what’s happening from an innovation perspective and therefore these are things that you might
0:08:45 want to anticipate that you need to think about.
0:08:49 So autonomous driving is a great example, right, which you mentioned is in order to
0:08:53 make that work in the United States, we probably need forward-looking governments to say there
0:08:58 are test zones or areas where we might have almost regulatory free zones for testing purposes,
0:09:01 right, that have proper supervision, but to enable something that otherwise might not
0:09:03 exist ahead of its time.
0:09:07 Obviously, you’ve got specific micro issues, I mean, they can be big in their impact like
0:09:11 driverless cars, but there is a specific thing, they’ve got specific issues attached to them.
0:09:18 But where does the tech sector go if it wants to engage on, you know, the bigger macro question
0:09:24 of how do you redesign government, by the way, as well as individual sectors, because
0:09:27 government itself is going to have to change.
0:09:28 That organization doesn’t exist today, at least.
0:09:30 I’m not aware of where you would do that.
0:09:34 And I think the other problem with it is we have to think beyond geographic and national
0:09:39 borders on this stuff, because technology and capital are free-flowing in our society.
0:09:44 You almost need a United Nations or some kind of, you know, type of organization to convene
0:09:45 to have those discussions.
0:09:46 Yeah.
0:09:47 I would say there’s a couple of things, too.
0:09:48 There’s a couple of factors.
0:09:54 One, there are obviously lobbying entities like the NVCA, there’s the Internet Association,
0:09:56 which a lot of major companies are a part of.
0:10:02 Then there is a group of players, like there’s a group of think tanks and a middle layer,
0:10:06 and then the government agencies themselves have been soliciting testimony.
0:10:10 Cooper has actually done testimony on all kinds of topics, from Cypheus to crypto to
0:10:12 various different topics.
0:10:18 But what’s really interesting to me, especially, is there’s organizations like 18F and USDS
0:10:23 in the US government, at least, where you have technologists doing literally rotating
0:10:24 apprenticeships.
0:10:28 It’s like the rotating missions, essentially, where they go for three years and they’re
0:10:30 contributing to actually reinventing government systems.
0:10:32 Now, this is a very important addition about it.
0:10:33 Yes.
0:10:34 I think it is, too.
0:10:36 And what’s really amazing is that it’s got tangible impacts.
0:10:41 So a specific example is we have a huge Veterans Administration that doesn’t get great healthcare.
0:10:46 So they redesigned the VA site in order to make sure that people who have accessibility
0:10:49 issues can use the site in a friendly way.
0:10:51 There’s many more applications of the types of things they’re doing.
0:10:53 We’ve actually had them on this podcast.
0:10:54 But I think those are some avenues.
0:10:56 But to Cooper’s point, there is no single entity.
0:11:01 I will say that at Wired, I edited a big set of op-eds around the ITU, which is sort of
0:11:03 like a UN for the internet.
0:11:06 And it was during the WC-12 hearings, which you might recall.
0:11:11 I think Hama Dune Touré was the head of the commission, and I edited him as well.
0:11:16 And what’s fascinating to me is that there’s a lagging versus a leading approach to it,
0:11:20 because you’re sort of taking the data that’s passed, not really looking forward.
0:11:23 And that was what I saw as a big drawback when I was working with the WC-12 op-ed.
0:11:28 So I’m curious for your take on how do we shift it, so you are listening to those being
0:11:34 affected by technology, but with the point of view that spins it forward for future generations.
0:11:38 Because if we had listened to all the farmers in the first wave of the Industrial Revolution,
0:11:43 we may not have many of the things today, but their grandkids are benefiting from those
0:11:44 things.
0:11:45 Yeah, no, absolutely.
0:11:48 So look, I think there are two caps that I see, and I just look at this from the side
0:11:54 of the, as a we’re ordinary politician, because I think there are initiatives that are happening
0:11:58 inside government where people or departments will get it, and therefore they’ll embrace
0:12:02 it and bring in smart people to help them and so on.
0:12:04 But I think there are two sort of lacunae.
0:12:10 One is your average politician does not understand a lot about this, and that is not sort of
0:12:12 a disrespect to your average politician.
0:12:18 It’s that it’s new, it’s complicated, it takes you time to get your head around it.
0:12:22 My eldest son is in technology, and I am always trying to get him to explain blockchain to
0:12:23 me.
0:12:24 We’re big on crypto.
0:12:25 I know.
