Abstract

Tim O'Reilly traces how the incentives of shareholder capitalism, from Milton Friedman's 1970 statement to what Sam Bankman-Fried called the number-go-up era, pushed Google and Amazon from organic search toward advertising and pay to play. He uses economic rents and the gap between the economies of capital and labor to explain the pattern, and asks whether the emerging artificial intelligence (AI) economy will repeat it. His remedy is mechanism design, changing the rules of the game as Unix and the Model Context Protocol (MCP) did, so that AI labs measure success by diffusion and not only by capability. Audience questions follow, including Matthew Connelly on universities, data and compute.

Chapters

Transcript

AI-generated transcript. It was produced from the recording by speech recognition and cleaned up with AI. It has not been edited by the speaker and may contain errors. The captions on YouTube are AI-generated in the same way. Each timestamp links to that moment in the video.

[0:12] Good afternoon everyone. I'm Louis Hyman. I'm the director of the Center for Economy and Society here at Johns Hopkins University and welcome to all those who are in person and those who are online. We are privileged today for our annual Margaret Levi Lecture in Political Economy to have Tim O'Reilly. Tim is the founder and CEO of O'Reilly Media and one of the most influential figures in the history of the Internet. Now, you might think he came out of computer science, but no, he came out of classics. He is a humanist by training, even though he's one of the most influential thinkers in Silicon Valley. He launched one of the very first commercial websites, Global Network Navigator, in 1993, and then went on to popularize open source software and coined the term Web 2 .0, which you may have heard of. Today, he brings that humanistic and entrepreneurial perspective to questions about artificial intelligence and the future of political economy in Silicon Valley.

[1:16] Thank you for being here, and please welcome Tim O'Reilly. Thank you very much for having me here. I'm really honored to be asked to give a lecture in Margaret Levi's name. I've been asked to reflect on changes in the political economy of Silicon Valley. I've spent 40-plus years there. I'm not a political economist. I'm not a historian. I'm just somebody who's lived through a lot of things. And I've principally known in some ways as a storyteller that has made sense of what has happened. And right now I'm spending my time trying to make sense of AI in the same way that I try to make sense of open source software on the web, so on. So, you know, Margaret understood that

[2:17] advances in the ability of humans to work together is really the heart of a better future. Cooperation is the human superpower. And if you think about this last generation of technology, in fact, just computer technology, it's an amazing advance in the ability of humans to work together to share our collective knowledge to build on each other's work. And so that's, you know, I think at the heart of Margaret's thinking about, you know, how we build systems to get better at doing these things. And technology has very much been part of that. But something is oddly changing, because the industry that pioneered advances in collective intelligence at scale used to gender love from its customers because it really solved their problems. And now it too often inspires fear and loathing.

[3:17] So how did things go so wrong? So let me start a little bit with my own history. I started my company. It's called O'Reilly Media. originally a book publisher and then became a conference producer and we have an online learning platform where we've taught generations of programmers about the latest and greatest technology and we still do that in the age of ai i started my company more than 40 years ago with just 500 worth of used furniture and we raised every penny from our customers you know we sold them things we built things they wanted we sold them and we grew uh that way um you know to a point where we have a business that generates hundreds of millions of dollars in profitable annual revenue you know we got there the slow way you know organic steady growth it's quite anathema to the rush to the top of whatever appears to be the latest hill that seems to characterize of Silicon Valley today. So my original corporate goal was stated as interesting work for interesting

[4:22] people. And we went from there. And eventually, though, about 2000, we actually came up with two key principles. One was we decided that our mission was to change the world by spreading the knowledge of innovators that describe what we do. And the other was to create more value than you capture. and this came out of a conversation that I'd had with several well-known early internet billionaires. You recognize the names, but I don't want to. And they said, hey, I started my company with an O'Reilly book. And I was joking at one of our management meetings. I said, we got 40 bucks and they got a billion. That's pretty cool. That's pretty cool. And my VP of marketing said, we create more value than we capture. And I went, that's a hell of a slogan. Let's use it. And I think it's really important because great companies, I think, do create more value than they capture. And some of my proudest moments, I remember

[5:24] talking to an entrepreneur and he said, I went through all these different ideas and I realized they all failed the Tim O'Reilly test. It was just like they were just an attempt to make money and they weren't actually doing this thing of creating value. So anyway, when we launched GNN, which was the first commercial website in 1993, it was the first ad-supported site on the World Wide Web. And it was really focused on how do we actually say, what's the best of this new breed of content? We were publishers and we were going, OK, yeah, we'll find the best sites. Well, it turned out that the world didn't quite work that way because you ended up with eventually, five years later, with a company, which we know now is Google, which figured out how to do this curation at scale. It was an early example of what you could consider the bitter lesson. How many people here are familiar with that term? Not too many of you.

