--:-- Marc Andreessen: The entire world that's going to use AI is going to benefit from this kind of hyper-competition that's going to potentially run for decades. And so I think if you put these CEOs in a truth serum, what they would say is that's actually their nightmare. That's why they're in Washington.
--:-- Ben Horowitz: Crypto is really our best answer for getting back to delivering the internet back to the people and away from the large tech monopolies. It is the one technology that can really do that. And if we don't do that, over the next five years, these monopolies are going to get much, much stronger. Probably some of them will be stronger than the U.S. government itself.
--:-- Marc Andreessen: We're definitely on the side of freedom to innovate. Having said that, that's not the same as saying no regulations of anything ever.
--:-- Ben Horowitz: Ignoring tech is no longer an option in the government. They've seen it impact elections and education and everything. Big tech has been present in Washington, but big tech's interests are not only very different than startup innovators' interests, but we think also divergent from America's interest as a whole.
--:-- Host: Hello, everyone. Welcome back to the A16Z podcast. In today's episode, you'll get to hear directly from A16Z's co-founders, Marc Andreessen and Ben Horowitz, as they discuss one of the most hot button issues of today: politics and the future of tech.
--:-- This is obviously a climactic year for both policy and technology, but the overlap of these two fields has been increasing for quite some time. So today, Marc and Ben sit down to answer audience questions around this topic, touching on the important and increasing distinction between big and little tech, AI regulation and funding, the role of tech in geopolitics, and even why the firm has chosen to get involved in tech policy.
--:-- They cover much more than that, including whether Marc or Ben would ever run for office themselves. You'll have to listen in to the end for that one. Enjoy.
--:-- [Disclaimer: 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 investor or potential investors in any a16z fund. Please note that a16z and its affiliates may maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com slash disclosures.]
--:-- Marc Andreessen: Welcome back, everybody. We are very excited for this episode. We are going to be discussing a lot of hot topics. The theme of today's show is tech and policy and politics. There's a tremendous amount of heat right now in the tech world about politics. There's a tremendous amount of heat in the political world about tech. And then we as a firm, and actually both Ben and I as individuals, have been spending a lot more time in policy and politics circles over the last several months. We as a firm have a much bigger push than we used to, which Ben will describe in a moment.
--:-- The big disclaimer that we want to provide up front is that we are a nonpartisan firm. We are 100% focused on tech politics and policy. Today in this episode, we are going to be describing a fair number of topics, some of which involve partisan politics. Our goal is to describe anything that is partisan as accurately as possible and to try to be fair-minded and represent multiple points of view as best we can. We are going to try very hard not to take any sort of personal political partisan position. So please, if you could, grant us some generosity of interpretation in what we say. We are trying to describe and explain as opposed to advocate for anything specifically partisan. We advocate for tech policy topics. We do not advocate for other partisan topics.
--:-- On that theme, Ben, we wrote a little while ago — you wrote a blog post about our firm's engagement in politics and policy. We laid out our goals and how we're going about it. We are actually quite transparent about this. I was hoping, as an introduction for people who haven't seen that, if you could walk through what our plan and strategy is and how we think about this.
--:-- Ben Horowitz: It starts with why now. Why get involved in politics now? Historically, tech has been a little involved in politics, but it's been relatively obscure issues: H-1B visas, stock option accounting, carried interest, things like that. But now the issues are much more mainstream. For most of the software industry's life, Washington just hasn't been that interested in tech or in regulating tech for the most part. But starting in the mid-2000s, as software ate the world and tech started to invade all aspects of life, ignoring tech is no longer an option in the government. They've seen it impact elections and education and everything. Policymakers really want to get in front of it — that's a term we hear a lot. "We need to be in front of these things this time, not like last time when we were behind the curve."
--:-- Tech really needs a voice, and in particular, little tech needs a voice. Big tech has been present in Washington, but big tech's interests are not only very different than startup innovators' interests, but we think they're also divergent from America's interest as a whole. That makes it quite imperative for us to be involved, not only to represent the startup community, but also to get to the right answer for the country.
--:-- For the country, this is a mission-critical effort. If you look at the last century of the world, why was America strong? Why was any country significant in terms of military power, economic power, cultural power? It was those countries that got to the industrial revolution first and exploited it best. Now at the dawn of the information age revolution, we need to be there and not fall behind, not lose our innovative edge. That's all really up for grabs.
--:-- The biggest way America would lose it — because we're still extremely strong from a capitalistic system standpoint, from an education standpoint, from a talent standpoint — would be bad or misguided regulation that forces innovation elsewhere out of the country and prevents America and the American government from adopting these technologies.
--:-- The thing driving bad tech regulation is big tech, whose goal is not to drive innovation or make America strong, but to preserve their monopoly. We've seen that play out in AI in a really spectacular way. Big tech has pushed for the banning of open source for safety reasons. You can't find anybody who's been in the computer industry who can tell you that any open source project is less safe. From a hacking standpoint, when you talk about things like prompt injection and new attacks, you would much more trust an open source solution. Also, for a lot of the concerns of the U.S. government about copyrights — where does this technology come from? — not only should the source code be open, but the data should probably also be open so we know what these things were trained on. That's also for figuring out what their bias is. How can you know if it's a black box?
--:-- This idea that closed source would be safer — and big tech actually got some of this language into the Biden administration executive order, literally under the guise of safety to protect themselves against competition — is really scary.
--:-- The other related driver is this combination of big tech pushing for fake safetyism to preserve their monopoly and a rather thin understanding of how the technologies work in the federal government. Without somebody bridging the education gap, we are, as a country, very vulnerable to these bad ideas.
--:-- We also think it's a critical point in technology's history to get it right. If you think about what's possible with AI, so many of our country's biggest challenges are very solvable now: education, better and more equal healthcare, thinning out the bureaucracy we've built and making the government easier to deal with, particularly for underprivileged people trying to get into business and become entrepreneurs. All these things are made much better by AI.
--:-- Similarly, crypto is really our best answer for getting back to delivering the internet back to the people and away from the large tech monopolies. It is the one technology that can really do that. If we don't do that, over the next five years, these monopolies are going to get much, much stronger. Probably some of them will be stronger than the U.S. government itself. We have this technology that can help us get to this dream of stakeholder capitalism and participation for all economically, and we could undermine the whole thing with poor regulation.
--:-- Finally, in the area of biology, we're at an amazing point. If you look at the history of biology, we've never had a language — much like we never had a language to describe physics for a thousand years. We didn't have a language to really model biology until now. The language for physics was calculus. The language for biology is AI. We have the opportunity to cure a whole host of things we could never touch before, as well as address populations that we never even did any testing on before and always put in danger. Again, you have big pharma whose interest is in preserving the existing system because it locks out all the innovative competition.
--:-- For all those reasons, we've massively committed the flag and the firm to being involved in politics. You've been spending a tremendous amount of time in Washington. I've been spending time in Washington. Many of our other partners like Chris Dixon and Vijay Pande have been spending time in Washington. We have a real lobbying capability within the firm — call it government affairs. They're registered lobbyists working to work with the government, set up the right meetings, and help us get our message across. We're deploying a really significant amount of money to pushing innovation forward, getting to the right regulation on tech that preserves America's strength. We are not only committed to doing that this year, but for the next decade. This is a big effort for us, and we thought it'd be a good idea to talk about it on the podcast.
--:-- Marc Andreessen: Thank you. That was great. The key point there at the end is double underlined: long-term commitment. There have been times with tech where people have cannonballed their way onto the political scene with large money bombs, and then they were single issue or whatever, and they're in and out. They thought they could have short-term impact, and two years later, they're gone. We're thinking about that very differently.
--:-- Ben Horowitz: That's why I brought up the historical lens. We really think that if America is going to be America in the next hundred years, we have to get this right.
--:-- Marc Andreessen: Good. We're going to unpack a lot of what you talked about and go into more detail. I will get going on the questions, which again, thank you everybody for submitting questions on X. We have a great lineup today.
--:-- I'm going to combine a bunch of these questions because there were some themes. Jared asks: "Why has tech been so reluctant to engage in the political process, both at the local and national level, until now?" And then Kate asks the opposite question, which I find this juxtaposition very interesting because it gets to the nature of how we've gotten where we've gotten. Kate asks: "Tech leaders have spent hundreds of millions lobbying in D.C. In your opinion, has it worked? And what should we be doing differently as an industry when it comes to working with D.C.?"
--:-- I wanted to juxtapose these two questions because I think they're both true. The way they're both true is that there is no single tech. To your point, there is no single tech. Maybe once upon a time there was. My involvement in political efforts in this domain started 30 years ago. I've seen a lot of the evolution over the last three decades. I was in the room for the founding of TechNet, which is one of the sort of legacy — John Chambers and John Doerr. I've seen a lot of twists and turns on this over the last 30 years.
--:-- The way I would describe it is, as Ben said, there was a distinction and a real difference of view between big tech and little tech 20, 30 years ago. Yes, there was. It's much wider now. That whole thing has really gapped out.
--:-- Ben Horowitz: You probably remember big tech companies in the 80s and 90s often didn't really do much in politics. Probably most famously, Microsoft. Everybody at Microsoft during that period would probably say they had underinvested, given what happened with the antitrust case that unfolded.
