00:00hi everyone welcome to the a6 & Z
00:02podcast I'm sonal today we're doing one
00:04of our book podcasts around the new book
00:06just out machine platform crowd the
00:09author's previously wrote the popular
00:11book the second Machine Age and before
00:13that their book was race against the
00:15machine hmm sensing a bit of a theme
00:16here so in this episode we cover those
00:19themes first starting with a bit of econ
00:20101 around network effects complements
00:23and other key concepts then we discussed
00:25how this all plays out organizationally
00:27especially given trends like machine
00:29learning blockchain and crowds and
00:30tackle the tricky question of whether
00:32networks can replace the firm and where
00:35are we in the classic question around
00:36the future of the firm and finally what
00:38can companies do more concretely frank
00:41chen joins the conversation in between
00:43as well to share his perspective on what
00:45he sees given his role as head of
00:46investing in research at a 6 and Z but
00:49our main guests on the episode both from
00:51MIT are Erik Brynjolfsson and Andrew
00:53McAfee who I'm gonna call Andy is that
00:55ok otherwise I'm gonna mistake you for
00:57my mom we kind of go way back in the
01:02sense that I met you years ago and not
01:04as far as I go with my mom so this is
01:09your third book together the real thrust
01:11of your work is that this is
01:12unprecedented and the speed at which
01:14we're changing and what the effects are
01:16and I think a great theme for this
01:17conversation is to sort of break down
01:18how those changes are gonna play out and
01:21where they're happening yeah yeah well
01:23but let me just push back on that first
01:25part a little bit because in Silicon
01:26Valley everybody agrees with that and
01:28and what and we agree with it that's
01:31very clear but we were reading people
01:33who didn't one of the things that got in
01:34writing our first book Race Against the
01:36Machine was there were people who were
01:38quote the great stagnation and how there
01:41were no good inventions anymore what
01:42nothing goes invented in particular
01:44Tyler Cowen it was a spot on that median
01:47income had been stagnating and that was
01:50kind of troubling for us because you
01:52know I had been taught this slogan that
01:54in productivity isn't everything but in
01:56the long run if we just have tech
01:57progress and everything else takes care
01:59of itself and when Tyler showed us that
02:01evidence we were like wow this is a real
02:03problem but we refuse to give up on the
02:05idea that technology which is doing we
02:07weren't gonna let a little evidence get
02:08in the way of now ya know God said no
02:11but fortunately we figured out a way out
02:13of it so and and the way out of it is
02:15that even though technology is making
02:17the pie bigger there's no economic law
02:19that everyone's going to benefit from it
02:21it's possible for some people to get
02:23left behind now to be clear that's not
02:25what happened for most of the past 200
02:27years but the past 10 20 years there
02:30really have been more and more people
02:31being left behind and so you could get
02:34stagnating median incomes even as some
02:37people may be in the top 1% got
02:39fabulously wealthy and that helped us
02:41reconcile these different perspectives
02:43and it led to a whole broader set of
02:45discussions about the way that
02:47organizations and society and business
02:49processes aren't keeping up with these
02:52amazing technologies and some of the
02:54dysfunctions that can create some of the
02:56opportunities they can create so what
02:57are some of the big well I think we
02:59should talk write down the fundamental
03:00building blocks of a lot of the
03:01arguments that you make throughout your
03:02work so let's talk about networks and
03:04one of the biggest questions I had
03:07reading your book was is a network gonna
03:10displace the firm in the future we talk
03:12a lot about network effects in our
03:13business so networks sometimes
03:15economists call them demand-side
03:16economies of scale and it's basically
03:18the idea that a product or service
03:20becomes more valuable the more other
03:22people that are using that product or
03:24service or classic examples you know a
03:25telephone or fax machine whatsapp
03:28Facebook and you could have supply-side
03:31economies of scale just to distinguish
03:33that that's when the cost get lower as
03:35more people use it and both of these
03:37things lead to the big companies winning
03:41and just for shorthand we tend to
03:43describe supply-side economies of scale
03:44is just economies of scale yeah
03:46economies of scale as a network that's
03:48the more common word generically and we
03:50used both sets of terminology it's
03:52sometimes useful to talk about
03:53supply-side demand-side because a lot of
03:55the economics become more intuitive once
03:57you understand that there's the demand
03:59side and the supply side and they both
04:00can get better as you get bigger and
04:02then to get add a little more layer of a
04:04subtlety to it you could have
04:05traditional single sided network effects
04:08like other people using the same
04:09telephone or you can have two sided
04:11network that's what really the platform
04:13revolution a lot of that has been
04:14triggered by the growth of so-called
04:16two-sided networks and the idea there is
04:17that it's not necessarily the people
04:19using the same product as you but it
04:21could be people on the other side using
04:23a different products so like driver
