Marc Andreessen on AI and Dynamism

0:03Hi CWT listeners, I'm Sam, one of the producers of the show. Today we're bringing you a bonus episode featuring a recent interview Tyler Held with Mark Andrewson on AI. This interview took place in late January at A16Z's 2024 American Dynamism Summit in Washington, DC. To hear more conversations from this event, visit A16Z .com slash AD Summit or use the link in the show notes below. Now onto the show!

0:37Conversations with Tyler is produced by the Mercatus Center at George Mason University, bridging the gap between academic ideas and real -world problems. Learn more at mercatis .org. For a full transcript of every conversation, enhanced with helpful links, visit ConversationsWithTyler .com.

1:05Hello Mark, if you're entrance music we're to be Beethoven, which symphony and why? For those of you who do care and know who Tyler Cohen is, this is how his podcast starts, this is how he intimidates his guests into submission. First of all, I just wanted to thank everybody for being here with us today. We're really grateful that you were all able to spend time with us and hopefully it's been useful. Second is I'm going to get new business cards printed up that say you either know who I am or you don't care. Or both. Or both. Well, let's see. I mean, I guess we have to rule out Beethoven's ninth symphony because that's the official music of the European Union, is that right?

1:38That's correct. That's the official anthem of the European Union. That's right. Which is just such a terrible mean thing for them to do. It's such a great piece of music like that. We should lodge a formal diplomatic protest. I guess probably Beethoven's fifth in retaliation. I would pick you as the fifth. Now, how will AI make our world different? Five years from now. What's the most surprising way in which it will be different? So there's a great breakdown on adoption in new technology that the science fiction author Douglas Adams wrote about years ago. He says any new technology has received differently by three different groups of people.

2:08If you're below the age of 15, it's just the way things have always been. If you're between the ages of 15 and 35, it's really cool and you might be able to get a job doing it. If you're above the age of 35, it's on holy and against the order of society and will destroy everything. AI, I think so far, is living up to that framework. What I would like to tell you is AI is going and be completely transformative for education. I believe that it will. Having said that, I did recently roll out Chad G. P. T. to my eight year old. I was very, very proud of myself, because I was like, wow, this is just gonna be such a great educational resource for him.

2:36And I felt like Prometheus bringing fire down from the mountain to my child. And I installed it on his laptop and said, son, this is the thing that you can talk to anytime and it will answer any question you have. And he said, yeah. And I said, well, no, this is like a big deal. It answers questions. He's like, well, what else would you use a computer for? And I was like, oh God, I'm getting old. So I actually think there's like a pretty good prospect that like kids are just going to like pick this up and run with it. And I actually think that's already happening, right? JGPT is fully out, you know, and Bard and banging all these other things.

3:06And so I think, you know, kids are going to grow up with basically, you know, you could use various terms assistant friend coach, mentor, tutor, but you know, kids are going to grow up in sort of this amazing kind of back and forth relationship with AI. And you know, and any time a kid is interested in something, if there's not a teacher who can help with something or they don't have a friend who's interested in the same thing, they'll be able to explore all kinds of ideas. And so I think it'll be great for that. You know, I think it's obviously going to be totally transformative and it feels like warfare and you already see that.

3:33You know, the concern quite honestly, I actually wrote an essay a while ago on sort of why I won't destroy all the jobs. And the sort of short version of it is because it's illegal to do that because so many jobs in the modern economy require licensing and are regulated. And so, you know, I think the concern would be that there's just so much sort of glue in the system now that prevents change, and it'll be very easy to sort of not have AI healthcare or AI education or whatever, because literally some combination of like, doctor licensing, teacher unions and so forth will basically outlaw it. So I think that's the risk.

4:04If we think of AI and its impact in sociological terms, large language models, who will gain in status and who will decline in status, and how should this affect how we think about policy? Yeah, so first of all, it's important to qualify sort of exactly what's going on with large language models, which is super interesting. And kind of this thing has happened that you kind of read about a lot in the press, which is kind of there was this general idea that there would be something called AI at some point. And then large language models appeared and everybody said, aha, that's AI, just like we thought it would be.

