Pod Insights a16z Podcast a16z Podcast | Companies, Networks, Crowds About this episode Is a network -- whether a crowd or blockchain-based entity -- going to replace the firm anytime soon? Not yet, argue Andrew McAfee and Erik Brynjolfsson in the new book Machine, Platform, Crowd. But that title is a bit misleading, because the real questions most companies and people wrestle with are more "machine vs. mind", "platform vs. product", and "crowd vs. core". They're really a set of dichotomies.
Yet the most successful systems are rarely all one or all the other. So how then do companies make choices, tradeoffs in designing products between humans and machines, whether it's sales people vs. chatbots, or doctors vs. AIs? How can companies combine the fundamental building blocks of businesses -- such as network effects, platforms, crowds, and more -- in a way that lets them get ahead on the chessboard against the Red Queen? And then finally, at a macro level, how do we plan for the future without falling for the "fatal conceit" (which has now, arguably flipped from radical centralization to radical decentralization) ... and just run a ton of experiments to get there?
We (Frank Chen and Sonal Chokshi) discuss all this and more with Brynjolfsson and McAfee, who also founded MIT's Initiative on the Global Economy -- and previously wrote the popular The Second Machine Age and Race Against the Machine. Maybe there's a better way to stay ahead without having to run faster and faster just to stay in place like Alice in a tech Wonderland.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Sonal? Hi, everyone. Welcome to the A6 and Z Podcast. I'm Sonu. Today, we're doing one of our book podcasts around the new book just out, Machine Platform Crowd. The authors previously wrote the popular book, The Second Machine Age, and before that, their book was Race Against the Machine. Sensing a bit of a theme here. So in this episode, we cover those themes, first, starting with a bit of Econ 101 around network effects, complements, and other key concepts. Then we discuss how this all plays out organizationally, especially given trends like machine learning, blockchain, and crowds, and tackle the tricky question of whether networks can replace the firm? And where are we in the classic question around the future of the firm? And finally, what can companies do more concretely? Frank Chen joins the conversation in between as well to share his perspective
0:44 Sonal? on what he sees, given his role as Head of Investing and Research at A6 and Z. But our main guests on the episode, both from MIT, are Erik Brynjolfsson and Andrew McAfee. Who I'm gonna call Andy. Is that okay?
0:56 Speaker 01 Otherwise, I'm gonna mistake you for my mom.
0:58 Sonal? Good. I don't want to be mistaken for your mom. I'm way too young to be your mom. We kind of go way back in the sense that I met you years ago.
1:03 Speaker 01 Not as far back as I go with my mom.
1:05 Sonal? No, no.
1:06 Speaker 01 You and I will do Andy, okay?
1:07 Sonal? All right, good. We're doing Andy. So this is your third book together. The real thrust of your work is that this is unprecedented in the speed at which we're changing and what the effects are. And I think a great theme for this conversation is to sort of break down how those changes are going to play out and where they're happening.
1:22 Speaker 02 Yeah, yeah. Well, but let me just push back on that first part a little bit because in Silicon Valley, everybody agrees with that. And, and what, and we agree with it.
1:31 Speaker 01 That's right.
1:31 Speaker 02 To be clear, but we were reading people who didn't. One of the things that guys writing our first book, Race Against the Machine, was there were people who were talking about, quote, "The great stagnation," and how there were no good inventions anymore. Nothing good was invented. In particular, Tyler Cowen, he was spot on that median income had been stagnating. And that was kind of troubling for us because, you know, I had been taught the slogan that productivity isn't everything, but in the long run, if we just have tech progress, everything else takes care of itself. And when Tyler showed us that evidence, we were like, oh, this is a real problem. But we refused to give up on the idea that technology was just doing amazing things.
2:08 Speaker 01 We weren't gonna let a little evidence get in the way of our beliefs, for God's sake.
2:10 Sonal? No way. No, damn it. We're not letting that happen.
2:12 Speaker 02 Fortunately, we figured out a way out of it. So, and, and the way out of it is that even though technology is making the pie bigger, there's no economic law that everyone's going to benefit from it. It's possible for some people to get left behind. Now, to be clear, that's not what happened for most of the past 200 years, but the past 10, 20 years. There really have been more and more people being left behind. And so you could get stagnating median incomes even as some people, maybe in the top 1%, got fabulously wealthy. And that helped us reconcile these different perspectives. And it led to a whole broader set of discussions about the way that organizations and society and business processes aren't keeping up with these amazing technologies and some of the dysfunctions that can create and some of the opportunities it can create.
2:57 Sonal? So what are some of the big, well, I think we should talk about, break down the fundamental building blocks of a lot of the arguments that you make throughout your work. So let's talk about networks. And one of the biggest questions I had reading your book was, is a network going to displace the firm in the future? We talk a lot about network effects in our business.
3:13 Speaker 02 So networks, sometimes economists call them demand-side economies of scale. And it's basically the idea that a product or service becomes more valuable the more other people that are using that product or service. The classic examples, you know, a telephone or fax machine, WhatsApp, Facebook. And you could have supply-side economies of scale, just to distinguish that. That's when the costs get lower as more people use it. And both of these things lead to the big companies winning.
3:41 Sonal? And just for shorthand, we tend to describe supply-side economies of scale as just economies of scale.