04:25and users are using slightly different
04:26apps and me as a user I don't really
04:29benefit when more users are also now I
04:32want more drivers and the drivers want
04:33more users so you care about the people
04:35on the other side of the network except
04:37when you're pooling cuz then you do care
04:39that's right and that's a case where you
04:41do want them on the same side exactly
04:43and then to make it even more
04:44complicated you can have two side in one
04:45side at the same time you could have
04:46economies of scale so you can layer them
04:48you mentioned the word building block
04:49let's start with these primitives and
04:51then you can start combining them in
04:52different ways this really starts to
04:54turn into three-dimensional chess
04:55because the right way to think about the
04:57app ecosystem and Apple is not any kind
05:00of one or two sided network it's an
05:02incited network but lots of different
05:05groups of people who value things on the
05:07other side but we don't decide what the
05:09sides aren't and we let the self
05:11selection happen and you just watch the
05:13the vortex form around that ecosystem
05:16and the only way to understand that is
05:18is by doing what Eric just did start
05:21with network effects 1-sided 2-sided 2
05:24goes to n value goes to n ok that's
05:26great let's probe on one big thing which
05:28is we talked about network effects but
05:29let's quickly define compliments in this
05:31because that's a term that's frequently
05:32used and I think it has a lot of
05:33misconceptions around it sure what are
05:35the key economic building blocks that we
05:36talked about is compliment and
05:37compliment is a very simple concept it's
05:39the idea that one product is more
05:42valuable in the presence of another so
05:43my left shoe is more valuable if I also
05:45have my right shoe software is more
05:49valuable with the right hardware and so
05:51you compliments can be physical or they
05:53can even be organizational well so you
05:56may have a system that taps into the
06:00crowd that's more valuable when you have
06:02a global internet that allows you to do
06:04that right so you can have
06:06organizational or technical or physical
06:08complements and you can sell products
06:11that are complementary to each other
06:13razor blade ya razor blades and and
06:15sometimes when you have products that
06:16are complementary to one another it
06:18actually can be profitable to give one
06:20away to increase the demand for the
06:22other one so people famously gave away
06:24razors to sell blades and this can
06:27interact with the network effects and
06:28the scale economies it's not a good
06:30strategy if you don't have those other
06:32things one of the things that you know
06:33makes us tear our hair out is that you
06:36know when MBA students yeah we'll just
06:38here is the underlying strategy oh so if
06:40you're just saying look I need to do
06:41freemium without really understanding
06:42that or underlying strategy disastrous
06:45and compliments are weirdly subtle
06:51clearly the econ 101 example but I
06:54always fall back on is hamburger meat
06:56and hamburger buns and so if the price
06:58of hamburger meat goes down demand for
07:00buns is going to go up even if the price
07:03of buns doesn't change that's the key
07:05thing the price of one good can stay the
07:07same and demand for it will go up the
07:10complements are so tricky that they
07:12actually tripped up Steve Jobs really
07:14badly this is not Laura this is fact he
07:16did not want to open up the App Store to
07:18any outside developers he thought he had
07:19to maintain super tight control over
07:21that digital environment and when the
07:24iPhone first released it did not have
07:26any external apps on it
07:27he fought boardroom battles for about a
07:29year with people who said no you need to
07:31open this up what may in cave
07:33pressure from really smart people inside
07:36and outside the company people on his
07:37board and executives at the company when
07:39he didn't fully realize is that if you
07:41open up the app store and you curate
07:43successfully you have just opened the
07:45door to this massive number of
07:46complements each one of which is going
07:49to nudge out demand for the iPhone and
07:51even if each one only nudges that demand
07:53outward like 99 cents worth just to be
07:58clear we're not talking about the
07:58literal money I never I still don't have
08:08an iPhone I have an Android but I still
08:09remember to this day the first thing
08:10people would say I don't like the iPhone
08:12that much and they're like it's not
08:13about the phone it's all about the app
08:16yeah and the only way to understand the
08:19value of opening up that app store is to
08:21understand that you are unleashing this
08:23tidal wave of complimentary goods that
08:26were priced at all different price
08:27points including zero which is awesome
08:29so zero is a really great price but the
08:31more fundamental thing I think is that
08:33it shifted out demanded nudged demand
08:35upward for the other complementary good
08:38the iPhone itself and once you once you
08:40grok into that then you say oh I got to
08:42find all kinds of different ways to do
08:44this and play three-dimensional chess
08:45with my platform is the corollary of all
08:47this that clothes will never win then no
08:49it's not nearly as simple as that
08:52it does show you that if you can
08:53leverage these complements you can
08:55create not just a one-time win but it
08:57whole ecosystem because it because Andy
08:59story turns into a virtuous cycle where