--:-- Marc Andreessen: Actually, the one issue we were united on was the stock option accounting, which interestingly we were against Warren Buffett on. Warren Buffett was absolutely wrong on it and won. It actually very much strengthened tech monopolies. I think it did the opposite of what people in Silicon Valley wanted, and what people in Washington and America would have wanted, which was to make these monopolies so strong, using their market cap to further strengthen their monopoly. Because we moved from stock options to — it's too esoteric to get into here, but let's just say, trust me, it was bad.
--:-- Ben Horowitz: Yes. It was very good for big companies, very bad for startups.
--:-- Marc Andreessen: That's another thing that happened in the 90s and 2000s. There's a fundamental characteristic of the tech industry, and in particular tech startups and tech founders — Ben and I would include ourselves in this group — which is we are idiosyncratic, disagreeable, iconoclastic people. There is no tech startup association. Every industry group in the country has an association with offices in D.C. and lobbyists and major financial firepower. You know them under names like the MPAA in the movie industry, the RIAA in the record industry, the National Association of Broadcasters, the National Oil and Gas Association. Every other industry has these groups where the industry participants come together and agree on a policy agenda, hire lobbyists, and put a lot of money behind it.
--:-- The tech industry, especially the startups, we've never been good at agreeing on a common platform. Ben, you just mentioned the stock option accounting thing. That's actually my view of what happened at TechNet. TechNet was an attempt to get startup founders and new dynamic tech companies together. But the problem was we all couldn't agree on anything other than basically two issues: stock option expensing and carried interest for venture capital firms. Carried interest tax treatment. TechNet early on got anchored on these pretty esoteric accounting and financial issues and just could not come to agreement on many other issues. I think a lot of attempts to coordinate tech policy in the Valley have had that characteristic.
--:-- The other side of it, Ben, you highlighted this, but I want to really underline it. The world has changed. Up until about 2010, I think you could argue that politics and tech were just never that relevant to each other. For the most part, what tech companies did was they made tools. Those tools got sold to customers. They used them in different ways. How do you regulate a database software or an operating system or a word processor or a router? It's like regulating a power drill or a hammer.
--:-- Ben Horowitz: Exactly.
--:-- Marc Andreessen: What are appropriate shovel regulations? It just wasn't that important. This is where I think Silicon Valley deserves its share of blame for whatever's gone wrong. As a consequence, I think we all just never thought it was that important to really explain what we were doing and to be really engaged in the process out there.
--:-- The other thing that happened was there was a love affair for a long time. There was a view that tech startups are purely good for society. Tech is purely good for society. There were really no political implications to tech. This continued interestingly up through 2012. People now know all the headlines about social media destroying democracy that kicked into gear after 2015, 2016. But even in 2012, social media had become very important in the 2012 election, but the narrative in the press was almost uniformly positive. It was specifically that social media is protecting democracy by making sure that certain candidates get elected. There were literally headlines from newspapers and magazines that are very anti-tech today that were very pro-tech at that point, because the view was tech helped Obama get reelected.
--:-- The other thing was the Arab Spring. There was this moment where it was like tech is not only going to protect democracy in the U.S., but it's going to protect democracy all over the world. Facebook and Google were the catalysts. At the time, we were viewed as the catalyst for the Arab Spring, which was going to bring a flowering of democracy to the Middle East.
--:-- Ben Horowitz: It didn't work out that way.
--:-- Marc Andreessen: Did not work out that way. The point is, it is relatively recent — in the last 10, 12 years — that everything has come together. All of a sudden, people in the policy arena are very focused on tech. People in the tech world have very strong policy and politics opinions. The media weighs in all the time. None of this is a U.S.-only phenomenon. We'll talk about other countries later on, but these issues are playing out globally in many different ways.
--:-- One thing I would add: when I'm in D.C. doing a fair amount on the non-political side, and when I'm in meetings involving national security or intelligence or civil policy of whatever kind, it's striking how many topics that you would not think are tech topics end up being tech topics. When the state exercises power now, it does so with technologically enabled means. When citizens resist the state or fight back against the state, they do so with technologically enabled means. We sometimes say we're the dog that caught the bus on this stuff. We all wanted tech to be important in the world. It turns out tech is important in the world. And it turns out the things that are important in the world end up getting pulled into politics.
--:-- Ben Horowitz: I think that's right. On the second part of the question — why has tech been so ineffective despite pouring all the money in — there are a few important issues around that. One is arrogance. I think we in tech, and a lot of the people who went in, are like, "Oh, we're the good guys. We're for the good. Everybody will love us when we get there, and we can just push our agenda on the policymakers without really putting in the time and the work to understand the issues and the things that you face as somebody in Congress or somebody in the White House in trying to figure out what the right policy is."
--:-- We are coming at that from our cultural value, which is we take a long deal of relationships. We try never to be transactional. That's especially important on policy because these things are massively complex. We understand our issues and our needs, but we have to take the time to understand the issues of the policymakers and make sure that we work with them to come up with a solution that is viable for everyone. I think that's thing one. I think tech has been very bad on that.
--:-- The second one is I think they've been partisan where it's been not necessary or even smart to be partisan. People have come in with whatever political bent — mostly Democratic Party — and they go in without understanding you and only work with Democrats because we're Democrats. Our approach is: we are here to represent tech. We want to work with policymakers on both sides of the aisle. We want to do what's best for America. If we can describe that correctly, we'll get support from both sides. That's a really different approach. Hopefully that's right, and hopefully we can make progress.
--:-- Marc Andreessen: Let's go to the next question. This is a two-part question. Sheen asks: "In what ways do you see the relationship between Silicon Valley and D.C. evolving in coming years, particularly in light of recent regulatory efforts targeting tech giants?" We'll talk about TikTok later on, but there have been big flashpoint events happening right now. For people who haven't seen, the DOJ just filed a massive antitrust lawsuit against Apple. Tech topics are very hot right now in D.C.
--:-- Ben Horowitz: That one is interesting. One of the things I've talked about is a lot of little tech is very much in alignment with some of the things the FTC is doing, but probably we would do it against a different set of practices and behaviors of some of the tech monopolies. It shows why more conversation is important on these things. What we think is the abuse of the monopoly and what the lawsuit is are not exactly the same thing.
--:-- Marc Andreessen: Let's talk about that for a moment, because this is a good case study of dynamics here. The traditional free market libertarian view is very critical of antitrust theory in general and is certainly very critical of the current prevailing antitrust theories, which are more expansive and aggressive than the ones of the last 50 years, as shown in things like the Apple lawsuit and many other actions recently. For people in business, there's a reflexive view that basically says businesses should be allowed to operate. But there are certainly people who have this view that any additional involvement of the political machine, especially the prosecutorial machine in tech, is invariably going to make everything worse. They sue Apple today and maybe you're happy because you don't like Apple today because they abused your startup. But if they win against Apple, they're just going to keep coming and coming and coming.
--:-- The opposite view would be that the interests of big tech and little tech have really diverged, and if there is not strong and vigorous investigation and enforcement, and then ultimately things like the Apple lawsuit, these companies are going to get so powerful that they may be able to really seriously damage little tech for a very long time. Ben, talk a little bit about how we think through that, because we even debate this inside our firm. Talk about how to process through that, what you think, and where you think those lines of argument are taking us.
--:-- Ben Horowitz: Full disclosure: we were at Netscape. We were certainly on the side of little tech against big tech. Microsoft at that time had a 97% market share on desktop, and it was very difficult to innovate on the desktop. It was just bad for innovation to have them in that level of position of power. I think that's happened on the smartphone now, particularly with Apple. The Epic case and the Spotify cases are really great examples. If I am fielding a product that's competitive with Spotify and I am charging Spotify a 30% tax on their product, that seems unfair from the standpoint of the world. It does seem like it's using monopoly power in a very aggressive way. It's certainly against our interest and the interest of new companies for the monopolies to exploit their power to that degree.
--:-- When the government gets involved, it's not going to be a clean surgical "here's exactly the change that's needed." But I also think with these global businesses with tremendous lock-in, you have to at least have the conversation and say, "What is this going to do for consumers if we let it run?" We need to represent that point of view from the small tech perspective.
--:-- Marc Andreessen: The big tech companies are certainly not doing us favors right now. They're certainly not acting in ways that are pro-startups, I think we could say as a generalist.
--:-- Ben Horowitz: No, no, no. The opposite.
--:-- Marc Andreessen: Quite the opposite. One of my ideas I kick around a lot is it feels like any company is either too scrappy or too arrogant, but never in the middle.
--:-- Ben Horowitz: It's like people.
--:-- Marc Andreessen: You're either the underdog or you're the overdog, and there's not a lot of reasonable dogs.
--:-- Ben Horowitz: Exactly.
--:-- Marc Andreessen: There's an inherent tension there. It seems very hard for these companies to reach a point of dominance and not figure out some way to abuse it.
