AI transcript
0:00:04 What will be the economic models for creative people in an AI world?
0:00:07 Don’t stop the inevitable, which is the technology progressing.
0:00:09 Lean into it and rethink those models.
0:00:13 That to me is the most exciting area for this intersection.
0:00:17 In the last few years, AI has been the talk of the town.
0:00:21 Founders have tivided, incumbents have plowed capital into new projects,
0:00:24 VCs have upended their investing theses.
0:00:27 All of this as part of the race to capitalize on what seems like
0:00:30 the biggest platform shift in decades,
0:00:34 and equally a new generation of the internet.
0:00:37 This generation is not only an opportunity to rethink the past,
0:00:42 but with parallel technology tracks from new hardware to crypto intersecting,
0:00:44 we can build things we never could before.
0:00:49 So what will the economic model of this wave be when so much is being upended?
0:00:51 You go to their websites, they give you an answer.
0:00:55 And so what happens to the billion other websites
0:00:57 if they aren’t getting traffic is a question, right?
0:01:00 When will we move past the skeuomorphic phase of this generation
0:01:02 to building net new behaviors?
0:01:06 And could crypto be the counterbalance to the centralized gravity of AI,
0:01:10 targeting more data, more compute, and more complex models?
0:01:12 Where we’re headed is a world where you have five big systems,
0:01:16 let’s call it three to five big AI systems.
0:01:19 Joining us to discuss all this and more are A16Z Growth,
0:01:21 general partner David George,
0:01:24 and A16Z Crypto founding partner Chris Dixon.
0:01:27 Last year, of course, Chris wrote his book Read Right Own,
0:01:29 building the next era of the internet,
0:01:34 all about how blockchains might finally bring us back to the early promise of the internet,
0:01:38 a decentralized democratic network of innovation, connection, and freedom.
0:01:41 So without further ado, let’s dive in.
0:01:45 By the way, if you did like this episode, it comes straight from our AI Revolution series.
0:01:48 And if you missed any of the previous episodes of that series,
0:01:51 with guests like AMD CEO Lisa Sue,
0:01:53 andthropic co-founder Dario Amade,
0:01:57 and the founders behind companies like Databricks, Waymo, Figma, and more,
0:02:02 head on over to a16z.com/airevolution.
0:02:06 As a reminder, the content here is for informational purposes only.
0:02:09 Should not be taken as legal, business, tax, or investment advice,
0:02:11 or be used to evaluate any investment or security,
0:02:16 and is not directed at any investors or potential investors in any A16Z fund.
0:02:18 Please note that A16Z and its affiliates
0:02:22 may also maintain investments in the companies discussed in this podcast.
0:02:24 For more details, including a link to our investments,
0:02:27 please see a16z.com/disclosures.
0:02:33 Chris, thanks for being here.
0:02:34 Yeah, thanks for having me.
0:02:35 I’ll always love hanging out with you.
0:02:37 Obviously, you spend most of your time on crypto today.
0:02:41 How do you generally see crypto and AI interacting?
0:02:44 Yeah, I mean, so I think, first of all, my kind of meta view is that
0:02:48 the technology waves tend to come in pairs or triples.
0:02:50 15 years ago, it was mobile, social, cloud.
0:02:53 And I’m always giving this speech to entrepreneurs.
0:02:54 They tend to reinforce each other.
0:02:57 And so mobile was what took computing
0:02:59 from hundreds of millions to billions of people.
0:03:02 Social was the killer app that hooked them.
0:03:04 And cloud was the infrastructure that made it possible, right?
0:03:06 And so you couldn’t really have all three of them.
0:03:09 And I remember back then, people having debates, which were better.
0:03:10 It turned out they were all better.
0:03:11 And they were all required.
0:03:12 They were all required.
0:03:14 And so I think of that with AI, crypto, and maybe new devices.
0:03:18 Yeah, they’re kind of probably robotics and self-taughting cars and VR and things.
0:03:20 I think of those as the three interesting things going on.
0:03:23 And I think they all kind of complement each other and work together.
0:03:26 It’s a new way to architect internet services, a new way to build networks
0:03:29 that has a bunch of different properties, which I argue are beneficial
0:03:33 for a bunch of reasons and can do a set of things you couldn’t do before, essentially.
0:03:36 And so I think a lot of people think of it as Bitcoin or meme coins or something.
0:03:38 And so that’s fundamentally not what it is to me.
0:03:40 Or I think to the kind of smart people working in the space.
0:03:44 There’s many different ways in which it intersects with AI.
0:03:46 So the first way, which is something we’ve invested a bunch in,
0:03:49 is just using this new architecture to build AI systems.
0:03:51 And so, for example, one of the core questions,
0:03:54 I think we’ve just talked a lot of this firm about the future of AI
0:03:57 is to what extent will AI be controlled by a small set of companies
0:03:59 or controlled by a broad community?
0:04:02 The obvious first question there is, is it open source?
0:04:02 Yes.
0:04:06 It’s negatively shocked me how closed source the world has become.
0:04:08 Ten years ago, everything was open and put in papers.
0:04:12 And then it all shut down and was closed.
0:04:14 And they said this was for safety reasons.
0:04:16 I think it just happened to be very good for their–
0:04:17 I just think it’s–
0:04:17 –offensibility.
0:04:18 –beneficial business reasons.
0:04:20 I don’t believe the safety thing.
0:04:23 But thankfully, there’s these ones like Lama and Flux
0:04:26 and Mistral and things who are open source.
0:04:28 I worry that’s a little fragile, because first of all,
0:04:29 I don’t know.
0:04:30 A lot of them don’t put their weights open.
0:04:31 Is it really open?
0:04:32 Some of it’s open.
0:04:33 Like the data pipeline’s not open.
0:04:35 Is it really reproducible?
0:04:36 They could switch it tomorrow.
0:04:38 These models get better every month.
0:04:39 And if they don’t, start doing the new frontier.
0:04:40 I don’t know.
0:04:40 So it’s like–
0:04:42 It’s very heavily dependent on one large company.
0:04:43 Yeah.
0:04:45 So one of the things we invested in
0:04:47 is a stack of internet services that
0:04:50 are built for the AI stack, but open services
0:04:51 is a different layer.
0:04:53 So as an example, there’s a project called
0:04:55 Jensen, which is building– think of it as crowdsourced
0:04:56 compute layer.
0:04:58 And so you, as a startup, can submit
0:05:01 a job that goes beyond the compute you control.
0:05:04 And it goes out to a network, kind of Airbnb style of people
0:05:05 that have access compute.
0:05:09 And the network manages that supply and demand, right?
0:05:10 And that’s the economic ledger.
0:05:11 Yeah.
0:05:12 That’s one example.
0:05:14 Another one is one called Story Protocol,
0:05:16 which is a new way to think about registering
0:05:17 intellectual property.
0:05:22 And so you could create image or video or piece of music.
0:05:25 And then you register it on a blockchain,
0:05:28 which keeps a record of the piece of media
0:05:29 and the rights around.
0:05:30 It uses existing copyright law.
0:05:32 So it actually– so the blockchain record
0:05:35 mirrors a legal agreement that’s been crafted
0:05:37 to work internationally.
0:05:38 And then anyone can come along.
0:05:41 And as long as they abide by your terms that you set,
0:05:42 you might say something like, you can use this.
0:05:43 You can remix it.
0:05:45 You can create derivative works.
0:05:47 But any revenue you make, you have to pay me 10% or whatever.
0:05:49 downstream, there’s a tick, yeah.
0:05:50 You set the terms.
0:05:51 But that creates this sort of open marketplace
0:05:54 where right now you have to call up some company
0:05:56 and try to do a BD deal and this and that.
0:05:58 And so you end up having this kind of thing where people either
0:06:00 basically steal it or don’t do it.
0:06:02 Or they’re scaled enough to make a deal or something,
0:06:04 like you have open AI going to Shutterstock
0:06:06 and they paid them $100 million.
0:06:08 But this is really just for the very high-end companies.
0:06:12 This is creating a broad democratic kind of resource
0:06:15 where anyone can, a small creator can set the terms.
0:06:18 And then ideally what you create, and this is a recurring theme
0:06:19 in the blockchain world,
0:06:21 is you have this kind of what we call composability.
0:06:23 I think the kind of core force behind the success
0:06:24 of open source software.
0:06:26 I mean, people forget this, but open source software,
0:06:28 certainly the most successful open computing movement
0:06:30 in the last, you know, 80 years.
0:06:33 But Linux went from 0% market share in the ’90s
0:06:34 to probably, I don’t know what,
0:06:36 90 plus percent market share today.
0:06:38 And a lot of that’s because of what we call composability,
0:06:40 which is basically all of these different people
0:06:42 coming along and contributing little pieces to the system
0:06:44 and the system collectively getting much better
0:06:45 in the same way that Wikipedia is,
0:06:47 a collective knowledge system.
0:06:48 And so something like Story Protocol,
0:06:51 you get the same kind of Lego brick effect with media.
0:06:52 So if someone comes along and they create a character,
0:06:53 someone else creates another character,
0:06:55 someone else remixes them, someone else,
0:06:57 and then you can use whatever AI tool,
0:07:00 you can create generative AI, and you can create your story.
0:07:01 I created a new superhero universe
0:07:02 where I use these other Lego bricks.
0:07:05 And as long as the money kind of waterfalls back,
0:07:06 that’s all okay.
0:07:07 I think it’s a really great vision
0:07:12 that both allows for people to embrace these new tools,
0:07:16 but also provides an economic model for creative people.
0:07:17 I think that’s a, for me,
0:07:19 that’s a recurring theme in our investing,
0:07:21 is like what will be the economic models
0:07:22 for creative people in an AI world?
0:07:24 Don’t stop the inevitable,
0:07:25 which is the technology progressing.
0:07:28 Lean into it and rethink those models.
0:07:31 That to me is the most exciting area for this intersection.
0:07:33 You go from social networking companies
0:07:35 which keep 100% of revenue for themselves
0:07:38 when creators create stuff effectively
0:07:41 to something where, hopefully,
0:07:44 the creator can capture an upfront amount that they set.
0:07:47 And then, ideally, the composability
0:07:49 allows for actually more creativity built on top.
0:07:50 That’s right.
0:07:53 Because of the economic incentive element, yeah.
0:07:54 We’re seeing people do interesting stuff
0:07:56 with kind of crowdsourced model evaluation.
0:07:58 Just think of it as all the data side of things.
0:08:00 Like you need more data,
0:08:02 and we have this crypto as a breakthrough
0:08:04 in new ways to design incentive systems.
0:08:05 And so you combine that,
0:08:07 and you say, well, how can you use new incentive systems
0:08:09 to get more data for these AI systems?
0:08:10 Data can either be an input
0:08:13 or it can be a model evaluation or whatever it might be.
0:08:15 So it’s kind of what these companies like Scale.AI do,
0:08:18 but in a crowdsourced way instead of a centralized way.
0:08:20 There’s a project that’s co-founded by Sam Altman
0:08:21 that we’re investing in called Worldcoin,
0:08:24 where the thesis is that in a world
0:08:28 where AI can replicate humans and content,
0:08:29 we need a way to prove you’re human.
0:08:31 And the best way to prove you’re human
0:08:34 is cryptographically using a blockchain.
0:08:36 And so the idea is they have an incentive system
0:08:37 for people to sign up.
0:08:40 And originally it was this orb that scanned your eyeballs.
0:08:41 Some people it was controversial.
0:08:43 They now have systems where you can identify yourself
0:08:45 in other ways, including your passport and other things.
0:08:47 But the idea is you prove who you are.
0:08:49 You get cryptographic proof on a blockchain.
0:08:52 And then you can use that for a bunch of different services.
0:08:53 Think of very simple examples.
0:08:54 Think of CAPTCHAs.
0:08:56 Today you have to go and play these puzzles,
0:08:58 which I think have gotten so complicated.
0:08:59 – Not AI proof anymore.
0:09:00 – I’m sorry, AI proof anymore.
0:09:01 And they may be human proof.
0:09:02 I have trouble with a lot of them,
0:09:05 but replace those with a set of systems like that
0:09:06 and other kinds of clunky fraud systems
0:09:07 have an actual cryptographic thing.
0:09:09 So I have a code, essentially.
0:09:10 This is how cryptography works.
0:09:12 And that code proves that I’m a human.
0:09:13 And then you can layer onto that
0:09:15 other kinds of things you prove on top.