09:01the more demand for my phone exactly
09:03it's a flywheel and so that can work
09:05very well but it's it's not like you
09:08always open up or you always right
09:10because I don't see a lot of the winners
09:11until now have been closed companies
09:14yeah and Apple was comparatively closed
09:15against exactly one of the things we say
09:19it's there is not one right answer there
09:21is not one recipe that you follow for
09:23success with machines platforms or crowd
09:26there are prints of their of principles
09:28and for entrepreneurs who are listening
09:29understanding complements and the way
09:32the people who are creating these
09:34ecosystems that have compliments is
09:35super important so we've been talking
09:37about compliments where the more phone
09:39the app the more apps in the App Store
09:41the more attractive and iPhone so think
09:44about that when you're thinking about
09:45development tools for these platforms
09:48Xcode Visual Studio are so important to
09:50Microsoft and Apple because they're
09:53creating these complements and therefore
09:54the desirability for their iPhone that's
09:56where they make all their money so if
09:58you think hey I'm gonna create a better
09:59development tool I'm gonna create a
10:01better Xcode like think again because
10:03Apple is gonna spend as much money as it
10:06needs depression complement universe the
10:08question that comes to mind for me is
10:09what this means for companies so one
10:11thing the conventional wisdom now is we
10:14fund companies whose defense ability is
10:16a network effect in other words we're in
10:18lyft and Airbnb precisely because once
10:21you have all the hosts you're gonna get
10:22all of the renters right and so one
10:25thing to think about is maybe in the
10:27future even the firm that creates the
10:29network effect gets decentralized who
10:31needs a firm why don't people just come
10:33together and we'll create the right set
10:35of incentives for the network to behave
10:37so you could imagine an eBay where there
10:39is no company there's just a network
10:41coming together with the right set of
10:42incentives that was how we wound up the
10:44book is is trying to grapple honestly
10:47with this question of in the universe
10:49that can be turbocharged by the fact
10:51that everyone's got a device that we've
10:53got this completely decentralized
10:54cryptocurrency system you could pay
10:56people with that we've got these
10:58technologies of radical decentralization
11:00like the blockchain like them exciting
11:02public distributed ledger yeah right
11:05you can stay everybody's distributor
11:06contracts and code into those things you
11:09can do a lot of the stuff that we used
11:10to need a company for the question gets
11:12teed up are we still gonna have
11:14companies in the future and as Eric and
11:15I started to think about all the stuff
11:17that we'd learned and tried to digest
11:18our answer was an unequivocal yes and
11:20the main reason for that is that
11:22ownership of a thing matters simply
11:26because almost while every economist I
11:28think that we've talked to would agree
11:30that you can never write a complete
11:31contract that will specify exactly what
11:33everybody's going to do in every Bunch
11:35engine cannot be accounted for and the
11:37reason for a firm is it gets to make the
11:39decisions that are not contractually
11:41specified elsewhere and it gets all the
11:45value that's not that's not a portioned
11:47apportioned elsewhere in the network and
11:49it starts with Ronald Coase 9000 he was
12:01he was he was 20 he was 26 I think he
12:04was but then and then more recently all
12:07over Hart who was my thesis adviser and
12:09banked homes from one of our other
12:10colleagues at MIT elaborate on that as
12:13Andy was saying with this so-called
12:15incomplete contracts theory one of the
12:17the blinders that a lot of people
12:19especially technologies have is they say
12:21hey we can just write everything down
12:22that engineering mindset will write a
12:24complete contract that covers all
12:26contingencies and the reality is is the
12:28world is just too complicated to cover
12:30every possible contingency so when you
12:33only not when you own a car you can sell
12:37that to someone else and whoever owns
12:39the car gets to have all where to call
12:41the residual rights of control
12:42everything that's not specified in the
12:44contract you want to change the color of
12:46it that's what ownership means and
12:47ultimately you take that to the level of
12:49a firm a firm is an aggregator of a
12:52bunch of assets and owns certain things
12:54and that means that gives you a certain
12:55power they give some certain incentives
12:57of how those objects are you as Eric and
12:59I were trying to reason our way through
13:00this and convinced ourselves to one view
13:02of the world here this amazing real-life
13:05experiment happened which was the Dow
13:06yeah and a quick terminology thing when
13:09you say that Dow you mean the
13:10corporation that was formed but that's
13:12very different than a Dow which is this
13:14autonomous organization or decentralized
13:16autonomous corporation which was
13:23intended to be a completely owner free
13:26completely decentralized organization
13:27along the lines that you just described
13:31somebody found out how to treat it like
13:32an ATM essentially and so to the extent
13:35there was a group of people kind of
13:37behind it they collectively freaked out
13:39and thought about what to do
13:40and then they made this fairly
13:41autocratic decision looks a lot like an