[6:25] Okay, this is a famous computer science paper by a guy named Richard Sutton in 2019, where he observed that, and he was mainly talking about chess and go AI there, but he basically said it turned out that human expertise was not the winning strategy. Brute force won over time. And so even when you had AlphaGo beat Lee Sedol, but then AlphaZero, which was trained with no human games, just learned by playing itself, beat AlphaGo. So he was saying this seems to be the bitter lesson is that human expertise doesn't matter. I'm spending a lot of my time today saying we have to stand up for humans. We have to find what is the high ground of human expertise. piece. And anyway, so, but I want to kind of go back to this, you know, this bit of history on the early web. You know, it was the playground of amateurs, you

[7:29] know, not entrepreneurs. You know, we have this whole narrative that we've been sold that innovation starts with people who are, you know, you know, they have an idea for a business, they go raise venture capital, and they start a company. Well, all of the great technology revolutions that I have been part of did not start that way. They started with amateurs. They started with people who were doing it for the fun of it. You know, the early PC enthusiasts, you've probably all seen pictures of the Homebrew Computer Club. You know, Steve Wozniak carved his first computer out of wood, you know? And, you know, the first Apple One, you know, his prototype. And then they put some circuit boards in it. you know you think about the early web you know nobody thought there was any money in it everybody was somewhere else they were all doing pc software at that point and you know the you know google only raised 35 million dollars right in 1998 you know that's all it took because they

[8:30] basically and there was just two guys who just had they thought there was a problem that needed solving right and so i think that we really need to understand that it's love and creativity that drives us to invent the future. And Silicon Valley has been selling us a bill of goods that says, no, no, it's all about capital. Anyway, sorry, I'm not following my script very well here. But anyway, I want to kind of come back to this idea of collective intelligence. In my 2005 essay, What is Web 2.0? So I described how the companies that survived the dot-com bust had all in one way or another figured out how to help people navigate the information explosion. I went back and I read Herbert Simon's original talk from 1971, the one where he said, well, what does information consume? Of course, it consumes attention. And he went on. Actually, it was in the discussion. It was a discussion after he presented his

[9:33] paper. your information will be so abundant, we will need machines to help us allocate our attention. And he was right. And if you think about all of the great companies since the dawn of the web, they have all been machines for allocating human attention. Google, Amazon, Facebook, Twitter, they're all algorithmic machines for helping us navigate untold amounts of information. And AI is in that same heritage. It's helping us to manage ever more information. So I called this, in that paper, I called it harnessing collective intelligence. I was looking at, well, why did some companies survive the dot-com bust and others didn't? I said, all the ones that did were doing this harnessing collective intelligence thing in one way or another. And it seems to me that this was, in economic terms, an incredible advance in network data market coordination. And it felt like a public good, even when it was privately owned.

[10:36] And if you think about this, Hal Varian, who was formerly Google's chief economist, told me that only about 6% of Google search results have any ads at all. 94% are just ad-free because there's nobody who wants to advertise on those topics. But they serve that up anyway. And so it's this incredible surplus that we all got. And Google's early advertising was sort of just a sidecar to the organic search. It was just this additional channel where you go, okay, well, if you didn't find what you want, somebody could say, well, here's an ad. And it was off to the side. But if you'd watched over time, it became more and more intrusive. Until today, you're basically scrolling down many pages before you get to the top organic result. But I want to just be in that first moment. The early days, it was all about solving this information problem. Amazon was like, we're going to find you the best product at the best price.

[11:36] So it kind of lasted from the mid -90s to about 2010. Shortly after the financial crisis, I think Silicon Valley all turned to the dark side. Google used to have this slogan, don't be evil. They actually retired it officially in 2019. I mean, why would you literally retire it? And they're now facing all the same surveillance capitalism charges as people like Facebook. you know advertising pushes organic search all the way down the page amazon i did had a project that University College London or what i was calling algorithmic rents where we were looking at the ways that you know amazon and and google now basically manipulate user attention in order to extract money from their suppliers and uh you know you have to advertise

[12:38] on amazon suddenly has dollar advertising business it's basically pay to play and here's this thing their organic search which was this incredible advance in market coordination is replaced by paying for the top spot and you know we did the research and we said okay you know we looked at what the organ you could find the location of the top organic search result on the page used to be at the top and now it's maybe 23 spots down. And the products that are at the top, which are paid for, are often ranked a third lower organically, and they cost 17% more. So if you're ever using Amazon, just scroll down until you find the top organic result, the one that does not say sponsored, and you're probably looking at the best product. But the fact that they have changed that model, What does that tell you about the incentives in Silicon Valley? You know, it's not that good people turn bad, although maybe they did

[13:42] some of that too. It's really the writing economic, overriding economic incentives of shareholder capitalism. You know, in 1970, you know, Milton Friedman said the social responsibility of business is to increase its profits. He had a whole rationale, and it was like the principal agent problem and blah, blah, blah, blah, blah. But the bottom line was we basically, eventually with Meckling and Jensen at Harvard, we ended up enshrining in corporate governance the only responsibility is to make money for your shareholders. And this was the beginning of what Sam Bankman-Fried later called the number-go-up era. That is the objective function. That's a term from computer science when they're training models. They're going, what are you training this thing to optimize for? Our companies are our AIs, slow AIs that are trained to make number go up. And they literally, number stops going up. The company gets punished by the market. They can't pay their employees in this