0:09:17 So I think there’s a bunch of this infrastructure layer
0:09:20 of like take AI systems that exist today in a centralized way
0:09:22 and decentralize them both in terms of code and services.
0:09:24 There’s new things you couldn’t do before
0:09:26 like machine to machine payments.
0:09:27 And then there’s these sort of really far off things
0:09:29 that I find the most exciting,
0:09:32 which are like what are new business models in this world.
0:09:33 – One of the things that you pointed out to me
0:09:35 right after the chat GPT moment is you’re like,
0:09:37 hey, we have the potential for sort of a break
0:09:39 in the pact of the internet.
0:09:40 – Oh, yeah, yeah.
0:09:41 – Which I think is a super fascinating.
0:09:44 – Yeah, yeah, there’s a chapter on this in the book
0:09:45 toward the end.
0:09:46 I call it a new covenant.
0:09:47 So like you think about the incentive system.
0:09:49 One of the main reasons the internet succeeds
0:09:51 is it had a very clever incentive system, right?
0:09:52 Like how do you get five billion people
0:09:54 to sort of opt into the system
0:09:57 without having a central authority tell them to, right?
0:09:58 This is because of the incentives of the internet.
0:10:01 And specifically there’s been a kind of what’s emerged
0:10:05 over the last 20-ish years is I call it an economic covenant
0:10:07 between the kind of the platforms,
0:10:09 specifically social networks and search engines
0:10:11 and all the people that create websites
0:10:13 that essentially those link to, right?
0:10:14 – Yeah, exactly.
0:10:17 – And so if you’re a travel website or a recipe website
0:10:21 or a artist who has illustrations,
0:10:22 there’s an implicit covenant you have,
0:10:23 let’s say with Google, right?
0:10:24 Which is you say to Google,
0:10:27 it’s okay if you crawl my content and you index me
0:10:30 and you show snippets in your search engine
0:10:32 if you send me traffic back.
0:10:33 This is how the internet has evolved, right?
0:10:35 And why do you want traffic back?
0:10:36 ‘Cause you have some business money.
0:10:38 Maybe it’s a free site, maybe it’s an ad-based site,
0:10:39 maybe it’s subscription-based site,
0:10:41 but whatever it is, somehow you have a way
0:10:42 to make money on traffic.
0:10:44 There’s some understanding, right?
0:10:45 – And it’s mutually beneficial.
0:10:46 – Mutually beneficial.
0:10:48 And occasionally that has been breached.
0:10:50 So if there was a thing Google does called one boxing,
0:10:52 which is they would take your content
0:10:54 and just put it like I was on the board of Stack Overflow
0:10:55 for a long time and they would do this,
0:10:56 where they would take, you type in a thing
0:10:58 for Stack Overflow instead of clicking on it,
0:10:59 they would just show you the answer
0:11:00 and remove the click,
0:11:01 they’ve done that with Wikipedia,
0:11:02 they did it with Lyric sites.
0:11:03 – Yeah, but they did it with Yelp,
0:11:04 they did it with Trial.
0:11:05 – And people get very upset, or they with Yelp,
0:11:07 they promote their own content on top.
0:11:10 And so there were issues, but it worked, right?
0:11:11 Now, the question in an AI world is
0:11:13 if you have these chatbots,
0:11:15 if you go and you say I want an illustration
0:11:16 and it just generates an illustration,
0:11:19 or you say I want a recipe and it gives you a recipe.
0:11:20 This might be a better user experience, by the way,
0:11:21 I’m not against it.
0:11:22 I think it’s probably better in the end
0:11:24 for the users of the internet.
0:11:25 But the problem is it breaks the covenant, right?
0:11:27 They took this data,
0:11:29 these systems were trained on data
0:11:32 that was put on the internet under the prior covenant.
0:11:34 – Under the premise that they’re gonna get traffic back
0:11:36 and they can monetize it correctly.
0:11:37 – And that was the premise,
0:11:38 and that was the promise, right?
0:11:40 And now you have a new system,
0:11:41 which may not send the traffic.
0:11:42 In fact, it probably won’t.
0:11:44 If these things can just tell you the answer,
0:11:46 why would you click through?
0:11:47 And so that’s probably where we’re headed
0:11:49 is a world where these, you have five big systems,
0:11:51 let’s call it, three to five big AI systems,
0:11:53 you go to their websites, they give you an answer.
0:11:56 And so what happens to the billion other websites?
0:11:59 If they aren’t getting traffic is the question, right?
0:12:02 And I’m surprised/disappointed that I don’t see anyone.
0:12:06 I feel like I’m the only person I’ve seen talking about.
0:12:07 I feel like I’m screaming to the abyss.
0:12:08 Like I’m a little bit surprised
0:12:11 that the AI people who just, it’s fine.
0:12:11 Like they took all the data
0:12:12 and they’ll be copyright lawsuits
0:12:14 and I’m not gonna apply on that.
0:12:16 – They’ve done some data deals here and there.
0:12:17 – Yeah, but aren’t we a little bit,
0:12:20 even forgetting about the societal questions
0:12:22 and all the small businesses that will be like,
0:12:23 don’t we worry about the internet?
0:12:24 ‘Cause like I worry about just the internet.
0:12:28 Like if you have an internet of five companies
0:12:30 and it becomes a broadcast TV in 1970s,
0:12:33 there’s four channels, is that the world we’re gonna live in?
0:12:36 Is that a world that’s pro startup, pro innovation?
0:12:39 – Yeah, there’s not gonna be a long tail of websites,
0:12:40 like that next generation of long tail websites.
0:12:42 – Yeah, how do you break out?
0:12:43 How do you create new things?
0:12:46 So I just worry without thinking it through.
0:12:48 And so to me, look, and I’m not saying
0:12:49 that I have the only answer to it
0:12:50 or you have to be a crypto answer.
0:12:51 I realize some people that’s controversial,
0:12:53 but I think that step one is we should say,
0:12:56 okay, wait, this breaks all the incentives of the internet.
0:12:57 And step two is, is that a good thing?
0:12:58 I don’t think so.
0:13:01 And then so what is the right answer
0:13:02 and should we create new incentives?
0:13:04 And this is why a lot of what I’ve been trying
0:13:06 to invest in and think about has been, okay,
0:13:08 like the example I gave a story protocol is,
0:13:12 let’s think about new incentive systems to layer on top.
0:13:14 – Yeah, one of the things you’ve talked about
0:13:17 is just this trifecta of technology products
0:13:19 that have come along at the same time.
0:13:22 So generative AI, crypto and new hardware platforms.
0:13:24 So how do you think about the three of those coming together?
0:13:26 – So yeah, and the analogy, of course,
0:13:27 is like mobile social cloud.
0:13:29 The last wave where they all ended up reinforcing each other.
0:13:30 So you’re already seeing some of this.
0:13:33 You have these new devices, the AR and VR glasses
0:13:35 and things which use a lot of AI
0:13:38 and sort of her style kind of stuff.
0:13:39 There’s a whole area of crypto I’m excited about
0:13:42 called D-PIN, which is decentralized physical infrastructure.
0:13:45 Most prominent example is a project called Helium.
0:13:49 And Helium is a community-owned crowdsourced telecom network
0:13:51 that tries to compete with Verizon and AT&T.
0:13:53 And so basically what they did is they created an incentive
0:13:56 system where anyone can put a Helium node up in their house
0:13:57 and that adds a little bit to the network.
0:13:59 It’s a wireless transmitter.
0:14:01 They got hundreds of thousands of people in the country
0:14:02 to put these networks up.
0:14:04 And now they offer a cellular service
0:14:06 that’s, I think, significantly cheaper
0:14:07 than some of the different Verizon.
0:14:09 It’s like 20 bucks a month instead of 70 bucks a month.
0:14:12 And it’s cheaper because much of the time
0:14:14 it’s using this homegrown network
0:14:16 that they didn’t have to spend tens of billions of dollars
0:14:17 to build it out.
0:14:19 But what’s interesting about it is crypto is very good
0:14:20 at creating incentive systems.
0:14:23 And traditionally, in networks,
0:14:25 the hardest part of a network is the bootstrap phase.
0:14:28 Once a network has critical mass, it’s clearly valuable.
0:14:31 Once I can sign up for a cellular network
0:14:32 and use it anywhere in the country,
0:14:34 clearly I’ll pay for that, right?
0:14:36 When you start it off and there’s only 10 houses
0:14:38 with the cellular access,
0:14:39 it’s not something you want to use.
0:14:40 Think of a dating site.
0:14:41 If there’s 10 people on a dating site,
0:14:41 you don’t want to use it.
0:14:43 If there’s millions, you do want to use it.
0:14:45 This is a classic problem with building networks
0:14:46 is how do you get over this early phase
0:14:49 when the network effects are weak, right?
0:14:50 And so crypto is the perfect complement
0:14:53 for that crypto is a great way to provide incentives
0:14:55 in the early areas of building a network.
0:14:56 And it turns out a lot of interesting networks
0:14:58 in the world are physical networks.
0:15:01 So there’s people doing this for climate weather modeling.
0:15:03 There’s people doing it for mapping,
0:15:05 self-driving data and mapping cars.
0:15:07 There’s people doing it for electric car charging,
0:15:09 for cellular networking.
0:15:12 We just did one that’s around energy metric monitoring.
0:15:14 And there’s people doing decentralized science,
0:15:16 which you mix it in with a more scientific application.
0:15:19 So one sort of simple heuristic is anywhere
0:15:20 where you want to build a network.
0:15:23 And as a challenge to build the early phases of the network,
0:15:26 crypto can be a really useful way to help bootstrap that.
0:15:27 Oh, interesting.
0:15:28 And so that’s one of my favorite areas
0:15:32 where the physical world and robotics intersecting
0:15:34 with data collection and all these other themes
0:15:36 that intersect with AI are relevant.
0:15:38 Mark actually gave me this framework,
0:15:39 which I like a lot,
0:15:41 which is is the AI frosting or sugar?
0:15:44 You know, if the AI is a core ingredient,
0:15:46 if it’s a frosting, all the incumbents are gonna win
0:15:49 ’cause you just slap a chat bot on your existing product
0:15:50 and you’ve got distribution.
0:15:52 You know, you have that like selling reference power,
0:15:54 incumbent relationships,
0:15:55 if it’s more fundamental of an ingredient,
0:15:58 like you can’t actually just slap AI into the product,
0:15:59 you have to build it from scratch
0:16:00 and that favors the newcomers.
0:16:01 It’s just very TV.
0:16:04 We haven’t seen anything that tells us what the answer is.
0:16:08 The more seal your thing, the more steamorphic it is,
0:16:10 which is early cycle thing,
0:16:12 the more it probably favors the incumbents.
0:16:14 Another way maybe to frame Mark’s thinking
0:16:16 is the Clay Christensen view,
0:16:19 where is it disruptive or sustaining?
0:16:20 And specifically, I think what people misunderstand
0:16:21 about Christiansen view, right?
0:16:23 Disruptive doesn’t just mean new.
0:16:25 It means misaligned with the incumbent business model.
0:16:26 Yeah, exactly.
0:16:27 That’s sort of the interesting part of his book, right?
0:16:30 Is it even when the smart incumbents sees it coming,
0:16:32 it’s very, very hard for them to react to it
0:16:34 because it’s not what their best customers are asking for.
0:16:35 Yeah, exactly.
0:16:37 And so that’s where I think
0:16:39 somewhat overlaps with Mark’s frosting icing thing.
0:16:40 Well, it could be that the business model
0:16:42 is a fundamentally shifted business model.
0:16:43 Yeah, so you come in and you’re like,
0:16:45 instead of databases,
0:16:47 it’s some radical new architecture that’s database free.
0:16:47 I don’t know why.
0:16:50 It’s something that cannibalizes the incumbent business model
0:16:51 and therefore makes it organizationally
0:16:55 and economically harder for the incumbents to layer it on.
0:16:56 We haven’t seen it yet.
0:16:58 We’ve seen people talk about outcome based pricing.
0:17:00 Well, let’s talk quickly about consumer.
0:17:01 So in consumer right now, at least,
0:17:03 I don’t think you see a lot of network effect businesses,
0:17:04 right?
0:17:05 Sure.
0:17:07 Like as successful as the Clause and ChatGPTs are,
0:17:09 I don’t think they have a network effect.
0:17:10 They’re switching costs are relative.