4:30And then everybody sort of extrapolates out. And that's true to a certain extent. But the success of large language models is very unexpected in the field. And actually the origin story of even chat GPT is, this is not what OpenAI actually started to do. They started to do something different. And then there's actually one guy named as I think Alec Redford. And he literally was like off in the corner of an open AI, like working on this in like 2018, 2019. And then it just basically was this revolution building on work that had been done at Google. So it was kind of this very surprising thing. And then it's important to sort of qualify like how it works, because it's not just like some sort of robot brain.

5:03What it is is you basically feed essentially, ideally, all known human generated information into a machine. And then you let it basically build a giant matrix of numbers and basically correlate everything. And I nutshell, that's what these things are. And then basically what happens is when you ask a question, or if you ask it to make a drawing or something, essentially does a search. It does a search across basically all of these words of sentences and diagrams and books and photos and everything that human beings have created. And it sort of tries to find the optimal kind of path through that, and that's how it sort of generates the answer that it gives you.

5:34And so it's philosophically, it's this kind of this really profound thing, I think, which is, it's like you as an individual, using this machine to kind of stare at the entirety of the creation of all human knowledge, and then and sort of have it played back at you. And so it sort of harnesses the creativity of thousands of years of human authors and artists and then sort of drives new kinds of answers, new kinds of images, whatever. But fundamentally, you're sort of in interaction with our civilization in a very profound way. In terms of who gains and who loses status, there's actually a very interesting thing happening in the research right now.

6:04There's a very interesting research question for the impact on job skill. For example, for people who work with words or work with images and are starting to use these technologies in the workforce. And so the question is, who benefits more, the high skilled worker, and think, you know, a lawyer, doctor, accountant, whatever, graphic designer, the high skilled person who uses these tools to become, right, an additional quantum leap high skilled, and that would be a theory of sort of separation. But the other scenario is the sort of average or even low skilled worker who gets upgraded. And of course, just kind of the nature of the economy, they're kind of more people in the middle.

6:37And at least there's been a series of research studies that have been coming back. it's actually the uplift average is actually more significant than the uplift to the high school level. And so actually what's used to be happening right now is it's actually a compression by kind of lifting people up. Social questions are often a zero -sum game of who gains and who loses, but there may be something here where just a lot of people just get better at what they do. Why is open source AI in particular important for national security? Yeah, so for a whole bunch of reasons. So one is it is really hard to do security without open source.

7:07There used to be, there's actually two schools of thought on kind of information security, computer security broadly that have played out over the last 50 years. There was one school of security that says you want to basically hide the source code. And this seems intuitive because you want to hide the source code so that bad guys can't find the flaws in it, right? And presumably that would be the safe way to do things. And then over the course of the last 30 or 40 years, basically what's evolved is the realization in the field and I think very broadly that actually that's a mistake. In the software field we call that security through obscurity, right?

7:35It's sort of we hide the code, people can't exploit it. The problem, of course, is, okay, but that means the flaws are still in there, right? And so if anybody actually gets to the code, they just basically have a complete index of all the problems, and there's a whole bunch of ways for people to get to code, they hack in. You know, it's actually very easy to steal software code from a company. You hire the janitorial staff to stick a USB stick into a machine at three in the morning. Software companies are like very easily penetrated. And so it turned out, security through obscurity was a very bad way to do it.

8:02But the much more secure way to do it is open source, basically put the code in public, and then basically build the code in such a way that when it runs, it doesn't matter whether somebody has access to the code. It's still fully secure, and then you just have a lot more eyes on the code to discover the problems. So in general, open source has turned out to be much more secure. I would start there if we want secure systems. I think this is what we have to do. What's the biggest adjustment problem governments will face as AI progresses? For instance, if drug discovery goes up by 3X, all of a sudden the FDA is overloaded.

