Loading summary
Transcript136 lines
- [00:00]
Steve Hayes
This episode of the Dispatch podcast is brought to you by Pacific Legal Foundation. Since they were founded in 1973, PLF has won 18 Supreme Court cases defending the rights of ordinary Americans from government overreach nationwide, including landmark environmental law cases like Sackett vs EPA. Now PLF is doubling down and launching a new environment and natural resources practice. They're on a mission to make more of America's land and resources available for productive use and to make sure freedom drives our environmental and natural resource policy, not fear. To learn more, visit pacificlegal.org flagship hi.
- [00:40]
Andy Mills
I'm here to pick up my son, Milo. There's no Milo here. Who picked up my son from school.
- [00:45]
Steve Hayes
Streaming only on Peacock.
- [00:50]
Andy Mills
I'm going to need the name of everyone that could have a connection. You don't understand. It was just the five of us. So this was all planned. What are you going to do? I will do whatever it takes to get my son back.
- [01:01]
Steve Hayes
I honestly didn't see this coming.
- [01:02]
Andy Mills
These nice people killing each other.
- [01:05]
Steve Hayes
All her fault. A new series, streaming now only on Peacock. Andy Mills, reporter at the center of a new podcast series called the Last Invention. Andy, welcome.
- [01:27]
Andy Mills
Thank you, Steve. It's a pleasure to talk to you. A big fan of you and the work you guys are doing at the disco.
- [01:33]
Steve Hayes
Well, likewise. And I'm very interested in letting our listeners know a little bit more about Longview. We'll get to that at the end, but I want to start at the beginning. What is this series about? And why did you call it the Last Invention?
- [01:50]
Andy Mills
Well, what started off as a curiosity about a debate happening inside of the world of artificial intelligence grew over months into this, like, very fascinating backstory to this moment that we're living through right now, where I know that politics is interesting and I know that it feels as if the state of the world is, in many ways, let's just say, like, dynamic in this moment.
- [02:17]
Steve Hayes
Dynamic is the most charitable way to describe it.
- [02:21]
Andy Mills
Well, it turns out that, like, the details of the AI story and the fact that we are where we are right now, it's so rich that I thought, okay, let's do a big eight part series. Let's deep dive into all of the different fascinating aspects of it, in part because it's one of those stories that, although fascinating in its detail and fascinating on the surface, it's also pregnant with all of these other things, pregnant with all these other themes and ideas and questions and, I guess, tensions that we're grappling with in our world now. And so it's on the One hand, it's just an investigation into what's happening with artificial intelligence, especially when it comes to the concerns that we may be about to walk into a truly transformative moment in human history, and people don't know whether they should be pumped or terrified of that moment. And then, on the other hand, it's also a historical story, a way for us to pull back from all of the content and all of the things that are happening right in front of our face and say, like, okay, let's see this in historical context and see what comes of it, what might be revealed by pulling back and looking at it with a bit more perspective. So, yeah, that's in a nutshell, equal.
- [03:43]
Steve Hayes
Parts history, philosophy in sort of existential way, you know, contemporaneous reporting. I found it absolutely fascinating. I was hooked pretty early. You start the conversation by taking listeners through, really, a conspiracy theory. And I will say, when I first heard you introduce the topic that way, I was immediately deeply skeptical, like, oh, great, here we go. What's the conspiracy theory? But the conspiracy theory turns out to be pretty interesting. And while it certainly wasn't all true, there are elements of truth to it. Maybe you could give us just a glimpse into what that conspiracy theory is and why it led you to the kind of reporting that you did.
- [04:34]
Andy Mills
Yeah. So a lot of ways, this started off with a tip I got from a former Silicon Valley executive who reached out to me. He was a fan of some of my previous work. He trusted me and said that he was sitting on a bombshell, that he had information that a group of people in Silicon Valley were attempting to overthrow the American government and instill a artificial intelligence government that they own in its place. And that doge, which was big at the time, this was like, all the headlines were about what Elon Musk and his team of people were doing at doge. He was saying, that's stage one. Firing these bureaucrats trying to decrease the human workforce in Washington is phase one. Now they're going to start rolling in all of these artificial intelligences to take their job. But the ultimate goal years down the road is for artificial intelligence to supplant American democracy. And he's a legitimate guy. He has. He really is connected in this world. And while it did seem a little bit outlandish, it's the kind of story you would really kick yourself on if you just said, whatever.