0:17:11 Maybe they learn your history.
0:17:12 But the question is right,
0:17:14 how do they avoid in the steady state
0:17:16 of having just like a model and price competition
0:17:18 to the price of the bottom, right?
0:17:19 Obviously they’re important big businesses,
0:17:21 but will they be dominant?
0:17:22 Yeah.
0:17:23 And then what’s the opportunity for new startups?
0:17:26 If you’re doing venture investing and AI consumer,
0:17:28 like you see a lot of these things that make your face
0:17:30 prettier, like these kind of fun apps
0:17:32 and they zoom up in the app chart
0:17:34 and then TikTok copies it and so forth, right?
0:17:37 ‘Cause it’s just not, ’cause again, no network effect.
0:17:39 And there’s just technique kind of strategy I like to talk
0:17:41 about called come for the tool, stay for the network.
0:17:43 And the idea is maybe you can use that,
0:17:46 make my face prettier and then use that as a hook
0:17:48 to get people into your new network,
0:17:50 like your social network, possibly,
0:17:51 although it just feels very, very hard today,
0:17:54 given the scale and power of these incumbents.
0:17:56 And that, by the way, we’ll intersect back to crypto
0:17:58 ’cause what crypto is and what I argue in my book
0:18:00 is that crypto is a new way to build networks.
0:18:00 And so, you know,
0:18:02 you sort of have the chocolate and peanut butter.
0:18:03 You have AI with all these really interesting use cases
0:18:06 and then you have this new technique for building networks.
0:18:08 AI, the interesting use cases, but no network effect.
0:18:09 And then you have this new thing that’s like,
0:18:11 all network effects are their interesting ways
0:18:12 to combine them.
0:18:13 So before I get to that,
0:18:15 I think it’s important to talk about
0:18:17 how big technologies roll out in multiple stages.
0:18:19 So there’s a distinction.
0:18:22 It’s not my distinction, but I’ve talked about a lot.
0:18:24 It’s sort of one way to think about technology
0:18:26 is that they can do one of two things.
0:18:28 They can do old things better
0:18:30 or they can do new things you couldn’t do before.
0:18:31 We call the first one skeuomorphic.
0:18:33 This is a Steve Jobs term,
0:18:36 which sort of refers to products and designs
0:18:38 that kind of harken back to a prior era
0:18:40 to make them more understandable.
0:18:42 And then there’s what we call native apps,
0:18:43 which are things which are the kind of new things
0:18:44 that couldn’t be done before.
0:18:46 And then there’s actually a third stage, I think,
0:18:48 which is second order effects,
0:18:49 which is you created the car
0:18:51 and now you have the highway system
0:18:53 and now you’re able to create suburbs
0:18:55 and trucking infrastructure, right?
0:18:58 Those are second order downstream effects.
0:19:00 There’s a famous line that good science fiction writers
0:19:02 predict the car, great science fiction writers
0:19:03 predict the traffic jam, right?
0:19:05 So like, it’s like that idea.
0:19:07 So it’s like, what are the second orders?
0:19:09 Like, Bitcoin is something that couldn’t have existed
0:19:11 before social networking, right?
0:19:12 So 30 years ago, you say someday people
0:19:13 are gonna have their own media
0:19:15 and you’re gonna remove these gatekeepers.
0:19:16 Who would have thought?
0:19:18 Then you’re gonna create these digital currencies.
0:19:19 – There would have been no way to create the community.
0:19:20 – Yeah, yeah, it would have been
0:19:22 in your time’s article saying it’s a stupid
0:19:24 and then it does the end of it, right?
0:19:26 There’s nowhere to get together and talk about it
0:19:28 and create, I mean, they’re really kind of religious
0:19:30 movements, you know, most token communities
0:19:32 and they need places to congregate and discuss it.
0:19:33 And now they have that.
0:19:34 And so there’s all these kind of second order.
0:19:36 I mean, we’re seeing effects in politics
0:19:37 and all these other things.
0:19:40 There’s the whole arguably our society and world
0:19:42 is changing as a second order effect of social networking.
0:19:43 So one way to think about AI.
0:19:45 So the first stage is the scumorphic phase,
0:19:48 which is this is the stuff you see talked about all the time
0:19:50 in the business and startup community
0:19:52 of like your customer service bots, right?
0:19:54 You take a job that’s currently done by a person
0:19:57 sitting in a call center and you replace that
0:19:59 with a AI voice and chat bot, right?
0:20:02 In the simplest case, it’s a one-to-one exchange.
0:20:04 It’s cheaper and it’s more systematic
0:20:07 and it will displace jobs.
0:20:08 Hopefully it will also create
0:20:11 equally or more jobs and better jobs.
0:20:12 But that’s sort of an obvious first stage.
0:20:14 And look, and this is, I think one of the reasons
0:20:15 people get so excited about the opportunity for AI
0:20:17 is you can just see that happening
0:20:20 in, I don’t know, tens of millions of jobs, I guess.
0:20:23 Like the whole laptop middle kind of section
0:20:26 of the economy, you can see many of those jobs.
0:20:29 Everyone, including us, who spend their days typing emails.
0:20:31 (laughing)
0:20:32 That’s the joke.
0:20:33 It’s like we can speculate on it,
0:20:35 but we’re part of that group too.
0:20:36 So that’s phase one, right, scumorphic.
0:20:39 But that’s phase one can last 20 years.
0:20:41 So just to be clear, the next phase is the native phase.
0:20:43 And that, to me, that’s what gets me more excited.
0:20:44 And by the way, let me give a little analogy
0:20:45 to the internet.
0:20:47 So the scumorphic phase was the ’90s.
0:20:48 And so basically, if you look at ’90s internet,
0:20:51 people were taking offline things
0:20:53 like magazines and catalogs and putting them online.
0:20:56 So you would go buy things, and it was much easier.
0:20:58 You could type in a website and go buy this rare book
0:21:01 on Amazon and it was much easier and it was convenient,
0:21:02 but it was fundamentally something
0:21:03 you could have done before.
0:21:04 It just would have been clumsy in getting
0:21:06 some weird magazine catalog or something.
0:21:08 But it wasn’t until the 2000s that people did things
0:21:09 like social networking.
0:21:10 And these things were just brand new things.
0:21:11 There’s no analog in the offline world
0:21:14 to a lot of these new behaviors that people created.
0:21:16 I talk a lot in detail about this in the book
0:21:17 if people aren’t interested.
0:21:19 So anyway, so you saw the internet play out that way.
0:21:20 ’93 was mosaic.
0:21:23 And 2000, I would say five-ish was sort of YouTube
0:21:25 and four, I think, was Facebook or whatever it was.
0:21:27 So it took at least a decade.
0:21:28 And by the way, one of the things you get
0:21:30 in the native phase, which is why it’s so exciting,
0:21:32 is you get new products, you get new forms of media.
0:21:34 So if you go back when photography
0:21:36 was growing in popularity,
0:21:39 there were all of these cultural art criticism,
0:21:42 think pieces about what will happen to art.
0:21:44 The famous, like Walter Benjamin,
0:21:46 the art in the age of mechanical reproduction.
0:21:47 There’s all these famous essays
0:21:49 where it was like, what’s gonna happen?
0:21:50 ‘Cause now that you can take a photo
0:21:52 and create a beautiful landscape,
0:21:53 what’s the role of the artist in that world, right?
0:21:55 And so people were worried about it.
0:21:56 In the same way, they’re worried today
0:21:57 about generative AI, right?
0:22:00 So like, what if you can now create a movie?
0:22:02 Looks like you can pretty soon, right?
0:22:03 I mean, images is there.
0:22:06 Images is there, and probably videos coming soon.
0:22:08 What happened in the case of photography
0:22:09 is that you had, I think two things happened.
0:22:12 Fine art went more abstract and went away from photography,
0:22:15 right, that leaned into what they were unique at.
0:22:16 And that’s when you had whatever,
0:22:18 cubism and all these other kinds of movements.
0:22:19 And then on the other side,
0:22:20 what I think was really interesting, right,
0:22:21 is you had the rise of film.
0:22:22 You had someone say, hey,
0:22:25 maybe you can use machines to replace photography,
0:22:27 but you can also now use machines
0:22:28 to create a brand new art form
0:22:30 that never could exist before, right?
0:22:31 You sort of had it with animation,
0:22:32 but now you can do it a really interesting
0:22:34 sophisticated way with film, right?
0:22:36 And so film would be,
0:22:37 what was the native form of media
0:22:40 in the age of mechanical reproduction, right?
0:22:41 – Oh, that’s a fascinating knowledge, yeah.
0:22:42 – And so I think to like today,
0:22:44 like so when you look at the generative AI,
0:22:45 like the negative way to look at it,
0:22:47 and you do see some of a lot of this negative sentiment
0:22:50 from like the art community and things on Twitter,
0:22:52 where they say, look, this is just a cheap replacement
0:22:54 for human creativity.
0:22:57 The positive way to look at it is this is the base layer,
0:22:59 in the same way that film was a base layer back then,
0:23:02 but now there’s this new canvas of human creativity
0:23:03 where you can create new art forms.
0:23:04 I don’t know what those are.
0:23:06 They may be virtual worlds or games
0:23:08 or new types of films and movies.
0:23:09 I don’t know.
0:23:10 – They may intersect with a new way
0:23:12 to consume the media altogether.
0:23:14 – Yeah, maybe there’s new interfaces,
0:23:15 and this is to me what’s so exciting
0:23:17 about the new native media, the native apps,
0:23:19 is that I won’t think of it,
0:23:21 because in my experience,
0:23:22 through watching some of these waves in the past,
0:23:25 is there, it really does take brilliant creative people
0:23:28 to come up with these new things,
0:23:29 and it surprises you in many cases.
0:23:31 And so I think that that’s gonna be the exciting phase
0:23:35 I’m looking for is not how do you just use this technology
0:23:37 to do the things you could do today, but do them cheaper,
0:23:39 but how do you use the technology to push the frontier
0:23:41 and do things that could never be done before
0:23:43 in the same way that film did that, right?
0:23:44 – Yeah.
0:23:47 – I think photography probably unlocked more opportunities
0:23:50 for creative people than it removed,
0:23:51 and I think this would be the hope in this kind of phase.
0:23:53 So that’s the media example,
0:23:56 but there’s probably that for consumer applications
0:23:59 and that for social networking, and that for productivity,
0:24:01 and so that will be the really exciting thing I think to see
0:24:03 is not just the replacing things we do today,
0:24:05 but come up with brand new behaviors
0:24:07 that are things we couldn’t do before.
0:24:09 And then the third thing is the second order effects, right?
0:24:10 So you create this new world,
0:24:12 so you’ve created this world of social networking.
0:24:13 As interesting to think with social networking,
0:24:14 we’ve seen it play out.
0:24:16 You know, you sort of have social networking rise
0:24:17 in the 2000s.
0:24:18 I think it hit a tipping point.
0:24:19 Maybe the Obama election.
0:24:20 – Yeah.
0:24:22 – Was that the 2008 and then ’12 too.
0:24:25 He really leaned into using that,
0:24:26 and I remember seeing all these news articles,
0:24:28 like, wow, this is different.
0:24:31 The bit had flipped from online as a secondary,
0:24:32 sort of online was primary,
0:24:35 but then we started seeing these kind of weirder things,
0:24:37 like I think the Trump movement and the populism
0:24:38 just surprised everybody,
0:24:41 and you just started seeing movements and just behave,
0:24:43 and I think we still haven’t really figured out
0:24:45 what’s going on, but where all this is headed,
0:24:48 and we’re in this disequilibrium state, I guess.
0:24:49 Anyways, those sort of second order effects
0:24:51 of social media will probably play out for,
0:24:53 as I mentioned, like crypto,
0:24:55 and I think a bunch of other interesting movements today
0:24:57 are second order effects of social media,
0:24:58 and that will probably play out for 20, 30 years,
0:25:00 and so that will probably be phase three
0:25:01 of the AI revolution.
0:25:03 – Yeah, and just think about the timelines.
0:25:04 – Yeah, and it’s probably gonna take a very long time.
0:25:06 For all, like, I’m always overly up,
0:25:07 stick on these things, historically.
0:25:09 I’m like, okay, we’re done with the skemorphic phase of AI,
0:25:10 now we’ll do the native phase,
0:25:13 but the reality is each phase probably takes a decade.