13:44ownership decision to me to reset the
13:46clock on the entire Dow they became de
13:48facto owners nay asserted those rights
13:50in a way and I de novo they said okay
13:52we're gonna do this and if enough you go
13:54along with this and this is what's going
13:55to happen is extraordinary decentralized
13:57organization it was kind of heavy I mean
13:59I love your saying something
14:00counterintuitive which is a firm is not
14:02gonna go away it's gonna actually look
14:03the same as it does now then but when we
14:05talk about the cost of the transaction
14:07cost of all this coordination and why
14:09you need management or even you have
14:10this incomplete contract theory and
14:12people you can't predict every
14:14contingency what if we have an
14:15algorithmic AI who's able to then
14:18account for every one of those
14:19contingencies versus like we don't base
14:22in our theories right on what we know
14:23already we don't know how it's gonna
14:24play out in the future
14:25well then we'll never say never yeah and
14:27yet if there's an AI that has magical
14:30properties that we can't imagine you
14:32know all bets are off of yes but we're
14:34talking about a world right now where
14:36the blockchain and related technologies
14:38allowing radical decentralization of
14:40lots of types of decisions and that's
14:42really important it's changing bring a
14:44lot of new opportunities but all it
14:46doesn't change everything and there are
14:47still some core things like this concept
14:49of incomplete contracts anything that's
14:51not explicit that you can't write down
14:53you maybe you can't anticipate and maybe
14:54the current ai's can't anticipate then
14:57you those are the residual and that's
14:59where ownership match that leads to
15:01something like company being an enduring
15:04part of the economics I would make it
15:06more basic which is human nature but
15:08people at the end of the day systems of
15:10networks that are online or in a company
15:12or any other form are made up of people
15:14and people are fallible right and
15:16emotional raksha's in the Bitcoin
15:20community yes and great and the civil
15:23war going on there okay one reason you
15:26have management is to say gang
15:27oh this way and not that way and
15:29disagree and then commit as opposed to
15:32disagree and then disagree and we all
15:35have bounded rationality Friedrich Hayek
15:37called it was the Fela conceit the idea
15:38that we could plan everything in
15:40excruciating detail the world is far too
15:42complicated for any one person or any
15:44one group of people to do that and
15:46there's even kind of a Red Queen
15:47phenomenon that the more sophisticated
15:48you are the more sophisticated your
15:50competitors are your customers are your
15:52suppliers are why is it called a Red
15:53Queen vanilla so if you get more
16:02sophisticated all those other parties
16:04are getting more sophisticated too
16:05you're not going to be able to
16:06completely anticipate what they all do
16:07because they'll be even whatever but
16:09think about how crazy this is I'm hired
16:11brought up the term the fatal conceit to
16:13demolish this idea that we could
16:15centrally plan an economy and at the
16:18time when a lot of intellectuals in the
16:20West were excited about soviet-style
16:21central planning Hayek wrote one paper
16:23and just demolished it there's a almost
16:25180 degree reverse perhaps fatal conceit
16:28going on among the fans of radical
16:31decentralization as opposed to radical
16:33separation so you're saying the same
16:34phenomena that play just in the Jews
16:35read something to you because I was
16:37gonna say there's now some claims out
16:39there that the power of simulation has
16:42gotten so good that we might be able to
16:44actually move to that that fatal conceit
16:46of being able to centrally plan an
16:48economy because of all these data
16:49machine learning you know sort of
16:51signals and whatnot so Alan Greenspan of
16:53all people I asked him about computers
16:55the ability to stimulate the economy and
16:57he was a chairman of the Federal Reserve
16:58you know set interest rates and
16:59everything and he said well yeah we can
17:03understand a lot better but all the
17:05companies are reacting that much faster
17:07as well and so it's exactly this Red
17:09Queen phenomenon that however much they
17:12the Federal Reserve advanced each
17:14company advanced all the other guys are
17:16doing the same thing if you could freeze
17:17the rest of the world and you were the
17:18only party that had access to cloud
17:20computing and Moore's Law etc yeah maybe
17:23you could stay one to ten steps ahead of
17:26them but that's not the way the world
17:27works there's a great story from the
17:28early days of AI on this fatal conceit
17:30idea which is in the late 80s Japan
17:33tried to organize their entire
17:34industrial policy around creating
17:37artificial intelligence for the
17:39generation that fifth generation have
17:41now built around expert systems
17:42optimized all the way down in the
17:44silicon so you can imagine silicon
17:45optimized for Lisp right yes so that we
17:48could build that's just imagine it was a
17:50complete failure precisely to this idea
17:53of you actually can't plan anything
17:55what happened out of the 80s was more
17:57the rise of client-server computing and
17:59Microsoft Windows nobody anticipated
18:00that yeah right and the idea that we're
18:03out of that world because of Moore's law
18:05because we have much more computational
18:07power now I find that ludicrous well
