Peter Thiel: AI: Intelligent, Conscious, or Merely Evil? (COSM, 2022)

I'm so thankful to have you here. You know, a great conference like COSM requires you. You're the best part of the conference, the people that you'll meet here, the conversations around the table, the discussions we'll have. COSM is a place for discussion and deliberation about the tech issues, about this convergence of technologies that is transforming our world. Artificial intelligence, 5G, cloud computing, blockchain, crypto, all these technologies are happening at once, and there's a question, what does that mean for the economy? What does it mean for our place geopolitically? And what does it mean for work, the future of work? What does it mean for lots of things? And so we'll be examining each of those things over the course of the day.

A great conference requires great speakers. And we have an incredible lineup. If you haven't already looked at the program, you're here, so I'm going to assume that you have. But I want to thank Peter Thiel in particular, because this is our third COSM Technology Summit. And Peter has helped us kick off COSM, all three of those conferences. And I'm so grateful for his time. He's going to be joining us virtually. He's going to have a back and forth with George Gilder. And then he'll take your questions. I'll point out the microphone for questions is here, and there's a camera over there, so you will, if you want to ask a question, if we have time, you'll come down here, ask the question. Peter will be able to see you, and we'll take questions and answers that way.

It requires a great staff, and this is a staff effort. I'm going to be recognizing them throughout. So grateful to our Discovery Institute staff for helping to put this together. It's not easy. And also our sponsors. And I think our AV team's gonna put up the sponsors. I wanna recognize them quickly. Microsoft from the beginning, Brad Smith has been a great friend to this conference. We're so appreciative. Amazon, BlockCelerate did a pre-conference. It was their own event, but they've been a great partner to us in the run-up to COSM 2022. Smead Capital Management right over here, Cole Smead, look for him later. The M.J. Murdock Charitable Trust. Zevenbergen Capital, Madrona Venture Group, Trilogy International Partners, INRIX, Lucas Creative, Dunn Lumber, Dan and Cindy Mader, and Byron and Joanne Nutley.

Can we give them a round of applause, please? All of those sponsors make this possible along with your involvement. And again, we're so appreciative. And I want to stress again, you're going to hear some things over the next day or two that you agree with. You're going to hear some things you disagree with. And that's what makes a conference fun. As George Gilder likes to say, a conference where everyone agrees is a boring conference. And COSM is not a boring conference. So we're so excited to entertain you over the next couple of days. Lastly, I want to introduce Matt McIlwain. Matt is the managing director of Madrona Venture Group, the leading VC firm in the Pacific Northwest. He is the chair of this year's COSM Technology Summit, and I'm so grateful to Matt.

Matt has really played a crucial role in succeeding Tom Alberg, who I know that Matt is going to talk about a bit. But I'm so grateful to Matt. So please join me in welcoming Matt McIlwain.

Well, you're right about these lights. They're quite bright, see? Well, welcome, everybody. Thank you so much for being here at the conference. I'm absolutely excited to hear all the different discussions throughout the next few days, and I'm incredibly honored to step into the shoes of my longtime colleague, great friend, and mentor, Tom Alberg. Many of you in the room know Tom, and there's some really fun ties to Tom and George and Bruce Chapman and others, but those three in particular, who go all the way back to their days at a little college out in Cambridge, Massachusetts. And I think some of the early ideation that led to the ideas and the desire to bring people together and to have conversations that all too often have been lost in different parts of our society, including in some places in the educational system.

Tom was this unique mix of humility and ideation. He was always thinking about, what can you start, what can you build? Things as diverse as building up Perkins Coie law firm here in Seattle, helping to build McCaw Cellular into AT&T Wireless, moving on from that into founding Madrona, our firm that's an early-stage venture capital firm, and leading the first outside investment round of Amazon, where he then went on to serve on the board for 23 years, and all the kinds of innovation that came from that. But he loved to talk about ideas. He loved to debate things, many of the topics that we're going to cover. And so it's just a real honor that we could think of him. He passed away a couple of months ago and was just an absolutely amazing person and inspiration and champion of this conference.

And one of the things he'd asked me to do was to step into this role for this season. And the reason that I love that opportunity is because of this embracing of diversity of thought. Steve just referenced it. It's so important in this day and age to have people have truly curious minds. What does it mean to have a curious mind? Well, curiosity, by definition, is an act of humility. It means that you don't know. You don't know for sure about something, and you're willing to explore alternative perspectives. You're willing to explore the facts. You're willing to explore the opinions. And through that diversity of thought, get to better ideas, better understanding of what is true and what could be true in the future.