0:25:15 – One of the interesting things you said
0:25:17 around these distinct phases,
0:25:18 obviously the internet took a long time,
0:25:20 partially because you had to build a network.
0:25:21 – Yeah.
0:25:22 – It was a supply and demand issue, right?
0:25:23 – A physical network, and then also–
0:25:25 – The literally laying cables and the wireless–
0:25:26 – Yeah, laying cables,
0:25:30 and sure you have to build large clusters of compute GPUs
0:25:31 here with networking,
0:25:34 but I think the constraining factor
0:25:37 for getting from that skemorphic phase to the native phase
0:25:39 is not necessarily capabilities themselves,
0:25:41 but like human creativity.
0:25:42 – Yeah, I think so.
0:25:43 – Even like ideas.
0:25:46 – I think the bottleneck will be humans and regulation,
0:25:48 which are obviously closely connected,
0:25:50 and I think humans on both the supply and the demand side,
0:25:51 probably more on the demand side.
0:25:52 So meaning supply side,
0:25:54 you need to have people come up with all the creative things,
0:25:56 but the world’s different now
0:25:58 in that I just think the startup world is different now.
0:26:01 It’s much more mature and much more sophisticated, honestly,
0:26:03 than like when I was coming up in it.
0:26:04 I mean, when I was starting off,
0:26:05 there were 10 venture firms,
0:26:07 now there’s thousands, the number of startups,
0:26:10 and honestly, there’s a lot of good, smart advice out there.
0:26:12 – Yeah, this is a more popular path for smart people
0:26:14 to go to. – Yeah, it’s like a thing you do,
0:26:16 like in places like Y Combinator
0:26:17 and other places have done a good job of this.
0:26:19 If you’re coming out of a top school,
0:26:22 I mean, even 10 years ago, this wasn’t like,
0:26:23 I knew people that were like, “Wow, you could do startups.”
0:26:25 I mean, definitely that was the case 15 years ago,
0:26:28 but now I think it’s like an established career path.
0:26:29 There’s an established set of mentors,
0:26:30 established set of funding.
0:26:33 There’s a canon of pretty good advice out there.
0:26:35 Like the standard advice used to be terrible advice,
0:26:36 now it’s good advice.
0:26:37 You can come out to San Francisco,
0:26:39 and I think relatively easily,
0:26:41 if you’re a smart network-friendly person,
0:26:43 get embedded pretty quickly.
0:26:44 And then, you know, and then still looking about,
0:26:45 it’s gotten just very good.
0:26:48 It’s throwing tons of capital energy against those problems.
0:26:49 So there’s the supply side.
0:26:52 I suspect the demand side is more like another meaning,
0:26:55 like changing organizational and human work
0:26:59 and behavior patterns, like getting an organization,
0:27:00 like take the video example we’re talking about.
0:27:02 Yeah, I mean, look, I wrote my book.
0:27:05 I wanted to have my own voice, use AI to read the book,
0:27:08 using my own voice, both the publisher and Audible,
0:27:11 the podcasting platform, ban AI completely.
0:27:14 And part of its unions and just a bunch of resistance.
0:27:15 I think people know this,
0:27:16 but like the capabilities are fully there to do that.
0:27:18 Yeah, I mean, like, look, Mark and Jason
0:27:19 had a great blog post, it’s like,
0:27:22 how do I know they’re gonna ban AI like medicine?
0:27:23 ‘Cause they already have, essentially.
0:27:25 I mean, essentially, like these things
0:27:26 are so heavily regulated.
0:27:28 And so many areas where it’s gonna have an impact
0:27:29 are so heavily regulated.
0:27:31 And just the organization, like,
0:27:32 look, take the Hollywood Gen AI thing.
0:27:35 You’d have to lay off a whole bunch of people, probably,
0:27:37 who you don’t wanna lay off, who are unionized.
0:27:39 So that means maybe there’ll be some fresh upstarts,
0:27:43 maybe in another country who create AI-native movie studios.
0:27:45 But that will take a very long time.
0:27:47 The right answer is probably to harness
0:27:48 all of that talent in Hollywood
0:27:51 and combine it with AI in some way.
0:27:53 There is a lot of very smart people and talent.
0:27:54 But how long will that take?
0:27:56 Culturally, it may take a whole generation
0:27:56 to really play out, right?
0:27:58 So that’s something by the demand side, right?
0:28:01 And then just human behavior, changing your workflow,
0:28:03 using an AI assistant, I don’t know.
0:28:04 Anyway, so.
0:28:05 Yeah, having like a co-pilot for everything you do,
0:28:06 like it feels like it’s-
0:28:09 Yeah, maybe that can be solved with interfaces and things.
0:28:09 I don’t know.
0:28:10 Then there’s the policy side,
0:28:12 which is there’s going to be this resistance,
0:28:14 I’m discussing already is,
0:28:16 going to be enshrined,
0:28:18 there’s going to be movements to enshrine it in law.
0:28:19 And that’s going to play out,
0:28:20 and I think in multiple levels,
0:28:22 it’s already starting to play out in the courts,
0:28:24 and it’s starting to play out in like state legislatures,
0:28:26 with like the California had the AI bill,
0:28:28 you know, you have a bunch of lawsuits around copyright.
0:28:31 My view is ultimately this will play out in Congress.
0:28:33 This is such a big issue when you have something
0:28:35 that affects tens of millions of jobs,
0:28:37 it is beyond something that people are going to allow
0:28:38 just to happen through.
0:28:40 Through free markets, yeah.
0:28:42 Yeah, and through regular court decisions.
0:28:44 Like the copyright thing is an example.
0:28:46 Like right now the question is,
0:28:48 when an AI system is trained on a piece of data,
0:28:49 is it copying that data,
0:28:50 or is it learning from that data?
0:28:51 Yeah. It’s a philosophical question.
0:28:52 There’s a fundamental question
0:28:54 across different media happening right now.
0:28:55 That’s right.
0:28:57 And so you could have five years from now,
0:29:00 some federal judge decide that philosophical question,
0:29:02 or I think more likely you’ll eventually have
0:29:05 some legislation, like congressional legislation,
0:29:06 that’s some kind of compromise
0:29:08 struck between the media industries and tech industries
0:29:09 that comes up with a solution
0:29:11 that both creates incentives for creators,
0:29:13 but also allows AI systems to exist.
0:29:15 I don’t know, but that thing will play out
0:29:16 over a very long time.
0:29:19 When will you be allowed to use AI in medical and finance?
0:29:21 And I mean, significant, what is it?
0:29:24 Probably 70% of our economy are regulated industries, right?
0:29:24 Yeah, of course.
0:29:25 You know, on the flip side,
0:29:28 like the stuff with the Waymo is really impressive.
0:29:30 I’m surprised they’re actually allowed in San Francisco.
0:29:31 And they’re–
0:29:33 Well, it turns out it’s seven to 10 times safer
0:29:34 than a human driver.
0:29:36 And there’s now millions of miles of game-filled–
0:29:38 So maybe that’s the playbook
0:29:40 to get this stuff adopted more broadly.
0:29:44 What is an ideal future state of the internet?
0:29:47 So there’s zero cost of creation and distribution,
0:29:49 transparent ownership, governance.
0:29:51 What does this look like?
0:29:52 I think that we’re at a crossroads
0:29:55 and there’s a real question as to whether it looks
0:29:57 more like its original vision,
0:29:58 which is the vision of the internet,
0:30:01 like the ’90s vision and the ’80s vision or something,
0:30:04 was an internet that was community-owned,
0:30:05 community-governed.
0:30:08 The money mostly flowed to the edges of the network
0:30:10 and not to the intermediaries in the middle.
0:30:11 Like, originally, in the ’90s,
0:30:13 the money flowed to the edges,
0:30:15 to small businesses, to innovators, to entrepreneurs.
0:30:16 If you looked at a map today,
0:30:18 it’s mostly flowing to the middle.
0:30:20 This is why the Seven Company–
0:30:21 It’s 200 million revenue.
0:30:22 So it’s not working at all.
0:30:24 Yeah, I think the top five internet companies
0:30:26 are something like more than half of the market cap,
0:30:28 not more, and it might be absolutely higher by now.
0:30:30 And so just you have all the green stuff
0:30:31 flowing into the middle.
0:30:32 I think of it as two kind of important things
0:30:34 that you want is power and money, control.
0:30:36 And my core argument in the book
0:30:39 is that those two questions are a product
0:30:41 of how you build these services.
0:30:42 The first sentence of the book
0:30:44 is your architecture is your destiny
0:30:44 or something like that.
0:30:46 Like, the architecture you choose
0:30:48 determines how it’s controlled and how the money flows.
0:30:51 And so, and I think we’re really at a kind of critical point.
0:30:53 In fact, I worry we’re approaching a point in no return
0:30:55 where it’s going to be an internet-controlled
0:30:56 by five companies.
0:30:57 And what’s happened is these networks
0:30:59 have all gotten to a certain scale
0:31:01 and they’ve just decided that the next kind of wave
0:31:03 is to keep you trapped there.
0:31:04 Well, there’s no way to grow users anymore.
0:31:05 They’ve captured all these users.
0:31:06 That’s right, that’s right.
0:31:08 They climb the ladder and they’re kicking it away.
0:31:09 And it’s really negative.
0:31:11 And this is why we as a firm have felt
0:31:12 that this is such an important topic.
0:31:14 Of being able to build internet services
0:31:16 with new architectures, like using blockchains,
0:31:19 is such an important topic for the future of small tech,
0:31:20 little tech, as we call it.
0:31:21 Along with open source AI,
0:31:23 the other kind of critical thing,
0:31:26 which is if startup has to pay this giant tax
0:31:29 to an incumbent to build competitive services,
0:31:30 they won’t be able to build services
0:31:31 that threaten those incumbents, right?
0:31:32 Yeah, we’ve seen that before, right?
0:31:34 Like you’ve talked about it as Shingo
0:31:35 was built on top of Facebook.
0:31:37 Yeah, it’s platform risk, right?
0:31:38 I mean, you’re building on quicksand.
0:31:41 So startups need access to distribution and networks
0:31:43 and they need access to modern software,
0:31:44 open source software.
0:31:46 And so I think those are the critical questions.
0:31:47 Those will be, I think a hugely important thing,
0:31:49 which is why we’ve invested so much time and money
0:31:50 and it is the regulatory side of this,
0:31:52 is like what policies are there?
0:31:54 And are they policies that encourage competition
0:31:56 and innovation and little tech?
0:31:59 And then I think just raising awareness of these topics
0:32:01 and having discussions about them are important.
0:32:03 What I’m worried about now is sort of backing ourselves
0:32:05 without having really thought it through
0:32:07 into a situation where there’s four companies
0:32:09 that control everything and it ends up,
0:32:10 we’re kind of eating our seed corn.
0:32:12 Like so much of what we benefit from today
0:32:15 is the startup innovation of the past
0:32:17 and we’ll risk losing that if we let these small set
0:32:18 of companies control everything.
0:32:20 Yeah, well, I’m optimistic.
0:32:22 Look, the bright side is through all the work
0:32:24 that you guys have done in our firm,
0:32:26 we’ve gotten the word out about little tech.
0:32:29 And I think understanding that building a new architecture,
0:32:32 new infrastructure, and then the importance of open source,
0:32:33 I think the word is getting out.
0:32:35 So this is awesome because thanks for being here.
0:32:36 I always love talking to you.
0:32:37 Thank you.
0:32:39 (upbeat music)
0:32:42 (upbeat music)
0:32:45 (upbeat music)
0:32:47 (upbeat music)
0:32:50 (upbeat music)
0:32:52 you
0:32:55 (gentle music)
0:00:07 Don’t stop the inevitable, which is the technology progressing.
0:00:09 Lean into it and rethink those models.
0:00:13 That to me is the most exciting area for this intersection.
0:00:17 In the last few years, AI has been the talk of the town.
0:00:21 Founders have tivided, incumbents have plowed capital into new projects,
0:00:24 VCs have upended their investing theses.
0:00:27 All of this as part of the race to capitalize on what seems like
0:00:30 the biggest platform shift in decades,
0:00:34 and equally a new generation of the internet.
0:00:37 This generation is not only an opportunity to rethink the past,
0:00:42 but with parallel technology tracks from new hardware to crypto intersecting,
0:00:44 we can build things we never could before.
0:00:49 So what will the economic model of this wave be when so much is being upended?
0:00:51 You go to their websites, they give you an answer.