0:12:31 I remember you sent me the other day saying, “This is the idiot’s guide to crypto currency,
0:12:33 and I still couldn’t understand it.”
0:12:35 I’m going to send you our crypto cannon.
0:12:36 We took a stab at Compound Water Resources.
0:12:37 Right.
0:12:38 But you probably shouldn’t test me on it.
0:12:39 But that is one lacunae.
0:12:44 Those people have to understand this is like the 19th century Industrial Revolution.
0:12:47 So you’ve got to get your ordinary politicians to understand it.
0:12:51 And then there’s another lacunae which is, I think, in getting the dialogue at the top
0:12:56 level between particularly the Americans and the Europeans, because I also think it would
0:13:00 be immensely helpful if we had a more transatlantic approach.
0:13:01 I think there’s a third piece.
0:13:03 There’s an incentives problem.
0:13:07 I would imagine if you did a survey of most politicians, they would say, “My fundamental
0:13:11 role is how do I improve long-term economic growth and job sustainability for my constituents?”
0:13:15 I mean, if people kind of cut through a lot of the politics, that’s really why they think
0:13:16 they’re there.
0:13:19 Look, they want to make a better life and make a better opportunity for their constituents.
0:13:23 The problem we have, though, is because their short-term incentive program is to get reelected,
0:13:25 which I understand is a good thing from a political perspective.
0:13:29 It’s very hard for them to take that long-term view because the shorter-term opportunity
0:13:34 is to say, “Look, I really need to do no harm to my constituents and by allowing technology,
0:13:39 which might be in the short-term, displacing and unsettling to job growth and other stuff,
0:13:41 particularly for different segments of the population.”
0:13:44 It’s very hard, I would imagine, as a politician to square those two things, which is how do
0:13:49 I help my constituents understand, you know, to Sonal’s point that, yes, over a period
0:13:54 of time, it was a good idea to have industrialized farming as opposed to pure manual agrarian
0:13:55 farming.
0:13:58 But that’s an incredibly unsettling thing, particularly in the U.S. here, if every two
0:14:01 years you have to get reelected or you go find a new job.
0:14:07 By the way, this happened in the 19th century and you had whole new politics created around
0:14:08 it.
0:14:10 And I think there are two things that are important here.
0:14:16 First of all, I think the technology will, in some way, provide solutions to what is
0:14:23 a constant dilemma for an ordinary politician, which is we need to do more for our people,
0:14:27 but we can’t just keep spending more and taxing more.
0:14:31 If the technology can help unlock part of that, that is something they’re prepared to
0:14:32 go for.
0:14:37 And secondly, with most politicians, if they’re able to see this within a longer-term perspective,
0:14:42 what you say to them, “Look, we’ll help you and guide you through this process of change,
0:14:44 but in the end, it’s a beneficial change.”
0:14:47 And what I found when I was in government is some of the most difficult reforms we put
0:14:54 through, for example, around education reform, healthcare reform, we were able, in some ways,
0:15:01 with at least some people, to say this short-term difficulty is going to be worth it.
0:15:03 How did you pull that off, though?
0:15:05 Was it the education, the explanation?
0:15:07 Was it consensus building?
0:15:12 I mean, let me take a very specific example, which, of course, is under attack now, but
0:15:18 we introduced tuition fees in the UK, but my point was very, very simple, that universities
0:15:22 are going to be major engines of economic growth in the future, in particular because
0:15:26 of the link between university and technology and the development of technology.
0:15:31 And therefore, we cannot afford for UK universities not to be able to get the best talent, and
0:15:34 they’re going to have to therefore have an extended funding base.
0:15:35 They can’t get it all from government.
0:15:39 And my point is, if you get it all from government in the end, some governments will start to
0:15:43 slice it away, and you’re always hand-to-mouth as universities.
0:15:49 And I reckon when we did that, it was very difficult, in fact, it was extremely difficult,
0:15:55 but in the end, you were able to say to people, “Look, if we want to save our position as
0:16:00 a country that, along with the US, probably has most high-quality universities in the
0:16:02 top 50 in the world, then we’ve got to be prepared to do that.”
0:16:07 Now, some people, by the way, rejected it, and today it’s a big political issue again,
0:16:12 but you can get to at least some form of alignment between long-term and short-term.
0:16:15 It’s a fundamental rethinking of what the role of government is, quite frankly, right?
0:16:19 Which is, again, if you take the premise that the overall objective for government is to
0:16:23 create economic conditions that hopefully generate long-term economic growth and sustainability
0:16:25 for individuals and companies, then you’re right.