8:32If regulatory comments are open, AI can write great regulatory comments. What does government have to do to get by in this new world? Yeah, so I think for every scenario, hopefully at least the first of those two scenarios happens, maybe also the second. For anything like this, what there should be is there should be a corresponding phenomenon happening on the other side. And so the government sort of correspondingly then should be using AI to evaluate new drugs. So company shows up in their drug design. There should be AI assist to the FDA to help them evaluate new drugs. A regulatory agency that has public comments should have AI assist for being able to be able to process all the information and be able to aggregate it and be able to reply back to everybody.

9:09There's a very interesting thing about AI. Every possible threat you can think of AI posing basically there is a corresponding defense that has to get built. I'll pick another one. Cybersecurity people are quite legitimately concerned that AI is going to make it easier to actually create and launch cybersecurity attacks. But correspondingly, there should be better defenses. There should be AI -based cybersecurity defenses. By the way, we see the exact same thing with drones. You know, weaponized AI autonomous drones are clearly a threat, as we see in the world today. And so we need AI defenses against drones.

9:38The cynical view would be this is just a classic arms race, you know, attack defense, attack defense, and kind of does the world get any better if there's just more threats and more defenses. I think the positive way of looking at it is we probably need these defenses anyway, right? So even if we didn't have AI drug discovery, I think we should be using AI to evaluate drugs, even if we didn't have AI drones, we should still have defense against standard missiles and enemy aircraft, even if we didn't have AI driven cyber attacks, we should have AI driven cyber defenses. So I think this is an opportunity for the defenders to not only keep up, but also build better systems for the present -day threat landscape also.

10:13The Biden AI directive. What's the best thing about it? What's the worst thing about it? Yeah, so the best thing about it is it didn't overtly attempt to kill AI. So that was good. You never know if these things, how much teeth they're going to try to put into it. And then, of course, there's always the question of whether that it stands up in court. There were things that were being discussed in the process that were much, much worse. And I think much, much more hostile to the technology than ended up being in it. So I think that's good news. I think it was quite benign in terms of it's just like flat out directives, which is good.

10:41The issue with it, people have different opinions. My opinion, the issue of it is it kind of greenlit, essentially 15 different regulatory agencies to basically put AI under their purview in sort of undefined ways. And so, you know, we will now have, I think, a relatively protracted process of many regulators and many agencies without explicit authority in the domain, basically inserting themselves into the space. And then, you know, presumably, at some point, there will be a determination of who, you know, has purview over what, but it seems like we're in for a period of quite a bit of confusion as a result.

11:09So how much more green energy do we need to an essence fuel all of this AI, and where will it come from? What do you see the prospect as like for the next 20 years? Yeah. The good news of AI and the good news with also, by the way, with crypto, because there's always a lot of controversy around crypto and Web 3 and blockchain around or on energy use. The good thing with these technologies, the good news from energy is these systems lend themselves to centralization of data centers, right? And so if we need a million to, going to 10 million, go to 100 million, a billion AI chips, could be distributed out all over the place, but they can also be highly centralized.

11:39Because you can highly centralize them, you can think not just in terms of building a server, you can think about building basically a data center or there's an integrated thing from the chip, basically all the way to the building or to the complex of buildings. And then the way those modern data centers are built by the leading edge companies now is they're sort of built on day one with an integrated strategy for energy and for cooling. And so basically any form of energy that you have, that you could do in a very efficient way in a very clean way, or new energy technologies, AI is a use case for developing and deploying that kind of power.

12:09And so just building what we've seen from internet data centers, that could be geothermal, that could be hydroelectric, that could be nuclear fission, that could be nuclear fusion, solar, you know, wind, big battery packs, and so forth. And so I think the aspirational hope would be this is sort of another catalyst to more advanced rollout of energy. And even if there's sort of net energy increase, the sort of motivation to get the higher levels of efficiency will be net good and helping us get to a better energy footprint. And which of those energy sources in your view is most underrated? Oh, I mean nuclear fission for sure is the most underrated today.