--:-- Ben Horowitz: You touched on an important point. In representing little tech, we're not a pure libertarian anti-regulatory force here. We think we need regulation in places. We certainly need it in drug development. We certainly need regulation in crypto and financial services. The financial services aspect of crypto is very important. It's very important to the industry that it be strong in America with a proper regulatory regime. We're not anti-regulation. We're pro the kind of regulation that will make both innovation strong and the country strong.
--:-- Marc Andreessen: We should also say, when we're advocating on behalf of little tech, there's self-interest as a component of that because we're a venture capital firm and we back startups. There's a straight financial interest there. I think, Ben, you'd agree with me, we also feel philosophically that this is a very pro-America position, very pro-consumer position. The reason is straightforward: as you've said many times in the past, the motto of any monopoly is, "We don't care because we don't have to."
--:-- Ben Horowitz: Exactly.
--:-- Marc Andreessen: You'd experience that if you've called customer service when one of these monopolies has kicked you off their platform.
--:-- Ben Horowitz: Yes, exactly.
--:-- Marc Andreessen: There is something in the nature of monopolies where if they no longer have to compete and if they're no longer disciplined by the market, they basically go bad. How do you prevent that? You force them to compete. In some cases, they compete with each other, although often they collude with each other, which is another thing. Monopoly and cartel are two sides of the same coin. In the history of the tech industry, it's really when they're faced with startup competition — when the elephant has a terrier at his heels nipping at him, taking increasingly big bites out of his foot — that big companies actually act and do new things.
--:-- Without healthy startup competition, there are many sectors of the economy where it's very clear now that there is not enough startup competition because the incumbents that everybody deals with on a daily basis are just practically intolerable. It's not in anybody's interest, ultimately, from a national policy standpoint, for that to be the case. Things can get bad where it's to the benefit of the big companies to preserve those monopolies, but very much not to anybody else's benefit.
--:-- Ben Horowitz: That's such a big impetus behind our political activity.
--:-- Marc Andreessen: That's right. Now we're going to future looking. In what ways do you see the relationship between Silicon Valley and D.C. evolving in the coming years?
--:-- Marc Andreessen: And then specifically, and again, we're not going to be making partisan recommendations here, but there is an election coming up and it is a big deal. What happens in the White House and in Congress is going to have big consequences for everything we've just been discussing. So how do we see the upcoming election affecting tech policy?
--:-- Ben Horowitz: Yeah, well, I think there are several issues that end up being really important to educate people on now. Whatever platform you run on as a congressperson or as a president, you want to live up to that promise when you get elected. A lot of these positions that will persist over the next four years are going to be established now.
--:-- In crypto in particular, we've been very active on this because we have a big donation to something called the Fair Shake PAC. We're identifying for citizens which politicians are on what side of these issues. Who are the flat out anti-crypto, anti-innovation, anti-blockchain, anti-decentralized technology candidates? Let's at least know who they are so that we can tell them we don't like it and then tell all the people who agree with us that we don't like it.
--:-- A lot of it ends up being that we want the right regulation for crypto. We've worked hard with policymakers to help them formulate things that will prevent scams, prevent nefarious uses of the technology for things like money laundering. And then enable the good companies, the companies that are pro-consumer, helping you own your own data and not have it owned by some monopoly corporation who can exploit it or just lose it, get broken into. You now have identity theft problems and so forth. It can help create a fairer economy for creatives so that there's not a 99% take rate on things that you create on social media. It's just important to educate the populace on where every candidate stands on these issues. We're really focused on that. Same true for AI, same true for bio.
--:-- Marc Andreessen: I'd also add, it's not actually the case that there's a single party in D.C. that's pro-tech and a single party that's anti-tech. Definitely not. And by the way, if that were the case, it would make life a lot easier. But it's not the case. I'll give a thumbnail sketch of at least what I see when I'm in D.C. and see if you agree with this.
--:-- The Democrats are much more fluent in tech. I think that has to do with who their elites are. It has to do with this very long-established revolving door — and I mean that in both the positive and pejorative sense — between the tech companies and the Democratic Party, Democratic politicians, political offices, congressional offices, White House offices. There's just a lot more integration. The big tech companies tend to be very Democratic, which you see in all the donation numbers and voting numbers. There's just a lot more tech fluent, tech aware Democrats, especially in powerful positions. Many of them have actually worked in tech companies. As an example, the current White House chief of staff is a former board member at Meta, where I'm on the board. There's a lot of connective tissue between those.
--:-- Having said that, the current Democratic Party, and in particular certain of its more radical wings, has become extremely anti-tech to the point of being arguably, in some cases, outright anti-business, anti-capitalism. So there's a real back and forth there.
--:-- Republicans, on the other hand, in theory and the stereotype would have you believe they are inherently more pro-business and more pro-free markets and should therefore be more pro-tech. But it's a mixed bag. Number one, a lot of Republicans just basically think Silicon Valley is all Democrats. If we're Republicans, that means they're de facto the enemy. They hate us. They're trying to defeat us. They're trying to defeat our policies. So they must be the enemy. There's a lot of some combination of distrust and fear and hate on that front. And again, with much less connective tissue, there are many fewer Republican executives at these companies, which means there are many fewer Republican officials or staffers who have tech experience. There's a lot of mistrust.
--:-- Of course, there have been flashpoint issues around this lately, like social media censorship, that have really exacerbated this conflict. And the other thing is, there are very serious policy disagreements. There are wings of the modern Republican Party that are actually quite economically interventionist. The term of the moment is industrial policy. There are Republicans who are very much in favor of a much more interventionist government approach towards dealing with business and in particular dealing with tech.
--:-- I guess to say, this is not an either/or thing. There are real issues on both sides.
--:-- Ben Horowitz: The way we think about that is that there's a real requirement to engage on both sides. There's a real requirement to educate on both sides. If you're going to make any progress in tech issues, there's a real need to have a bipartisan approach because you do have to actually work with both sides.
--:-- Marc Andreessen: Yeah, and I think that's absolutely right.
--:-- Ben Horowitz: Just to name names a little, on the Democratic side, you've got people like Richie Torres out of the Bronx. A huge swath of the Congressional Black Caucus sees that crypto is a real opportunity to equal the financial system, which has historically been documented as racist against a lot of their constituents, and also the creatives, which they represent a lot, to get a fair shake. On the other hand, you have Elizabeth Warren, who has taken a very totalitarian view of the financial system and is moving to consolidate everything in the hands of a very small number of banks and basically control who can participate and who cannot in finance. These are just very different views out of the same party. We need to just make the specific issues really clear.
--:-- Marc Andreessen: Yeah, and the same thing. We could spend a long time also naming names on the Republican side.
--:-- Marc Andreessen: Yes, which we'll do later.
--:-- Ben Horowitz: I should do it right now just to make sure that we're fair on this.
--:-- Marc Andreessen: There are Republicans who are full-on pro-free market, very opposed to all current government efforts to intervene in markets like AI and crypto. By the way, many of those same Republicans are also very negative on antitrust action. They're very ideologically opposed to antitrust. They would also be opposed to things like the Apple lawsuit that a lot of startup founders might actually like. On the flip side, you have folks like Josh Hawley that are quite vocally irate at Silicon Valley and very in favor of much more government intervention and control. I think a Hawley administration would be extremely interventionist in Silicon Valley and would be very pro-industrial policy, very much trying to set goals and have government management of more of tech, but also much more dramatic action against perceived or real enemies. So it's the same kind of mixed bag.
--:-- Anyway, I wanted to go through that. This is the long, winding answer to the question of how will the upcoming election affect tech policy. There are real issues with the Biden administration, in particular with the agencies and with some of the affiliated senators as just described. Under the Trump administration, the agencies would be headed by very different kinds of people. Having said that, it's not that a Trump presidency would necessarily be a clean win. There are many people in that wing who might be hostile in different ways, or actually might be hostile in some cases in the same ways.
--:-- Ben Horowitz: And by the way, Trump himself has been quite a moving target on this. He tried to ban TikTok and now he's very pro-TikTok. He has been negative on AI. He was originally negative on crypto, now positive on crypto. It's complex. Which is why I think the foundation of all of this is education. Why we're spending so much time in Washington is to make sure that we communicate all that we know about technology so that at least these decisions that the politicians make are highly informed.
--:-- Marc Andreessen: Good. Okay, so moving forward. Three questions in one. Alex asks, as tech regulation becomes more and more popular within Congress, which is happening, do you anticipate a lowering in general of the rate of innovation within the industry? Number two, Tyler asks, what is a key policy initiative that if passed in the next decade could bolster the U.S. for a century? And Elliot Parker asks, what's one regulation that if removed would have the biggest positive impact on economic growth?
--:-- I don't know that there's a single regulation or a single law or a single issue. There are certainly individual laws or regulations that are important. But I think the thematic thing is a much bigger problem. The things that are coming are much more serious than the things that have been.
--:-- Ben Horowitz: I think that's correct.
--:-- Marc Andreessen: Okay, we'll talk about that.
--:-- Ben Horowitz: If you look at the current state of regulation, if it stayed here, there's not anything that we really feel like a burning desire to remove in the same way that things that are on the table could be extremely destructive. If we ban large language models in general, or we force them to go through some kind of government approval, or if we ban open source technology, that would have a devastating effect. It would basically take America out of the AI game and make us extremely vulnerable from a military standpoint, extremely vulnerable from a technology standpoint in general. That's devastating.