18:09tell me why if we have this accelerating
18:11growing fast happening thing and I don't
18:13want to make it a crutch to say like we
18:15can't predict the future blah blah blah
18:16we already know that but yeah what why
18:18not it's a lot of things that were tried
18:20before it didn't work because it was a
18:21wrong time why wouldn't that be possible
18:22now I can't simulation work there yeah I
18:25mean speaking as an investor you know
18:28it's trying to predict the future and
18:29often gets it wrong as you should you
18:31know yeah it's hard to imagine a better
18:34system than the one we have which is
18:36let's spend a little money and run a ton
18:38of experiment exactly on businesses to
18:40figure out what people want because
18:41until you have it in the world you're
18:43not sure what that people want and
18:45that's not called simulation in the face
18:46of massive computational power
18:48that's called entrepreneurship and
18:50capitalism the data is going the
18:54opposite direction which you're seeing
18:55less planning and predicting less 5-year
18:58plans but we do this and a lot more
18:59experimenting testing fail-fast that
19:03seems to be Amala works a lot better but
19:05the other thing I was gonna say is like
19:06I look at countries like China and
19:07they're incredibly coordinated efforts
19:10and while I agree that past central
19:12industrial planning efforts have failed
19:14for various reasons I don't know I think
19:17there might be something to it this time
19:18I just want to make sure you guys really
19:19disillusion me of that because let help
19:21me let it go then our colleagues durin s
19:23mo glory Gems Robinson wrote this
19:24amazing book called why nations fail
19:26okay and their and their answer was very
19:28was really straightforward nations fail
19:30because they have extract extractive
19:32institute of institutions were an elite
19:34grabs power and they just surround the
19:36value of model and you know that's why
19:38companies that filled that's a good
19:43analogy yeah right the next book and
19:45their descendants and they just make
19:47sure the vapor over the rules of the
19:49game to benefit themselves that's as
19:50opposed to inclusive institutions where
19:52you have an honest shot of making the
19:55in capital now which one is China they
19:57took big steps in the direction of India
19:59by opening up to a market economy would
20:01we call that authoritarian state one of
20:04actually inclusive institutions I would
20:06not I think that's a legitimate thing to
20:08say okay so just going back to this idea
20:09of extractive institutions so I do think
20:11it's interesting that there are now
20:12networks that are coming up that are
20:15letting people participate differently
20:16as owners for sure in different ways and
20:19that is where I think this topic of icos
20:20and token launches is really interesting
20:22part of the power is as Hayek would have
20:24said is that you decentralize some of
20:26the local knowledge they have
20:28information that nobody else has and
20:29that's right exactly if you can move the
20:34decision rights to where that knowledge
20:36is you're gonna be better off and one of
20:37the great things that technology is
20:38allowed us to do is move around decision
20:40rights move around ownership so
20:42hopefully if you do it right you get a
20:43better match between the incentives and
20:45the decision rights the whole entire
20:47third section of our book is about this
20:49rebalancing necessary between the core
20:51institutions of a company and the crowd
20:53available over the internet now how much
20:55more room there's very likely ahead of
20:57us yeah with crowdfunding with
20:59crowdsourcing with different ways to tap
21:02into what people can do to give them an
21:04ownership stake to get them bought in
21:06and pointed in the right direction have
21:08we scratched the surface of that let's
21:10talk a little bit about Joy's law that
21:11no matter what company you work for most
21:13of the smart people the world work for
21:14somebody else I used to be limited what
21:17you could do about that because there's
21:18only so far you can communicate but now
21:20for the first time in history a majority
21:22of the world's people are connected with
21:24a digital network so they can access all
21:26the world's knowledge and part of it
21:28isn't necessary that they're smart or
21:29else out there part of it just comes
21:31from the raw variety that the diversity
21:34the variance within a company you tend
21:37to have people who are like-minded
21:38they've trained the same way that's who
21:40they get hired and maybe the way to
21:41solve a problem is with an entirely
21:43different approach and that may be
21:45somebody from a different culture a
21:47different way of looking at the world
21:48and you're very unlikely to have that
21:50diversity inside of a company it works
21:51against it but if you can find a way to
21:53tap into it one of our colleagues Karim
21:56Lakhani he's now at Harvard Business
21:56School it was a PhD student at MIT has
22:00done just case study after case study of
22:02examples we're tapping into the crowd
22:04blew away what companies were able to do
22:06internally he worked with the now
22:08Institutes of Health to try to improve
22:09the speed and accuracy of sequencing
22:12human white blood cell genomes which are
22:14really complicated but important to
22:15sequence the National Institutes of
22:17Health which I would call the core of
22:19the medical established core in the
22:20sense of core versus crowd they had an