And in an era where we live in the combination of a real world and a kind of a spiritual dimension that we all want to try to understand better, and then increasingly a synthetic and virtual world, it's harder and harder and harder to know—let's say it another way—one person's misinformation or disinformation is another person's truth. So let's dive into all these different topics over the course of the next couple of days. Explore the facts, listen to the opinions, be respectfully open to the different ideas, and see what we can all learn together. Finally, I'm especially excited about this time of what's going on in a broad variety of areas. I'm going to have an opportunity to host a panel tomorrow on artificial intelligence.

And there's many debates and topics around artificial intelligence, from, 'What is artificial general intelligence?' I'm guessing that Peter, when he comes on here in a minute or two, might reference this topic. But is there such a thing? Can AI be sentient or not? And what's even more interesting about this time, not only at that philosophical level, but at the level of what we would call intelligent applications. We all live with intelligent applications every day in our lives, whether it's with a search engine, whether it's a recommender system on something like Amazon or Spotify or Netflix. Intelligent applications are pervasive. But there's a whole new wave of intelligent applications that are coming.

We'll unpack this tomorrow in the session. Those are what we refer to as generative applications. And the generative applications, unlike intelligent applications, are built on something called foundation models, things like GPT-3 and DALL-E and Stable Diffusion. And so that is a very interesting area where I might have a point of view that the human's always gonna be in the loop, you might have a point of view that the human's gonna be written out of that loop over time, but let's have those kinds of discussions because AI, generative apps, intelligent apps, is a very important part of every aspect of life going forward in the future. Peter, this is Matt McIlwain. It's great to see you again.

Let me just take a second to introduce you. I was just talking about generative apps and foundation models and speculating that you might have a thing or two to say about that in this, maybe more broadly in your talk here. Peter, of course, was the co-founder of PayPal, co-founder of Palantir, still serves on the board there. And even beyond that, has been a very active investor both individually as well as through Founders Fund, which is the fund that he, again, created several years back. Companies that you may have heard of like Airbnb and LinkedIn and so many others that he's been involved with, again, either directly and or through Founders Fund. He's also done a couple of other things that I have found personally very inspirational.

One of those being the Thiel Fellows, which is a group of individuals that have been encouraged to go off and pursue their dreams with resources that the Thiel Fellows program has provided. One of them most famously here of late, Dylan. His company, Figma, is being acquired by Adobe for $20 billion, so not too shabby to leave college. He left Brown and pursued his dreams as a Thiel Fellow. That's Dylan Field. Dylan Field. There's a number of things that Peter has done, and he's also taken the time. The book I'm going to mention specifically has been quite inspirational to me, is Zero to One. It's a lot harder to go from zero to one than to go from one to a hundred or a thousand or a million. And so I'm sure he's gonna touch on some of these different themes and without further ado, Peter Thiel.

Awesome. Thanks so much for having me. And I thought I would do another AI or anti-AI talk. It was a great conversation I had last year with George Gilder. I think one of the challenges in this field is always to figure out the right, it's not just to get a diversity of views, but it's even just to figure out what the right questions to ask. And so the high-level question I want to ask today is basically, how should we think about AI? Should we think of it as intelligent, conscious, or merely evil? And so the first step of perhaps tackling this question is to tackle another question, which Gilder asked me. It was the first question he asked me last year in the Q&A part. And it was sort of like, why do so many people in Silicon Valley believe in the simulation hypothesis that the entire universe, the cosmos, is just a computer simulation?

Why do they believe something as crazy as this? And I thought about this question some more. There are a few answers. And as I will explain, I think this question is actually somehow entangled in an interesting way with the question about AI, intelligent, conscious, or merely evil. Now, one way in which you can question the premise of this question is, as I will explain, I think probably the peak belief in the simulation hypothesis was maybe something like a decade ago, maybe circa 2012 to 2015. And it has probably faded some. You still have Gen Z people say that things are a glitch in the simulation, or there's still some sort of passive reference to it. But it's a little bit less intense. So the question of why we believe it is sort of like, why did it gain so much momentum over the last 20 years?