0:16:29 Maybe the ancillary role of government is, how do we deal with short-term issues that
0:16:32 have market dislocations for people?
0:16:35 Maybe that’s a more proper way to describe what the role of government is, in many cases.
0:16:38 I think the other thing would be, I think there’s another question for politics which
0:16:43 would be very challenging, because what would be weird is if the whole of the world is undergoing
0:16:46 this revolution and politics is just kind of staying fixed.
0:16:49 The type of people who go into politics, what happens often is people leave university,
0:16:53 they’re going to become a researcher for an MP, and then they become an MP, and then they
0:16:55 become a minister, but they have no experience of the outside world, right?
0:16:59 So that’s one, and it becomes a constraint over time, and then the types of people who
0:17:00 work in government.
0:17:04 So you were saying something about the people who’ve been brought into, say, the Veterans
0:17:06 Administration here.
0:17:12 So how do you actually open up public service and then get a greater interoperability between
0:17:15 public service and the private sector?
0:17:20 Because all of the sort of pressure certainly coming from the media has been not to allow
0:17:25 that to happen, and not to allow politicians to have anything other than they’re usually
0:17:26 just focused on…
0:17:29 Okay, let’s say we all agree, which I think we do, that there needs to be a connection
0:17:32 between all the entities working together, no question.
0:17:36 More engagement, more explanation, more understanding, thinking of consequence.
0:17:37 I think those are all table stakes.
0:17:41 The question now is, how do you then think about unintended consequences?
0:17:46 Because the story, to me, is not that bad things have bad consequences.
0:17:51 It’s that often the worst consequences come from very well-intended things.
0:17:53 And quite frankly, the perfect example that comes to mind is GDPR.
0:17:57 Yeah, to make it concrete, there’s been a, over the last several months, and I’m sure
0:18:00 probably more so in Europe as well, there’s been a number of articles talking about when
0:18:05 you look at kind of the broad impact of GDPR, essentially, it’s endured largely to the benefit
0:18:08 of the very large incumbents, which was probably not what’s intended to do.
0:18:09 Because they’ve got the resource to better handle.
0:18:10 That’s exactly right.
0:18:14 And the analogy we have here in the States was the Dodd-Frank legislation that came out
0:18:18 of the global financial crisis, where financial institutions had to comply with a whole new
0:18:20 set of regulations.
0:18:24 What it really did here in the U.S. was, it really entrenched those incumbents very well,
0:18:27 and it made it very hard for startup financial institutions to grow.
0:18:31 It was very hard for a new institution to get a banking license for many years, in part
0:18:33 because of the regulatory cost of doing so.
0:18:35 And so, how do you balance that?
0:18:38 And maybe the answer is, look, it’s an education problem, but well-meeting and politicians
0:18:43 certainly expect and intend that regulation is the appropriate way to deal with these things.
0:18:47 It does, in some cases, interfere with the overall goal of entrepreneurship in a startup
0:18:48 formation.
0:18:53 That’s why I think that the attitude of the technology sector to engagement with government
0:18:54 is so important.
0:18:58 Because if you’re engaging with government saying, look, we understand there’s a massive
0:19:01 set of issues here, and we’re really going to sit down and work with you as to how we
0:19:07 get the right answer, then government’s in a position where they regard you as a partner.
0:19:11 But I think for this moment in time, a bit like, actually, the aftermath of the financial
0:19:17 crisis, government kind of regards, you’re looking after yourselves, but we got to look
0:19:19 after the public.
0:19:21 And that’s where it leads to poor regulation.
0:19:28 I mean, poor regulation is nearly always the consequence of a failure on the regulating
0:19:31 side to really understand what’s going on.
0:19:35 And on the founder’s side, what I’m hearing is to really communicate the benefits of the
0:19:36 technology upfront.
0:19:40 And to be so defensive that you’re just thinking all the time, how can we ward these people
0:19:41 off?
0:19:46 But here’s the thing, you can sometimes, if you have wealth, which a lot of these big
0:19:51 tech players do, and power, and you also have access, and you can go and see whoever they
0:19:57 want to see, it can sometimes mask your essential underlying vulnerability.
0:19:58 Interesting.
0:19:59 Right.
0:20:05 And your vulnerability is the comes a point when suddenly the mood flips, and it doesn’t
0:20:10 matter how much wealth and power and access you have, you’re the target.
0:20:12 So that point, everything changes.