[14:45] appreciating currency of stock. All kinds of bad things happen, right? and so you know our whole uh you know in a lot of ways corporate governance was our trial run of ai governance you know and you look at that how's it going you know uh Eric Ries uh recently uh who's famous for a book called uh uh the uh um the lean startup which by the way he came and asked me to publish and i said eric uh i think you'd do better with a new york publisher this is a kind of book that you know really could go become a bestseller and you know we've been friends ever since to go literally you turned down this amazing book because you thought i would do better if i went somewhere else you know and i feel like that's how we should be working anyway but point is you know if you know it the problem were just villains you could you you try to solve it but yeah anyway back to eric his book is called incorruptible and he talks about a

[15:45] bunch of that can keep companies holding to their values even when they face the pressure from the market. But very few companies actually pull that off, very few. We built a system that rewards bad behavior. And I think it's enshrined in corporate governance, enshrined in the tax system. So there's all kinds of things that have to change. It's not just Silicon Valley that's the problem. But there is a piece that Silicon Valley shows us very deeply, and it's kind of like a cycle of innovation. And I describe it through the language of economic rents. Now, I'm not an economist, and I'm probably using things incorrectly from the point of view of economics. and but i you know i find that these these ideas often help you see the world freshly you know the the essential idea of a rent

[16:48] is it's when the reward to an actor in the economy is greater than the value they create you know and there's a lot of different kinds of of rents but the classical economists just thought about us they were looking at the landowners right the land thought that you know um they they thought that all the value was created by the the farmers right and it was appropriated by the landowners and a lot of ways they were right and then later on they went well actually maybe it was appropriate you know it was like manufacturing was valuable too and then later oh maybe trade is valuable but the point is they didn't think that just owning something and getting the the bulk of the value was um uh you know the way to go that was what they saw all these people as rentiers. And, you know, they were sort of like a persistent consequence and source of inequality. And I think you have to actually ask, where are the rents in today's world? Now, a lot of economists who talk about rent-seeking today, you know,

[17:53] mean what George Stigler called regulatory capture. And they think about, you know, what public choice talks about, intervention, you know, at the behest of lobbyists and so on. And if we could only get government out of the picture, we would be, you know, we'd live in an economy that was rent-free. But that's actually not true. The rents that we're dealing with today are rents that are very much asymmetries of power, just like they were in the days when the robber baron, you know, who would say, okay, hey, I'm going to take half of your crop this year. Google or Meta says, I'm going to take your information. I'm going to show you the stuff that I want you to see that's most profitable for me. I'm no longer going to show you the stuff that you came and asked for. That's kind of the old kind of asymmetry of power leading to asymmetry of results. so um so i i but i i think there are a kind of of rising tide rents which are kind of related to what

[18:58] joseph schumpeter called you know schumpeterian profits the idea that innovators who were ahead of the game you know could have uh you know could extract a higher than a normal rate of return because they knew something that other people didn't but he thought those kinds of advantages were eventually competed away. And he was right. It used to be that way. But now everybody talks about how do we get a moat? How do we keep our profits? And so what I've seen in my career is that you have this initial experience where a company really creates value. Go back to the early days of the PC, Microsoft, a computer on every desk and in every home. They were really expanded the market for computing enormously, but they, they turned to the dark side. They went, okay, we're going to get, you know, as my mother once said, when we were talking about Bill Gates versus open source software, she said, my, he sounds like somebody who'd come over to your house for dinner and sit down and say, you know, I think I'm going to have all the mashed

[20:00] potatoes, you know? And I kind of feel like, okay, so, so we, we went through that and we had this revolution with the web and we had a new set of people who said, no, no, we're create value for everyone. And eventually they too said, we think we're going to have all the mashed potatoes. And so, but there is that virtuous period. And it's also something, it's not just in tech. You just think about like government builds a new road or a new school in some area and the schools are better. And then more people go to live there. And there's this sort of rising tide where home values go up. And those who actually own homes benefit even more than people who just live there and rent. They get some benefit, but the homeowners really get this big rent appreciation, which is this kind of economic rent, unearned income. And in a similar way, when new technology comes along, we all became richer because of Google and Amazon and these companies that built these new advances.

[21:01] But then they came to a point where the market starts to flatten out. and they start to say, well, the market says that we have to keep increasing our profits. So what's a company to do? They can't give you your fair share of the value anymore. They have to take more. They suddenly feel like, oh, we have to have all the mashed potatoes. It's necessary because our stock price has to go. We built a machine where these companies depend on their this critical asset the smartest people by having a rising stock price that's how they pay them so if the stock price stops going up the whole thing falls down so it's it's it's really something in in the you know incentive structure that we built into our economy and i don't think it's going to get better until um uh i i suspect that we're going to have some major

[22:05] you know uh economic you know collapse before we basically go okay this system is really broken because it really is broken and uh i think we've also got a narrative coming from people like Marc Andreessen that uh Andreessen Horowitz where he's just talking about how government's the enemy and it just the enemy of innovation and go, wait a minute, you know, Tesla had loan guarantees. Google got, you know, grants from, you know, the NSF, you know, SpaceX had NASA as its founding customer. You know, are you sure that, you know, you guys did this all by yourselves? I don't think so. And so anyway, I guess the point is that I do think that there is this wonderful early period have what Cory Doctorow calls enshittification, where companies go, okay, we got to have more, and we'll make things worse. But there's another problem with Silicon Valley, and