And then also, why did it lose some steam? So the first very dumb answer to the simulation hypothesis question is that it's just sort of a sociological status game between the computer science and the physics people, and that in the 2000s, 2010s, computer science became the most important field. And so you could sort of say that you were showing that you were more important than the physics people. You got to determine what ultimate reality was, and it wasn't particles and matter and fields, but it was just bits in a computer, and that this is sort of a physics versus computer science type dynamic and reflected something about that. But if we want to give sort of the more fundamental answer on cosmology, is that something had gone very haywire in physics, that you had sort of

You had sort of the multiverse, where sort of a lot of the Big Bang inflationary cosmology had gradually gone into this infinite multiverse where basically anything goes. And I think that probably on some level, if you want to critique the multiverse as a theory of science, it was one where the physicists couldn't think through enough questions like where, um, you know, the, the being lost in all these infinities is tricky. If you cannot do induction, you know, are you still doing science? Is it a universe that's too big for science? And, uh, and, and that's why I think you should often think of the multiverse as a gateway drug to all these, uh, these very different things. And once you have the multiverse, you can also have the Matrix or a simulation. You can have Boltzmann brains. You can have, you know,

Um, you can have all kinds of, uh, strange possibilities for the nature of the universe. And so that, uh, you know, even if we say the simulation hypothesis is kind of crazy on some level, uh, is it really crazier than, than the multiverse? And sort of link—now, there's a third answer I'm going to give to why the simulation universe, uh, theory gained so much traction, but that will take a little bit more time to develop, um. Now, but as part of this, we should look back on the last 17, 20 years of AI, AGI. And if we went back 20 years in time, it was super optimistic in all these ways, where it's going to be cornucopian. 2005 was the year Kurzweil wrote *The Singularity Is Near*—is basically you're going to have accelerating technological progress, runaway technological progress.

In some ways, it felt like you didn't have to do much. So it was somewhat passive for the humans. You just had to sit back, eat some popcorn, watch the movie of the future unfold. And then there were certainly all these versions where there was going to be so much growth, we would need basic income as a safety net because there'd be no more labor market. You'd have incredible discoveries in all these fields. There was certainly some latent question about how you'd have to make sure that AGI was friendly. Of the narrative in the 2000s, early 2010s was in this sort of, you know, optimistic, utopian, cornucopian direction of, of, um, what AI would, would, uh, would look like. But now if we sort of look at what's actually happening with AI, not where it's going.

I'm not going to speculate on AGI or even where it's going in the near future. But if you look at what is actually happening with AI, we get something like this guy. This is a TikTok video. And one of my colleagues went through, constructed a whole series of profiles on TikTok where it's an AI optimization engine that feeds you content that you want to see. That's supposedly what it does. And basically, he came up with all these different profiles, Midwestern housewife, all sorts of people with different kinds of interests. But part of what the algorithm fed you was this fairly deranged content. And this particular person—I was going to play the video, but we decided there were too many naughty words in it that I might get in trouble for.

But he basically explains how, um, he exploits employment law, bounces from job to job, sues all his employers for health and safety violations, and it's sort of designed to make, you know, it's designed to sort of polarize race relations in the U.S. and make, you know, Black people angry because if you have an honest job, you're a sucker. It's designed to make white people angry. And there's sort of other versions like this that come up throughout TikTok. This is sort of what I would say, front and center, what cutting-edge AI looks like in the U.S. today in 2022. And of course, we don't really know what the full intentionality is. We don't know whether this is just sort of an emergent property, that it's going to take what people want, push it to extremes, derange it, derange the discourse, so you'll have

you'll intensify the race contradictions, you'll intensify the economic contradictions, how the U.S. has the worst healthcare system in the world. You'll have someone else talking about, you know, how you should call pedophiles a MAP person, a minor-attracted person, because we shouldn't discriminate against those people. So it's all sort of like designed to, you know, intensify the wokeness, anti-wokeness, polarize and derange our society in one way or another. Um, you can view it as emergent, you can view it as full-on intentional, where you should think of TikTok as a sort of Chinese communist weapon that is, um, being used to derange our, our society. That, you know, the, the book that I, I think is interesting is the one Wang Huning, who's the number four guy in the, in the entire communist government. He's sort of this, the, the professor

theorist of Xi Jinping Thought. And he wrote this book 32 years ago, America versus America. And it's basically a roadmap for how to derange our society by sort of heightening these sort of Hegelian contradictions. And I would submit that if, if you go with the full intentional version, TikTok is basically a weapon that's designed to derange us through decentralized and heightened contradictions. And of course, this particular AI is up against another self-destructive communist Chinese AI, which is the centralized one that's being imposed on China itself, where basically you have perfect face recognition. Everybody knows they're being monitored. There is sort of an important way in which China has become North Korea.