0:20:16 So if you want to avoid that, I think it’s got to be a dialogue that’s structured and
0:20:22 it requires not just things happening between the tech sector and government, but for people
0:20:28 like my own institute, to use our sort of convening power, the political side, to say
0:20:30 to the politicians, look, let’s get our heads around this.
0:20:32 Here’s my essential challenge.
0:20:37 How do you take this technological revolution and as a politician, weave it into a narrative
0:20:39 of optimism about the future?
0:20:40 Yes.
0:20:41 I want that too.
0:20:42 Right.
0:20:43 Yeah.
0:20:44 So what’s driving the populism is pessimism.
0:20:46 If people are pessimistic, then they look for someone to blame.
0:20:52 If people are optimistic, they look for something to aspire to, and that’s the essential difference.
0:20:54 It’s really interesting also, Sonal and I have been having this conversation, and we’ve
0:20:58 been having this conversation in the U.S. about ESG, right, which obviously, you know, certainly…
0:20:59 Environmental social governance.
0:21:00 Right.
0:21:02 Which Europe is way ahead of the U.S. there, and Larry Fink, who’s the head of BlackRock
0:21:05 here, has written this letter, you know, appealing to CEOs, and it really goes to the same issue
0:21:09 you’re talking about, which is fundamentally, what is the role of the corporation and how
0:21:12 do corporations think about obviously enhancing value for their shareholders, but also to your
0:21:17 point recognizing that they impact constituents in many other ways, and I think that’s kind
0:21:21 of the dialogue we ought to be having with politicians, which is, look, we can create
0:21:25 a world where it’s compatible to have, you know, maximizing shareholder opportunity,
0:21:28 but also recognizing and being a part of the broader community discussion about the impact
0:21:29 on society.
0:21:35 The other thing is to recognize that when we create these things, we have some obligation
0:21:36 to share.
0:21:40 It comes out of fundamental macroeconomics, right, which is we can improve growth for
0:21:44 a country by either population growth and/or productivity growth, right?
0:21:47 Those are the two levers in theory that we can impact, and if we can frame the discussion
0:21:50 around technology, that’s a lot of where the U.S. has done well, right?
0:21:54 We’ve generally, obviously times are changing, but we’ve generally been very open to immigration
0:21:58 and thought about population growth as a way to help improve the lot for people generally,
0:22:01 and we’ve also been very open to productivity growth, right, in the form of technology and
0:22:05 automation, and if we can frame it that way, but also to your point, recognize that there
0:22:09 are going to be disintermediations along the way, and part of our responsibility is to help
0:22:14 from a training and education perspective, and even potentially the role of government
0:22:19 in subsidizing the transition from less automated to more automated society.
0:22:21 What happens to education in all of this?
0:22:23 I don’t think we have a singular point of view on it.
0:22:27 We have talked about education a lot on this podcast and shared a diversity of views, but
0:22:32 I think a couple of the high level things are that universities are huge drivers, of course,
0:22:36 as you mentioned, of innovation, and in every study of regional innovation, every innovation
0:22:43 cluster is successful because of the collaboration between universities, government, local, entrepreneurial
0:22:44 communities.
0:22:47 The other key point, however, is it’s a combination of top-down and bottom-up.
0:22:52 People who have tried top-down, industrial, planned, smart cities or things like Silicon
0:22:53 Valley never work.
0:22:56 The only bottom-up ones alone don’t necessarily work.
0:22:57 You need a combination of the two.
0:23:00 That’s the number one finding, but the second thing, and this is a big topic we talk about
0:23:05 on this podcast, is the importance of apprenticeship and a new type of education that really thinks
0:23:07 about skills-based education.
0:23:12 We have this elitist attitude that education has to be a certain way when, in fact, in
0:23:16 this day and age, especially with increased automation and the need for jobs, we might
0:23:19 want to be really thinking about very specific skills-based education.
0:23:24 It’s actually fascinating because, in fact, my eldest son’s got a company that’s on apprenticeships
0:23:28 and technology, so that’s exactly what he does.
0:23:32 I think it’s really, really interesting because of the idea that you don’t necessarily have
0:23:33 to go to university.
0:23:38 Well, there are alternative universities coming about too, like we’re investors in
0:23:39 Udacity.
0:23:40 There’s just Lambda School.
0:23:44 There’s all these interesting types of containers where people can get what they call nano-degrees
0:23:46 or microskills or specific skills.