[23:06] this really has to do with the structure of capital versus labor. If you think about, I guess this really came to my attention when Elon Musk first became the world's richest man. and I did this analysis and I looked at Bill Gates and I looked at Larry and Sergey and I looked at a number of other internet billionaires and I said, what if they had been following the same rules like of the days of Rockefeller or JP Morgan or whatever, they had a kind of a company where they kept their share of the profits versus they they had this sort of stock value and it turned out Larry and Sergey would have been worth if they had just kept their share of Google's profits over the years that they'd owned it. Whereas Elon Musk, here he was, the world's richest man, he would have been worth maybe under $100 million instead of $160 billion. He was getting this incredible lift from the fact that people were giving him

[24:10] credit for 600 years of future profits. you know and meanwhile you know at that point google was at 34 and amazon was at 45 or something like that still pretty good you know and the average you know p multiple you know is about 10 right but even 10 you think about that you know you buy a share of something and you're about you get 10 for every dollar of profit you get well it's a huge lever there right so of course you want to go make the profits go up number go up right because it drives the system and whereas an ordinary person, getting paid a salary or running a small business. You get a dollar, it's worth a dollar. But Elon Musk was getting a dollar and it was worth $600 or $1,000. And the problem is that, first of all, that's one of the biggest drivers of inequality. And I think until we realize that there's really two

[25:11] economies, there's this economy of capital and there's an economy of labor. And they're really disjoint. And the systems of government, for example, reinforce this instead of trying to counteract it. We literally pay lower taxes on capital, even though capital is now so abundant and the return on capital is so high. We have this broken idea that somehow we have to give even more incentives. When I started in Silicon Valley, I thought it was pretty good to become a millionaire. And now you have all these You know, kids who are becoming billionaires, sent to billionaires, you know. And they haven't actually even made a company that has turned any profits at all, you know. And so there's this kind of weird thing going on. Speculative bets on the future are important. It's how we fund innovation, you know. And so I don't think that that's, you know, wrong. And, you know, Tesla ended up being a real company. They kick-started the electric vehicle revolution,

[26:14] SpaceX commercial launch. I mean, these are real companies. And Elon Musk successfully used the hype machine to actually make real stuff happen. But there are a lot of people who use the hype machine to not make real things happen. You guys remember Bored Apes, right? And the whole Web3 thing and the metaverse and Mark Zuckerberg with his stupid little avatars. You know, but there were other examples. Back in the run-up to Uber's IPO, they did an acquisition of, you know, supposedly a new small company that was doing self-driving cars. I had just had a demo. And here's what it could do. It was it got under the back of a truck. There were three guys with a server rack back there. They got on the freeway on one stretch of San Francisco. and the driver said, okay, you can turn it on now. They drove on the freeway for about a year, for about a mile, staying in one lane. He's going, okay, we can get off now. Turn it off. Turn it off, we can get off.

[27:17] And Uber paid something like $900 million for that. And I was talking to John Krafcik, who at the time was the CEO of Waymo, and I said, John, don't people get that that was stock manipulation? And he says, oh yeah, I wish somebody would talk about that. I can't do it because it's, you know. But like, you know, Waymo was sitting there this is a hard problem we're working on, maybe we'll get there in 10 years. And Uber is basically running up to their IPO and saying, we're not going to need drivers. We're going to be a much more profitable company because we're doing self-driving. So it was worth it to pay $900 million for something that made your valuation go up by tens of billions. And so this is kind of like the sick thing that we have now. Hype has become more and more central. I think it's important to understand the role of hype in today's AI market as well. It's like super intelligence is coming. It's going to be so overwhelming. It's all a narrative to make number go up.

[28:24] So I think one of the things, it's not to say there isn't reality there. Unlike, for example, Web3, there's a lot of real substance to AI. You guys all know it's an incredible new But we're in a world where capital appreciation has gone nuts, where people are pushing. Again, Google had to raise $35 million, and they became profitable. And now we're in a situation where companies are raising tens of billions of dollars, being valued at tens of billions of dollars, when they haven't really done very much. And it's really a big betting economy. and I'm not quite sure long term whether that is kind of like a total aberration and it won't last or that is the new shape of the economy that capital has become so abundant that betting is what people do it's just like fast fashion and that investing is

[29:33] is not really investing anymore. It's just betting, you know? And there is this sort of current still where there are people who are investing in real businesses trying to make real things happen. But my experience in Silicon Valley is there's way too many people, you know, like you just go out there and you ask people, how's your business doing? And they tell you about, well, we just raised our C round at blah, blah, blah valuation. They don't talk about their customers. They don't talk about the, you know, and people don't talk about the products. They don't talk about the value they're delivering to customers with AI. They just talk about, well, we have this cool technology. And so I think we're kind of, you know, broken. It's a real transformative technology, but we are pretty early still in figuring it out. And, you know, this is, I was saying this earlier to Louis, you know, history helps having lived through several of these waves and I go, Oh yeah. I remember when Marc Andreessen, you know, who was the young out of college founder