22:22algorithm that could do a run in about
22:24four hours with about 70% accuracy there
22:26was a faculty member at Harvard med
22:28school who made a big improvement to
22:29that algorithm he developed one that got
22:32them up to about seventy five percent
22:34accuracy kareem then worked with the nih
22:36and top coder to make this an
22:38algorithmic challenge and open up to the
22:40crowd and the best solutions got down to
22:42about 10 seconds and about 80% accuracy
22:45four hours to 10 seconds so we called up
22:48cream and he is about average when I
22:50when I run a crowdsourcing tournament
22:52this is the magnitude of improvement I
22:53expect to see well the last part of that
22:55story that that continued to blow us
22:57away is that they interviewed the best
22:59performers that we submitted the top
23:01performing algorithms none of them had a
23:03life sciences background there was not a
23:06biologist among them so crowds and
23:08prediction markets are similar what's a
23:10different I would say a prediction
23:11market is one way to harness crowd right
23:13markets do a really good job overall on
23:16aggregating knowledge markets tap into
23:19the crowd google taps into the crowd
23:21because their search algorithm basically
23:23exploits the link structure that all of
23:24us can wrap it where we make pages there
23:26are lots of ways of tapping into the
23:28crowd but being clever about how to
23:30reach them motivate them aggregate them
23:32it's a lot of work to be done on that
23:34let's talk about the nature of work
23:35because I think what people do in that
23:37firm either inside or outside probably
23:40changes a lot so we have this idea that
23:43human decision-making is sort of
23:45fundamentally flawed in that like
23:46there's biases that you bring to your
23:48decision making but you don't even
23:49understand right so when you're thinking
23:50it through you're still gonna make the
23:52same mistake you guys if you don't
23:53understand that you have that bias after
23:54all we're walking you through your
23:56decision-making process is your brain
23:57that came off that flawed
23:58decision-making process in the first
24:00place right it's not gonna catch its own
24:01mistakes right so it's a permanent blind
24:03spot and by contrast you would have sort
24:06of assumed that a machine learning
24:08algorithm trained with a carefully
24:10selected broad set of data sets will
24:13have a decision-making efficiency or
24:16effectiveness better than you know
24:18flawed humans so if that's the case what
24:21people in firms do like how do you
24:23prepare for this world where there's
24:25gonna be machine learning algorithms
24:26that can in general make pretty good
24:28decisions and then there's this idea
24:30that like maybe the talent is better
24:32outside your company that inside your
24:34company so what should you do should you
24:35join a company it's just breathtaking
24:38but it is far far from being AI complete
24:42being able to do everything that humans
24:43can do there's a certain class of
24:44problems that it's kicking butt on but
24:46that's a tiny sliver of what human
24:48decision-making is even just defining
24:50what the problem is and exactly what
24:52needs to be done that's half the battle
24:53but you need humans to do that there's a
24:55quote that we had from the book from
24:56Picasso computers are useless all they
24:59do is give you answers I was a little
25:01shocked when I pricasso was alive when
25:02computers actually said that we were
25:05investigated that one but he well he's a
25:09really guy in a lot of different ways
25:11and obviously he didn't know much about
25:12the latest neural networks today but but
25:15his understanding was spot-on that
25:18simply giving the answer isn't
25:19necessarily the most interesting
25:21important part of solving Kevin Kelley
25:23actually makes this argument about
25:24inevitable get him on the podcast that
25:26the number one job of the future for
25:28humans that humans preserve and this is
25:30I think what you're getting at is that
25:31we ask the questions and computers
25:33answer but I should just agree with the
25:36that a little bit because I'm seeing a
25:38new class of generative AI that makes me
25:41wonder if they're gonna be asking new
25:43questions that make us want to answer
25:45differently there's all kinds of
25:47our brains are made of a delay and so
25:49our computers so I'm not going to say
25:51that that there's some things that they
25:52just can never touch but but I agree
25:54which is that on average our wetware is
25:57amazing but it's got a host of bugs and
26:00biases and glitches in it that machine
26:02learning systems and properly configured
26:04algorithms in general do not have so if
26:06you could only pick one of those two
26:08entities to help you the good news is
26:10that's a false choice we don't have to
26:12make that choice and and I think the art
26:14going forward is being more clear about
26:18what are we actually good at versus what
26:20the machines are actually good at the
26:22happy news is that they have very
26:23different failure modes yeah and I think
26:25that's exactly the key point it's a
26:27matter of how we can leverage each of
26:28them because machines have biases as
26:30well yeah thumbs are biased by
26:32definition it's not just somebody
26:33designed them but all
26:34the training data that they get I mean
26:36if you if you decide to give loans based
26:38on all the loans that have been approved
26:39or rejected in the past that could have
26:41some bias he's built into it
26:43and some of these neural nets could have