0:23:50 There’s so much that’s actually in play, because the point I want to raise here, this is kind
0:23:55 of an underlying theme to me, is that technology, as you pointed out, you can take an optimistic
0:23:56 view.
0:23:59 It also gives you the means to address many of the problems that we are complaining about
0:24:04 because when I think of some of the trade-offs between waiting for a government to update
0:24:12 policy, what I love is that a mass of users on a platform can essentially vote with saying
0:24:18 leave that platform, and immediately that platform is going to act the next day in a
0:24:20 way that a lawmaker cannot overnight.
0:24:22 Yes, from a political perspective.
0:24:24 You want this thing at least to have some sort of rational-
0:24:25 Of course.
0:24:26 It shouldn’t be mobbed.
0:24:31 But I think the other thing is, if you take, so a lot of what drove, for example, Brexit
0:24:35 in the UK is, apart from the immigration issue, was this idea of communities, people left
0:24:36 behind.
0:24:41 So what is it that technology would do to go into those communities and help people gain
0:24:45 better education, get connectivity to the world, because in the end, this is what it’s
0:24:46 all about.
0:24:47 If you’re not connected, you’re not really–
0:24:48 If you’re left behind.
0:24:49 Right.
0:24:55 So I think one big question is, how does the technology sector help us as policymakers
0:25:00 reach those people for whom the conversation we’ve just been having may be sort of scratching
0:25:02 their heads and thinking about what these guys are on.
0:25:03 That’s a fantastic question.
0:25:05 And actually, it’s interesting because we’re investors in NationBuilder, which is one of
0:25:10 the companies that mobilize a lot of the communities that actually organize for pro and for con
0:25:11 around these things.
0:25:14 So a quick thing, I do want to make sure we actually give answers because we’re asking
0:25:15 a lot of questions.
0:25:19 So can you both give a little bit more on what concretely we need to do?
0:25:23 So from the point of view of my institute, what we’re doing is we’re creating a tech
0:25:25 and public policy center.
0:25:29 And the idea is to bring a certain number of people who really understand the tech side
0:25:32 and a certain number of people who come from the public policy side, put them together
0:25:33 in a structured way.
0:25:35 I will kind of curate that.
0:25:43 And out of it should come what I call serviceable policy and a narrative about the future,
0:25:46 which makes sense of this technological revolution.
0:25:50 And then to link up with politicians, not just in the UK and Europe, but actually over
0:25:55 here and create a sense that this technological revolution should be at the center of the
0:25:56 political debate.
0:25:57 How do we handle it?
0:26:00 How do we, as I say, mitigate its risks and access its opportunities?
0:26:03 So that’s one very specific thing.
0:26:07 And then I think the other thing, frankly, is just to be out there, myself and a number
0:26:13 of other people, at least of access to the airwaves, to say, guys, we’ve got to switch
0:26:14 the conversation.
0:26:18 You’ve got to put this technology question at the heart of the political debate.
0:26:22 Now the solution, some people may go to the left, some people may go to the right.
0:26:24 Some people will never be in between.
0:26:25 But make it the conversation.
0:26:29 Put it at the top four of those priorities for every country, every organization.
0:26:30 I think that’s right.
0:26:32 Fundamentally, we’re talking around the issues.
0:26:36 It’s either immigration or its income inequality or other things that drive the debate.
0:26:40 But the fundamental question is exactly that, which is how do we move forward with broader
0:26:41 economic growth initiatives?
0:26:46 So sitting here in Silicon Valley, any individual company is probably better off, quite frankly,
0:26:49 taking the break glass first and then ask for forgiveness later.
0:26:53 And so it’s, I think, the idea of having kind of solving that collective action problem
0:26:56 through a convening organization makes a lot of sense.
0:26:59 But you come to the issues of very traditional income inequality.
0:27:03 Now there is a perfectly good question as to whether you raise the minimum wage, and
0:27:04 if so, by how much?
0:27:07 And my government was coming to introduce the minimum wage in the UK, so I’m very familiar
0:27:10 with all those arguments.
0:27:17 But in the end, there is a whole other dimension to that individual, which is about the world
0:27:20 that’s changing and their place in it, and whether they’re going to have the skills and
0:27:21 the aptitude to be able to.
0:27:25 So you’re just saying completely at every level reframe that technology is at the center
0:27:26 of that.
0:27:27 Right.