12:39Yes, wave of magic one. We ought to be doing what Richard Nixon proposed in 1971, right? We ought to build what he called project independence, which was built, build a thousand new nuclear power plants in the U .S. and then cut the entire U .S. grid over to nuclear and electricity, go to all electric cars, do everything else. Richard Nixon's other, you know, great, you're corresponding, you know, creation of the nuclear regulatory commission. Of course, guarantees that won't happen. The plan is exactly, exactly, on track. But we could. And so, either with existing, you know, nuclear -fishing technology, or there's actually a significant number, you know, now of new nuclear -fishing startups, as well as fusion startups working on new designs.

13:11And so this would certainly be a great use case for that. So if the nations that will do well in the future are strong in AI and strong in energy, thinking about this in terms of geopolitics, which countries rise in importance? For better or worse? OK, so different things. Add a couple more things to that, which companies are in a position to best invent these new technologies, and then there's a somewhat separate question of who's in the best position to deploy. Because it doesn't help you that much to invent it if you can't deploy it. And so I would put that in there. But I mean, look, I would give the US very, very high marks on the invention side.

13:44I think we're the best. I think we have the best RAD innovation capability in the world in most fields, not all, but most. And I think that's certainly true of AI. And I think that's at least potentially true in energy. I don't know whether it actually is, but it could be. And so we should be able to forge ahead on that. China is clearly the other country with critical mass in all of this. You could quibble about the level of invention versus fast follow and talk about kind of IP, IP acquisition, things like that. But nevertheless, whatever your view is, they're moving very quickly and aggressively and have critical mass, big internal domestic market and a huge number of researchers in a lot of state support.

14:17So I think by and large, we're looking for sure on AI and then I think probably also in energy, we're probably looking at primarily a bipolar world for quite a while, and then spheres of influence kind of going out. I would say Europe is sort of a dark horse in a sort of a strange way in that the EU seems absolutely determined to ban everything. So I've sort of put a blanket ban on capitalism within that BAN AI and BAN Energy. On the other hand, we have this incredible AI company called Mistral in France, which is the leading open source AI company right now, and one of the best AI companies in the world, and the French government has actually really been stepping up to help the ecosystem in Europe.

14:49And so I would actually like to see sort of a tri -polar world. I like to see that you fully punch in, but I'm not sure how realistic that is. So let's say you're in charge of speeding up deployment in the United States. What does it you do? State level, local level, Feds, what should we be all be doing? of AI specifically? Everything. Because it's all increasingly interrelated, right? AI, energy, biomedicine, everything. Yes, yes. Well, an AI takes you straight to chips, which takes you straight to the chips act, which is not yet resulted in the creation of energy plants, although it might someday.

15:19The most basic observation is maybe the most but all, which is, you know, stagnation is a choice, decline is a choice. As Tyler's written at Great Length, you know, the US economy downshifted its rate of technological change, you know, basically since the 1960s. And technological change is measured by productivity growth in the economy was much faster prior to the last 50 years than the most recent 50 years. And that's just the result of just sort of a big argument as to exactly what caused that. But a lot of it is just an imposition of just blankets and blankets and blankets of regulation and restrictions and controls and processes and procedures and all the rest of it.

15:51So yeah, so you know, and you can start by saying step one is you know harm. And so this is sort of our approach on AI regulation, which is, you know, don't regulate the technology, you don't regulate AI as a technology any more than you regulated microchips or software or anything like operating systems or databases instead regulate the use cases. And the use cases are generally regulated anyway. It's no more legal to field a new AI design drug without FDA approval than it is a standard design drug. And so apply the existing regulations as opposed to hamstringing the technology. So that's one energy exploitation.

16:20Again, energy is just pure choice. We could be building a thousand nuclear plants tomorrow. My favorite idea there, which always gets me in trouble. and so I can't resist is the Democratic administration should give Koch industries the contract to build a thousand nuclear reactors Right, everybody gets revenge and everybody else the Democrats get Charles Koch to fix climate change and then Charles gets all the money for the contracts And so if everybody ends up happy Nobody yet has bit on that idea when I pitched it, but maybe I'm not talking to the right people So you know look, we could be doing that, you know, we'll see if we choose to look that the chips plant thing is gonna be fascinating to watch There was this really, you know, we passed the chips act in a theory the funding is available and you know the American chip companies are generally pretty aggressive and I think trying pretty hard to build new capacity in the US.