3:46 Speaker 02 Yeah.
3:46 Sonal? And demand-side economies of scale as network effects.
3:48 Speaker 02 That's the more common way.
3:49 Sonal? More generically.
3:49 Speaker 02 And we use both sets of terminology. It's sometimes useful to talk about supply-side, demand-side, because a lot of the economics become more intuitive once you understand that there's the demand side and the supply side, and they both can get better as you get bigger. And then to get, add a little more layer of subtlety to it, you could have traditional single-sided network effects, like other people using the same telephone, or you can have two-sided network.
4:11 Sonal? Two-sided.
4:12 Speaker 02 And that's really the platform revolution. A lot of that has been triggered by the growth of so-called two-sided networks. And the idea there is that it not necessarily is people using the same product as you, but it could be people on the other side using a different product. So like drivers and users are using slightly different apps. And me, as a user, I don't really benefit when more users are also using it.
4:31 Sonal? No, right.
4:32 Speaker 02 I want more drivers, and the drivers want more users. So you care about the people on the other side of the network.
4:37 Sonal? Except when you're pooling. Because then you do care.
4:39 Speaker 02 That's right.
4:40 Sonal? And that's a case where you do want them on the same side.
4:42 Speaker 02 Exactly. And then to make it even more complicated, you can have two-sided and one side at the same time. You can have economies of scale. So you can layer them. You mentioned the word building block. Let's start with these primitives and then you can start combining them in different ways.
4:53 Speaker 01 This really starts to turn into three-dimensional chess because the right way to think about the app ecosystem in Apple is not any kind of one or two-sided network. It's an N-sided network.
5:03 Sonal? Multi-sided.
5:03 Speaker 01 Lots and lots of different groups of people who value things on the other side, but we don't decide what the sides are in advance. And we let the self-selection happen. And you just watch the vortex form around that ecosystem. And the only way to understand that is by doing what Eric just did. Start with network effects. One-sided, two-sided. Two goes to N. Value goes to N.
5:26 Sonal? Okay, that's great. And then let's probe on one big thing, which is we talk about network effects, but let's quickly define complements in this because that's a term that's frequently used and I think it has a lot of misconceptions around it.
5:34 Speaker 02 Sure. One of the key economic building blocks that we talk about is complements. And complement is a very simple concept. It's the idea that one product is more valuable in the presence of another. So my left shoe is more valuable if I also have my right shoe.
5:46 Sonal? Well, that's an obvious example.
5:47 Speaker 02 Yeah. Software is more valuable with the right hardware.
5:50 Sonal? Ah, right.
5:51 Speaker 02 And so complements can be physical. They can even be organizational. Well, so you may have a system that taps into the crowd. That's more valuable when you have a global internet that allows you to do that.
6:04 Sonal? Right.
6:05 Speaker 02 So you can have organizational or technical or physical complements. And you can sell products that are complementary to each other.
6:13 Sonal? Razor blade is a good example.
6:13 Speaker 02 Yeah, razors and blades. And sometimes when you have products that are complementary to one another, it actually can be profitable to give one away to increase the demand for the other one. So people famously gave away razors to sell blades. And this can interact with the network effects and the scale economies. It's not a good strategy if you don't have those other things. One of the things that, you know, makes us tear our hair out. Is that, you know, when MBA students say, oh yeah, we'll just give it away. Like, where is the underlying strategy?
6:40 Sonal? Oh, so if you're just saying like, I need to do freemium without really understanding the or underlying strategy for what we're trying to accomplish.
6:45 Speaker 02 Exactly.
6:45 Speaker 01 Disastrous. And compliments are weirdly subtle. And Eric just explained-
6:49 Sonal? This is why I want to ask about it, because it's a very nuanced concept.
6:51 Speaker 01 Eric just explained them super clearly. The Econ 101 example that I always fall back on is hamburger meat and hamburger buns. And so if the price of hamburger meat goes down, demand for buns is going to go up even if the price of buns doesn't change. That's the key thing. The price of one good can stay the same and demand for it will go up. The compliments are so tricky that they actually tripped up Steve Jobs really badly. This is not lore, this is fact. He did not want to open up the App Store to any outside developers. He thought he had to maintain super tight control over that digital environment. And when the iPhone first released, it did not have any external apps on it. He fought boardroom battles for about a year with people who said, no, you need to open this up.
7:32 Sonal? What made him cave?
7:34 Speaker 01 Pressure from really smart people inside and outside the company, people on his board and executives at the company. What he didn't fully realize is that if you open up the App Store and you curate successfully, you have just opened the door to this massive number of compliments, each one of which is going to nudge out demand for the iPhone. And even if each one only nudges that demand outward-
7:54 Sonal? Like 99 cents worth.
7:55 Speaker 01 Yeah.
7:56 Speaker 02 Or no, even less.
7:56 Speaker 01 Even less.
7:57 Speaker 02 Just to be clear, we're not talking about the literal money that they get from the-
8:00 Sonal? Yeah, I know, I know.
8:00 Speaker 02 Yeah, we're talking about the fact that it makes-
8:02 Sonal? It's a relationship that makes the entire-
8:03 Speaker 02 It makes the phone-
8:04 Sonal? It makes people want the phone more. I remember the early days of the iPhone. I never, I still don't have an iPhone, I have an Android, but I still remember to this day, the first thing people would say, and I'm like, I don't really like the iPhone that much. And they're like, oh, it's not about the phone, it's all about the app.