Those of us who are anti-communist tend to conflate that it was a communist country 10 years ago and is now. But there is a way in which the AI technology, the surveillance technology, has really, really transformed it. It is, again, not AGI. In many versions, it's barely AI at all. It's just cameras, ways to do big data on this. This is what the Kai-Fu Lee book talks about, too, that China will win at AI through these big data, uh, algorithms. It's not about the sort of cutting-edge, futuristic stuff that people in Silicon Valley talk about, things about like TikTok or what China has done to themselves. And so you can basically, uh, you know, one way to think of the rivalry between the U.S. and China is that it's, uh, it's, uh,

It's sort of a question which society will be destroyed faster by the somewhat dystopian AI that's being imposed on it. And we could have sort of a long debate about that. So this is sort of the framing of where AI actually is, where it actually is being implemented in the most powerful, dramatic ways today. Let's go back to our three questions about AI. First off, is it intelligent? On the first two questions, I'm going to sort of say they're above my pay grade, but it seems to set a low bar for intelligence. And the rhetorical point I would make is that it's often just a filler word when, you know, we're talking about something quite different. Um, there was a 2016 Obama administration study about the transformative importance of AI entitled, quote, the title of the whole paper, "The National Artificial Intelligence Research and Development Strategic Plan."

And, um, and basically, if you went through this paper and if you replaced every use of the word of AI with software or even just computers, the meaning wouldn't change at all. And, um, and I think this is sort of a tell that, uh, you know, that maybe the first approximation when you hear AI, you should just think software or computers. It's, um, you—AI is probably not intelligent. AGI, that's somewhere in the future. Don't know. In a similar way, I would say the question of whether it's conscious is probably hard to say. My strong suspicion is that it's not. You, of course, have the epistemological problems. Thomas Nagel, what is it like to be a bat? Searle's Chinese room problem. And then, of course, and so it's hard to say.

And for Mr. Lemoine, who I think is talking later, it was literally hard to say that AI might be conscious. And so as a contrarian, I'm always a little bit biased to say that things that you're not allowed to say might be true. So I don't want to dismiss the possibility entirely that it's conscious. But it's probably the wrong sort of question on, on some level, for us to be asking. The question whether it's intelligent, whether it's conscious, are just the wrong questions. The, um, you know, I always go back to, uh, what I don't like even about Descartes, where if you think about Cartesian dualism as the, the origins of, um, of the problem of consciousness, um, uh, the way, um, the way consciousness worked for Descartes was that it was meant to be, um, you know, a smart person in the 17th century was supposed to become a priest and use his brains to think about God, and Descartes came up with this very mysterious, different thing called the mind, and it was sort of an attention redirection distraction mechanism. And, uh, and I always think we should, we should, uh, um—

Remember the 17th century context where consciousness was not something that was mystical or spiritual or dualist in sort of the way we might think of these categories in the 21st century, but it was meant to be anti-theological. But maybe more generally, the problems of consciousness or even of intelligence are somehow the wrong question. Go to my third one: Is AI evil? And this one seems more straightforwardly answerable. Certainly, I think that the TikTok algorithm is evil. I think what China is doing to itself is clearly evil. You can talk about evil in all kinds of different versions. There's, of course, the kind of disembodied brain in C.S. Lewis's book, That Hideous Strength, who turns out to be a demon.

I'm not sure it's literally demonic in quite that sort of a way, although certainly I don't think that we've had an exorcist at Google to check that out and make that determination. So I think even that possibility couldn't quite be ruled out here. But it's evil. The creepy-looking woman on the upper right is—this is this image, Loab, who seems to be sort of a strange attractor that comes up in a number of the art projects that DALL-E, the AI art program, has generated. And it's sort of, if you do, if you ask what is the opposite of Marlon Brando, you get this somewhat abstract painting. And then if you ask what's the double negative of that, this sort of creepy woman, Loab, uh, emerges, um, and, uh, um, and, uh, and, and what's, what's, what's sort of interesting, that she emerged on sort of a number of things for sort of this strange attractor, and you can, so you can think of it as, uh, maybe it is a kind of occult knowledge where we're, we're learning something. Did we really need to know that the double negative of Marlon Brando was a witch-like woman?