0:00:55 And so what happens to the billion other websites
0:00:57 if they aren’t getting traffic is a question, right?
0:01:00 When will we move past the skeuomorphic phase of this generation
0:01:02 to building net new behaviors?
0:01:06 And could crypto be the counterbalance to the centralized gravity of AI,
0:01:10 targeting more data, more compute, and more complex models?
0:01:12 Where we’re headed is a world where you have five big systems,
0:01:16 let’s call it three to five big AI systems.
0:01:19 Joining us to discuss all this and more are A16Z Growth,
0:01:21 general partner David George,
0:01:24 and A16Z Crypto founding partner Chris Dixon.
0:01:27 Last year, of course, Chris wrote his book Read Right Own,
0:01:29 building the next era of the internet,
0:01:34 all about how blockchains might finally bring us back to the early promise of the internet,
0:01:38 a decentralized democratic network of innovation, connection, and freedom.
0:01:41 So without further ado, let’s dive in.
0:01:45 By the way, if you did like this episode, it comes straight from our AI Revolution series.
0:01:48 And if you missed any of the previous episodes of that series,
0:01:51 with guests like AMD CEO Lisa Sue,
0:01:53 andthropic co-founder Dario Amade,
0:01:57 and the founders behind companies like Databricks, Waymo, Figma, and more,
0:02:02 head on over to a16z.com/airevolution.
0:02:06 As a reminder, the content here is for informational purposes only.
0:02:09 Should not be taken as legal, business, tax, or investment advice,
0:02:11 or be used to evaluate any investment or security,
0:02:16 and is not directed at any investors or potential investors in any A16Z fund.
0:02:18 Please note that A16Z and its affiliates
0:02:22 may also maintain investments in the companies discussed in this podcast.
0:02:24 For more details, including a link to our investments,
0:02:27 please see a16z.com/disclosures.
0:02:33 Chris, thanks for being here.
0:02:34 Yeah, thanks for having me.
0:02:35 I’ll always love hanging out with you.
0:02:37 Obviously, you spend most of your time on crypto today.
0:02:41 How do you generally see crypto and AI interacting?
0:02:44 Yeah, I mean, so I think, first of all, my kind of meta view is that
0:02:48 the technology waves tend to come in pairs or triples.
0:02:50 15 years ago, it was mobile, social, cloud.
0:02:53 And I’m always giving this speech to entrepreneurs.
0:02:54 They tend to reinforce each other.
0:02:57 And so mobile was what took computing
0:02:59 from hundreds of millions to billions of people.
0:03:02 Social was the killer app that hooked them.
0:03:04 And cloud was the infrastructure that made it possible, right?
0:03:06 And so you couldn’t really have all three of them.
0:03:09 And I remember back then, people having debates, which were better.
0:03:10 It turned out they were all better.
0:03:11 And they were all required.
0:03:12 They were all required.
0:03:14 And so I think of that with AI, crypto, and maybe new devices.
0:03:18 Yeah, they’re kind of probably robotics and self-taughting cars and VR and things.
0:03:20 I think of those as the three interesting things going on.
0:03:23 And I think they all kind of complement each other and work together.
0:03:26 It’s a new way to architect internet services, a new way to build networks
0:03:29 that has a bunch of different properties, which I argue are beneficial
0:03:33 for a bunch of reasons and can do a set of things you couldn’t do before, essentially.
0:03:36 And so I think a lot of people think of it as Bitcoin or meme coins or something.
0:03:38 And so that’s fundamentally not what it is to me.
0:03:40 Or I think to the kind of smart people working in the space.
0:03:44 There’s many different ways in which it intersects with AI.
0:03:46 So the first way, which is something we’ve invested a bunch in,
0:03:49 is just using this new architecture to build AI systems.
0:03:51 And so, for example, one of the core questions,
0:03:54 I think we’ve just talked a lot of this firm about the future of AI
0:03:57 is to what extent will AI be controlled by a small set of companies
0:03:59 or controlled by a broad community?
0:04:02 The obvious first question there is, is it open source?
0:04:02 Yes.
0:04:06 It’s negatively shocked me how closed source the world has become.
0:04:08 Ten years ago, everything was open and put in papers.
0:04:12 And then it all shut down and was closed.
0:04:14 And they said this was for safety reasons.
0:04:16 I think it just happened to be very good for their–
0:04:17 I just think it’s–
0:04:17 –offensibility.
0:04:18 –beneficial business reasons.
0:04:20 I don’t believe the safety thing.
0:04:23 But thankfully, there’s these ones like Lama and Flux
0:04:26 and Mistral and things who are open source.
0:04:28 I worry that’s a little fragile, because first of all,
0:04:29 I don’t know.
0:04:30 A lot of them don’t put their weights open.
0:04:31 Is it really open?
0:04:32 Some of it’s open.
0:04:33 Like the data pipeline’s not open.
0:04:35 Is it really reproducible?
0:04:36 They could switch it tomorrow.
0:04:38 These models get better every month.
0:04:39 And if they don’t, start doing the new frontier.
0:04:40 I don’t know.
0:04:40 So it’s like–
0:04:42 It’s very heavily dependent on one large company.
0:04:43 Yeah.
0:04:45 So one of the things we invested in
0:04:47 is a stack of internet services that
0:04:50 are built for the AI stack, but open services
0:04:51 is a different layer.
0:04:53 So as an example, there’s a project called
0:04:55 Jensen, which is building– think of it as crowdsourced
0:04:56 compute layer.
0:04:58 And so you, as a startup, can submit
0:05:01 a job that goes beyond the compute you control.
0:05:04 And it goes out to a network, kind of Airbnb style of people
0:05:05 that have access compute.
0:05:09 And the network manages that supply and demand, right?
0:05:10 And that’s the economic ledger.
0:05:11 Yeah.
0:05:12 That’s one example.
0:05:14 Another one is one called Story Protocol,
0:05:16 which is a new way to think about registering
0:05:17 intellectual property.
0:05:22 And so you could create image or video or piece of music.
0:05:25 And then you register it on a blockchain,
0:05:28 which keeps a record of the piece of media
0:05:29 and the rights around.
0:05:30 It uses existing copyright law.
0:05:32 So it actually– so the blockchain record
0:05:35 mirrors a legal agreement that’s been crafted
0:05:37 to work internationally.
0:05:38 And then anyone can come along.
0:05:41 And as long as they abide by your terms that you set,
0:05:42 you might say something like, you can use this.
0:05:43 You can remix it.
0:05:45 You can create derivative works.
0:05:47 But any revenue you make, you have to pay me 10% or whatever.
0:05:49 downstream, there’s a tick, yeah.
0:05:50 You set the terms.
0:05:51 But that creates this sort of open marketplace
0:05:54 where right now you have to call up some company
0:05:56 and try to do a BD deal and this and that.
0:05:58 And so you end up having this kind of thing where people either
0:06:00 basically steal it or don’t do it.
0:06:02 Or they’re scaled enough to make a deal or something,
0:06:04 like you have open AI going to Shutterstock
0:06:06 and they paid them $100 million.
0:06:08 But this is really just for the very high-end companies.
0:06:12 This is creating a broad democratic kind of resource
0:06:15 where anyone can, a small creator can set the terms.
0:06:18 And then ideally what you create, and this is a recurring theme
0:06:19 in the blockchain world,
0:06:21 is you have this kind of what we call composability.
0:06:23 I think the kind of core force behind the success
0:06:24 of open source software.
0:06:26 I mean, people forget this, but open source software,
0:06:28 certainly the most successful open computing movement
0:06:30 in the last, you know, 80 years.
0:06:33 But Linux went from 0% market share in the ’90s
0:06:34 to probably, I don’t know what,
0:06:36 90 plus percent market share today.
0:06:38 And a lot of that’s because of what we call composability,
0:06:40 which is basically all of these different people
0:06:42 coming along and contributing little pieces to the system
0:06:44 and the system collectively getting much better
0:06:45 in the same way that Wikipedia is,
0:06:47 a collective knowledge system.
0:06:48 And so something like Story Protocol,
0:06:51 you get the same kind of Lego brick effect with media.
0:06:52 So if someone comes along and they create a character,
0:06:53 someone else creates another character,
0:06:55 someone else remixes them, someone else,
0:06:57 and then you can use whatever AI tool,
0:07:00 you can create generative AI, and you can create your story.
0:07:01 I created a new superhero universe
0:07:02 where I use these other Lego bricks.
0:07:05 And as long as the money kind of waterfalls back,
0:07:06 that’s all okay.
0:07:07 I think it’s a really great vision
0:07:12 that both allows for people to embrace these new tools,
0:07:16 but also provides an economic model for creative people.
0:07:17 I think that’s a, for me,
0:07:19 that’s a recurring theme in our investing,
0:07:21 is like what will be the economic models
0:07:22 for creative people in an AI world?
0:07:24 Don’t stop the inevitable,
0:07:25 which is the technology progressing.
0:07:28 Lean into it and rethink those models.
0:07:31 That to me is the most exciting area for this intersection.
0:07:33 You go from social networking companies
0:07:35 which keep 100% of revenue for themselves
0:07:38 when creators create stuff effectively
0:07:41 to something where, hopefully,
0:07:44 the creator can capture an upfront amount that they set.
0:07:47 And then, ideally, the composability
0:07:49 allows for actually more creativity built on top.
0:07:50 That’s right.
0:07:53 Because of the economic incentive element, yeah.
0:07:54 We’re seeing people do interesting stuff
0:07:56 with kind of crowdsourced model evaluation.
0:07:58 Just think of it as all the data side of things.
0:08:00 Like you need more data,
0:08:02 and we have this crypto as a breakthrough
0:08:04 in new ways to design incentive systems.
0:08:05 And so you combine that,
0:08:07 and you say, well, how can you use new incentive systems
0:08:09 to get more data for these AI systems?
0:08:10 Data can either be an input
0:08:13 or it can be a model evaluation or whatever it might be.
0:08:15 So it’s kind of what these companies like Scale.AI do,
0:08:18 but in a crowdsourced way instead of a centralized way.
0:08:20 There’s a project that’s co-founded by Sam Altman
0:08:21 that we’re investing in called Worldcoin,
0:08:24 where the thesis is that in a world
0:08:28 where AI can replicate humans and content,
0:08:29 we need a way to prove you’re human.
0:08:31 And the best way to prove you’re human
0:08:34 is cryptographically using a blockchain.
0:08:36 And so the idea is they have an incentive system
0:08:37 for people to sign up.
0:08:40 And originally it was this orb that scanned your eyeballs.
0:08:41 Some people it was controversial.
0:08:43 They now have systems where you can identify yourself
0:08:45 in other ways, including your passport and other things.
0:08:47 But the idea is you prove who you are.
0:08:49 You get cryptographic proof on a blockchain.
0:08:52 And then you can use that for a bunch of different services.
0:08:53 Think of very simple examples.
0:08:54 Think of CAPTCHAs.
0:08:56 Today you have to go and play these puzzles,
0:08:58 which I think have gotten so complicated.
0:08:59 – Not AI proof anymore.
0:09:00 – I’m sorry, AI proof anymore.
0:09:01 And they may be human proof.
0:09:02 I have trouble with a lot of them,
0:09:05 but replace those with a set of systems like that
0:09:06 and other kinds of clunky fraud systems
0:09:07 have an actual cryptographic thing.
0:09:09 So I have a code, essentially.
0:09:10 This is how cryptography works.
0:09:12 And that code proves that I’m a human.
0:09:13 And then you can layer onto that
0:09:15 other kinds of things you prove on top.
0:09:17 So I think there’s a bunch of this infrastructure layer
0:09:20 of like take AI systems that exist today in a centralized way
0:09:22 and decentralize them both in terms of code and services.
0:09:24 There’s new things you couldn’t do before
0:09:26 like machine to machine payments.
0:09:27 And then there’s these sort of really far off things
0:09:29 that I find the most exciting,
0:09:32 which are like what are new business models in this world.
0:09:33 – One of the things that you pointed out to me
0:09:35 right after the chat GPT moment is you’re like,
0:09:37 hey, we have the potential for sort of a break
0:09:39 in the pact of the internet.
0:09:40 – Oh, yeah, yeah.
0:09:41 – Which I think is a super fascinating.
0:09:44 – Yeah, yeah, there’s a chapter on this in the book
0:09:45 toward the end.
0:09:46 I call it a new covenant.
0:09:47 So like you think about the incentive system.