[30:37] of Netscape, you know, had this idea that they was going to ride the web to become the next Microsoft. Sound like Sam Altman with chat GPT. Yeah, it does. You know, and, and he thought, wow, we're going to control everything. We can win it all. And of course, within a few years, they were, you know, sold off to AOL, which was another company that thought they could have it all. And they also didn't succeed. So I mean, this thing, and if you think about the web, four or five years into the web, we were at the Yahoo stage, where Yahoo was the big deal. We had AltaVista. We didn't have Google yet. Amazon was just getting started. So I feel like we're pretty early in this web thing. And yet we are, we have built an economy in which the entrepreneurs and the venture capitalists can actually already extract all the money. I mean, look at this. SpaceX went public with very small

[31:41] real profits at value to trillions of dollars. How crazy is that? And I do think some of these things are real businesses and they're important, but they're so overvalued. we have to actually understand that this is a game that will not continue. And we have to ask ourselves, where is the real value being created? And I think we need to push harder on the companies and their narratives. But more than that, we also have to think very deeply about what game we want to be playing i've been really getting um super interested in um this idea of mechanism design and just like with rents i i'm stealing a term from the economists and and changing it and using it more broadly um you know i was talking with Erik Brynjolfsson who's a an economist and i was telling him my idea of mechanism design he says tim you know that's not how

[32:42] economists use the term. I go, yeah, I know, but it's actually a really useful way of thinking about it. So for economists, mechanism design is, you know, some, you know, economists usually has figured out a way to change the shape of a market. And this can be pretty great. You know, Al Roth got a Nobel Prize for figuring out a new way to structure kidney transplant marketplaces, You know, like how do you make a long chain of trust? And it was great stuff, you know. But when I think about mechanism design in technology, I think, oh, actually so many things that I would say would be deserving of a Nobel Prize were just, we took for granted, you know. I don't know how technical all of you are. Do you guys know who Thompson and Ritchie are? Ken Thompson and Dennis Ritchie, they were the creators of Unix. back in 1970, right? Unix was the first portable operating system.

[33:43] And a lot of people think the story of open source software started in 1984 with Richard Stallman and free software and this whole sort of licensing business. But I think it really started with Unix in 1970, where they said, okay, we're going to build a portable operating system that will run on many, many different types of computers, which of course meant that you had to distribute source code. binaries, right? Also, they had an architecture with a small kernel surrounded. Then there was a shell, which was kind of like a harness for AI, right? And a bunch of utilities. And there was this explosion of universities around the world writing new utilities that would get added into Unix. It had what I came to call the architecture of participation, which is small modular pieces that can be replaced, added to, swapped out. And it was that mechanism. That's mechanism design. They totally changed the shape of the software development market.

[34:45] And then there were other mechanisms that followed on. For example, they got into the first source code control systems. And, of course, Linus Torvalds is known for writing the new Linux kernel, which was put together with a re-implementation of all the Unix utilities to become Linux. But he also wrote Git, which became, later somebody else came along and built a central hub, GitHub for that. And those mechanisms became the coordinating institutions. You can think there's a mechanism, which is this modular architecture. And then there's a new set of mechanisms and institutions. and the institution was not the open source initiative which managed the licenses. GitHub turned out to be the coordinating institution and the knowledge object was actually just the source code repository. And it turned out that software licenses were not even that important. Most of the code on GitHub carries no license at all. People just put it up there and share it. And so you kind of have this

[35:50] interesting interplay you know, and they really changed the shape of markets. And this has happened in AI because, okay, here we have the big labs and Sam starts, Sam Altman starts with, this is a winner takes all race for super intelligence. Dario then is racing and Dario is beating him. And that's the story that we're all being told. It's a race for super intelligence. It's going to, the winner is going to be so rich. You know, we're going to do all useful economic activity. The rest of you will be on the dole, effectively, some kind of UBI. We haven't figured out how that will work, but don't worry, we'll be so rich that we'll figure it out. AI will do so much. And that whole narrative, that's the game that they're telling us. It's just why people don't really trust AI. People are literally saying, we are building a product that might kill you, that will almost certainly put you out of work and you should love us. No, no, no. That's a bad game.

[36:52] And that's where mechanism design comes in mind. Leonid Hurwicz, who was one of the founders of the field, called it reverse game theory. He said, instead of trying to figure out what the rules of the game are and how you play it, ideally, you say, what game do we want to be playing? What outcome do we want? What rules will produce that outcome? And so I think a lot of mechanism design as saying, okay, well, if we want AI to produce human flourishing, we have to say that's the game. That's the game that we want. The game that we want is not the race for superintelligence, the ultimate number-go-up game where the winners who get the most capital deployed on this project will win and everybody else will lose. That's a terrible game to play. Why are we playing that as a society? effectively through our government, supporting that craziness. And so I feel like we really need to go rethink the game and say, what

[37:55] mechanism would change the shape of that game? And I think we see some signs of it. And just like with Unix, it wasn't AT&T's corporate strategy to create this portable operating system. It was just some group of guys at Bell Labs, right? Create a portable operating system in a portable language C. And in a similar way, this guy, this small team at Anthropic creates MCP, the model context protocol. And then suddenly, it's like, wow, here's a way that anybody can create an AI service and other AI agents can consume it. And suddenly it wasn't just, wow, you're going to call the APIs of the big providers. It could be like anybody could provide AI. And we start to have an AI economy that looks more like the participatory web that we had in the previous generation. So again, I look at that and I say, I call that mechanism design. It's changing the rules of the game so that you get a better outcome. And so I'm really interested in that.