And something like that. But of course, maybe the closest analog to sort of a demon is an idol, a pagan god, where we worship the god. Um, the god seems to tell us what to do, um, and it's unclear if it's actually telling us these things or if it's just somehow some kind of psychosocial effect that's, uh, uh, creating some kind of mass hallucination, um, and, um, and, and that leads, leads to this. And this is sort of, uh, this is sort of where I've, I've suggested that, you know, maybe you should think of the European Union as the closest thing we have to functioning AI in government, in a way, where the goal is just to prevent human thought. It tells us very basic, simple things we should do. But it functions these ways.

But now let's come back. If we say that there is a lot in AI that is straightforwardly evil, that is merely evil, that is simply about stopping humans from thinking, from using their capacities and things like this. Let me use this to come back to the very big cosmological question about the simulation. And so let me give an alternate sort of explanation of why the simulation hypothesis gained so much traction in the 2000s and early 2010s. And it's something like this. As we were building AI, as we were building towards AGI, it seemed potentially dangerous. AGI, in the full utopian sense, was going to be this superhuman mind. Could we really be confident that it was going to be aligned with human beings, that it was not going to be aligned?

There seemed to be a lot of risk in that. And I think that a lot of those of us who are skeptical of AI or skeptical of AGI often underestimate how troubling the alignment arguments are, how it's not straightforward. If you can have such a thing as AGI, it's not at all straightforward to get the AGI to be friendly. If you have a Darwinian view of the world or just a Machiavellian view of the world where you could say the core axiom is that there is no such thing as a selfless being, a purely selfless being, and therefore the alignment problem is fundamentally difficult to solve. So how do you get to friendly AI? Not, um, not super straightforward, um, but if we say, um, and maybe, maybe AGI ends up being a kind of Great Filter where, um, you know, if you get it wrong, it will, um, it will destroy the world. And this is where, you know, this is where there seemed to be a very big difference between the multiverse and the simulation theories, because in the multiverse

The AGI is simply in the future. And whatever Great Filter the AGI represents, whatever threat it represents to all of humanity, wipe out all of humanity, it's in the future and seems quite dangerous. Whereas if you have a simulation theory, you also have this cosmic AI or cosmic AGI that created our universe. And in some sense, there was a Great Filter in the past. And so there is a way that perhaps you could think that, that the cosmic AI isn't entirely hostile since we're here having this conversation. Perhaps the cosmic AI is guiding the development of the AGI and we can infer that it will sort of be—I think this is the way you have to think of the simulation theory. It was in the context of a lot of these concerns about friendly versus unfriendly AI, and it shifted the problem from the future where it is in the multiverse to the past and seemed to solve it.

Now, the thing that is very different from when Kurzweil was writing about this in 2005 is we've seen some of the progress, and somehow this illusion has become very hard to maintain. And we have the sort of AI, the thing that perhaps is tracking towards, um, you know, autonomous weapons systems, cyber warfare, um, uh, you know, a runaway AGI, it doesn't seem very good for humans. And this is both in its cutting-edge centralized and decentralized forms. And as a result, I would say we've been sort of inclined to flip the causation, that the emergent AI or emergent AGI is what's telling us something about the AI that built the universe. And if it's the nature of the emergent AI to be fundamentally or merely evil, then perhaps we should not be so assured that the cosmic AI that created the simulation was fundamentally good.

And we should extrapolate from one AI to the other and assume that it's also self-interested, not aligned with humans, not fundamentally beneficial to the human world. And this is why I think the simulation theory maybe was a fake way to solve this problem, but it's not at all working anymore. I think one way to think about this is there's sort of all these, all these kinds of debates about the meaning and nature of AI that map onto these theological controversies from the Middle Ages. And it's always sort of interesting to try to map it onto these past theological debates as a way to just sort of understand the nature of the argument. And you can think of, you can think of the cosmic AI as sort of analogous to a form of strict monotheism, like Judaism or Islam, where it is the oneness of God.

And then the problem with extreme monotheism is that you cannot speculate on the attributes of God. You ultimately do not know much about the nature of God. To have a science of God, you need a plurality. If you have too many gods, of course, they're probably not gods. If you think of Christ and Trinitarian Christianity as telling us something about the nature of God, you go from the God in history to tell us about the God outside of history. And I think there's roughly a similar move that's happened with the emergent AI, which is the sort of, you know, the idol, the demon idol, whatever you want to call it, that's emerging in history, that it is telling us that if there was some demiurge or something like that that built the simulation, we should also infer that it's not that well-aligned.