0:27:31 So it’s not that you displace traditional questions of taxation and inequality, but the
0:27:38 truth of the matter is it’s going to be probably in the long term more significant for that
0:27:43 individual and for the society if this technological revolution is handled properly.
0:27:47 So if you had a debate in the UK at the moment about our healthcare system, our national
0:27:53 health service, it would be, should we spend 10 billion pounds a year more on it or 5 billion?
0:27:59 But how do you change the whole of the way we implement care for people because of technology
0:28:00 is going to have a much bigger impact?
0:28:01 I agree.
0:28:05 I guess the only thing I would add to this, because I think about this a lot, interestingly,
0:28:10 is that we treat technology like this word, this homogenous, nebulous entity.
0:28:16 And the reality is that every single instance so depends on the specific technology.
0:28:19 So my call to action, I guess, would be to think about it very specifically.
0:28:25 The way we think about AI, that’s such a broad phrase and it’s a very scary phrase that
0:28:30 suggests everything from generalized intelligence to very specific automation that gets your
0:28:33 bank account updated automatically.
0:28:34 So I think there’s two things to this.
0:28:38 One that we need to be incredibly specific about what technology we’re talking about,
0:28:39 in what context.
0:28:43 And then B, we also dial in the right degree to what we’re talking about at what point
0:28:44 because I don’t really make a difference.
0:28:45 I completely agree with that.
0:28:50 I mean, I think the only thing I would say is right now we’re actually far away from
0:28:53 even getting the specifics.
0:28:54 Yeah, you’ve got a good point.
0:28:55 That’s fair.
0:28:59 You know, I was just saying to people in politics when we were campaigning and they’d say, you
0:29:03 know, I’d say, right, because we were, we campaigned in the slogan 99% right, new labor,
0:29:04 new Britain.
0:29:05 Right.
0:29:06 And they say, no, but it’s much more complicated now.
0:29:09 I say, okay, guys, it is, but actually it’s complicated, but sometimes you need to go
0:29:10 straight.
0:29:11 I hear you.
0:29:15 You’re saying that you’re saying that when you go back to your old constituency in constituency
0:29:19 in Northern England, they don’t care about the specifics, they just need to have their
0:29:20 fears established.
0:29:22 The first thing that you need to persuade them of, you’ve got to say to them, guys, technology
0:29:24 is going to change the world and we’ve got to prepare for it.
0:29:25 They’re not there yet.
0:29:29 When you get them there, then obviously in all sorts of different ways, but this is where
0:29:35 I think that the gulf that there is between the technology sector and the policies and
0:29:36 therefore the people is so big.
0:29:40 How do we deal with the fundamental challenges that we have as we talked about earlier from
0:29:44 an incentive perspective and short tenures on, you know, office and people’s ability
0:29:45 to be in office?
0:29:48 Is that a conversation you think that the country and the nations are prepared to have?
0:29:50 I think so, but it’s a very good question.
0:29:55 I would also say that in all of the change that’s going to happen, I mean, this is a
0:29:58 whole topic for another podcast, probably with the different people.
0:30:08 But how you exchange information and the validation of that information is an essential part of
0:30:12 having a democratic debate and that is a big problem in today’s world.
0:30:19 So I think it is possible to have that conversation with people, but all political conversations
0:30:25 today are extremely difficult because they happen in such a fevered environment with
0:30:31 so much polarization and the interaction between conventional and social media makes a rational
0:30:33 debate occasionally extremely difficult.
0:30:35 With that qualification I answer this.
0:30:39 I think the better way to approach that problem is to say, how do we make the U.S. and/or Europe
0:30:45 or other places attractive to entrepreneurship and encourage people to think about the regulatory
0:30:49 framework and the economic framework as, you know, want to be participants in these markets
0:30:53 as opposed to the anti, you know, kind of policies we have, which is let’s make it harder for
0:30:56 free flow of capital and try to stave off those opportunities.
0:31:00 It used to be that 90% of venture capital and entrepreneurship happen in the U.S. literally
0:31:05 almost as early as 20 years ago, and if you look at those numbers today, it’s about 50%.
0:31:09 And so the amount of kind of capital that’s kind of been distributed globally and therefore
0:31:12 the amount of opportunity set distributed globally is interesting.
0:31:15 We have to think about this beyond kind of regional borders.
0:31:18 We will have talent in people that are free flowing across geographies.
0:31:21 And so we have to think about this from a broader, you know, global initiative.
0:31:25 Well, you guys, I just want to say thank you for joining the A6NZ podcast.
0:31:26 Thank you.