8:14 Speaker 02 Yeah.
8:15 Sonal? It's all about the app. So it was their line all the time.
8:16 Speaker 02 Angry Birds.
8:17 Speaker 01 Yep. And the only way to understand the value of opening up that App Store is to understand that you are unleashing this tidal wave of complimentary goods that were priced at all different price points, including zero, which is awesome. So zero is a really great price, but the more fundamental thing, I think, is that it shifted out demand. It nudged demand upward for the other complimentary good, the iPhone itself. And once you, once you grok into that, then you say, oh, I gotta find all kinds of different ways to do this and play three dimensional chess with my platform.
8:46 Sonal? Is the querrillary of all this that clothes will never win, then?
8:49 Speaker 01 No, it's not nearly as simple as that.
8:52 Speaker 02 But it does show you that if you can leverage these complements, you can create not just a one-time win, but a whole ecosystem because Andy's story turns into a virtuous cycle.
9:01 Sonal? Right.
9:01 Speaker 02 Where the more demand for the iPhone. Exactly. It's a flywheel. And so that can work very well. But it's not like you always open up or you always.
9:10 Sonal? Right. Because I would say a lot of the winners until now have been closed companies.
9:13 Speaker 02 Yeah, absolutely.
9:14 Speaker 01 Yeah. And Apple was comparatively closed against Google and the Android ecosystem. One of the things we say is there is not one right answer. There is not one recipe that you follow for success with machines, platforms or crowd. There are principles.
9:27 Speaker 02 There are principles.
9:28 Frank Chen? And for entrepreneurs who are listening, understanding complements and the way the people who are creating these ecosystems that have complements is super important. So we've been talking about complements where the more phone, the more apps in the App Store, the more attractive an iPhone. So think about that when you're thinking about development tools for these platforms.
9:47 Speaker 01 Yeah.
9:48 Frank Chen? Xcode, Visual Studio are so important to Microsoft and Apple because they're creating these complements and therefore the desirability for their iPhone. That's where they make all their money. So if you think, hey, I'm going to create a better development tool. I'm going to create a better Xcode. Like, think again because Apple is going to spend as much money as it needs to defend the complement universe.
10:08 Sonal? The question that comes to mind for me is what this means for companies.
10:11 Frank Chen? So one thing, the conventional wisdom now is we fund companies whose defensibility is a network effect. In other words, we're in Lyft and Airbnb precisely because once you have all the hosts, you're going to get all of the renters. Right? And so one thing to think about is maybe in the future, even the firm that creates the network effect gets decentralized. Who needs a firm? Why don't people just come together and we'll create the right set of incentives for the network to behave? So you can imagine in eBay where there is no company. There's just a network coming together with the right set of incentives.
10:43 Speaker 01 That was how we wound up the book is trying to grapple honestly with this question of in the universe that can be turbocharged by the fact that everyone's got a device, that we've got this completely decentralized cryptocurrency system you could pay people with, that we've got these technologies of radical decentralization.
11:01 Frank Chen? Like the blockchain.
11:02 Speaker 01 Like blockchain.
11:02 Speaker 02 Public distributed ledger.
11:04 Speaker 01 Yeah.
11:04 Speaker 02 Right.
11:04 Speaker 01 Every transaction, everybody sees.
11:06 Speaker 02 Distributed autonomous organization.
11:07 Speaker 01 Contracts and code into those things. You can do a lot of the stuff that we used to need a company for. The question gets teed up. Are we still going to have companies in the future? And as Eric and I started to think about all the stuff that we'd learned and tried to digest, our answer was an unequivocal yes. And the main reason for that is that ownership of a thing matters. Simply because almost, well, every economist I think that we've talked to would agree that you can never write a complete contract that will specify exactly what everybody's going to do in every
11:34 Sonal? Every possible contingency cannot be accounted for.
11:37 Speaker 01 And the reason for a firm is it gets to make the decisions that are not contractually specified elsewhere. And it gets all the value that's not, that's not apportioned elsewhere in the network.
11:49 Speaker 02 And it starts with Ronald Coase.
11:51 Sonal? Of course, the classic nature of the firm paper, what, 1937 or something?
11:55 Speaker 01 He's a geek hero. He was nine years old when he wrote that. He was, he was in his twenties or something.
12:00 Sonal? Did you say 90 or 90?
12:01 Speaker 01 No, he was, he was, he was in his twenties.
12:02 Speaker 02 Yeah, he was in his twenties, 26, I think he was. But then, and then more recently, Oliver Hart, who was my thesis advisor, and Bankt Holmstrom, one of our other colleagues at MIT, elaborate on that, as Andy was saying, with this so-called incomplete contracts there. One of the blinders that a lot of people, especially technologists, have is they say, hey, we can just write everything down. Engineering mindset, we'll write a complete contract that covers all contingencies. And the reality is, is the world is just too complicated to cover every possible contingency. So when you own it, when you own a car, you can sell that to someone else and whoever owns the car gets to have all what are called the residual rights of control. Everything that's not specified in a contract. You want to change the color of it? That's what ownership means.