--:-- Similarly, if we don't get proper regulation around crypto, the trust in the system and the business model is going to fade. It's going to be in jeopardy. It's not going to be the best place in the world to build crypto companies and blockchain companies, which would be a real shame. The analog would be the creation of the SEC after the Great Depression, which really helped put trust into the U.S. capital markets. Trust into the blockchain system as a way to invest, participate, be a consumer, be an entrepreneur are really important and necessary, and very important to get those right.
--:-- Marc Andreessen: Okay, and then speaking of that, let's move straight into the specific issues. Lenny asks, what form do you think AI regulation will take over the next two administrations? B. Sacondi asks, will AI regulation result in a concentrated few companies or an explosion of startups and new innovation? E. Ray asks, how would you prevent the AI industry from being monopolized, centralized with just a few tech corps? And our friend Beth Jesus asks, how do you see the regulation of AI compute and open source models realistically playing out? Where can we apply pressure to make sure we maintain our freedom to build and own AI systems?
--:-- Ben Horowitz: It's really interesting because there's a regulatory dimension of that, and then there's the technological version of that. They do intersect. If you look at what big tech has been trying to do, they're very worried about new competition to the point where they've taken it upon themselves to go to Washington and try and outlaw their competitors. If they succeed with that, then I think it is super concentrated AI power, making the concentrated power of social media or search pale in comparison. It would be very dramatic if there were only three companies that were allowed to build AI. That's certainly what they're pushing for.
--:-- In one regulatory world where big tech wins, there's very few companies doing AI. Probably Google, Microsoft, and Meta. Microsoft having basically full control of OpenAI as they demonstrated — they have the source code, they have the way to say that we own everything, and they control who the CEO is, as they demonstrated beautifully. It will all be owned by three, maybe four companies.
--:-- If you follow the technological dimension, what we're seeing play out has been super exciting. We were all wondering, would there be one model that ruled them all? Even within a company, I think we're finding that there's no current architecture that's going to gain on a single thing — a transformer model, a diffusion model — that's going to become so smart in itself that once you make it big enough, it's just going to know everything. What we've seen is that even the large companies are deploying a technique called a mixture of experts, which implies you need different architectures for different things. They need to be integrated in a certain way and the system has to work. That just opens the aperture for a lot of competition because there are many ways to construct a mixture of experts, to architect every piece of that.
--:-- We've seen little companies like Mistral field models that are highly competitive with the larger models very quickly. There are other factors like latency, cost, that factor into this. And then there's also "good enough" — when is a language model good enough? When it speaks English, when it knows about what things, what are you using it for? And then there's domain-specific data. I've been doing medical research for years, and I've got data around all these genetic patterns and diseases. I can build a model against that data that's differentiated by the data.
--:-- I think we're likely to see a great Cambrian explosion of innovation across all sectors — big companies, small companies — provided that the regulation doesn't outlaw the small companies. That would be my prediction right now.
--:-- Marc Andreessen: I'd add a bunch of things to this. One is, even on the big model side, there's been this leapfrogging that's taking place. GPT-4 was the dominant model not that long ago. It's been leapfrogged in significant ways recently by both Google's Gemini Pro, especially the one with the long context window where you can feed it 700,000 words or an hour of full motion video as context for a question, which is a huge advance. And then Anthropic's big model, Claude, a lot of people now are finding that to be a more advanced model than GPT-4. One assumes OpenAI is going to come back. This leapfrogging will probably happen for a while. Even at the highest end, these companies are still competing with each other.
--:-- And then, as you articulate very well, there is this giant explosion of models of all kinds of shapes and sizes. Our company, Databricks, just released another big leapfrog on the smaller model side. I think it's the best small model now in the benchmarks, and it's so efficient it will run on a MacBook.
--:-- Ben Horowitz: And they have the advantage of, as an enterprise, you can connect it to a system that gives you not only enterprise quality access control, but also the power to do SQL queries, the power to create a catalog so that you can have a common understood definition of all the weird corporate words you have. One of which is "customer" — there's almost no two companies that define customer in the same way. In most companies, there are several definitions of customer.
--:-- Marc Andreessen: I think if you put the CEOs of the big companies under truth serum, what they would say is their big fear is that AI is actually not going to lead to a monopoly for them. It's going to lead to a commodity. It's going to lead to a race to the bottom on price. You see that a little bit now — people who are using one of the big model APIs are able to swap to another big model API from another company pretty easily. The main business model for these big models, at least so far, is an API — pay per token generated or per answer. If these companies really have to compete with each other, it may be that it actually is a hyper competitive market. It may be the opposite of a search market or an operating system market. It may be a market where there's continuous competition and improvement and leapfrogging and constant price competition.
--:-- The payoff from that to everybody else in the world is an enormously vibrant market where there's constant innovation happening and constant cost optimization happening. As a customer downstream of this, the entire world that's going to use AI is going to benefit from this kind of hyper-competition that could potentially run for decades. If you put these CEOs in a truth serum, what they would say is that's actually their nightmare.
--:-- Ben Horowitz: Well, that's why they're in Washington.
--:-- Marc Andreessen: That's why they're in Washington.
--:-- Ben Horowitz: That is what's actually happening. That is the scenario they're trying to prevent. They are trying to shut off competition.
--:-- Marc Andreessen: There is a funny thing. Tech is so historically bad at politics that I think some of these folks think they're being very clever in how they go about this. They show up in Washington with the public service narrative or end of the world narrative. I think they think they're going to very cleverly trick people in Washington into giving them cartel status. And the people in Washington don't realize until it's too late. But it actually turns out people in Washington are actually quite cynical. They've been lobbied before.
--:-- There is this thing I get from them off the record a lot, especially after a couple of drinks, which is basically if you've been in Washington for longer than two minutes, you have seen many industries come to Washington. Many big companies come to Washington and want monopoly or cartel regulatory protection. If you're in Washington, you've seen this play out dozens or hundreds of times.
--:-- Nine months ago, there was a moment where it seemed like the big tech companies could kind of get away with this. I'm still concerned and we're still working on it, but I think the edges come off a little bit because I think the cynicism of Washington in this case is actually correct. I think they're onto these companies. If there's a unifying issue, there are basically two unifying issues in Washington. One is they don't like China and the other is they don't like big tech. This is a winnable war on behalf of startups and open source and freedom and competition. I'm worried, but I'm feeling much better about it than I was nine months ago.
--:-- Ben Horowitz: Well, we had to show up. That's the other thing. It's taught me a real lesson. You can't expect people to know what's in your head. You've got to go see them. You've got to put in the time. You've got to say what you think. If you don't, you don't have any right to wring your hands about how bad things are.
--:-- Marc Andreessen: I just wanted to note one more thing. You kind of mentioned the big companies being Microsoft, Google, Meta. It is worth noting Meta is on the open source side of this. Meta is actually working quite hard. This is a big deal because it's very contrary to the image people have of Meta over prior issues, correctly or not. But on the open source AI topic and on freedom to innovate, at least for now, Meta is very strongly on that side.
--:-- Ben Horowitz: I think that's right. It's a very interesting point. Essential for people to understand is that the way Meta is thinking about this and the way they're actually behaving and executing is very similar to how Google thought about Android. Their main concern was that Apple not have a monopoly on the smartphone. Not so much that they make money on the smartphone themselves. A monopoly on the smartphone for Apple would mean that Google's other business was in real jeopardy. They ended up being an actor for good. Android has been an amazing thing for the world, including getting smartphones in the hands of people who wouldn't be able to get them otherwise all over the world.
--:-- Meta is doing a very similar effort. In order to make sure that they have AI as a great ingredient in their products and services, they are willing to open source it and give all of their very large investment in AI to the world so that entrepreneurs and everybody can keep them competitive, even though they don't plan to be in the business of AI in the same way that Google is in the business of smartphones to some extent, but it's not their key business. Meta doesn't have a plan to be in the AI business. Maybe to some extent they will, but that's not the main goal.
--:-- I would put one other company on the concerning side on this. It's truly to tell where they're going to shake out. Amazon just announced they're investing a lot more money in Anthropic. Basically Amazon is to Anthropic what Microsoft is to OpenAI.
--:-- Marc Andreessen: I think that's an honest ability, yep.
--:-- Ben Horowitz: Anthropic is very much in the group of big tech, new incumbent big tech, lobbying very aggressively for regulation and regulatory capture in D.C. It's an open question whether Amazon is going to pick up that agenda as Anthropic increasingly becomes effectively a subsidiary of Amazon.
--:-- Marc Andreessen: This is another place where we're on the side of Washington, D.C. and the current regulatory motion. The big tech companies have done this thing which we thought was illegal because we observed it occur at AOL and people went to jail. What they've done is they invest in startups huge amounts of money. Microsoft and Amazon and Google are all doing it — billions of dollars with the explicit requirement that those companies then buy GPUs from them at not the discount that they'd ordinarily get, but at a relatively high price and then be in their clouds. In the Microsoft case, even more aggressive: give me your source code, give me your weights. That's extremely aggressive.