And so this is sort of where we are seemingly at these dead ends with the progress of AI. I'm not going to solve this problem today, but it seems to me that surrendering control to AI, blindly worshiping AI, the emergent AI, letting it dominate and control our societies leads to one of two catastrophic outcomes. It's sort of decentralized runaway violence, which is the derangement of TikTok, America versus America, and then centralized totalitarian one-world state, worse than North Korea or China. And that the challenge for us is to find some kind of a third path where we make progress in areas other than AI. We find a way to get back to the future. The runaway apocalyptic violence, the centralized totalitarian one-world state, they are seemingly exclusive possibilities.

I don't think they're exhaustive. I think there should be a third way, and the challenge is for us to, to find a way to build it. Uh, you know, I'm not a fan of B. F. Skinner, the behaviorist psychologist, but the quote, as I always like to cite, is, you know, 'The real problem is not whether machines think, but whether men do.' And, um, we need to get back to thinking ourselves, uh, and regain control of our future. Thank you very much.

Peter, thank you so much for that stimulating and philosophical and prophetic oration. It was really worthy of Zero to One. This conference is going to be—one of our prime themes is the development of superabundance. This is really being demonstrated by this new book by Marian Tupy and Gale Pooley that really documents on the basis of time prices that abundance is steadily increasing at an accelerating pace. And this conflicts in important ways with your vision, Peter, that you have previously expressed, that in some way technology is becoming less fruitful and less creative and less responsive to real human needs. So I wondered whether you can—what sort of insights you can have that transcends this apparent conflict between the idea that science is growing stagnant, technology is becoming sterile, and the demonstration of this magnificent new book that

That poverty is being overcome everywhere, that the price of commodities is plummeting, that everything's becoming more abundant, that science actually is offering new bounties every day. That would be my...

well i i don't i don't agree with the book on any level so uh so it's um okay but i think look i think i think i think even something as basic as commodity prices are you know at um you know there was a hundred year decline trend in the 20th century and then you know um and if they go much higher than they are now it's like uh that we've had a 20-year bull market and it will go up in a way that suggests the whole decline trend is broken. And then there are all these different reasons you can do this. We have certainly the macroeconomic version is always to look at real inflation, inflation versus real you know, wages, are people's wages going up faster than inflation? And the felt sense is that that's not happening.

And then you can make, you know, and then, you know, the superabundance argument is somehow that the government is understating the inflation and that there's less inflation than it looks. And, you know, I don't think they're overstating it massively, but, but at the margins, I, I believe the government is—I, I—the superabundance argument is that they're overstating inflation, there's less inflation, there's more real growth. My argument would be, you know, at the margins, they're probably understating inflation, so there's more inflation and actually, you know, even, even less productivity growth. But, but the way—if I had to sort of reconcile these two views, it would be along the lines of the talk I just gave you, which is, let us say that there are some dimensions where there is a reasonably—

rapid amount of progress. There has been, you know, maybe not as much progress in the world of atoms as I would like. There has not been progress on energy, or, you know, we don't have nuclear power plants, we don't have, uh, you know, we haven't had—less progress in futuristic medicines than I would like, but, you know, we had—we've had, you know, a lot of progress around computers and, um, and the internet, the mobile internet, and then, of course, all these things that get loosely categorized under AI. And then you have the macroeconomic question, how much that progress lifts our human society generally. I would say it's less than said. But then I think the other dimension you have to ask is, is it the sort of progress that people think of as simply good?

And the futuristic AGI, it was pitched to me in 2005, is you have no idea what sort of, we'll be able to cure aging. We'll be able to find all these fantastic medical treatments. We still have not gotten those. If we had gotten those, I would score it as at least more positive. What we have gotten, we got TikTok. And yeah, it's valuable for TikTok. It's valuable for the company that sort of does this AGI. But I would score it as a form of technology that is, and I don't want to sound overly Luddite, but even if it's happening rapidly, it's deranging us. It's making us go crazy. Or the surveillance state in China. That is a form of technological progress over a decade ago, but it has really deranged that society.

It has disabled the humans. It's a less happy, less functional, less free place than it was 10 years ago. Even if we say that it somehow shows up in the economic statistics, which it doesn't, but even if you can jigger the econ statistics to make the superabundance show up, we should ask this question: Is it evil or is it good?