12:47 Speaker 02 And ultimately, you take that to the level of firm. A firm is an aggregator of a bunch of assets and owns certain things. And that means, that gives them certain power. It gives them certain incentives of how those objects are used.
12:59 Speaker 01 As Eric and I were trying to reason our way through this and convince ourselves to one view of the world here, this amazing real life experiment happened, which was the DAO.
13:06 Sonal? Yeah. And let's do a quick terminology thing. When you say the DAO, you mean the corporation that was formed, but. That's very different than a DAO, which is a decentralized autonomous organization or decentralized autonomous corporation.
13:17 Speaker 01 Yes, there's a thing called
13:18 Sonal? The DAO, the entity.
13:19 Speaker 01 This is a thing called the DAO.
13:20 Sonal? The proper noun, not the generic noun.
13:21 Speaker 01 The DAO, which was intended to be a completely owner-free, completely decentralized organization along the lines that you just described. And it got hacked. Somebody found out how to treat it like an ATM, essentially. So to the extent there was a group of people kind of behind it, they collectively freaked out and thought about what to do. And then they made this fairly autocratic decision, looks a lot like an ownership decision to me, to reset the clock on the entire DAO.
13:47 Speaker 02 They became de facto owners and they asserted those rights in a way.
13:50 Speaker 01 And de novo, they just said, okay, we're going to do this. And if enough of you go along with this, then this is what's going to happen. It was extraordinary for a very decentralized organization. It was kind of heavy.
13:59 Sonal? I mean, I love your saying something counterintuitive, which is a firm is not going to go away. It's going to actually look the same as it does now, then. But when we talk about the cost of the transaction cost of all this coordination and why you need management, or even you have these incomplete contract theory and people, you can't predict every contingency. What if we have an algorithmic AI who's able to then account for every one of those contingencies versus like we don't, we only, we're basing our theories right now on what we know already. We don't know how it's going to play out in the future.
14:25 Speaker 02 Well, we'll never say never. Okay. And yeah, if there's an AI that has magical properties that we can't imagine, you know, all bets are off, of course. But we're talking about a world right now where the blockchain and related technologies allowing radical decentralization of lots of types of decisions. And that's really important. It's changing, creating a lot of new opportunities, but all, it doesn't change everything. And there are still some core things like this concept of incomplete contracts, anything that's not explicit that you can't write down, maybe you can't anticipate, and maybe the current AIs can't anticipate, then you, those are the residual and that's where ownership matters.
15:00 Speaker 01 That leads to something like company. Being an enduring part of the economic landscape.
15:05 Sonal? I would even make it more basic, which is this human nature that people, at the end of the day, systems of networks that are online or in a company or any other form are made up of people and people are fallible.
15:16 Speaker 01 Right.
15:16 Sonal? And they're emotional.
15:17 Speaker 01 And fractious, right?
15:18 Sonal? Yeah, they want to fight.
15:19 Speaker 01 And you look at the breakdown in the Bitcoin community.
15:21 Sonal? Yes, there's forks.
15:23 Speaker 01 And the civil war going on there, okay, one reason you have management is to say, gang, we're going to go this way and not that way and disagree and then commit as opposed to disagree and then disagree.
15:33 Sonal? Yeah.
15:34 Speaker 02 And we all have bounded rationality. Friedrich Hayek called it was the fatal conceit, the idea that we could plan everything in excruciating detail. The world is far too complicated for any one person or any one group of people to do that. And there's even a kind of a Red Queen phenomenon that the more sophisticated you are, the more sophisticated your competitors are, your customers are, your suppliers are.
15:53 Sonal? Why is it called a Red Queen phenomenon?
15:54 Speaker 02 Oh.
15:55 Sonal? From the Victoria's Secret novel?
15:57 Speaker 02 No, from Alice in Wonderland.
15:58 Sonal? Oh, from Alice in Wonderland, of course.
15:59 Speaker 02 You have to run faster and faster just to stay in place.
16:01 Sonal? Gotcha.
16:01 Speaker 02 So if you get more sophisticated, all those other parties are getting more sophisticated too, you're not going to be able to completely anticipate what they all do because they'll be even more clever.
16:09 Speaker 01 But think about how crazy this is. Hayek brought up the term the fatal conceit to demolish this idea that we could centrally plan an economy. And at the time when a lot of intellectuals in the West were excited about Soviet-style central planning, Hayek wrote one paper and just demolished it. There's a almost 180-degree reverse, perhaps fatal conceit going on among the fans of radical decentralization as opposed to radical centralization.
16:33 Sonal? Right, so you're saying the same phenomenon's at play just in a different direction, but I'm going to add something too because I was going to say, there's now some claims out there that the power of simulation has gotten so good that we might be able to actually move to that fatal conceit of being able to centrally plan an economy because of all these data, machine learning, you know, sort of signals and whatnot.
16:52 Speaker 02 So Alan Greenspan, of all people, I asked him about computers and the ability to simulate the economy and he was a chairman of the Federal Reserve, you know, set interest rates and everything. And he said, well, yeah, we can understand a lot, lot better, but all the companies are reacting that much faster as well. And so it's exactly this Red Queen phenomenon that however much they, the Federal Reserve... Reserve advance, each company advance, all the other guys are doing the same thing. If you could freeze the rest of the world and you were the only party that had access to cloud computing and Moore's Law, et cetera. Yeah, maybe you could stay one, two, 10 steps ahead of them, but that's not the way the world works.