--:-- They're moving money from the balance sheet to their P&L in a way that, at least from an accounting standpoint, was our understanding wasn't legal. The FTC is looking at that now. It'll be interesting to see how that plays out.
--:-- Ben Horowitz: That's one area. Another issue people should watch is around tripping. The other one is just consolidation. If you own half of a company and you get to appoint the management team, is that not a subsidiary? There are rules on that. At what point you own the company equity, you own the intellectual property of the company, and you control the management team — is that not your company? If you're not consolidating it, is that legal? The SEC is going to weigh in on that. To the extent that some of these companies have nonprofit components to them, there are tax implications to the conversion to for profit and so forth. The stakes in the legal, regulatory, and political game that's being played here are quite high.
--:-- Marc Andreessen: Quite high, yes. Ben and I are old enough where we do know a bunch of people who went to jail. So some of these issues turn out to be serious.
--:-- Gabriel asks, what would happen if there was zero regulation of AI, the good, the bad, and the ugly?
--:-- Marc Andreessen: And this is actually a really important topic. So we're vigorously arguing in DC that there should be, basically, anybody should be completely capable of building AI, deploying AI. Big companies should be allowed to do it. Small companies should be allowed to do it. Open source should be allowed to do it. And look, a lot of the regulatory pushes we've been discussing that comes from the big companies and from the activists is to prevent that from happening and put everything in the hands of the big companies. So we're definitely on the side of freedom to innovate. Having said that, that's not the same as saying no regulations of anything ever. And so we're definitely not approaching this with a hardcore libertarian lens.
--:-- The interesting thing about regulation of AI is that it turns out when you go down the list of the things that reasonable people consider to be concerns around AI on both sides of the aisle, the implications they're worried about are less the technology itself and are more the use of the technology in practice, either for good or for bad. Ben, you brought up, for example, if AI is making decisions on things like granting credit or mortgages or insurance, then there are very serious policy issues around how those answers are derived, which groups are affected in different ways. The flip side is if AI is used to plan a crime or a bank robbery or a terrorist attack, that is obviously something people focused on national security and law enforcement are very concerned about.
--:-- Our approach on this is very straightforward: it seems completely reasonable to regulate uses of AI that would be dangerous. The interesting thing about that is, as far as I can tell, every single use of AI to do something bad is already illegal under current laws and regulations. It's already illegal to be discriminatory in lending. It's already illegal to redline in mortgages. It's already illegal to plan bank robberies. It's already illegal to plan terrorist attacks. These things are already illegal, and there are decades or centuries of case law and regulation and law enforcement and intelligence capabilities around all of these. We think it's completely appropriate that those authorities be used. And if there are new laws or regulations needed due to other bad uses, that makes total sense. But the issues that people are worried about can be contained and controlled to the level of the use, as opposed to somehow saying, as some of the Doomer activists do, that we need to literally prevent people from doing linear algebra on their computers.
--:-- Ben Horowitz: Yeah, I think that's important to point out. What is AI? It turns out to be math, specifically a mathematical model. You can think of it like an equation: y equals x squared plus b. That equation can model the behavior of something in physics, so you can predict something happening like the speed an object will drop. AI is that, but with huge computer power applied so you can have much bigger equations with 300 billion variables instead of two or three or four.
--:-- If you get into regulating math and you say math is okay up to a certain number of variables, but at the two billionth and first variable it's dangerous, then you're in a pretty bad place. You're going to prevent everything good from the technology from happening as well as anything you might think is bad. You really do want to be in the business of regulating the applications of the technology, not the math. In the same way that nuclear power is very potentially dangerous and nuclear weapons are extremely dangerous, you wouldn't want to put parameters around what physics you could study in order to prevent somebody from getting a nuke. "You can no longer study physics in Iran because then you might be able to build a nuke" would be the conclusion. That has been what big tech has been pushing for, not because they want safety, but because they want a monopoly. I think we have to be very, very careful not to do that.
--:-- I do think there will probably be some cases that come up that are enabled by AI that do need to be regulated. For example, I don't know that there's a law that if you recreate something that sounds exactly like Drake and put out a song that sounds like a Drake song, that's illegal. Maybe that should be illegal. I think those things need to be considered for sure. I also think we need technological solutions, not just regulatory solutions, for things like deepfakes that help us get to what's human and what's not human. A lot of those are viable now based on blockchain crypto technology.
--:-- Marc Andreessen: Yeah, so let's just on the voice thing real quick. It actually is not currently possible to copyright a voice. You can copyright lyrics, music, tunes, melodies. But not a voice. That seems like a perfect example where it probably is a good idea to have a law that lets you copyright your voice.
--:-- Ben Horowitz: I feel that way, particularly if people call their voice "Drake squared" or something. It could get very dodgy. Although you can trademark your name, so you could probably prosecute on that. But having said that, this also gets to the complexity of these things. There is an issue around copyrighting a voice: how close to the voice of Drake does it have to be? There are a lot of people with a lot of voices in the world. How close do you have to get before you're violating copyright? What if my natural voice actually sounds like Drake? Am I now in trouble? Do we outlaw Jamie Foxx imitating Quincy Jones?
--:-- Marc Andreessen: Agreeing violently with you on this — that seems like a great topic that needs to be taken up and looked at seriously from a legal standpoint. It's an issue elevated by AI, but it's a general concept of being able to copyright and trademark things, which has a long history in U.S. law.
--:-- Let's talk about the decentralization and the blockchain aspects of this. Goose asks, how important is the development of decentralized AI? And how can the private sector catalyze prudent and pragmatic regulations to ensure the U.S. retains innovation leadership in this space? Let's talk about decentralized AI. The default way AI systems are being built today is with supercomputer clusters in a cloud. You have a single data center with 10,000 or 100,000 chips and systems interconnect them. A company owns and controls that. These AI companies are raising a lot of money to do that now. To train a state-of-the-art model, you're at $100 million plus. To train a small one like the Databricks model that just came out, it's on the order of $10 million. These are large centralized efforts. We all think the big models are going to end up costing a billion and up in the future.
--:-- This raises the question: is there an alternate way to do this? We believe strongly there is, with a decentralized approach, particularly a blockchain-based approach. It's the kind of thing the blockchain web3 method seems like it would work with very well. We are already backing companies and startups that are doing this. There are at least three obvious layers you could decentralize that seem increasingly important. Actually, let me say four.
--:-- There's the training layer — building the model. There's the inference layer — running the model to answer questions. There's the data layer, Ben, to your point on opening up the black box of where the data is coming from. There should probably be a blockchain-based system where people who own data can contribute it for training of AIs, get paid for it, and track all that. And then there's a fourth that you alluded to: deepfakes.
--:-- I should pause and say, in my last three months of trips to D.C., the number one issue politicians are focused on with AI is deepfakes. It's the one that directly affects them. Every politician who's thought about this has a nightmare scenario: three days before their reelection campaign, a deepfake goes out with them saying something absolutely horrible, it's so good, the voters get confused, and they lose the election. That's the thing that has the most potency right now.
--:-- A lot of people, including the politicians, say we need a way to detect deepfakes. Either the AI systems need to watermark AI-generated content, or we need scanners like the ones being used in some schools to detect that something was AI-generated. Our view as technologists and investors is that the methods of detecting AI-generated content after the fact are basically not going to work. They're not going to work because AI is already too good at this. If you happen to have kids in a school running one of those scanner programs to detect whether your kid used ChatGPT to write an essay, those really don't work in a reliable way. There are a lot of false positives and false negatives. Those are very bad ideas. For the same reason, detection of AI-generated photos, videos, and speech is not going to be possible.
--:-- Our view is you have to flip the problem. You have to invert it. What you have to do instead is have a system in which real people can certify that content about them is real, where content has provenance. Go ahead, Ben, describe how that would work.
--:-- Ben Horowitz: One of the amazing things about crypto blockchain is it deploys something known as a public key infrastructure, which enables every human to have a key that's unique to them where they can sign. If I was in a video or a photo or I wrote something, I can certify that yes, this is exactly what I wrote. You cannot alter it to make it into something else. As that thing gets transferred to the world — let's say it's a song that you sell — you can track it. Just like with a work of art, we track the provenance, or with a house, who owned it before you. That's also an easy application on the blockchain. That combination of capabilities can make this whole program much more viable in terms of knowing what's real, what's fake, where it came from, where it started, where it's going.
--:-- Going back to the data one, I think it's really important. One of the things these systems have done that is dodgy — and there has been big pushback against it with Elon trying to lock down Twitter and the New York Times suing OpenAI — these systems have gone out and just slurped in data from all over the internet and from people's businesses and trained their models on them. I think there's a question of whether the people who created that data should have any say in whether the model is trained on that data. Blockchain is an unbelievably great system for this because you can permission people to use it. You can charge them a fee. It can be all automated in a way where you can say, sure, come train. Training data ought to be of this nature where there's a data marketplace and people can say, yes, take this data for free, I want the model to have this knowledge, or no, you can't have it for free but you can have it, or no, you can't have it at all. Rather than what's gone on, which is this very aggressive scraping. You have these very smart models where these companies are making enormous amounts of money taken from data that certainly didn't belong to them. Maybe it's in the public domain, but that ought to be an explicit relationship. It's not today. That's a very great blockchain solution. Part of the reason we need the correct regulation on blockchain and we need the SEC to stop harassing and terrorizing people trying to innovate in this category.