So, Peter, before we have some audience questions, I'm going to ask a question, too, as I think about some of these contemporary examples like in the area of science, things like AlphaFold or DeepFold, you know, where we've been able to predict and understand the structure of proteins coming out of the DeepMind team. Or we, you know, and so we have the productivity, you know, Copilot that's been recently launched by Microsoft and the claim by Satya and Charles Lamanna and others there is a 30% improvement in productivity in software development. Or you take Stable Diffusion and the creativity that that unlocks for humans. In every one of those cases, of course, humans in the loop in many, many different ways.

So no AGI here. And I guess the question is, therefore, can AI both be evil and good?

Sure. But, you know, all the examples, you know, like there's always a fast answer: it's just a technology and it just, you know, and as such, it's pretty neutral and it's up to humans what we do with it. But I think one can—I would still say one should ask the question more, you know, is it, you know, you know, how do we sort of weight these different kinds of applications? You know, the protein folding is, um, is interesting. It's, it's, it's only valuable, I would say, if it actually leads to, um, new medical interventions, new cures. And, you know, when I, when I push the, the AI people on that, they always, they don't, they don't want to engage in that conversation. And, um, you know, so, um, and then, um, and then if it's saving 30% of coding time, that would be very impressive.

It's not clear that that's showing up in any of the stats of the big software companies at this point. I would say all of Silicon Valley has this problem where they have to pay the coders, the computer programmers, more and more. And so, yeah, there's definitely a need for technology to replace the people and would make the businesses more profitable. And I would score it as probably a productivity-enhancing, generally positive thing. It doesn't show up in the computer science labor market. And then I think, look, the big thing, the biggest AI technology that's actually being used is TikTok. And we need to be talking about that. And I gave you two very different ones: one is that it's emergent.

It's not like an intentional weapon. It's just emergent. And it's just we have a tendency to get deranged. And this is what happens, even though it seems to derange us. Um, and then, um, and then, but nobody, nobody's in China to double-check the algorithms. The ones they're training on the US are very different from the ones they train on people in China. You don't get, you don't get videos that, um, that make people, you know, radically, radically undermine the belief in the society in China like you do in the US. And that, that, that suggests to me we should at least be asking this intentional question.

Very fair point. And I'm certainly not trying to make the argument that TikTok is a productivity helper. And it could well be insidious and evil as well.

My intuition is that it's the biggest thing in AI. So if you had to weight them, how big they are, I think TikTok is the biggest thing.

We have a microphone here if folks are interested in asking questions to Peter. And if you can look this way towards the cameras, that would be helpful as well.

Thank you, and thank you, Peter, for that most precious of gifts, your time. And I, for one, am willing to pay the time price for it. Given our addiction— Who are you? Announce your— I'm nobody. No, you aren't. My name is Frank Stevens. I went to your first conference.

I know, but you've got to give your name when you speak at the microphone. We don't believe in anonymity around here.

Very good, and you go to the core of the question, which is the individual versus the identity. I stand as an individual to ask you this question. Given our addiction to narratives like AI, and given our aversion to knowledge in favor of all kinds of misinformation, and given our emotional overload of the past few years, as a student of René Girard, my question is, will we have a mimetic pandemic where violence is the vengeance, and could AI help or hinder it?

There's one for you. This is certainly a read on what is very haywire about TikTok, that it just gives people what they mistakenly think they want. And then that's often somehow a bad thing. I don't know. Look, I think there are ways in which I wouldn't—I'm hesitant to sort of blame tech for everything that's wrong in our society. But I do think there's something about, you know, and it's wrong to scapegoat it and say it's the single thing that causes everything to go haywire. But at the margins, you know, is it helping us? There's something about a lot of these sort of short packet content forms that has deranged discourse. I think TikTok is by far the worst. And this is sort of a sense in which Silicon Valley is getting some of the blame for all this stuff.

And then I think the alternatives also didn't work because the alternatives were a centralized media system, in which everything was controlled. So I think you have to frame the social problem. It's too much totalitarian centralization versus deranged decentralization. And you have to think of both problems. And there are instances of both that we had in the COVID epidemic the last two, three years. We had, you know, decentralized conspiracy theories that were not helpful, and we had centralized narratives that cut off, uh, cut off much-needed debate. Hi, thank you. Uh, I'm Dr. Jeff Gunnason from Anchorage, Alaska. This is my second COSM, um...