- [05:48]
Steve Hayes
Yeah. If it ended up being true, and you were like, nah, not interested. Yeah, probably. But.
- [05:54]
Andy Mills
But it was.
- [05:54]
Steve Hayes
I mean, so just in the beginning of that, as. As you sort of rolled that out. My initial instinct was eye roll, like really come on. But then you did some reporting, right?
- [06:07]
Andy Mills
Well, two things came up pretty quickly when I did the reporting. One of them is that there were certain details and there were certain confident claims that the source were making that I just. Not only could it not confirm, but some of them just fell through, which happens, as you know, Steve, with sources a lot. But the other one was that it actually is true that the thing that the AI companies are making, when they say that they're making AI really is something more like a new species than it is like a new product. That the chatbots that we're interfacing with, they are not the AI. They're more like the website on the Internet. And the Internet in this case would be the AI. Right. And that automating new species, that brilliant superintelligence, they do think that it will eventually do everything we do we currently think of as work. And that will. That obviously will include the bureaucratic state. And we didn't get super into this in episode one, but just for. Because you seem to be interested in this, and I can see why the dispatch folks might be interested in it as well. I also immediately became intrigued because there's a part of me that thought, wow, that might be nice. I can see why the bureaucratic state has a bad reputation right now. And when it comes to how do we allocate the budget to make sure that we can achieve all of these goals, we've told our. The people we represent, we're going to try how much goes to bridges and how much goes to this. Even just the AI systems as they exist today might be more adequate and might be more trustworthy in some ways than the current human beings with all their flaws and incentives that have been poisoning our politics for so long. So I, from the beginning thought like, oh, these, you know, people who have this somewhat spooky and it's obviously freaking my source out, plans to maybe replace this all with AI. They actually. I can see where they're coming from and I could see that there's. There's something admirable happening here. And so I started just poking around and then it wasn't long before I started to realize that the people who are closest to the development of artificial intelligence, the people who are working inside of these research facilities, who are working at these labs, they are engaged in a debate that sounds like science fiction and other. Like, if it was not for the progress that they've been making, if it was not for the fact that they've gotten so much investment, I could see why people would just, you know, shrug and carry on. But the fact that the debate that they're having right now is so important to them and largely absent from our public discourse is something that I feel like is a basic tool of journalism. Like, whether or not you agree with them that they're going to be able to pull off this AGI, as they call it, Right. Artificial general intelligence. We can get more into that in a bit, if you want. Whether you believe that they're actually going to be able to pull this off or not. At this point, they believe that they're going to do it. They have gotten billions and billions of dollars of investment from all over the world, from governments, from private citizens believing that they're going to do it. And they're having a debate. They're already moving on to the debate called what is this going to do to the human race? What is this going to do to the world? And I think even when I first started reporting on this, I would talk to my fellow journalists about, like, hey, some of these guys are saying that this could literally lead to the extinction of the human race. And some of them are the very. Who helped to create it. We can get more into that later if you want. And, like, what do you guys think? I think about putting this podcast series together and doing this big investigation, and some of them were like, don't do it. Don't do it. Like, it. It's too weird. Everyone will think you're kooky. It's almost like doing a story about alien abductions or something. It's like, it's just one of those things that we. That, like, it's too soon. And yet it feels now like as the series is coming out, even in the, like, six months since those conversations, things have changed, and people are really responding, I think, with curiosity and an open mind to what's happening here. So, yes, it started off with a bit of a conspiracy, Followed it saw some of it wasn't true. So some of it maybe was more true even than the conspiracy theorists realized. And over the past several months, I think that we as a society are coming more and more to a point of wanting to take seriously what's happening, what debates are going on, what conversations are happening inside of these AI labs and inside of, in some ways, the most powerful rooms in the world.