--:-- Then you have training and inference. The push against decentralized training and inference is, well, you need this very fast interconnect and you need it all in one place technologically. I think that's true for people who have more money than time — startups and big companies. But for people in academia who have more time than money, they're getting completely frozen out of AI research. There's not enough money in all of academia to participate anymore in AI research. Having a decentralized approach where you can share all the GPUs across your network — yeah, maybe it takes a lot longer to train your network or serve it. But you still can do your research. You can still innovate, create new ideas, new architectures, and test them out at large scale. That will be amazing if we can do it. Again, we need the SEC to stop illegally terrorizing every crypto company and trying to block laws from being put in place that help us enable this.
--:-- Marc Andreessen: The college thing actually really matters. We have a friend who is very involved in one of the big computer science programs at one of the major American research universities. A lot of the technology we're talking about was developed at American research universities.
--:-- Ben Horowitz: And Canadian ones too. Toronto.
--:-- Marc Andreessen: Canadian ones and European ones. Historically, as with every other wave of technology in the last hundred years, the research universities have been the gems, the wellsprings of a lot of the new technology empowering the economy and everything else around us. This friend said a while ago that his concern was that his university would be unable to fund a competitive AI cluster — a compute grid that would actually let students and professors work in AI because it's now getting too expensive and universities are just not funded to have CapEx programs that big. More recently, his concern has been that all research universities together might not be able to afford to do that. All universities together might not be able to have cutting edge AI work happening on the university side.
--:-- I was in D.C. in a bipartisan House meeting the other day on these topics. One of the Democratic Congresswomen asked me a question that comes up: how do you get more members of underrepresented groups involved in tech? I found myself giving the same answer I always give: the most effective thing you need to do is go upstream and have more people coming out of college with computer science degrees who are skilled, qualified, trained, and mentored to participate in the industry. You and I, Ben, both came out of state schools with computer science programs where we were able to have the careers we've had. I answered the question saying we need more computer science graduates from every group. But in the back of my head, I was thinking it's going to be impossible to do that because none of these places are going to be able to afford to have the compute resources to actually have AI programs in the future.
--:-- Maybe the government can fix this by dumping a ton of money on top of these universities. Maybe that's what will happen. The current political environment seems like maybe it's not quite feasible for a variety of reasons. The other approach would be a decentralized approach, a blockchain-based approach that everybody could participate in, if that were something the government were willing to support, which right now it's not. I think there's a really central, important, vital issue here that is being glossed over by a lot of people that should really be looked at.
--:-- Ben Horowitz: No, I think it's absolutely critical. Going back to our original thing, it's so important to the country being what America should be to get these issues right. We're definitely in danger of that not happening. I think people are taking much too narrow a view of some of these technologies and not understanding their full capabilities. We get into "oh, the AI could say something racist, therefore we won't cure cancer." We're getting into that kind of dumb idea. We need to have a tech-forward kind of solution to some of these things, and then the right regulatory approach to make the whole environment work.
--:-- Let's go to the next phase now, which is the global implications. I'm going to conjoin two different topics on purpose. Michael Frank Martin asks, what could the U.S. do to position itself as the global leader of open source software? Do you see any specific legislation or regulatory constraints hampering the development of open source projects? Arda asks a similar question: what would an ideal AI policy for open source software models look like? And then Sarah Holmes asks the China question: do you think we will end up with two AI tech stacks, the West and China, and ultimately companies will have to pick one side and stay on it?
--:-- Marc Andreessen: And so, look, I would say this is where you get to the really, really big geopolitical long-term issue. My understanding of things is as follows: for a variety of reasons, technological development in the West is being centralized into the United States, with some in Canada and some in Europe, although quite frankly, a lot of the best Canadian and European tech founders are coming to Silicon Valley. Jan LeCun, a hero in France, teaches at NYU and works at Meta, both American institutions. So there's an American, or American plus European, tech vanguard wedge in the world. And then there's China. It's actually quite a bipolar situation.
--:-- Marc Andreessen: The dreams of tech being fully democratized and spreading throughout the world have been realized on the use side, but not nearly as much on the entrepreneurship side or the invention side. Immigration is a great virtue for the countries that are the beneficiaries, but the other side is it makes other countries less competitive because their best and brightest move to the US. So we are in a bipolar tech world, primarily the US and China. This is not the first time we have been in a bipolar world involving geopolitics and technology.
--:-- Marc Andreessen: The US and China have two very different systems. The Chinese system has all the virtues and downsides of being centralized. The US system has all the virtues and downsides of being more decentralized. There is a very different set of views on how society should be ordered, what freedom means, and what people should be able to do. Both the US and China have visions of global supremacy and agendas to carry forward their points of view on the technology of AI and its societal implications throughout the world. So there is this Cold War.
--:-- Marc Andreessen: In DC, it's crystal clear that there's a dynamic where Republicans and Democrats are trying to leapfrog each other every day on being more anti-China. Our friend Neil Ferguson is using the term "Cold War 2.0." Whether we want to or not, we're in Cold War 2.0, a dynamic similar to the one with the USSR 30, 40, 50 years ago. To Sarah Holmes' question, it's 100% going to be the case. There are two AI tech stacks, two AI governance models, and two AI deployment systems. There are two ways AI dovetails into everything from surveillance to smart cities to transportation, self-driving cars, drones, who controls what, who gets access to what, who sees what. The degree to which AI is used as a method for population control involves very different visions. These are national visions and global visions, and there's a very big competition developing.
--:-- Marc Andreessen: It certainly looks to me like there's going to be a winner and a loser. I think it's overwhelmingly in our best interest for the US to be the winner. For the US to be the winner, we have to lean into our strengths. The downside of our system is that we are not as well organized and orchestrated top-down as China is. The upside of our system, at least historically, is that we benefit from decentralization, competition, a market economy, and a private sector. We have a much larger number of smart people making lots of small decisions to get to good outcomes, as opposed to a dictatorial system with a small number of people trying to make decisions. This is how we won the Cold War against Russia: our decentralized system just worked better economically, technologically, and ultimately militarily. So it seems fairly obvious that we have to lean into our strengths. If we think we're just going to be another version of a centralized system without all the advantages China has, that seems like a bad formula.
--:-- Marc Andreessen: Let me pause there and see what you think.
--:-- Ben Horowitz: Yeah, I know for sure. I think that would be disastrous. I think this is why it's so clear that if there's one regulatory policy we would enact to ensure America's competitiveness, it would be open source. The reason is that, as you said, this enables the largest number of participants to contribute to AI, to innovate, and to come up with novel solutions. You're right about China. They're going to pick one because they can, and they're going to drive all their wood behind that arrow in a way we could never do because we just don't work that way. They're going to impose that on their society and try to impose it on the world. Our best counter to that is to put it in the hands of all of our smart people. We have so many smart people from all over the world. As we like to say, diversity is our strength. We have tremendous different points of view and different kinds of people in our country. The more we can enable them, the more likely we'll be competitive.
--:-- Ben Horowitz: I'll give you a tremendous example. If you go back to 2017 and read any foreign policy magazine, there wasn't a single one that didn't say China was ahead in AI. They have more patents, more students going to universities. They're ahead in AI. Then ChatGPT comes out and goes, "Oh, I guess we're not behind in AI. We're ahead in AI." The truth is, what China was ahead on was integrating AI into their government, their one AI, into their government in that way. We're working on doing a better job of that with American dynamism, but we're never going to be good at that model. That's the model they're going to be great at, and we have to be great at our model. If we start limiting that, outlawing startups and outlawing anybody but the big companies from developing AI, we'll definitely shoot ourselves in the foot.
--:-- Ben Horowitz: I would say another important point in the safety of the world is that when you talk about two AI stacks, it's very important that countries aligned with America or China can align AI to their values. I'll give you one really important example. I've been spending a lot of time in the Middle East. If you look at the history of a country like Saudi Arabia, they're coming from a world of fundamentalism and a set of values they're trying to modernize. They've done tremendous things with women's rights, but there are still people who don't want to go to that future so fast. They need to preserve some of their history to avoid a revolution or extreme violence. We're seeing Al-Qaeda re-spark up in Afghanistan, and Al-Qaeda is the real enemy of modern Saudi just as much as they are an enemy of America. If Saudi can't align an AI to their current values, they could literally spark a revolution in their country. It's very important that the technology we develop not be totally proprietary closed source, but modifiable by our allies who need to progress at their pace to keep their country safe and keep us safe in doing so. This has great geopolitical ramifications. If we go to the China model that Google and Microsoft are advocating for, this Chinese model of only a few can control AI, we're going to be in big trouble.