Sixty years ago, C.S. Lewis wrote an article in the Saturday Evening Post called "Screwtape Proposes a Toast," where he criticized the dumbing down of American education. And I chair an educational foundation, and I'm very concerned about the quality of U.S. education, as you talked about, and what can we do about educating our youth, at least in public education. All my colleagues send their kids to private schools now. And so I just wonder if you could comment on that.

You know, I mean, I think there are a lot of people who articulated the issues quite, quite, quite strongly. I don't have, you know, I don't have, I know I've got much to add to it. I, I do think I do think it's sort of like always a question, you know, what are people doing? What is the teleology of it? And, you know, I think a healthy, you know, Primary, secondary, tertiary education system is, you know, it's supposed to make you a well-rounded, educated person, become a functioning citizen in our society. You know, if you go into research or academia or certain industries, it is for you to become, be a creative person who sort of pushes the frontiers of knowledge. And in some sense, it has somehow gotten deranged.

And again, I don't want to blame it on AI or make this the single focus, but if you have a narrative out there that in the future, AI will do all the thinking for you, and you don't need to think for yourself in any way. You know, maybe, you know, you don't need to learn as much stuff. You don't need to memorize things. You don't—there's sort of all this, these things would be considered education that seem to be much less important. And I, yeah, and I do wonder, that's sort of, that's the broader context in which a lot of these things are operating. It's sort of like, you know, if you, if you say the public school system is not completely deranged, it's just trying to sort of make people these passive cogs in this very large machine, that's sort of the way in which it fits into this dystopian AI narrative.

And we need to tackle this problem on all these different levels.

Hi, Peter. My name is Rick with Mivium. I heard you give a talk at Stanford a few months back. And so this is the second time I heard you mention a lack of progress on the atom level. So coming from the semiconductor material science field, I wanted to see what you mean by that because all we do is manipulate atoms on a day-to-day basis at scale from atomic layer depositions, molecular beam epitaxial growth to mechanochemistry. That's the only way we can make these next-gen semiconductors, these new materials that can basically replace silicon and free us from any kind of dependence on China and raw materials. So for me, in my field, it's not a lack of progress. It's more the incumbent thought, school of thought out there is all chemistry-based.

Everybody wants to deal with chemistry and toxic chemicals and things like that, whereas we're trying to do it a different way, but we face a lot of resistance. The barrier of entry and the cost is also very high. So I wanted to know what you mean by lack of progress.

Well, it's certainly slowed in a lot of fields. Look, I think the complicated version I would tell is that we've had a decent amount of progress in the world of bits the last 40, 50 years. I think semiconductors are sort of the in-between thing. And then most other fields have been disappointingly slow. When I was an undergraduate at Stanford, I was Class of '89. You know, in retrospect, it wasn't obvious at the time, but in retrospect, you were supposed to study computer science. All the engineering fields were bad fields to go into. It was a bad idea to go into Aero/Astro. It was a bad idea to go into nuclear engineering. People knew that. They didn't do that. It was a bad idea in mechanical, chemical.

You know, all the sort of world of atom stuff was bad. I think the one that was still in between that worked okay, but much less well than computer science, was electrical engineering. And so if we look at how well have the companies done, how well have the people gotten paid, EE has done better than other fields, but a lot less well than computer science. I think on some level it reflects the ways in which the progress is hard. It's slower than it was in the '80s and '90s, but by various measures. It requires enormous scale. So the role for the individual is less. You have to have a $500 million ASML, you know, machine to do the lithography. You know, it's harder to start—you know, it's hard to start a new semiconductor company.

And so, yeah, the ecosystem has shifted towards, you know, much bigger businesses dominating it. And, you know, certainly as a venture capitalist, we've done virtually no investing in semiconductors. I think I should, but we do so little that we don't know enough about it to do it. And that makes me think that there's some things happening, but it's still a lot slower than it was in the past. This is the challenge with China. China is copying the West. They will eventually catch us up. And so if we are progressing slowly, they will eventually be able to copy things and converge. And I think the sort of rate at which we're doing new things is not so great that China will not be able to converge.

Yeah, but China has no access to the ASML machines.

They're still doing the chemical method. It's possible that it's still going to be non-trivial for them to catch up. But it's not as though we have this exponentially growing lead. It's not quite as strong as that.

Thank you, Peter, for your provocative remarks that are going to precipitate many discussions for the next two days, including about all these issues surrounding technological progress. And thank you. Thank you.