17:27 Frank Chen? There's a great story from the early days of AI on this fatal conceit idea, which was in the late 80s, Japan tried to organize their entire industrial policy around creating artificial intelligence.
17:38 Speaker 02 Fifth generation.
17:39 Frank Chen? The fifth generation.
17:40 Sonal? It's like what happened in China right now.
17:41 Frank Chen? Built around expert systems, optimized all the way down into silicon. So you can imagine silicon optimized for Lisp, right? So that we could build.
17:48 Speaker 02 Not just imagine, they built it, did it, right?
17:49 Frank Chen? Yeah. And it was a complete failure precisely to this idea of you actually can't plan anything, right? What happened out of the 80s was more the rise of client server computing and with Microsoft Windows, nobody anticipated that.
18:01 Speaker 02 Yeah.
18:02 Speaker 01 And the idea that we're out of that world because of Moore's Law, because we have much more computational power now, I find that ludicrous.
18:08 Sonal? Well, tell me why, if we have this accelerating, growing, fast happening thing, and I don't want to make it a crutch to say like, we can't predict the future, dot, dot, blah, blah, blah. We already know that. But what, why not? A lot of things that were tried before didn't work because it was the wrong time. Why wouldn't that be possible now? Like, can't simulation work there?
18:24 Frank Chen? Yeah. I mean, speaking as an investor, you know, is trying to predict the future and often gets it wrong.
18:31 Sonal? As you should.
18:31 Frank Chen? You know, it's hard to imagine a better system than the one we have, which is let's spend a little money and run a ton of experiments.
18:39 Speaker 02 Exactly.
18:39 Frank Chen? On businesses to figure out what people want, because until you have it in the world, you're not sure what people will want.
18:45 Speaker 01 And that's not called simulation in the face of massive computational power. That's called entrepreneurship and capitalism. It's a very different approach.
18:51 Sonal? I agree with you guys. I find that.
18:53 Speaker 02 The data is going the opposite direction. We're seeing less planning and predicting, less five-year plans, we're going to do this, and a lot more experimenting, testing, fail fast. That seems to be a model that works a lot better.
19:05 Sonal? But the other thing I was going to say is like, I look at countries like China. And they're incredibly coordinated efforts. And while I agree that past central industrial planning efforts have failed for various reasons, I don't know. I think there might be something to it this time. I just want to make sure you guys really disillusion me of that because let, help me let it go.
19:22 Speaker 01 And our colleagues, Daron S. Mogul or James Robinson, wrote this amazing book called Why Nations Fail.
19:26 Sonal? Okay.
19:27 Speaker 01 And their, and their answer was very, was really straightforward. Nations fail because they have extractive.
19:32 Speaker 02 Extractive institutions.
19:33 Speaker 01 Extractive institutions. Where an elite grabs power and they just suck up the value from all.
19:36 Sonal? Yeah.
19:37 Speaker 01 And, you know, they.
19:38 Sonal? Arguably, that's why companies that fail, companies fail too. Their users, if they're too extractive, they abandon.
19:43 Speaker 02 That's a good analogy. You should write the next book.
19:45 Speaker 01 And they hand down power to their descendants and they just make sure that they pervert the rules of the game to benefit themselves. That's as opposed to inclusive institutions where you have an honest shot of making the most of your human capital. Now, which one is China? They took big steps in the direction of inclusion by opening up to a market economy. Would we call that authoritarian state, one of actually inclusive institutions? I would not.
20:06 Sonal? I think that's a legitimate thing to say. Okay. So just going back to this idea of extractive institutions. So I do think it's interesting that there are now networks that are coming up that are letting people participate differently as owners.
20:17 Speaker 01 For sure.
20:17 Sonal? In different ways. And that is where I think this topic of ICOs and token launches is really interesting.
20:22 Speaker 02 Part of the power is, as Hayek would have said, is that you decentralize some of the local knowledge. They have information that nobody else has and they take advantage of that.
20:30 Sonal? Or resources, like if it's a computing power.
20:32 Speaker 02 Exactly. If you can move the decision rights to where that knowledge is, you're going to be better off. And one of the great things that technology has allowed us to do is move around decision rights, move around ownership. So hopefully, if you do it right, you get a better match between the incentives and the decision rights.
20:46 Speaker 01 The whole entire third section of our book is about this rebalancing necessary between the core institutions of a company and the crowd available over the internet now. How much more room there's very likely ahead of us.
20:57 Sonal? Yes.
20:57 Speaker 01 With crowdfunding, with crowdsourcing, with different ways to tap into what people can do, to give them an ownership stake, to get them bought in and pointed in the right direction. Have we scratched the surface of that?
21:09 Speaker 02 Let's talk a little bit about Joy's Law, that no matter what company you work for, most of the smart people in the world work for somebody else.
21:15 Sonal? Oh, yeah.