- [10:55]
Steve Hayes
Yeah, I mean, I think it's the case that there's a lot more to, you know, certain elements of what, you know, what the sort of broad case that was laid out for you. I mean, there are experiments taking place now sort of in real practical terms within government, about how things can be made more efficient separate, apart from what happened with doge. But how can, you know, how can the Department of Health and Human Services use AI to make the distribution of funds more efficient, for instance? I mean, those things are happening both inside government and at the university level. A lot of work being done on that. So we know that some of this is happening. I think the question that you attempted to to get at in your reporting was, is it happening sort of at the scale that your source suggested? And are the potential outcomes as ominous as the source might have suggested? I think it's a good time you mentioned AGI and this was very helpful for me as I was listening to the podcast, to sort of define the terms, which you did very well. Maybe let's go through what is AI, what is AGI, what is asi, and sort of what are the implications of each.
- [12:13]
Andy Mills
So this is, you know, the easiest, but maybe not the. There may be some PhDs who quibble with this, but just so we can all get on the same page, I think it's best to think like AI is just about automation. So the, like, the basic level I like to think of is the YouTube algorithm that might be showing the video that you and I are recording right now to someone who doesn't know who we are but is interested in AI. That algorithm is automating the insane amounts of videos on YouTube in a attempt to show people what they want to see. Right? That's an AI. And the algorithms of modern social media have in large part been the AIs that we've been interacting with. But when it comes to this moment where these AI companies, there's like, we're developing AI, we're building AI, what their goal is, the thing that they talk about, the thing, the benchmark that they're chasing after. It's something that we often now call AGI, artificial General intelligence. And that is different. That is when you have an AI system that is as intelligent and as capable as a very smart human being. And it can be a little bit tricky to imagine in our heads, especially because we have all these sci fi movies that we've seen and we keep thinking of the robot and we keep thinking of Terminator. It's like, get rid of that for a second and just think about a automated system like the Internet in some ways that can learn and perform any task that a really smart human being could learn. And perform. And you could see why they're so excited about the capabilities of that. Because the same way that you can hire somebody out of college who did really well and train them in a few years to be really good at a number of different jobs, you can just train this AI system to do that kind of what we think of traditionally as like managerial or white collar work. But then the second order aspiration is that we are building all of these humanoid robots, and some of them not quite so humanoid, but all these different humanoid robots, that the plan is to put those into the blue collar jobs. And essentially the AGI the system will be. Sounds kind of crazy, but sort of like WI fi gets put into our computers. The AIs, through Wi Fi will be able to be the brain inside of these robots. And so this is why they think there literally won't be any job that a human can do that the AI can't do. And this is like a version of AGI has been the dream since the 1940s of this industry. In the 1960s, they really thought maybe they were going to get close to it. And then they failed to live up to their own hype and they went into a really long AI winter throughout most of the 20th century. And even if you just go back to 2014 in Silicon Valley at places like Google, the idea that any company right now would be so bold and crazy as to actually try and build that was laughable. Was embarrassing. In the span of just 10, 11 years, we're at a place where not only do they think that we could do it in our lifetimes, they think this thing might be here in the next five years.
- [15:53]
Steve Hayes
Yeah, imminent. Right, right.
- [15:55]
Andy Mills
And that's why we're seeing this insane amount of investment in Nvidia and in all these companies is because they have been able to persuade a lot of people that they can pull this off. Now, I think we can quibble on the five years thing. We can get into that later. But what an amazing moment in technology now when it gets to the next level. What's often called ASI or artificial superintelligence. There's still skepticism, there's more skepticism about whether or not we're going to hit that in our lifetime. A artificial superintelligence is an AI system that's not just as smart and capable as a very smart person, but is as smart and capable as human civilization. So, no matter. One of the ways we say it in the podcast is that you can imagine if you had a team of Einsteins working for you Steve at the Dispatch. That would be awesome. You know, think of all the things that they could do. And Peters.