--:-- Marc Andreessen: Yeah, and I just want to close on the open source point because it's so critical. This is where I get extremely irate at the idea of closing down open source, which a number of these people are lobbying for very actively. I'm going to name one more name. We even have VCs lobbying to outlaw open source, which I find to be completely staggering. Vinod Khosla. This is incredible to me. He's the founder of Sun Microsystems, which was in many ways a company built on open source, built on open source Unix out of Berkeley, and itself built a lot of critical open source. The internet was all built on open source. And Vinod has been lobbying to ban open source AI. He denies he's been doing this, but I saw him with my own eyes when the Congressional China Committee came to Stanford. I was in the meeting where he was with 20 or 30 congressmen lobbying actively for this. I've seen him do it myself. He's got a big stake in OpenAI. Maybe it's financial self-interest. Maybe he's a true believer in the dangers.
--:-- Ben Horowitz: Well, I think he proved on Twitter he was not a true believer in the dangers. I'll get into that.
--:-- Marc Andreessen: Yeah. So even within little tech, even within the startup world, we are not uniform in this. I think that's extremely dangerous. What is open source software? It is quite literally the technological equivalent of free speech, which means it's the technological equivalent of free thought. It is the way the software industry has developed to build many of the most critical components of the modern technological world. As you said earlier, it allows us to secure those components and have them be safe and reliable. It provides the transparency we've talked about so you know how they work and how they're making decisions. And to your last point, it allows for customized AI in many different environments. You don't end up with a world where you just have one or a couple AIs, but a diversity of AIs with lots of different points of view and capabilities. The open source fight is at the core of this.
--:-- Marc Andreessen: The reason people with an eye towards monopoly or cartel want to ban this is that open source is a tremendous threat to monopoly or cartel. It's a guarantee that monopoly or cartel can't last. It is absolutely required for the furtherance of a vibrant private sector and a vibrant startup sector. Going back to the academia point, without open source, university college kids won't be able to learn how the technology works. They'll be completely boxed out. A world where open source is banned is bad on so many fronts. It's incredible to me that anybody's advocating for it. It needs to be recognized as the threat that it is.
--:-- Ben Horowitz: Yeah. And on Vinod, it was such a funny dialogue between you and him. I'll give you a quick summary. He was arguing for closed source, you for open source. His core argument was that this is the Manhattan Project and therefore we can't let anybody know the secrets. You countered by saying, if this is the Manhattan Project, is the OpenAI team locked in a remote location? Do they screen all their employees very carefully? Is it airlocked? Is there super high security? Of course, none of that is close to true. I'm quite sure they have Chinese nationals working there. Probably some are spies for the Chinese government. There's no strong security at OpenAI or Google or any of these places anywhere near the Manhattan Project, where they built a whole city that nobody knew about so they couldn't get into it. Once you caught him in that, he said nothing. Then he said, "Well, it costs billions of dollars to train these models. You just want to give that away? Is that good economics?" That was his final counterpoint, which basically said, "Oh, yeah, I'm trying to preserve a monopoly here. What are you doing? I'm an investor." I think that's true for all these arguments.
--:-- Marc Andreessen: The kicker to that story is that three days later, the Justice Department indicted a Chinese national Google employee who stole Google's next generation AI chip designs, which is quite literally the family tools for an AI program. It's the equivalent of stealing the design for the bomb. That Google employee downloaded those chip designs and took them to China. By definition, that means he took them to the Chinese government, because there's no distinction in China between the private sector and the government. It's an integrated thing. The government owns and controls everything. 100% guaranteed that went straight to the Chinese government and Chinese military. Google, which has a big information security team, did not realize, according to the indictment, that that engineer had been in China for six months.
--:-- Ben Horowitz: Amazing.
--:-- Marc Andreessen: It gets better. This is the same Google with the same CEO who refused to sell Google proprietary AI technology to the US Department of Defense. So they're supplying China with AI and not supplying the US. That goes back to the point: if it's not open source, we're never going to compete. We've lost the future of the world right here, which is why it's the single most important AI issue for sure.
--:-- Ben Horowitz: Yeah, and you're not going to lock this stuff up. Nobody's locking it up. It's not locked up. These companies are security Swiss cheese. You'd have a debate about the tactical relevance of chip embargoes, but the horse has left the barn on this. These companies are without a doubt riddled with foreign assets and are very easy to penetrate. We have to be very realistic about the actual state of play here. We have to play in reality. We have to win in reality. The win is we need innovation, competition, free thought, and free speech. We need to embrace the virtues of our system and not shut ourselves down in the face of the conflicts that are coming.
--:-- Host: Another one. Andreas asks, why are US VCs so much more engaged in politics and policy than their global counterparts? I really appreciate that question because it basically means that VCs outside the US must not be engaged at all, because US VCs are barely engaged. And then, what do you believe the impact of this is on both the VC ecosystem and society in general? And then related, Vincent asks, are European AI companies becoming less interesting investment targets for US-based VCs due to the strict and predictably unpredictable regulatory landscape in Europe? Would you advise early stage European AI companies to consider relocating to the US as a result?
--:-- Marc Andreessen: Great question. I think it goes back to a little of what you said earlier. In the startup world, there's the United States and then there's everywhere else. The United States is bigger than everywhere else combined. So it's natural. In these political things, it starts with the leader. The US is the leader in VC. We feel like we're the leaders in US VCs, so we need to go to Washington. Until we go, nobody's going. That's a lot of the reason why we started things.
--:-- Marc Andreessen: On European regulatory policy, I think generally regulatory policy is likely to dictate where you can build these companies. We've seen some interesting things. France turns out to be leading a revolution in Europe on AI regulatory policy, where they're basically telling the EU to pound sand. A large reason is because they have a company there, Mistral. It's a national jewel for the country, and they don't want to give it up because the EU has some crazy safetyism thing going on.
--:-- Ben Horowitz: Yeah. And you'd also note France is playing the same role with nuclear policy in Europe. They're the one country. They're probably one of the cleanest countries in the world as a result.
--:-- Marc Andreessen: Right. They had been staunchly pro-nuclear and trying to hold off attempts throughout the rest of Europe, especially from Germany, to ban civilian nuclear power. The UK has been flip-flopping on AI policy, and we'll see where they come out. Brussels has been ridiculous, as they have been on almost everything.
--:-- Marc Andreessen: The big thing I note here is a really big philosophical distinction. It's rooted in the difference between the Anglo-American approach to law and the continental European approach. The legal terms are common law and civil law. The difference is basically: that which is not outlawed is legal, versus that which is not explicitly legal is outlawed. By default, do you have freedom and then impose law to have constraints? Or by default, do you have no ability to do anything and then the law enables you to do things? This is a fundamental philosophical, legal, political distinction. It shows up in a lot of these policy issues with the precautionary principle, which is the traditional European approach. The precautionary principle says new technologies should not be allowed to be fielded until they are proven to be harmless.
--:-- Marc Andreessen: The precautionary principle is a hallmark of the European approach to regulation and increasingly the US approach. Its origin was described and given that name by the German Greens in the 1970s as a means to ban civilian nuclear power, with catastrophic results. We could spend a lot of time on that, but I think everybody at this point agrees, including the Germans, that was a big mistake. It has led to Europe funding Russia's invasion of Ukraine through the need for imported energy because they keep shutting down their nuclear plants. A catastrophic decision. But the precautionary principle has become extremely trendy. It sounds great: why would you want anything released in the world if it's not proved to be harmless? How can you possibly support anything that's going to cause harm? The obvious problem is that with that principle, you could never have deployed technology such as fire, electric power, internal combustion engines, cars, airplanes, or the computer. Every single piece of technology that powers modern civilization has some way it can be used to hurt people. Technologies are double-edged swords. You can use fire to protect your village or to attack the neighboring village. If we had applied the precautionary principle historically, we would still be living in mud huts.
--:-- Marc Andreessen: The idea of imposing the precautionary principle today, from an Anglo-American perspective or a freedom to innovate perspective, is incredibly horrifying. It's guaranteed to stall out progress. This is very much the mentality of the EU bureaucrats in particular, and it's the mentality behind a lot of their recent legislation on technology issues. France does seem to be the main counterweight against this in Europe. The UK has been a counterweight in some areas, but they've also received a full dose of this programming.
--:-- Ben Horowitz: Yeah, they have that tendency.
--:-- Marc Andreessen: In AI in particular, I think they've been on the wrong side of that, which hopefully they'll reconsider. This is a really important issue. If the surface level thing of "this technology might be able to be used for some harmful purpose" is allowed to be the end of the discussion, nothing new is ever going to happen in the world. That will cause us ultimately to stall out completely. If we stall out, that will over time lead to regression. This is happening. The power is going out. German industrial companies are shutting down because they can't afford the power that's resulted from the imposition of this policy in the energy sector. This is a very, very important thing. I think the EU bureaucracy is lost on this. It's going to be up to the individual countries to directly confront this if they want to. I really applaud what France has done, and I hope more European countries join them in being on the right side on this.
--:-- Ben Horowitz: Yeah, it always is funny to me to hear the EU and the economists say, "Oh, the EU may not be the leader in innovation, but we're the leaders in regulation." I'm like, well, you realize those go together. One is a function of the other.