0:09:49 One of the main reasons the internet succeeds
0:09:51 is it had a very clever incentive system, right?
0:09:52 Like how do you get five billion people
0:09:54 to sort of opt into the system
0:09:57 without having a central authority tell them to, right?
0:09:58 This is because of the incentives of the internet.
0:10:01 And specifically there’s been a kind of what’s emerged
0:10:05 over the last 20-ish years is I call it an economic covenant
0:10:07 between the kind of the platforms,
0:10:09 specifically social networks and search engines
0:10:11 and all the people that create websites
0:10:13 that essentially those link to, right?
0:10:14 – Yeah, exactly.
0:10:17 – And so if you’re a travel website or a recipe website
0:10:21 or a artist who has illustrations,
0:10:22 there’s an implicit covenant you have,
0:10:23 let’s say with Google, right?
0:10:24 Which is you say to Google,
0:10:27 it’s okay if you crawl my content and you index me
0:10:30 and you show snippets in your search engine
0:10:32 if you send me traffic back.
0:10:33 This is how the internet has evolved, right?
0:10:35 And why do you want traffic back?
0:10:36 ‘Cause you have some business money.
0:10:38 Maybe it’s a free site, maybe it’s an ad-based site,
0:10:39 maybe it’s subscription-based site,
0:10:41 but whatever it is, somehow you have a way
0:10:42 to make money on traffic.
0:10:44 There’s some understanding, right?
0:10:45 – And it’s mutually beneficial.
0:10:46 – Mutually beneficial.
0:10:48 And occasionally that has been breached.
0:10:50 So if there was a thing Google does called one boxing,
0:10:52 which is they would take your content
0:10:54 and just put it like I was on the board of Stack Overflow
0:10:55 for a long time and they would do this,
0:10:56 where they would take, you type in a thing
0:10:58 for Stack Overflow instead of clicking on it,
0:10:59 they would just show you the answer
0:11:00 and remove the click,
0:11:01 they’ve done that with Wikipedia,
0:11:02 they did it with Lyric sites.
0:11:03 – Yeah, but they did it with Yelp,
0:11:04 they did it with Trial.
0:11:05 – And people get very upset, or they with Yelp,
0:11:07 they promote their own content on top.
0:11:10 And so there were issues, but it worked, right?
0:11:11 Now, the question in an AI world is
0:11:13 if you have these chatbots,
0:11:15 if you go and you say I want an illustration
0:11:16 and it just generates an illustration,
0:11:19 or you say I want a recipe and it gives you a recipe.
0:11:20 This might be a better user experience, by the way,
0:11:21 I’m not against it.
0:11:22 I think it’s probably better in the end
0:11:24 for the users of the internet.
0:11:25 But the problem is it breaks the covenant, right?
0:11:27 They took this data,
0:11:29 these systems were trained on data
0:11:32 that was put on the internet under the prior covenant.
0:11:34 – Under the premise that they’re gonna get traffic back
0:11:36 and they can monetize it correctly.
0:11:37 – And that was the premise,
0:11:38 and that was the promise, right?
0:11:40 And now you have a new system,
0:11:41 which may not send the traffic.
0:11:42 In fact, it probably won’t.
0:11:44 If these things can just tell you the answer,
0:11:46 why would you click through?
0:11:47 And so that’s probably where we’re headed
0:11:49 is a world where these, you have five big systems,
0:11:51 let’s call it, three to five big AI systems,
0:11:53 you go to their websites, they give you an answer.
0:11:56 And so what happens to the billion other websites?
0:11:59 If they aren’t getting traffic is the question, right?
0:12:02 And I’m surprised/disappointed that I don’t see anyone.
0:12:06 I feel like I’m the only person I’ve seen talking about.
0:12:07 I feel like I’m screaming to the abyss.
0:12:08 Like I’m a little bit surprised
0:12:11 that the AI people who just, it’s fine.
0:12:11 Like they took all the data
0:12:12 and they’ll be copyright lawsuits
0:12:14 and I’m not gonna apply on that.
0:12:16 – They’ve done some data deals here and there.
0:12:17 – Yeah, but aren’t we a little bit,
0:12:20 even forgetting about the societal questions
0:12:22 and all the small businesses that will be like,
0:12:23 don’t we worry about the internet?
0:12:24 ‘Cause like I worry about just the internet.
0:12:28 Like if you have an internet of five companies
0:12:30 and it becomes a broadcast TV in 1970s,
0:12:33 there’s four channels, is that the world we’re gonna live in?
0:12:36 Is that a world that’s pro startup, pro innovation?
0:12:39 – Yeah, there’s not gonna be a long tail of websites,
0:12:40 like that next generation of long tail websites.
0:12:42 – Yeah, how do you break out?
0:12:43 How do you create new things?
0:12:46 So I just worry without thinking it through.
0:12:48 And so to me, look, and I’m not saying
0:12:49 that I have the only answer to it
0:12:50 or you have to be a crypto answer.
0:12:51 I realize some people that’s controversial,
0:12:53 but I think that step one is we should say,
0:12:56 okay, wait, this breaks all the incentives of the internet.
0:12:57 And step two is, is that a good thing?
0:12:58 I don’t think so.
0:13:01 And then so what is the right answer
0:13:02 and should we create new incentives?
0:13:04 And this is why a lot of what I’ve been trying
0:13:06 to invest in and think about has been, okay,
0:13:08 like the example I gave a story protocol is,
0:13:12 let’s think about new incentive systems to layer on top.
0:13:14 – Yeah, one of the things you’ve talked about
0:13:17 is just this trifecta of technology products
0:13:19 that have come along at the same time.
0:13:22 So generative AI, crypto and new hardware platforms.
0:13:24 So how do you think about the three of those coming together?
0:13:26 – So yeah, and the analogy, of course,
0:13:27 is like mobile social cloud.
0:13:29 The last wave where they all ended up reinforcing each other.
0:13:30 So you’re already seeing some of this.
0:13:33 You have these new devices, the AR and VR glasses
0:13:35 and things which use a lot of AI
0:13:38 and sort of her style kind of stuff.
0:13:39 There’s a whole area of crypto I’m excited about
0:13:42 called D-PIN, which is decentralized physical infrastructure.
0:13:45 Most prominent example is a project called Helium.
0:13:49 And Helium is a community-owned crowdsourced telecom network
0:13:51 that tries to compete with Verizon and AT&T.
0:13:53 And so basically what they did is they created an incentive
0:13:56 system where anyone can put a Helium node up in their house
0:13:57 and that adds a little bit to the network.
0:13:59 It’s a wireless transmitter.
0:14:01 They got hundreds of thousands of people in the country
0:14:02 to put these networks up.
0:14:04 And now they offer a cellular service
0:14:06 that’s, I think, significantly cheaper
0:14:07 than some of the different Verizon.
0:14:09 It’s like 20 bucks a month instead of 70 bucks a month.
0:14:12 And it’s cheaper because much of the time
0:14:14 it’s using this homegrown network
0:14:16 that they didn’t have to spend tens of billions of dollars
0:14:17 to build it out.
0:14:19 But what’s interesting about it is crypto is very good
0:14:20 at creating incentive systems.
0:14:23 And traditionally, in networks,
0:14:25 the hardest part of a network is the bootstrap phase.
0:14:28 Once a network has critical mass, it’s clearly valuable.
0:14:31 Once I can sign up for a cellular network
0:14:32 and use it anywhere in the country,
0:14:34 clearly I’ll pay for that, right?
0:14:36 When you start it off and there’s only 10 houses
0:14:38 with the cellular access,
0:14:39 it’s not something you want to use.
0:14:40 Think of a dating site.
0:14:41 If there’s 10 people on a dating site,
0:14:41 you don’t want to use it.
0:14:43 If there’s millions, you do want to use it.
0:14:45 This is a classic problem with building networks
0:14:46 is how do you get over this early phase
0:14:49 when the network effects are weak, right?
0:14:50 And so crypto is the perfect complement
0:14:53 for that crypto is a great way to provide incentives
0:14:55 in the early areas of building a network.
0:14:56 And it turns out a lot of interesting networks
0:14:58 in the world are physical networks.
0:15:01 So there’s people doing this for climate weather modeling.
0:15:03 There’s people doing it for mapping,
0:15:05 self-driving data and mapping cars.
0:15:07 There’s people doing it for electric car charging,
0:15:09 for cellular networking.
0:15:12 We just did one that’s around energy metric monitoring.
0:15:14 And there’s people doing decentralized science,
0:15:16 which you mix it in with a more scientific application.
0:15:19 So one sort of simple heuristic is anywhere
0:15:20 where you want to build a network.
0:15:23 And as a challenge to build the early phases of the network,
0:15:26 crypto can be a really useful way to help bootstrap that.
0:15:27 Oh, interesting.
0:15:28 And so that’s one of my favorite areas
0:15:32 where the physical world and robotics intersecting
0:15:34 with data collection and all these other themes
0:15:36 that intersect with AI are relevant.
0:15:38 Mark actually gave me this framework,
0:15:39 which I like a lot,
0:15:41 which is is the AI frosting or sugar?
0:15:44 You know, if the AI is a core ingredient,
0:15:46 if it’s a frosting, all the incumbents are gonna win
0:15:49 ’cause you just slap a chat bot on your existing product
0:15:50 and you’ve got distribution.
0:15:52 You know, you have that like selling reference power,
0:15:54 incumbent relationships,
0:15:55 if it’s more fundamental of an ingredient,
0:15:58 like you can’t actually just slap AI into the product,
0:15:59 you have to build it from scratch
0:16:00 and that favors the newcomers.
0:16:01 It’s just very TV.
0:16:04 We haven’t seen anything that tells us what the answer is.
0:16:08 The more seal your thing, the more steamorphic it is,
0:16:10 which is early cycle thing,
0:16:12 the more it probably favors the incumbents.
0:16:14 Another way maybe to frame Mark’s thinking
0:16:16 is the Clay Christensen view,
0:16:19 where is it disruptive or sustaining?
0:16:20 And specifically, I think what people misunderstand
0:16:21 about Christiansen view, right?
0:16:23 Disruptive doesn’t just mean new.
0:16:25 It means misaligned with the incumbent business model.
0:16:26 Yeah, exactly.
0:16:27 That’s sort of the interesting part of his book, right?
0:16:30 Is it even when the smart incumbents sees it coming,
0:16:32 it’s very, very hard for them to react to it
0:16:34 because it’s not what their best customers are asking for.
0:16:35 Yeah, exactly.
0:16:37 And so that’s where I think
0:16:39 somewhat overlaps with Mark’s frosting icing thing.
0:16:40 Well, it could be that the business model
0:16:42 is a fundamentally shifted business model.
0:16:43 Yeah, so you come in and you’re like,
0:16:45 instead of databases,
0:16:47 it’s some radical new architecture that’s database free.
0:16:47 I don’t know why.
0:16:50 It’s something that cannibalizes the incumbent business model
0:16:51 and therefore makes it organizationally
0:16:55 and economically harder for the incumbents to layer it on.
0:16:56 We haven’t seen it yet.
0:16:58 We’ve seen people talk about outcome based pricing.
0:17:00 Well, let’s talk quickly about consumer.
0:17:01 So in consumer right now, at least,
0:17:03 I don’t think you see a lot of network effect businesses,
0:17:04 right?
0:17:05 Sure.
0:17:07 Like as successful as the Clause and ChatGPTs are,
0:17:09 I don’t think they have a network effect.
0:17:10 They’re switching costs are relative.
0:17:11 Maybe they learn your history.
0:17:12 But the question is right,
0:17:14 how do they avoid in the steady state
0:17:16 of having just like a model and price competition
0:17:18 to the price of the bottom, right?
0:17:19 Obviously they’re important big businesses,
0:17:21 but will they be dominant?
0:17:22 Yeah.
0:17:23 And then what’s the opportunity for new startups?
0:17:26 If you’re doing venture investing and AI consumer,
0:17:28 like you see a lot of these things that make your face
0:17:30 prettier, like these kind of fun apps
0:17:32 and they zoom up in the app chart
0:17:34 and then TikTok copies it and so forth, right?
0:17:37 ‘Cause it’s just not, ’cause again, no network effect.
0:17:39 And there’s just technique kind of strategy I like to talk
0:17:41 about called come for the tool, stay for the network.