26:44billions of connections getting it to
26:47short out how exactly not gonna be one
26:48of those says okay discriminate against
26:50women but there may be some very subtle
26:52interactions that are hard to anticipate
26:53or explain that said at least the
26:56machines can be tested and improved and
26:58it's often easier to do that with human
27:00than it is with humans we are really
27:02resistant to having our wet we're
27:03tweaked we really just don't like to be
27:05told glitchy and here's the fix and just
27:08go do that now it's really really hard
27:14to do there's a concept from linguistics
27:16that that I find incredibly helpful for
27:18helping understand but what I think some
27:19of the most durable human advantage in a
27:22world full of machines will be and it's
27:23a concept called the intuition of the
27:25native speaker and what they mean by
27:27that is if I look at any English
27:29language sentence I can immediately tell
27:31if it's grammatically perfect or not you
27:33just hear it in your hang up we are the
27:35native speakers of the human created
27:37world of computers are doing this as
27:38their second language we I believe we
27:40have a massive advantage we are the
27:42native speakers about this reality
27:44around us rather than trying to build a
27:46system that does everything from soup to
27:47nuts you get some kind of a division of
27:49labor Sebastien's run describe assistant
27:51to us recently there was just
27:52fascinating he's at Udacity and a lot of
28:10but you know Sebastian described how
28:13they get incoming traffic in their chat
28:15rooms of people asking about their
28:17offerings they decided let's take this
28:19data we'll see which of these
28:20conversations lead to sales which ones
28:22don't lead to sales and label them that
28:24way and then train a neural net about
28:27which caught which replies were
28:28successful and then what they took with
28:30those replies I didn't try to have a
28:31standalone chat bot that nine then
28:33talked to customers instead they had the
28:36human sales people keep interacting but
28:38when they saw one of these more common
28:40error modes they would gently prompt the
28:43not-so-good salesperson you know maybe
28:45you want to give them this set of
28:46answers or this other set so that's kind
28:48an idea absolutely because there's a
28:51long tail of other questions that the
28:52bots had no clue what they were about so
28:55it could help with the most common sets
28:56of queries and this is I think a pattern
28:58that you see lots and lots you see it
29:00among radiologists you combine the two
29:02and you end up having fewer false
29:03positives and fewer false negatives yeah
29:05I love this idea of sort of machines and
29:07humans working together and I think it's
29:09only a matter of time before we walk
29:11into a doctor's office or a lawyer's
29:12office where that isn't the fundamental
29:14interaction and we'll just be horrified
29:16like where's your AI companion why are
29:18you trying to do this yourself with your
29:20biases look I couldn't agree more
29:22why on earth would I expect my GP who's
29:25a really good doctor to be on top of the
29:28accumulated mass of human medical
29:30knowledge and keeping up-to-date with
29:31the latest developments and all the
29:32fields that might relate to what I walk
29:34in the door with that's an absurd
29:35request on a human being now I want that
29:38person to be well trained even more I
29:40want them to be able to empathize with
29:42me and get me to go along with the
29:44course of treatment and get me to buy
29:45into what's going on because that AI in
29:47the background that's got access to my
29:49tests and my lab results again assess
29:50the jaundice in my skin and you know how
29:52white the sclera of my eyes are that's
29:55going to be the diagnostic expert in the
29:57not AI not in the backroom I want it in
30:01the room with me when I'm doing this
30:04it's a theme that comes up again and
30:06again we talk about mind and machine
30:08product and platform core and crowd and
30:11we don't give people the mistaken idea
30:12that you just cross off the first words
30:15of those lists and only do the second
30:16one the mantra that I've learned is that
30:19tech progress rewrites the business
30:20PlayBook and what what the two of us
30:22believe is that the way the PlayBook is
30:24being rewritten these days is in favor
30:26of machines platforms and crowds so the
30:29balance needs to shift more in those
30:31directions so the PlayBook is in favor
30:33of machine platform and crowds as
30:35opposed to my product and core right so
30:39each of them is really a dichotomy and
30:41the most successful systems are rarely
30:43all one or all of us right a couple of
30:46threads that we didn't get to pool one
30:47question I had when we were talking
30:48about not all the talent is inside your
30:50company and you know a lot of people
30:52talk about open innovation as a way to
30:54kind of get around that like yes open
30:55source communities etc what does that
30:57mean for business concretely what does
30:59that mean for core in the way that you
31:01defining core and and deploying the
31:03power of the crowd like does a business
31:04whose main strategy is their core
31:06business does that mean that all their
31:08innovation is now outsourced to the
31:09crowd or is it the other direction
31:11what's the ideal framework I think way
31:12too many even successful companies today
31:14are over weighting their core they're
31:16probably spending too much of their