[38:57] So I've been talking a lot about, well, what are the missing mechanisms of the AI economy. We need portable memory. We need portable context. We need open-weight harnesses. And I think a lot of the conversation about open-source AI has bought into the idea that somehow has to do with the models. Like, oh, we have to have open-weight models. And I think that's necessary, but not sufficient. Yes, we do want to have cheaper inference. We want to be able to bring our own model to the game and have it be a first-class citizen. But if you look at the history of open source, it's really up, it keeps going up the stack. And already we have this, we see the value of harnesses. And so, you know, so here's Pi as an open source harness. That's a super important, you know, part of the story. There's all this invention that's starting to happen with tools for evaluation or controlling models.

[39:59] And there's this whole world that's starting to be built that is more participatory. So it seems to me that the game that Silicon Valley would like us to play is this winner takes all game. It's basically a recipe for congestive heart failure of the economy. We're going to have all of the value come in to a few winners and their investors. And those investors are piling on massive, massive amounts of capital with this idea that they will come away insanely rich. And they're not thinking at all about the circulatory economy. And I think back to this wonderful quote from William Blake. He said, sorry, I have it actually written down. I'm going to get it wrong. It's something like the creative, let me get it right.

[41:07] So he said, the prolific would cease to be prolific unless the devourer as a seed received the excess of his delights. You know, like you think about this. here's this world where we're going to put all our customers out of work and somehow we'll get very rich through this sort of capital appreciation and they haven't really thought about the circular economy like at what point unless you think about the mechanism by which your customers get rich you're not actually building a real economy and so i think we really have to be you know get the AI labs to be thinking about what the new economy looks like. What's the circulatory system? They have to measure success by diffusion, not just capability. They have to ask how their customers are going to make money, not just how they themselves will. They have to actually think about, well, all these problems potentially with the labor transition. It's a core problem to be solved. And they have to be willing to win in the marketplace. They have to deliver real value. and so you know i kind of feel like

[42:11] um you know we're in a place where we think we're in the end game and i think we're in the beginning you know it's a little bit like churchill said you know this is not even the the end of the beginning so what is that it's not the beginning of the end it may be the end of the beginning you know and i think we're in the end of the we're heading for the end of the beginning. And every major technology wave arrives in two stages. The first is the invention of the technology itself. The second is the invention of the critical economic mechanisms that allow it to become part of the economy. So you think about the web, but it really wasn't until Google came up with PageRank, which really made search good. And then Bill Gross came up with pay-per-click advertising, and then Google perfected it. And that totally was this new, again, once again, mechanism design, pay-per-click advertising, totally transformative economic mechanism.

[43:14] Or think even later about YouTube content ID, an economic mechanism that allowed the creator, the video, there were all these lawsuits just like there are today with AI. Lawsuits, take it down, take it down, take it down. as creators were using music in unauthorized ways. And Google said, how about we help you monetize it instead? And I think the AI labs need to think that way. They need to be building mechanisms for a human economy, not just this one-way colonialist flow where they take all the resources, they take our content, they take our value, and they basically go, well, we'll give it to ourselves and our customers, and sorry that you were exploited. it's the only way we could build this amazing thing. And so, you know, I kind of feel like we're in this stage where we're, we're waiting for the truly transformative economic mechanisms of AI to arrive.

[44:15] And, you know, I think if we understand that that is the challenge of Silicon Valley today, we'll be much better off. And in that regard, I just want to kind of finish with reference to something you should all read. I don't know if you've seen Henry Farrell, Alison Gopnik, Cosma Shalizi, and James Evans' article in Science called AI as a Social and Cultural Technology. Really worth reading. Because it really does put AI in this context of this is about human cooperation. It's not about an independent superintelligence. intelligence. It's really about this is a tool for humans to work together in new ways. And, you know, so I think, you know, so this idea, just going back to Margaret Levi, about this is a new kind of commons. And that's how we should be trying to see it and how to use it. But I also see it as being kind of a medium. You know, each of us, when we're in conversation with AI or setting an agent in

[45:19] motion, we're doing something different than the next person trying to do the same task. It's just like I write something and you write something about the same topic. They're going to come out differently. And that's kind of wonderful. And you think about artists. Leonardo was great. So was Van Gogh. They did totally different things with paint, but they had the same medium. And I think in a similar way, we're all going to do things with AI and we should look at it that way, not as, wow, it's going to be the super intelligence that's going to take over. And so the thing that I keep waiting to hear, I'll know that this generation of technology has arrived when I start hearing people talk about the transformative apps that they built, not the fact that they're just doing lots and lots of coding really fast. They go, where are the things that you look around and go, oh yeah, that's so great. it really works. It's making my life better. Because making the lives of people better, solving human problems is the heart of

[46:23] what this is for. So our job is to draw something more and something better out of the ocean of possibilities that AI provides to us. AI was built from our collective intelligence, and it should not be designed to replace us. It should be designed to depend on us and to help us co-create value with it. Thank you. Thank you. Thank you. Okay. Introduce yourself. Tim O'Reilly, terrific talk. Thank you very much. You're welcome. I love that you zeroed in on the valuation as this key mechanism because of which we're experiencing this inflation of value