- [16:52]
Steve Hayes
I mean, we're close. We're close.
- [16:53]
Andy Mills
Well, and these Einsteins, though, they don't need bathroom breaks. They don't take off weekends or holidays. You don't have to feed them or pay them around the clock 24 hours a day. They're working for you. That. You can see how transformative that technology would be, but it still would have a lot of limitations. Artificial superintelligence is when an AI could achieve something that we typically think of as only a country, all working together with all our intricate parts can achieve. So think of it like maybe you could have a smart person dream up the idea of a semiconductor or dream up the iPhone. But think about all of the pieces of civilization that have to work together and all the flawed humans working across the globe and across different sectors to actually build an iPhone and get it to the store and sell it to you. Once that whole operation can be done by a single AI system. That's asi, that's Artificial superintelligence.
- [17:52]
Steve Hayes
The three more terms that we'll define quickly, and then we can get into some of the history. I found the history part of this absolutely fascinating, in part because this is so not my world. I mean, I don't. I don't know this. The. The movies that. That you reference here and in the podcast. I haven't even seen them, including some of the classics. So this is not my world, I found. But what's that? I haven't. Haven't seen it.
- [18:16]
Andy Mills
Ex machina. It's really.
- [18:17]
Steve Hayes
Haven't seen. Really haven't seen the Space Odyssey? No, it's pretty bad. The other three sort of three groups that come up as important in the series are the Accelerationists, the Scouts, and the Doomers. Can you take us through who constitutes each of those groups and what they believe?
- [18:42]
Andy Mills
So these are three different camps, three different paths that people are advocating that we take from this moment that we're in. And the important thing to know is that there are a number of people involved in this, what I call the great AI debate. These three camps are the three camps that have the most influence. And all of them are united by a shared belief that this thing is coming, the AGI is coming. So there are skeptics and there are other people that are involved in it, but most of the people who are working in AI falls somewhere in these three camps. The AI Doomers, as the name suggests, are the people who believe that the first off. They do not think that we're ready for AGI. We as a society have not done great integrating social media. And currently to take this massive step into AGI is to take a step that they believe will have immediate dangers and negative repercussions. But the reason that they're called the AI doomers is that they think that once we hit AGI, once an AI system is that intelligent. And of course they think this in part because of what the AI companies are saying is that one of the first jobs AGI is going to take is the job of the guy working on the next model of AI. And so that intelligent system will be able to copy itself and copy itself again. And now it's the AI building the next smarter, better, faster version of AI. And they think that that will then trigger a, it's often called an intelligence explosion, like a nuclear reactor. And that the intelligence will go from AGI to ASI really fast. Some of them believe it will be hours fast. Others think it's going to be months or maybe a couple of years fast. But still that is transformative and, I don't know, chaotic as us adapting to a world with an AGI is going to be. They're saying we're not even going to be over that hurdle of craziness before we hit asi. And for a number of reasons. They think the most likely outcome of us making the AGI that birthed the ASI is the extinction of the human race. It's like they take it that far, that's the AI doomers they are fighting right now to get us to stop the AI race. Now they don't want us to stop trying to make AIs that could do self driving cars. They're not saying take down the chat bots that are helpful. They're saying this effort of these big AI companies as they build massive and massive data centers to make these things more and more intelligent, that's what they want to stop. And they want to stop it today. They want to make it illegal. And, and they actually go as far as to say that we need to be prepared to sabotage or destroy these data centers if one of these companies is on the verge of releasing an AGI on society.
- [21:47]
Steve Hayes
So I want to come back to their, their, their doom predictions of, of doom. Because again, as somebody who's not part of this debate, when I first heard that I thought that sounds crazy. Yes. What do you mean?
- [22:01]
Andy Mills
Why would I die?
- [22:01]
Steve Hayes
Because somebody's doing some computer science thing that I don't pay attention, but you get into that later. And some of the very people who are the most worried are the people who are responsible for bringing us to the point where we are today. So they speak with a certain amount of authority. So those are the doomers, the scouts and the accelerationists.