21:16 Speaker 02 It used to be limited what you could do about that. Because there's only so far you can communicate. But now, for the first time in history, a majority of the world's people are connected with a digital network. So they can access all the world's knowledge. And part of it isn't necessarily that they're smarter out there. Part of it just comes from the raw variety, the diversity, the variance. Within a company, you tend to have people who are like-minded, they've trained the same way, that's who they get hired. And maybe the way to solve a problem is with an entirely different approach. And that may be somebody from a different culture, a different way of looking at the world. And you're very unlikely to have that diversity inside of a company, it works against it. But if you can find a way to tap into it. One of our colleagues, Kareem Lakani, he's now at Harvard Business School, he was a PhD student at MIT,
21:59 Speaker 02 has done just case study after case study of examples where tapping into the crowd blew away what companies were able to do internally.
22:07 Speaker 01 He worked with the National Institutes of Health to try to improve the speed and accuracy of sequencing a human white blood cell genomes, which are really complicated but important to sequence. The National Institutes of Health, which I would call the core of the medical establishment.
22:20 Speaker 02 Core in the sense of core versus crowd.
22:22 Speaker 01 They had an algorithm that could do a run in about four hours with about 70% accuracy. There was a faculty member at Harvard Med School who made a big improvement to that algorithm. He developed one that got them up to about 75% accuracy. Kareem then worked with the NIH and TopCoder to make this an algorithmic challenge and open it up to the crowd. And the best solutions got down to about 10 seconds and about 80% accuracy.
22:45 Speaker 02 From four hours to 10 seconds.
22:47 Speaker 01 So we called up Kareem and he goes, it's about average. When I run a crowdsourcing tournament, this is the magnitude of improvement I expect to see.
22:54 Sonal? Wow.
22:54 Speaker 01 The last part of that story that continues to blow us away is that they interviewed the best performers who submitted the top performing algorithms. None of them had a life sciences background.
23:04 Sonal? Oh, that's the best part of the story.
23:05 Speaker 01 There was not a biologist among them.
23:07 Sonal? So crowds and prediction markets are similar, what's the difference?
23:11 Speaker 01 I would say a prediction market is one way to harness crowd markets. Do a really good job overall on aggregating knowledge.
23:18 Speaker 02 Markets tap into the crowd, Google taps into the crowd because their search algorithm basically exploits the link structure that all of us contribute wherever we make pages. There are lots of ways of tapping into the crowd. But being clever about how to reach them, motivate them, aggregate them. Still a lot of work to be done on that.
23:34 Frank Chen? Let's talk about the nature of work because I think what people do in that firm, either inside or outside, probably changes a lot. So we have this idea that human decision making is sort of fundamentally flawed and that like there's biases that you bring to your decision making that you don't even understand.
23:49 Speaker 02 Right.
23:50 Frank Chen? So when you're thinking it through, you're still going to make the same mistake because you don't understand that you have that bias.
23:54 Speaker 01 After all, walking you through your decision making process is your brain that came up that flawed decision making process in the first place.
24:00 Frank Chen? Right.
24:00 Speaker 01 It's not going to catch its own mistakes.
24:02 Frank Chen? Right, so it's a permanent blind spot. And by contrast, you would have sort of assumed that a machine learning algorithm trained with a carefully selected broad set of data sets will have a decision making efficiency or effectiveness better than flawed humans. So if that's the case, what do people in firms do? Like, how do you prepare for this world where there's going to be machine learning algorithms that can in general make pretty good decisions? And then there's this idea that like maybe the talent is better outside your company than inside your company. So what should you do? Should you join a company?
24:36 Speaker 02 It's just breathtaking what it can do. But it is far, far from being AI complete, being able to do everything that humans can do. There's a certain class of problems that it's kicking butt on, but that's a tiny sliver of what human decision making is. Even just defining what the problem is, exactly what needs to be done, that's half the battle. But you need humans to do that. There's a quote that we had in the book from Picasso, "Computers are useless. All they do is give you answers."
25:00 Sonal? I was a little shocked when Picasso was alive when computers were...
25:03 Speaker 01 He actually said that. We went down and quote investigated that one. He said that.
25:07 Sonal? I know, I just never associated Picasso and computers.
25:09 Speaker 02 Well, he's a brilliant guy in a lot of different ways. And obviously, he didn't know much about the latest neural network systems.
25:14 Sonal? Right.
25:15 Speaker 02 But but his understanding was spot on that simply giving the answer isn't necessarily the most interesting or important part of solving a problem.
25:22 Sonal? Kevin Kelly actually makes this argument inevitable. We had him on the podcast, that the number one job of the future for humans that humans preserve, and this is, I think, what you're getting at, is that we ask the questions and computers answer. But I have to say, I actually disagree with that a little bit because I'm seeing a new class of generative AI that makes me wonder if they're going to... I'd be asking new questions that make us want to answer differently. I mean, there's all kinds of interesting things.
25:47 Speaker 02 Our brains are made of atoms and so are computers. You know, I'm not going to say that there's some things that they just can never touch.
25:53 Speaker 01 But I agree, which is that on average, our wetware is amazing, but it's got a host of bugs and biases and glitches in it that machine learning systems and properly configured algorithms in general do not have. So if you could only pick one of those two entities to help you. The good news is that's a false choice. We don't have to make that choice. And I think the art going forward is being more clear about what are we actually good at versus what the machines are actually good at. The happy news is that they have very different failure modes.
26:24 Speaker 02 Yeah, and I think that's exactly the key point. It's a matter of how we can leverage each of them because machines have biases as well.
26:31 Sonal? Yeah, algorithms are biased by definition.