[47:24] that isn't aligned to human experience. At this conversation, we opened with a plenary where a philosopher of history has been encouraging this community of historians who work with data and think about data to consider ourselves as working for the common good, as experts who understand the time course of data. So there's a way in which the valuation problem is our problem. It's the problem in this community. Yeah. And historical materialism is the cure. we understand that there is a reality principle. It's captured in different kinds of data that the stock market is measuring. And some of the data that we can capture and analyze could offer alternative metrics to correct the stock as the sole indicator of health. You know, it's interesting when you, you know, I've been reading a lot. How many of you have been reading, Terry Tao has been writing about math and AI,

[48:28] the amazing stuff. But it really applies to all fields. I love the way he talks about that the output of mathematics is not proof. It's actually understanding. So OpenAI throws tens of millions of dollars of compute and says, we solved the problem. And you go, but we can't actually follow the explanation. We didn't actually learn anything, you know? And so, you know, this idea, and you think about this, I think about it like with something like history, right? You know, why are we still going over and over again, the same history, right? Because we're trying to understand it better. We're trying to learn it better. It's not like we just proved it once and it's done. And I think human, you know, experience is always about, you know, engaging. change. And again, when I think about mechanism design, all the AI narrative is they're designing the game of cognitive surrender, telling

[49:28] us that that's what we should do. We should just, hey, the bitter lesson, it's going to be better than us. Real quick, I didn't get to the question. The question is, historians don't know where the door is to this conversation. Is there a door to the conversation where historians can enter? Thanks. You know, I think the door is where you make it. You know, it's always surprising. I feel like when I have spent time with any of the ideas that I've kind of been associated with, I just kind of describe the world as I see it. And people, you know, it starts out like when I first started talking about, I did a conference before I came up with Web 2.0, which was in 2004. I did a conference in 2001 called Building the Internet Operating System. I became convinced that open source software was actually about network enabled collaboration more than it was about, you know, licenses. And so I was then, oh, wait, there's all these other forms of network enabled

[50:30] collaboration. collaboration. Napster came along and SETI at Home, which was distributed computation, and then web services. And I went, oh, all this is using the internet as a platform. And I told a story, and that story didn't quite catch. In fact, at my first Web2.0 conference, Bram Cohen, who was the creator of BitTorrent, said, oh, the internet's nothing like an operating system. And I was like, yeah, not like the old ones, but there's this new thing happening. And all I'm saying is i just kept telling the story until it made sense to people and all i'm saying is like historians above all are sense makers so just keep trying to make sense of the world and uh eventually uh you know if you if you tell a story that's useful to people it will break through and i find i've go in these five six year cycles of like i'm trying to explain the world and Finally, I get it right. And everybody goes, oh, yeah, yeah, that. Yeah, so I'll just piggyback on what Jo was just saying.

[51:31] So what's that? Oh, Matt Connelly. I'm a historian from Columbia. So a lot of academics and archivists, librarians, we've been immensely frustrated for the last almost three years because beginning with OpenAI, they begin launching these products into our classrooms. learned. They apparently either didn't do any research to figure out if that was true, or they're not telling us what they found out. So for almost three years now, the rest of us have had to do our own research, mainly with our own money and time, to try to figure out how we can make these products work for our students. And so, you know, there are a lot of us who would think, you know, if we wanted to build comparable systems, right, what do you need? You need, you know, you need data, you need people who know how to build these systems, and you need compute. most valuable data there is. Like higher education, we are in the business of like creating knowledge, preserving knowledge and sharing it. And we have the, we train the engineers and

[52:31] the scientists, you know, who, who built these systems. The only thing that we lack is the compute and these companies have over-invested in compute. And so when is somebody going to come along and say, we should, in higher education, we should be building the systems that we need, you know, to enhance human knowledge, to create more knowledge. You know, and when I say this, everyone's like, that's so naive. You know, why would that ever happen? Well, I mean, why would we allow them to keep doing this to us? They're eating their seed grain. Like, where do they think the next ideas are going to come from if they're making it possible for us to teach our students and learn from them? Yeah. Yeah. I think it's, you know, these transitions are always difficult and there are people who are dispossessed in each of them. you know, this incredible asymmetry of compute is actually going to go away in a way. It's because, you know, we're getting to a level of capability where, yeah, sure, there may be frontier

[53:36] uses of AI that are still, you know, require even more capability than we have today. But, you know, the analogy I'm using is we have cars that go 250 miles an hour. and we race them at, you know, Indianapolis Speedway or whatever, we don't drive them. And I think, you know, this idea that we keep pushing the frontier, I don't think that's going to be the ultimate driver. We're going to get to a level where we go, okay, this is as much AI as we can digest as a society. And we're getting to the point where that level is getting cheaper all the time. And so I don't know that it's necessarily going to be the case. there's always going to be you know oh yeah the people with the most money win and I you know I think and I see it even you know like in my own company you know just it's part of it is just understanding better you know like like I you know I'm looking at things that we built and I challenged my company I