0:17:43 And the idea is maybe you can use that,
0:17:46 make my face prettier and then use that as a hook
0:17:48 to get people into your new network,
0:17:50 like your social network, possibly,
0:17:51 although it just feels very, very hard today,
0:17:54 given the scale and power of these incumbents.
0:17:56 And that, by the way, we’ll intersect back to crypto
0:17:58 ’cause what crypto is and what I argue in my book
0:18:00 is that crypto is a new way to build networks.
0:18:00 And so, you know,
0:18:02 you sort of have the chocolate and peanut butter.
0:18:03 You have AI with all these really interesting use cases
0:18:06 and then you have this new technique for building networks.
0:18:08 AI, the interesting use cases, but no network effect.
0:18:09 And then you have this new thing that’s like,
0:18:11 all network effects are their interesting ways
0:18:12 to combine them.
0:18:13 So before I get to that,
0:18:15 I think it’s important to talk about
0:18:17 how big technologies roll out in multiple stages.
0:18:19 So there’s a distinction.
0:18:22 It’s not my distinction, but I’ve talked about a lot.
0:18:24 It’s sort of one way to think about technology
0:18:26 is that they can do one of two things.
0:18:28 They can do old things better
0:18:30 or they can do new things you couldn’t do before.
0:18:31 We call the first one skeuomorphic.
0:18:33 This is a Steve Jobs term,
0:18:36 which sort of refers to products and designs
0:18:38 that kind of harken back to a prior era
0:18:40 to make them more understandable.
0:18:42 And then there’s what we call native apps,
0:18:43 which are things which are the kind of new things
0:18:44 that couldn’t be done before.
0:18:46 And then there’s actually a third stage, I think,
0:18:48 which is second order effects,
0:18:49 which is you created the car
0:18:51 and now you have the highway system
0:18:53 and now you’re able to create suburbs
0:18:55 and trucking infrastructure, right?
0:18:58 Those are second order downstream effects.
0:19:00 There’s a famous line that good science fiction writers
0:19:02 predict the car, great science fiction writers
0:19:03 predict the traffic jam, right?
0:19:05 So like, it’s like that idea.
0:19:07 So it’s like, what are the second orders?
0:19:09 Like, Bitcoin is something that couldn’t have existed
0:19:11 before social networking, right?
0:19:12 So 30 years ago, you say someday people
0:19:13 are gonna have their own media
0:19:15 and you’re gonna remove these gatekeepers.
0:19:16 Who would have thought?
0:19:18 Then you’re gonna create these digital currencies.
0:19:19 – There would have been no way to create the community.
0:19:20 – Yeah, yeah, it would have been
0:19:22 in your time’s article saying it’s a stupid
0:19:24 and then it does the end of it, right?
0:19:26 There’s nowhere to get together and talk about it
0:19:28 and create, I mean, they’re really kind of religious
0:19:30 movements, you know, most token communities
0:19:32 and they need places to congregate and discuss it.
0:19:33 And now they have that.
0:19:34 And so there’s all these kind of second order.
0:19:36 I mean, we’re seeing effects in politics
0:19:37 and all these other things.
0:19:40 There’s the whole arguably our society and world
0:19:42 is changing as a second order effect of social networking.
0:19:43 So one way to think about AI.
0:19:45 So the first stage is the scumorphic phase,
0:19:48 which is this is the stuff you see talked about all the time
0:19:50 in the business and startup community
0:19:52 of like your customer service bots, right?
0:19:54 You take a job that’s currently done by a person
0:19:57 sitting in a call center and you replace that
0:19:59 with a AI voice and chat bot, right?
0:20:02 In the simplest case, it’s a one-to-one exchange.
0:20:04 It’s cheaper and it’s more systematic
0:20:07 and it will displace jobs.
0:20:08 Hopefully it will also create
0:20:11 equally or more jobs and better jobs.
0:20:12 But that’s sort of an obvious first stage.
0:20:14 And look, and this is, I think one of the reasons
0:20:15 people get so excited about the opportunity for AI
0:20:17 is you can just see that happening
0:20:20 in, I don’t know, tens of millions of jobs, I guess.
0:20:23 Like the whole laptop middle kind of section
0:20:26 of the economy, you can see many of those jobs.
0:20:29 Everyone, including us, who spend their days typing emails.
0:20:31 (laughing)
0:20:32 That’s the joke.
0:20:33 It’s like we can speculate on it,
0:20:35 but we’re part of that group too.
0:20:36 So that’s phase one, right, scumorphic.
0:20:39 But that’s phase one can last 20 years.
0:20:41 So just to be clear, the next phase is the native phase.
0:20:43 And that, to me, that’s what gets me more excited.
0:20:44 And by the way, let me give a little analogy
0:20:45 to the internet.
0:20:47 So the scumorphic phase was the ’90s.
0:20:48 And so basically, if you look at ’90s internet,
0:20:51 people were taking offline things
0:20:53 like magazines and catalogs and putting them online.
0:20:56 So you would go buy things, and it was much easier.
0:20:58 You could type in a website and go buy this rare book
0:21:01 on Amazon and it was much easier and it was convenient,
0:21:02 but it was fundamentally something
0:21:03 you could have done before.
0:21:04 It just would have been clumsy in getting
0:21:06 some weird magazine catalog or something.
0:21:08 But it wasn’t until the 2000s that people did things
0:21:09 like social networking.
0:21:10 And these things were just brand new things.
0:21:11 There’s no analog in the offline world
0:21:14 to a lot of these new behaviors that people created.
0:21:16 I talk a lot in detail about this in the book
0:21:17 if people aren’t interested.
0:21:19 So anyway, so you saw the internet play out that way.
0:21:20 ’93 was mosaic.
0:21:23 And 2000, I would say five-ish was sort of YouTube
0:21:25 and four, I think, was Facebook or whatever it was.
0:21:27 So it took at least a decade.
0:21:28 And by the way, one of the things you get
0:21:30 in the native phase, which is why it’s so exciting,
0:21:32 is you get new products, you get new forms of media.
0:21:34 So if you go back when photography
0:21:36 was growing in popularity,
0:21:39 there were all of these cultural art criticism,
0:21:42 think pieces about what will happen to art.
0:21:44 The famous, like Walter Benjamin,
0:21:46 the art in the age of mechanical reproduction.
0:21:47 There’s all these famous essays
0:21:49 where it was like, what’s gonna happen?
0:21:50 ‘Cause now that you can take a photo
0:21:52 and create a beautiful landscape,
0:21:53 what’s the role of the artist in that world, right?
0:21:55 And so people were worried about it.
0:21:56 In the same way, they’re worried today
0:21:57 about generative AI, right?
0:22:00 So like, what if you can now create a movie?
0:22:02 Looks like you can pretty soon, right?
0:22:03 I mean, images is there.
0:22:06 Images is there, and probably videos coming soon.
0:22:08 What happened in the case of photography
0:22:09 is that you had, I think two things happened.
0:22:12 Fine art went more abstract and went away from photography,
0:22:15 right, that leaned into what they were unique at.
0:22:16 And that’s when you had whatever,
0:22:18 cubism and all these other kinds of movements.
0:22:19 And then on the other side,
0:22:20 what I think was really interesting, right,
0:22:21 is you had the rise of film.
0:22:22 You had someone say, hey,
0:22:25 maybe you can use machines to replace photography,
0:22:27 but you can also now use machines
0:22:28 to create a brand new art form
0:22:30 that never could exist before, right?
0:22:31 You sort of had it with animation,
0:22:32 but now you can do it a really interesting
0:22:34 sophisticated way with film, right?
0:22:36 And so film would be,
0:22:37 what was the native form of media
0:22:40 in the age of mechanical reproduction, right?
0:22:41 – Oh, that’s a fascinating knowledge, yeah.
0:22:42 – And so I think to like today,
0:22:44 like so when you look at the generative AI,
0:22:45 like the negative way to look at it,
0:22:47 and you do see some of a lot of this negative sentiment
0:22:50 from like the art community and things on Twitter,
0:22:52 where they say, look, this is just a cheap replacement
0:22:54 for human creativity.
0:22:57 The positive way to look at it is this is the base layer,
0:22:59 in the same way that film was a base layer back then,
0:23:02 but now there’s this new canvas of human creativity
0:23:03 where you can create new art forms.
0:23:04 I don’t know what those are.
0:23:06 They may be virtual worlds or games
0:23:08 or new types of films and movies.
0:23:09 I don’t know.
0:23:10 – They may intersect with a new way
0:23:12 to consume the media altogether.
0:23:14 – Yeah, maybe there’s new interfaces,
0:23:15 and this is to me what’s so exciting
0:23:17 about the new native media, the native apps,
0:23:19 is that I won’t think of it,
0:23:21 because in my experience,
0:23:22 through watching some of these waves in the past,
0:23:25 is there, it really does take brilliant creative people
0:23:28 to come up with these new things,
0:23:29 and it surprises you in many cases.
0:23:31 And so I think that that’s gonna be the exciting phase
0:23:35 I’m looking for is not how do you just use this technology
0:23:37 to do the things you could do today, but do them cheaper,
0:23:39 but how do you use the technology to push the frontier
0:23:41 and do things that could never be done before
0:23:43 in the same way that film did that, right?
0:23:44 – Yeah.
0:23:47 – I think photography probably unlocked more opportunities
0:23:50 for creative people than it removed,
0:23:51 and I think this would be the hope in this kind of phase.
0:23:53 So that’s the media example,
0:23:56 but there’s probably that for consumer applications
0:23:59 and that for social networking, and that for productivity,
0:24:01 and so that will be the really exciting thing I think to see
0:24:03 is not just the replacing things we do today,
0:24:05 but come up with brand new behaviors
0:24:07 that are things we couldn’t do before.
0:24:09 And then the third thing is the second order effects, right?
0:24:10 So you create this new world,
0:24:12 so you’ve created this world of social networking.
0:24:13 As interesting to think with social networking,
0:24:14 we’ve seen it play out.
0:24:16 You know, you sort of have social networking rise
0:24:17 in the 2000s.
0:24:18 I think it hit a tipping point.
0:24:19 Maybe the Obama election.
0:24:20 – Yeah.
0:24:22 – Was that the 2008 and then ’12 too.
0:24:25 He really leaned into using that,
0:24:26 and I remember seeing all these news articles,
0:24:28 like, wow, this is different.
0:24:31 The bit had flipped from online as a secondary,
0:24:32 sort of online was primary,
0:24:35 but then we started seeing these kind of weirder things,
0:24:37 like I think the Trump movement and the populism
0:24:38 just surprised everybody,
0:24:41 and you just started seeing movements and just behave,
0:24:43 and I think we still haven’t really figured out
0:24:45 what’s going on, but where all this is headed,
0:24:48 and we’re in this disequilibrium state, I guess.
0:24:49 Anyways, those sort of second order effects
0:24:51 of social media will probably play out for,
0:24:53 as I mentioned, like crypto,
0:24:55 and I think a bunch of other interesting movements today
0:24:57 are second order effects of social media,
0:24:58 and that will probably play out for 20, 30 years,
0:25:00 and so that will probably be phase three
0:25:01 of the AI revolution.
0:25:03 – Yeah, and just think about the timelines.
0:25:04 – Yeah, and it’s probably gonna take a very long time.
0:25:06 For all, like, I’m always overly up,
0:25:07 stick on these things, historically.
0:25:09 I’m like, okay, we’re done with the skemorphic phase of AI,
0:25:10 now we’ll do the native phase,
0:25:13 but the reality is each phase probably takes a decade.
0:25:15 – One of the interesting things you said
0:25:17 around these distinct phases,
0:25:18 obviously the internet took a long time,
0:25:20 partially because you had to build a network.
0:25:21 – Yeah.
0:25:22 – It was a supply and demand issue, right?
0:25:23 – A physical network, and then also–
0:25:25 – The literally laying cables and the wireless–
0:25:26 – Yeah, laying cables,
0:25:30 and sure you have to build large clusters of compute GPUs
0:25:31 here with networking,
0:25:34 but I think the constraining factor
0:25:37 for getting from that skemorphic phase to the native phase
0:25:39 is not necessarily capabilities themselves,
0:25:41 but like human creativity.
0:25:42 – Yeah, I think so.
0:25:43 – Even like ideas.
0:25:46 – I think the bottleneck will be humans and regulation,
0:25:48 which are obviously closely connected,
0:25:50 and I think humans on both the supply and the demand side,
0:25:51 probably more on the demand side.