26:32 Speaker 02 Not just somebody designed them, but also the training data that they get. I mean, if you decide to give loans based on all the loans that have been approved or rejected in the past, that could have some biases built into it. And some of these neural nets could have billions of connections. Getting it to sort out how exactly, it's not going to be one of those says, okay, discriminate against women. But there may be some very subtle interactions that are hard to anticipate or explain. That said, at least the machines can be tested and improved. And it's often easier to do that with humans than it is with humans.
27:02 Speaker 01 We are really resistant to having our wetware tweaked. We really just don't like to be told that we're glitchy and here's the fix and just go do that now.
27:09 Sonal? That's the story of most marriages.
27:10 Speaker 01 Yeah, most marriages and most everything, right? It's really, really hard to do. There's a concept from linguistics that I find incredibly helpful for helping understand what I think some of the most durable human advantage in a world full of machines will be. And it's a concept called the intuition of the native speaker. And what they mean by that is if I look at any English language sentence, I can immediately tell if it's grammatically perfect or not.
27:33 Speaker 02 You just hear it in your head.
27:34 Speaker 01 Yeah, we are the native speakers of the human created world. Computers are doing this as their second language. We, I believe we have a massive advantage. We are the native speakers about this reality around us.
27:45 Speaker 02 Rather than trying to build a system that does everything from soup to nuts, you get some kind of a division of labor. Sebastian Thrun described this system to us recently. It was just fascinating. Fascinating. He's at Udacity and a lot of he gets incoming traffic.
27:56 Speaker 02 Another A6G company. Yeah.
27:58 Speaker 01 Rock on.
27:59 Speaker 02 All right. We've been good investment.
28:00 Speaker 01 Listeners at home, we are not, we don't have a list of A6G companies.
28:04 Speaker 02 Yeah, I was gonna say, this is all natural organic. We do have a list of cool companies, which seems to overlap for some reason. But, but, you know, Sebastian described how they get incoming traffic in their chat rooms of people asking about their offerings. They decided, let's take this data, we'll see which of these conversations lead to sales, which ones don't lead to sales. And label them that way, and then train a neural net about which replies were successful. And then what they took with those replies, they didn't try to have a standalone chatbot that then talked to customers. Instead, they had the human salespeople keep interacting. But when they saw one of these more common error modes, they would gently prompt the not so good salesperson. You know, maybe you want to give them this set of answers or this other set of responses. So it's kind of getting at the augmentation idea.
28:49 Speaker 02 It's absolutely augmentation, because it's a long tail of other questions that the bot had no clue what they were about. So it could help with the most common sets of queries. And this is, I think, a pattern that you see lots and lots. You see it among radiologists. You combine the two and you end up having fewer false positives and fewer false negatives.
29:05 Frank Chen? Yeah, I love this idea of sort of machines and humans working together. And I think it's only a matter of time before we walk into a doctor's office or a lawyer's office, where that isn't the fundamental interaction and we'll just be horrified. Like, where's your AI companion? Why are you trying to do this yourself with your biases?
29:20 Speaker 01 Look, I couldn't agree more. Why on earth would I expect my GP, who's a really good doctor, to be on top of the accumulated mass of human medical knowledge? And keeping up to date with the latest developments in all the fields that might relate to what I walk in the door with? That's an absurd request on a human being. Now, I want that person to be well trained. Even more, I want them to be able to empathize with me and get me to go along with a course of treatment and get me to buy into what's going on. Because that AI in the background that's got access to my tests and my lab results, it can assess the jaundice in my skin and, you know, how white the sclera of my eyes are. That's going to be the diagnostic expert in the not too distant future.
29:58 Frank Chen? That's exactly right. And I want AI not in the back room. I want it in the room with me when I'm doing the.
30:03 Sonal? With the doctor.
30:04 Speaker 02 Well, yeah, right.
30:04 Sonal? Seat of the table.
30:05 Speaker 02 It's a theme that comes up again and again. We talk about mind and machine, product and platform, core and crowd. And we don't want to give people the mistaken idea that you just cross off the first word on each of those lists and only do the second one.
30:17 Speaker 01 The mantra that I've learned is that tech progress rewrites the business playbook. And what the two of us believe is that the way the playbook is being rewritten these days is in favor of machines, platforms and crowds. So the balance needs to shift more in those directions.
30:31 Sonal? So the playbook is in favor of machine, platform and crowd.
30:35 Speaker 02 As opposed to.
30:36 Speaker 01 Mind, product and core.
30:38 Speaker 02 Right. So each of them is really a dichotomy. And the most successful systems are rarely all one or all the others.
30:45 Sonal? A couple of threads that we didn't get to pull. One question I had when we were talking about not all the talent is inside your company. And you know, a lot of people talk about open innovation as a way to kind of get around that, like.
30:55 Speaker 01 Yes.
30:55 Sonal? Open source communities, et cetera. What does that mean for business concretely? What does that mean for core in the way that you're defining core and deploying the power of the crowd? Like, does a business whose main strategy is their core business, does that mean that all their innovation is now outsourced to the crowd? Or is it the other direction? What's the ideal framework?
31:12 Speaker 01 I think way too many, even successful companies today are overweighting their core. They're probably spending too much of their total budget on it, way too much of their managerial bandwidth on it. And at the risk of being a little bit cute, I think a core capability for most organizations going forward is going to be interfacing with the crowd, harnessing its energy and its abilities, and then finding out how to bring that back into the organization without setting off all kinds of antibodies and resistance and nonsense.