[54:36] say we are we are competing with these guys we have to be better and I'm starting to see errors where I go oh yeah actually now we're better you know like I did this fine tuning project with Fable And I put it in step-by-step mode so I could learn and watch, okay, well, what are the stages? But it didn't really explain things unless I knew to ask. Whereas the O'Reilly coach, I said, I want to do a fine-tuning project. He says, okay, go read this stuff about what a good training data set looks like. And then build the training data set. And I went, oh, that's why my project with Fable didn't work. Because I didn't build a very good training data set because it didn't tell me what it would look like. And so it doesn't necessarily take a lot to be better. And I think that, again, it's the mythology that Silicon Valley is selling. The raw capability is the magic. And we're increasingly finding in our work that the magic, there's actually a lot of magic still in human expertise.

[55:37] My CTO tells this story about we've launched this new product called Expertise. intelligence and it's really related to the idea by the way you guys know who dr strange is you know you never know he when he calls it it does a spell he says something like you know uh by the hoary hosts of hoggoth or by the you know power of the dread door mamma i'm gonna do blah blah blah these are things i remember from my comic books from my childhood more from the avengers movies but um he he uh you know there's these magic words that invoke things anyway so So the reason I tell you that is my CTO was, he'd written this analysis tool so that people would say, you know, I need a new hire. And he'd go kind of, this analysis tool would go look at all the GitHub and the JIRA and all the software tools that told them what the workload was. And it would say, yeah, they're overloaded or whatever. And then he said, huh, let me connect the O'Reilly expert intelligence MCP. And he said, what would the O'Reilly experts say about this? And the thing was like, oh, actually came up with a completely different answer.

[56:40] Said I looked at the Google SRE handbook and there's this concept called operational toil, which is useless work overhead. And I did a toil analysis on this project. And you need to reorganize the group. You don't need to hire another person. So it gave a completely different answer. So you can invoke the spell of human expertise like Dr. Strange invokes the Dread Dormammu and you will get a different result out of an LLM than if you just leave it on its own. And I think that's underappreciated because we are the lens through which this, you know, just think of it. It's like, is this light and you have a little magnifying glass and you point at something and you make fire burn. And, you know, like swipe people get AI slop because they got no lens to focus it. And so I guess I just feel like all of us have, we're going to increasingly be in a world where this capability is available to all of us and we will get to use it. And some people will be better at it. And I think about this with my friend Doug Eck, who works on Google's AI music

[57:44] stuff. And he says, oh yeah, all of the Google engineers, we were such hot shit with this stuff. And then we brought in a couple of real musicians. Some of them were initially skeptical. And then once they started using it, they were doing things that we never dreamed of because we're not really that good. We're not as good as they are as musicians. So human expertise, you can pull more out of it. you know and i i think the labs don't want us to know that because their narrative is no no the power is all in this machine we're building that's why we're so valuable and i think the value is going to end up being in what we build with it and we're going to get to a level of capability that's enough and we will run with it and people will invent and create just like we always have. Tim, one last question. Yeah, I'll try to be as brief as possible. Steve Harvey, I work with the Bloomberg School of Public Health, speaking of capitalists. I think if we counted the words in your talk,

[58:46] probably the word you said more than anything else was we. And I want to go back to a reference you made sort of halfway through about we need to the rules of the game so that we're focused on human flourishing as opposed to super intelligence. I did a quick search on DuckDuckGo while you were talking, because I'm trying not to use Google as a search engine anymore. And there are different amounts that are listed that Elon Musk has spent on political campaigns in the last couple of months, but it's somewhere in the neighborhood of $360 million. My question is, why do you think that we have the agency to make a difference? And how do we go about doing that? Because I don't feel very agentic at all in

[59:48] this situation and it's nice to say we need to do this and we need to do that but who's the we and how do we get it done in an environment in which there's so much power and so much spending trying to create a completely different vision of society than the one that you've outlined thank Thank you. Yeah. You know, it's interesting because when I wrote a book back in 2016, published in 2017 called WTF, What's the Future and Why It's Up to Us. And Kara Swisher interviewed me and asked me, who is this we you're talking about? The story I told her was the story of the American Revolution. I said, somehow we decide, and there's a famous story, which is actually, I believe, historically accurate, that when George III heard that George Washington was not declared king in America after they won, he said, if he has done that, he

[1:00:50] is the greatest man that has ever lived. Because we believed something different than the British did. We believed that we're not going to have kings. And I think even today, we come to believe things collectively. We are a collective organism in certain ways. Human beings and the internet and AI, this actually helped humans to share our ideas. And it isn't always going to be the fact that, you know, I mean, money does not necessarily buy belief. You know, you look at all the young people who are sitting there going, we don't like AI. They are making a different decision about that. And I don't know how that will play out. I mean, it's clear that we have a rigged political system that's so dependent on money. I'm all with Larry Lessig and his power to get money out of politics. maybe one day we will decide to do that but i i don't i

[1:01:53] think that you know uh you know we'll see what happens in this coming election you know do um do we decide to do something different than we did you know two years ago i i hope we will thank you so much for joining us and then please enjoy the reception and uh thanks again to Tim O'Reilly for that amazing talk