31:17total budget on it way too much of their
31:19managerial bandwidth on it and at the
31:22risk of being a little bit cute I think
31:23a core capability for most organizations
31:25going forward is going to be interfacing
31:28with the crowd harnessing its energy and
31:30its abilities and then finding out how
31:32to bring that way into the organization
31:35without setting off all kinds of
31:36antibodies and resistance and and
31:39it's part of the same lesson we learned
31:40from the mind machine trade-off is that
31:42defining the problem is important
31:43whether you define it for the machine or
31:45whether you define it for the crowd
31:47understanding what the problem is you're
31:48really trying to solve if you can define
31:50it well enough then these contests work
31:51great the contests don't work great if
31:53you just say hey guys you know tell us
31:55stuff right tell yeah yeah you give them
31:58a really price ever worked for a reason
32:04so then there's another question though
32:06for me which is um if you take the
32:09innovation from the crowd and you said
32:10earlier that there's this escalating
32:11effect where everyone has access to the
32:12same tools and they're they're all
32:14catching up really fast with each other
32:15and you can't you're always a red queen
32:17right you have to run faster than
32:18everybody else but if everyone has
32:19access the same crowd how does the
32:21company get advantage in this space but
32:23then honestly it's a matter of where
32:25your leadership throws its attention how
32:27firmly you believe in these new kinds of
32:29energies out there not how willing you
32:32are to open the checkbook and spend
32:33money on technology but how willing you
32:35are and forgive me to open up your
32:37brains and rethink your business model
32:38in the face of this craziness and use
32:41these tools more effectively just like
32:42who can use the cloud more effectively
32:44yeah I mean it's a matter it's like what
32:46I always used to just set a weapons out
32:47there and some people have a better
32:48strategy some people have better
32:50techniques the companies that failed
32:51during the transition from steam power
32:54over to electric power almost none of
32:56them failed because they refused to
32:58invest in electricity that was not the
32:59failure mode right the failure mode was
33:01they refused to rethink what a factory
33:03could be and they refuse how to really
33:05absorb and they refused seriously the
33:07idea of an overhead crane or an assembly
33:11your belt I'm just thinking about the
33:13the statistic when you said this thing
33:15about this antibody that organizations
33:17naturally have which is essentially they
33:19just immediately reject this non
33:21invented here syndrome basically buddies
33:23are the best news possible for your
33:24industry but the research has shown over
33:26and over and over again that it is
33:28practically impossible for big companies
33:30to absorb start up successfully unless
33:32they keep them isolated and one of the
33:33questions I have is the next follow up
33:35basically what happens when you when you
33:37leverage this crowd how do you then
33:38really bring them into the company so
33:40that you don't have these antibodies do
33:41you have any concrete advice I would
33:43look to do that in some of the most
33:44forward-thinking parts of the
33:45organization as Eric said in parts of
33:47the organization where the problem can
33:49be most clearly defined and where you've
33:51got people at the helm of that part of
33:53the organization who are willing to take
33:55the innovation the algorithm whatever
33:56that the crowd comes up with and slot
33:58that into the work of the organization
34:00there's a role for the core to be able
34:03to define that in our world a perfect
34:04example of the core leveraging cloud is
34:06the classic enterprise software company
34:08yeah so in the old days basically you
34:10wrote software it was all proprietary
34:12you won Gartner Magic Quadrant then you
34:14sent your Rolex wearing direct
34:16salesperson to go sell it to them but
34:17the new enterprise company is let me
34:20create an open-source project yeah let
34:22me get a lot of contributors let me get
34:24contributors to get downloads and that's
34:28my path to market right right so the
34:31core needs to be there cuz they got it
34:33what's the project exactly what problem
34:35are we trying to solve but the crowd
34:37comes in and into it to basically lend
34:39legitimacy and support and enthusiasm
34:42for the project so if you can be that
34:44scarce compliment to the abundant crowd
34:46you can create a lot of value then you
34:48become the linchpin that is create that
34:51it's capturing a lot of the value as
34:53well as creating it ultimately we are in
34:55an economy of creative destruction and
34:57one of the strengths of the United
34:59States and other dynamic economies is
35:01that we had this constant turnover and
35:03one of the things that discourages us is
35:05that there's actually less fewer
35:08startups less innovation fewer young
35:10firms in America today the new or 10 or
35:1220 absolutely we are all for trying to
35:17make the bigger companies more nimble
35:19they may understand this but the bigger
35:21way that come that the economy innovates
35:25innovative set of new startups that rise
35:28and adopt some of the new technologies
35:31yeah I've got to have both and we'd like
35:33to see progress on both dimensions
35:34that's great thank you for joining the
35:36a6 and Z podcast thanks for having us on
35:38this is fantastic it's a real pleasure