17:01But there was this actually very outstanding article in the New York Times some months back by Ezra Klein where he sort of goes through and he says okay even suppose the money is available to build chip plants like is it actually possible to build chip plants in the US? And he sort of talks about all of the different regulatory and legal, you know basically requirements and obligations to get layered on top. and it was sort of speculating us to whether any of these plants were actually get built. And so again, I think we have here, we have just sort of a level of fundamental choices in society, which is, you know, do we want to build new things?

17:29I can't say how exciting it's been, at least in the West Coast, how exciting it's been for Las Vegas to get the sphere, because it's now impossible to visit Las Vegas without, like, like, everybody's always complaining, the Egyptians built the pyramids, like, we're pyramids, and it's like, ah, we have a sphere. And so just like flying into Vegas just gets your juices flowing, like it's all fired up, because like this thing is like amazing. And by the way, I'm just talking about the view from the outside. I understand the thing on the inside is also amazing. And so, you know, we clearly can do that, you know, at least in Vegas.

17:57We've been lives now, you know, in London, I think they just gave up on building the sphere. So that's, you know, that's the other side of it. And so, you know, we do have to decide whether we want these things to happen, you know, is a little bit disfiering to see the liquid natural gas decision that just came down. But are the roots of this stasis quite general and quite cultural because parents coddle their children much more their higher rates of mental illness amongst the young young people it seems have less sex along a lot of cultural variables have the percent of old music people listen to compared to new music there seems to be more general stagnation so how would you pinpoint our loss of self -confidence or dynamism where's that coming from yeah well so first of all to be clear we're very much in favor of young people not dating because is that's very distracting from their work at our startups.

18:42So that works out fine and fortunately, in our industry, we have a long experience without having to have a dating life when we're young. So that works out well. So it's not all bad. So it is really interesting. I mean, it looks Silicon Valley has all kinds of problems and we're kind of a case study for a lot of the, I mean, look, it's not like you can build anything in Silicon Valley, right? So we've got, I mean, our politicians absolutely hate us and they don't let us do anything if they can avoid it. So we have our issues. The view from the valley is a lot of kids are being brought up and trained to basically adopt a fundamentally pessimistic or sort of how to put it like stagnation oriented in nerd, like have very low expectations.

19:20Basically a lot of what passed this for education now is teaching people how to complain, which they're very good at. The complaining has reached operatic levels lately. And so there is a lot of that. Having said that, look, I'm also actually really optimistic. And in particular, I'm actually quite optimistic about the new generation coming up. I think Gen Z, and then I think it's Gen Alpha, and then it's, whatever what it rolled is, we're seeing more and more kids that are coming up, and they're being sort of exposed to a full load of basically programming, sort of cultural programming education programming that says you should be depressed about everything, you should be upset about everything, you should have low ambitions, you shouldn't try to do these things.

19:54And they're coming out with sort of a very radical, you know, kind of hard shove in the other direction. And they're coming out with just basically coming up with, you know, tremendous energy, and tremendous enthusiasm to actually do things. And so, which is very natural, right? Because kids rebel. And so if the system is teaching stagnation, then kids, at least some kids, will come up the other way and decide they really want to do things in the world. And so I think entrepreneurs in their 20s now are a lot better than certainly my generation. And they're frankly more aggressive than the generation that preceded them and they're more ambitious.

20:23And so, now, we're not dealing with them. We're dealing with a minority, not a majority, but I think there's quite a bit, yeah. Every hour I get what I can spend at 20 year olds is actually very encouraging. One emotional sense I get from your walk on music, Beethoven's Fifth Symphony, is those that the stakes are remarkably high. Now if we're looking for indicators to keep track of whether in essence things are going your way, greater dynamism, freedom to build, willingness to build, American dynamism, what should we track, what should we look at, how do we know if things are going well? Yeah.