31:39 Speaker 02 It's part of the same lesson we learned from the mind machine trade off is that defining the problem is important, whether you define it for the machine or whether you define it for the crowd, understanding what the problem is you're really trying to solve. If you can define it well enough, then these contests work great. The contests don't work great if you just say, hey, guys, you know, tell us stuff.
31:55 Speaker 01 Right.
31:56 Speaker 02 You know?
31:57 Sonal? Yeah.
31:57 Speaker 02 You give them a really precise.
31:58 Sonal? Which is what people used to do with the olden days. Remember when companies used to do these crowds or innovation boxes? And yeah, it never worked for a reason. So then that begs another question, though, for me, which is. If you take the innovation from the crowd, and you said earlier that there's this escalating effect where everyone has access to the same tools and they're all catching up really fast with each other and you can't, it's always a red queen.
32:17 Speaker 01 Right.
32:18 Sonal? You have to run faster than everybody else. But if everyone has access to the same crowd, how does the company get advantage in this space?
32:23 Speaker 01 Then honestly, it's a matter of where your leadership throws its attention, how firmly you believe in these new kinds of energies out there, not how willing you are to open the checkbook and spend money on technology, but how willing you are, forgive me, to open up your brains and rethink your business model in the face of this craziness.
32:40 Frank Chen? Who can use these tools more effectively? Just like, who can use the cloud more effectively?
32:44 Sonal? Yeah.
32:44 Frank Chen? I mean, it's a matter, it's like what it always is, there's a set of weapons out there and some people have a better strategy, some people have better techniques.
32:50 Speaker 01 The companies that failed during the transition from steam power over to electric power, almost none of them failed because they refused to invest in electricity. That was not the failure mode.
33:00 Sonal? Right.
33:00 Speaker 01 The failure mode was they refused to rethink what a factory could be and they refused
33:04 Sonal? And how to really absorb it into the core of their business.
33:07 Speaker 01 And they refused to take seriously the idea of an overhead crane or an assembly line or a conveyor belt.
33:12 Sonal? I'm just thinking about the statistic, because when you said this thing about this antibody that organizations naturally have.
33:18 Speaker 01 Yep.
33:18 Sonal? Which is essentially, they just immediately reject this non-invented here syndrome, basically.
33:22 Speaker 01 Look, those antibodies are the best news possible for your industry.
33:25 Sonal? But the research has shown over and over and over again that it is practically impossible for big companies to absorb startups successfully unless they keep them isolated. And one of the questions I have is the next follow-up, basically what happens when you leverage this crowd? How do you then really bring them into the company so that you don't have these antibodies? Do you have any concrete advice?
33:42 Speaker 01 I would look to do that in some of the most forward-thinking parts of the organization. As Eric said, in parts of the organization where the problem can be most clearly defined and where you've got people at the helm of that part of the organization who are willing to take the innovation, the algorithm, whatever that the crowd comes up with, and slot that into the work of the organization.
34:01 Frank Chen? There's a role for the core to be able to define that. In our world, a perfect example of the core leveraging of the cloud is the classic enterprise software company.
34:08 Speaker 01 Yeah.
34:09 Frank Chen? So in the old days, basically, you wrote software, it was all proprietary, you won Gartner Magic Quadrant, then you sent your Rolex-wearing direct salesperson to go sell it to someone. The new enterprise company is, let me create an open source project. Let me get a lot of contributors. Let me get contributors to get downloads, and that's my path to market.
34:29 Sonal? Right.
34:29 Frank Chen? Right?
34:29 Sonal? The open source becomes like that.
34:31 Frank Chen? And the core needs to be there because they got to. What's the project?
34:34 Speaker 02 Exactly.
34:35 Frank Chen? And what problem are we trying to solve? But the crowd comes in into it to basically lend legitimacy and support and enthusiasm for the project.
34:43 Speaker 02 So if you can be that scarce complement to the abundant crowd, you can create a lot of value, then you become the linchpin that is creating, that is capturing a lot of the value as well as creating it. Ultimately, we are in an economy of creative destruction. And one of the strengths of the United States and other dynamic economies is that we have this constant turnover. And one of the things that discourages us is that there's actually less, fewer startups, less innovation, fewer young firms in America today than there were 10 or 20 years ago.
35:13 Sonal? Oh, yeah, we talk about this phenomenon. It worries us, too.
35:15 Speaker 02 Absolutely. We are all for trying to make the bigger companies more nimble.
35:19 Speaker 01 Amen.
35:19 Speaker 02 And understand this. But the bigger way that the economy innovates is by having this innovative set of new startups that rise and adopt some of the new technologies.
35:31 Speaker 01 Yeah.
35:31 Speaker 02 You've got to have both. And we'd like to see progress on both dimensions.
35:35 Sonal? That's great. Thank you for joining the A6 and Z podcast.
35:37 Speaker 01 Thanks for having us on. This is fantastic.
35:38 Speaker 02 It's a real pleasure.
Previous Company Building in Crypto: A Conversation between Brian Armstrong and Chris Dixon Next What's Next for Education Startups Page updated October 26, 2026 · transcript and summary are machine-generated