0:25:52 So meaning supply side,
0:25:54 you need to have people come up with all the creative things,
0:25:56 but the world’s different now
0:25:58 in that I just think the startup world is different now.
0:26:01 It’s much more mature and much more sophisticated, honestly,
0:26:03 than like when I was coming up in it.
0:26:04 I mean, when I was starting off,
0:26:05 there were 10 venture firms,
0:26:07 now there’s thousands, the number of startups,
0:26:10 and honestly, there’s a lot of good, smart advice out there.
0:26:12 – Yeah, this is a more popular path for smart people
0:26:14 to go to. – Yeah, it’s like a thing you do,
0:26:16 like in places like Y Combinator
0:26:17 and other places have done a good job of this.
0:26:19 If you’re coming out of a top school,
0:26:22 I mean, even 10 years ago, this wasn’t like,
0:26:23 I knew people that were like, “Wow, you could do startups.”
0:26:25 I mean, definitely that was the case 15 years ago,
0:26:28 but now I think it’s like an established career path.
0:26:29 There’s an established set of mentors,
0:26:30 established set of funding.
0:26:33 There’s a canon of pretty good advice out there.
0:26:35 Like the standard advice used to be terrible advice,
0:26:36 now it’s good advice.
0:26:37 You can come out to San Francisco,
0:26:39 and I think relatively easily,
0:26:41 if you’re a smart network-friendly person,
0:26:43 get embedded pretty quickly.
0:26:44 And then, you know, and then still looking about,
0:26:45 it’s gotten just very good.
0:26:48 It’s throwing tons of capital energy against those problems.
0:26:49 So there’s the supply side.
0:26:52 I suspect the demand side is more like another meaning,
0:26:55 like changing organizational and human work
0:26:59 and behavior patterns, like getting an organization,
0:27:00 like take the video example we’re talking about.
0:27:02 Yeah, I mean, look, I wrote my book.
0:27:05 I wanted to have my own voice, use AI to read the book,
0:27:08 using my own voice, both the publisher and Audible,
0:27:11 the podcasting platform, ban AI completely.
0:27:14 And part of its unions and just a bunch of resistance.
0:27:15 I think people know this,
0:27:16 but like the capabilities are fully there to do that.
0:27:18 Yeah, I mean, like, look, Mark and Jason
0:27:19 had a great blog post, it’s like,
0:27:22 how do I know they’re gonna ban AI like medicine?
0:27:23 ‘Cause they already have, essentially.
0:27:25 I mean, essentially, like these things
0:27:26 are so heavily regulated.
0:27:28 And so many areas where it’s gonna have an impact
0:27:29 are so heavily regulated.
0:27:31 And just the organization, like,
0:27:32 look, take the Hollywood Gen AI thing.
0:27:35 You’d have to lay off a whole bunch of people, probably,
0:27:37 who you don’t wanna lay off, who are unionized.
0:27:39 So that means maybe there’ll be some fresh upstarts,
0:27:43 maybe in another country who create AI-native movie studios.
0:27:45 But that will take a very long time.
0:27:47 The right answer is probably to harness
0:27:48 all of that talent in Hollywood
0:27:51 and combine it with AI in some way.
0:27:53 There is a lot of very smart people and talent.
0:27:54 But how long will that take?
0:27:56 Culturally, it may take a whole generation
0:27:56 to really play out, right?
0:27:58 So that’s something by the demand side, right?
0:28:01 And then just human behavior, changing your workflow,
0:28:03 using an AI assistant, I don’t know.
0:28:04 Anyway, so.
0:28:05 Yeah, having like a co-pilot for everything you do,
0:28:06 like it feels like it’s-
0:28:09 Yeah, maybe that can be solved with interfaces and things.
0:28:09 I don’t know.
0:28:10 Then there’s the policy side,
0:28:12 which is there’s going to be this resistance,
0:28:14 I’m discussing already is,
0:28:16 going to be enshrined,
0:28:18 there’s going to be movements to enshrine it in law.
0:28:19 And that’s going to play out,
0:28:20 and I think in multiple levels,
0:28:22 it’s already starting to play out in the courts,
0:28:24 and it’s starting to play out in like state legislatures,
0:28:26 with like the California had the AI bill,
0:28:28 you know, you have a bunch of lawsuits around copyright.
0:28:31 My view is ultimately this will play out in Congress.
0:28:33 This is such a big issue when you have something
0:28:35 that affects tens of millions of jobs,
0:28:37 it is beyond something that people are going to allow
0:28:38 just to happen through.
0:28:40 Through free markets, yeah.
0:28:42 Yeah, and through regular court decisions.
0:28:44 Like the copyright thing is an example.
0:28:46 Like right now the question is,
0:28:48 when an AI system is trained on a piece of data,
0:28:49 is it copying that data,
0:28:50 or is it learning from that data?
0:28:51 Yeah. It’s a philosophical question.
0:28:52 There’s a fundamental question
0:28:54 across different media happening right now.
0:28:55 That’s right.
0:28:57 And so you could have five years from now,
0:29:00 some federal judge decide that philosophical question,
0:29:02 or I think more likely you’ll eventually have
0:29:05 some legislation, like congressional legislation,
0:29:06 that’s some kind of compromise
0:29:08 struck between the media industries and tech industries
0:29:09 that comes up with a solution
0:29:11 that both creates incentives for creators,
0:29:13 but also allows AI systems to exist.
0:29:15 I don’t know, but that thing will play out
0:29:16 over a very long time.
0:29:19 When will you be allowed to use AI in medical and finance?
0:29:21 And I mean, significant, what is it?
0:29:24 Probably 70% of our economy are regulated industries, right?
0:29:24 Yeah, of course.
0:29:25 You know, on the flip side,
0:29:28 like the stuff with the Waymo is really impressive.
0:29:30 I’m surprised they’re actually allowed in San Francisco.
0:29:31 And they’re–
0:29:33 Well, it turns out it’s seven to 10 times safer
0:29:34 than a human driver.
0:29:36 And there’s now millions of miles of game-filled–
0:29:38 So maybe that’s the playbook
0:29:40 to get this stuff adopted more broadly.
0:29:44 What is an ideal future state of the internet?
0:29:47 So there’s zero cost of creation and distribution,
0:29:49 transparent ownership, governance.
0:29:51 What does this look like?
0:29:52 I think that we’re at a crossroads
0:29:55 and there’s a real question as to whether it looks
0:29:57 more like its original vision,
0:29:58 which is the vision of the internet,
0:30:01 like the ’90s vision and the ’80s vision or something,
0:30:04 was an internet that was community-owned,
0:30:05 community-governed.
0:30:08 The money mostly flowed to the edges of the network
0:30:10 and not to the intermediaries in the middle.
0:30:11 Like, originally, in the ’90s,
0:30:13 the money flowed to the edges,
0:30:15 to small businesses, to innovators, to entrepreneurs.
0:30:16 If you looked at a map today,
0:30:18 it’s mostly flowing to the middle.
0:30:20 This is why the Seven Company–
0:30:21 It’s 200 million revenue.
0:30:22 So it’s not working at all.
0:30:24 Yeah, I think the top five internet companies
0:30:26 are something like more than half of the market cap,
0:30:28 not more, and it might be absolutely higher by now.
0:30:30 And so just you have all the green stuff
0:30:31 flowing into the middle.
0:30:32 I think of it as two kind of important things
0:30:34 that you want is power and money, control.
0:30:36 And my core argument in the book
0:30:39 is that those two questions are a product
0:30:41 of how you build these services.
0:30:42 The first sentence of the book
0:30:44 is your architecture is your destiny
0:30:44 or something like that.
0:30:46 Like, the architecture you choose
0:30:48 determines how it’s controlled and how the money flows.
0:30:51 And so, and I think we’re really at a kind of critical point.
0:30:53 In fact, I worry we’re approaching a point in no return
0:30:55 where it’s going to be an internet-controlled
0:30:56 by five companies.
0:30:57 And what’s happened is these networks
0:30:59 have all gotten to a certain scale
0:31:01 and they’ve just decided that the next kind of wave
0:31:03 is to keep you trapped there.
0:31:04 Well, there’s no way to grow users anymore.
0:31:05 They’ve captured all these users.
0:31:06 That’s right, that’s right.
0:31:08 They climb the ladder and they’re kicking it away.
0:31:09 And it’s really negative.
0:31:11 And this is why we as a firm have felt
0:31:12 that this is such an important topic.
0:31:14 Of being able to build internet services
0:31:16 with new architectures, like using blockchains,
0:31:19 is such an important topic for the future of small tech,
0:31:20 little tech, as we call it.
0:31:21 Along with open source AI,
0:31:23 the other kind of critical thing,
0:31:26 which is if startup has to pay this giant tax
0:31:29 to an incumbent to build competitive services,
0:31:30 they won’t be able to build services
0:31:31 that threaten those incumbents, right?
0:31:32 Yeah, we’ve seen that before, right?
0:31:34 Like you’ve talked about it as Shingo
0:31:35 was built on top of Facebook.
0:31:37 Yeah, it’s platform risk, right?
0:31:38 I mean, you’re building on quicksand.
0:31:41 So startups need access to distribution and networks
0:31:43 and they need access to modern software,
0:31:44 open source software.
0:31:46 And so I think those are the critical questions.
0:31:47 Those will be, I think a hugely important thing,
0:31:49 which is why we’ve invested so much time and money
0:31:50 and it is the regulatory side of this,
0:31:52 is like what policies are there?
0:31:54 And are they policies that encourage competition
0:31:56 and innovation and little tech?
0:31:59 And then I think just raising awareness of these topics
0:32:01 and having discussions about them are important.
0:32:03 What I’m worried about now is sort of backing ourselves
0:32:05 without having really thought it through
0:32:07 into a situation where there’s four companies
0:32:09 that control everything and it ends up,
0:32:10 we’re kind of eating our seed corn.
0:32:12 Like so much of what we benefit from today
0:32:15 is the startup innovation of the past
0:32:17 and we’ll risk losing that if we let these small set
0:32:18 of companies control everything.
0:32:20 Yeah, well, I’m optimistic.
0:32:22 Look, the bright side is through all the work
0:32:24 that you guys have done in our firm,
0:32:26 we’ve gotten the word out about little tech.
0:32:29 And I think understanding that building a new architecture,
0:32:32 new infrastructure, and then the importance of open source,
0:32:33 I think the word is getting out.
0:32:35 So this is awesome because thanks for being here.
0:32:36 I always love talking to you.
0:32:37 Thank you.
0:32:39 (upbeat music)
0:32:42 (upbeat music)
0:32:45 (upbeat music)
0:32:47 (upbeat music)
0:32:50 (upbeat music)
0:32:52 you
0:32:55 (gentle music)
Technology doesn’t grow in isolation—it evolves in waves. Just as mobile, cloud, and SaaS shaped the internet of the past 20 years, so too could crypto, AI, and new hardware usher in an era of the internet that’s pro-innovation, pro-startup, and pro-creator.
Speaking with a16z Growth General Partner David George, a16z crypto Founder and Managing Partner Chris Dixon breaks down his vision for a new internet, from using crypto to decentralize AI infrastructure and kickstart network effects, to why AI will be this era’s native form of media just as film was in the 1930s. He also explores why the internet’s original covenant—where content creators traded free access for search traffic—is breaking today, and how a better internet could introduce entirely new business models for creators.
Right now, we have a choice to make: will the next era of the internet be shaped by a handful of centralized players, or transformed into an open ecosystem where power and control flow to creators across the globe?
Resources:
Watch the conversation here: https://youtu.be/gioxu1CVjhM
Read more, including the full transcript, here: https://a16z.com/ai-crypto-internet-chris-dixon/
Chris’s recent article on blockchain innovation: https://a16zcrypto.com/posts/article/blockchain-ai-internet/
Find Chris’s book, Read Write Own: Building the Next Era of the Internet:
Penguin Random House: https://www.penguinrandomhouse.com/books/744504/read-write-own-by-chris-dixon/
Penguin UK: https://www.penguin.co.uk/books/459860/read-write-own-by-dixon-chris/9781804949245
For more resources on AI & crypto visit: https://a16zcrypto.com/posts/?tag=ai-crypto,web2-to-web3
Stay Updated:
Let us know what you think: https://ratethispodcast.com/a16z
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
Subscribe on your favorite podcast app: https://a16z.simplecast.com/
Follow our host: https://twitter.com/stephsmithio
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.