--:-- Host: Okay, good. Let's do one more global question. Lep Gong Leong asks, are there any other countries that could be receptive to techno-optimism? For example, could Britain, Argentina, or Japan be ideal targets for our message and mission?
--:-- Ben Horowitz: Yes. We'll look for work on that in Britain. We've got some pretty good reception from the UK government. There are a lot of very smart people there. We're working with them tightly on their AI and crypto efforts, and we're hoping that's the case. Japan, having spent a lot of time there, has obviously shown that capability over time. There's a lot about the way Japanese society works that holds them back from that at times. Without getting into all the specifics, they have a very unusual and unique culture that has a great deference for the old way of doing things, which sometimes makes it hard to promote the new way of doing things.
--:-- Ben Horowitz: Around the world, the Middle East is very on board with techno-optimism. The UAE, Saudi, Israel, of course. Many countries out there are very excited about these ideas, taking the world forward, and creating a better world through technology. With our population growth, if we don't have a better world through technology, we're going to have a worse world without technology. That's very obvious. It's a very compelling message. Oh, and South America. There are a lot of countries who are really embracing techno-optimism now in South America, with some great new leadership there pushing that.
--:-- Marc Andreessen: I would also say, if you look at the polling on this, what you find is that the younger countries are more enthusiastic about technology. I don't mean younger literally in when they were formed, but two things. One is how recently they've emerged into what we would consider modernity, embracing concepts like democracy, free market capitalism, innovation, or global trade. The other is the demographics: countries with a large number of young people. Those are often the same countries. They have the reverse demographic pyramid we have, where they actually have a lot of young people. Young people both need economic opportunity and are very fired up about new ideas.
--:-- Ben Horowitz: By the way, this is true in Africa as well. In many African countries like Nigeria, Rwanda, and Ghana, techno-optimism is taking hold in a real way. Some of them need governance improvements, but they definitely also have young populations. Saudi, I think 70% of the population is under 30. So just to your point, that's very hopeful in those areas.
--:-- Host: Ben, Goltra asks, do you think the lobbying efforts by good faith American crypto firms will be able to move the needle politically in the next few years? What areas make you optimistic as it relates to American crypto regulation, crypto, blockchain, Web3?
--:-- Ben Horowitz: Yeah, I think I'm hopeful. I'm as hopeful as I've ever been. There are a bunch of things that have been really positive. First of all, the SEC has lost, I think, five cases in a row. Some of their arbitrary enforcement of things that aren't laws is not working. Secondly, there was a bill that passed through the House Financial Services Committee, which is a very good bill on crypto regulation. Hopefully that will eventually pass the House and the Senate. We've seen Wyoming adopt really good new laws around DAOs. There's some progress there.
--:-- Ben Horowitz: We've been working really hard to educate members of Congress and the administration on the value of the technology. There are strong opponents to it, as I mentioned earlier, and that continues to be worrisome. But I think we're making great progress. The Fair Shake PAC has done a tremendous job of backing pro-crypto candidates. With great success, there were six different races on Super Tuesday that they backed, and all six won. So, another good sign.
--:-- Host: Fantastic. I'll hit a couple other topics here quickly to get under the wire. Father Time asks, can you give us your thoughts on the recent TikTok legislation? If passed, what does this mean for big tech going forward?
--:-- Host: And so I'll just give a quick swing at that. So the TikTok legislation being proposed by the U.S. Congress and currently being taken up in the Senate, which, by the way, and President Biden has already said he'll sign it if the Senate and the House pass it. This is legislation that would require a divestment of TikTok from its Chinese parent company, ByteDance. And so TikTok would have to be a purely American company or would have to be owned by a purely American company. And then failing that, it would be a ban of TikTok in the U.S.
--:-- Host: This bill is a great example of the sort of bipartisan dynamic in D.C. right now on the topic of China. This bill is being enthusiastically supported by the majority of politicians on both sides of the aisle. I think it passed out of its committee like 50 to zero, which is basically impossible to get anybody in D.C. to agree on anything right now except basically this. So this is super bipartisan. The head of that committee is a Republican, Mike Gallagher, and he worked in a bipartisan way with his committee members. But the Democratic White House immediately endorsed his bill. So this bill has serious momentum.
--:-- Host: The Senate is taking it up right now. They're likely to modify it in some way, but it seems reasonably likely to pass based on what we can see, with overwhelmingly bipartisan support.
--:-- Host: The argument for the ban or the divestment is a couple different ways. Number one, an app on the phones of a large percent of Americans with the surveillance and potential propaganda aspects certainly has people in Washington concerned. And quite frankly, there's an underlying industrial dynamic: U.S. internet companies can't operate in China. So there's an unfair asymmetry underneath this that really undercuts a lot of the arguments from ByteDance.
--:-- Host: It has been striking to see that there are actually opponents of this bill who have emerged, on the further right and further left in their respective parties. On the further left, I think there are people who feel like TikTok is a really important and vital messaging system for them to use with their constituents who tend to be younger and very internet centric. On the further right, there are a fair number of people who are very worried that the U.S. government is so prone to abuse any regulatory capability with respect to tech, and especially with respect to censorship, that if you hand the U.S. government any new regulatory authority to come down on tech, it will inevitably be used not just against the Chinese company, but also against the American companies.
--:-- Host: So there's some drama surfacing around this, and we'll see whether the opponents can pull it through. Quite frankly, I think this is one of those cases where there are excellent arguments on all three sides. There are very legitimate questions here. So I think it's great that the issue is being confronted, but it's also great that the arguments have surfaced and that we're going to hopefully figure out the right thing to do.
--:-- Host: Couple closing things. Let's close on a hopefully semi-optimistic note. John Potter asks, how do you most effectively find common ground with groups and interests that you benefit from working with, but with which you are usually opposed ideologically or otherwise?
--:-- Ben Horowitz: I mean, I think there's this term in Washington, common ground. I think you always want to start by finding the common ground because in politics generally, most people have the same intention. In Washington, people want life to be fair. They don't want people to go hungry. They want citizens to be safe but have plenty of opportunity. So there's a lot of common ground. The differences lie not in the intent, but how you get there, what is the right policy to achieve the goal. So I think it's always important to start with the goal and then work our way through why we think our policy position is correct. We don't really have a lot of disagreements on stated intent, at least.
--:-- Ben Horowitz: I mean, there are some intentions that are very difficult in Washington. The intention to control the financial system from the government, or nationalize the banks, or achieve the equivalent of nationalizing the banks. When you have that intent, that's tough. But most intentions are, I think, shared between us and policymakers on both sides.
--:-- Host: And then we'll close on this great question. Zach asks, would either of you ever consider running for office? And for fun, what would be your platform?
--:-- Marc Andreessen: I want to just say, look, I think being a politician requires a certain kind of skill set and attitude and energy from certain things that I don't possess, unfortunately.
--:-- Host: Do you have a platform you would run on if you did run?
--:-- Marc Andreessen: Yeah, I do.
--:-- Host: Okay, yeah, let's hear your platform.
--:-- Marc Andreessen: The American dream. I won't do it now, but I like to put up this chart that shows the change in prices in different sectors of the economy over time. What you basically see is the price of television sets, software, and video games are crashing hard in a way that's great for consumers. I saw a 75 inch flat screen, ultra high def TVs now are down below $500. It's amazing. When technology is allowed to work its magic, prices crash in a way that's just great for consumers. It's the equivalent of a giant raise. So it makes human welfare a lot better.
--:-- Marc Andreessen: The three elements of the economy that are central to the American dream are healthcare, education, and housing. To have the American dream means to be able to buy and own a home. It means being able to send your kids to great schools, get a great education, to have a great life. And then it means great healthcare to be able to take care of yourself and your family. The prices on those are skyrocketing, straight to the moon. And of course, those are the sectors that are the most controlled by the government, where there's the most subsidies for demand from the government, the most restrictions on supply from the government, and the most interference with the ability to field technology and startups.
--:-- Marc Andreessen: The result is we have an entire generation of kids who, I think, are quite rational in looking forward and saying, I'm never going to be able to achieve the American dream. I'm never going to be able to own a home. I'm never going to be able to get a good education or send my kids to a good education. I'm not going to be able to get good healthcare. I'm not going to be able to live the life that my parents lived or my grandparents lived. And I'm not going to be able to fundamentally form a family and provide for my kids.
--:-- Marc Andreessen: I think that's the underlying theme to what has gone wrong socially, politically, psychologically in the country. That's what's led to this intense level of pessimism. That's what's led to this attraction to very zero-sum politics, recrimination over optimism and building. So I would confront that absolutely directly. And then, of course, I would point out that I don't think anybody in Washington is doing that right now. Either I would win because I'm the only one saying it out loud, or I would lose because nobody cares. But I've always wondered whether that actually would, both on the substance and on the message, whether that would be the right platform.
--:-- Ben Horowitz: Yeah, no, it would certainly be the thing to do. It's very complex in that healthcare policy is largely national, but education policy and housing policy also has a very large local component. So it'd be a complicated set of policies that you'd have to enforce.
--:-- Host: We still have a ton of questions, so we may do part two on this at some point. But we really appreciate your time and attention, and we will see you soon.
--:-- Marc Andreessen: Okay, thank you.