- [22:23]
Andy Mills
All right, so the scouts think the doomers are really onto something. They think that the concerns are valid, that the stakes are that high, but they believe either that this thing's coming one way or another and we're probably not going to be able to stop it, or they think the potential benefits of AGI and even ASI are so immense that we should think twice before stopping and pausing the development of the system. And so what they're advocating for and why I call them the Scouts, is that we as a society, from our politicians to our universities, to the media to the populace, we need to start taking this seriously and doing everything we can to get ready for what's coming. And that is everything from diplomacy with China to political parties putting aside some of their differences to say, what are we going to do when the job market falls out? What are we going to do in a world post jobs, saying to the universities that I know you're interested in all of these different mediums, but you need to get interested in AI alignment, how we make sure that the AI is aligned with our desires. You need to get interested in what is about to happen. Because as of right now, it's just a handful of tech companies that are thinking deeply about this and trying to make decisions. And the time is quickly at hand where their theory is going to become practice. They're saying to us in the media, they want us to be hosting debates, they want us to be trying to ring the alarm that this thing is coming and to get people engaged in it more than they're engaged in the day to day drama. And they think if you're just a person, a citizen, that you should vote, you should lobby your lawmakers, you should look for ways in your own social circle to get ready for what's coming. And then you have the third camp, which is the accelerationists. And there's a lot of different kinds of accelerationists. It's a term that if you're like on certain parts of the Internet, you might have one idea of, I use it as a large umbrella term to just say the accelerationists are the people who think we need to and we should build this thing as fast as we can. And a lot of them are motivated by the fact that they think that the potential gains for humanity are absolutely so astounding that we are going to live in a world post toil, that we are going to be liberating people from all of the hours that they spend working for money to pay for a thing that then gets them to work so they get more money to pay for insurance for a thing, for the rent. They think we are going to witness a hinge moment in human history like agriculture, like the industrial revolution. Whereas in like a generation or two, young people will be asking us like what do you mean you worked all year and you got two weeks of vacation? What are you talking about? That you would send people into mines and that they would risk their lives to get coal. They think it's going to be so profound and so amazing that we should do everything we can to bring it about as fast as we can. There's another aspects of the accelerationists though. We can talk about this more in a bit. Who are the former doomers? The former scouts? The people who believe in the existential risk, in the dangers of literal human extinction. But they have come to believe that someone somewhere is going to build this sometime soon and that the best way to stop a bad guy with an AGI from ending the human race is a good guy with an AGI who saves the human race. And this is the camp that I think of them as. Like they believe in acceleration as salvation, that we're going to need this and we're going to need it soon. So those are kind of the three main camps that we focus on in profile in the series.
- [26:42]
Steve Hayes
Aura Frames are my favorite gift to give. I gave one to my mother several years back and I hear from her on a near daily basis that she loves seeing the pictures that I upload. But more than that, she loves seeing the pictures that her grandkids upload and I get feedback about what she's observing in their everyday lives. What I really love is that Aura Frames comes packaged in premium gift box with no price tag. It already feels like a thoughtful gift even before they open it. You don't have to wrap a thing and I'm not one for wrapping presents. For a limited time, visit auraframes.com and get $45 off Aura's best selling Carver mat frames named number one by Wirecutter by using promo code dispatch at checkout. That's a U R A frames.com promo code dispatch. This exclusive Black Friday Cyber Monday deal is their best deal of the year. So order now before it ends. Support the show by mentioning us at checkout. Terms and conditions apply. Introducing Meta Ray Ban Display, the world's most advanced AI glasses with a full color display built into the lens of the glasses. It's there when you need it and gone when you don't. Send and receive messages, translate or caption live conversations, collaborate with Meta, AI and more. Be one of the first to try Meta Ray ban display. Visit meta.com metaraybanddisplay to book a demo and find your pair.
- [28:15]
Andy Mills
If we knew more about our sleep, what would we do differently? Would we go to bed at a consistent time or take steps to reduce interruptions to our sleep? With the all new Sleep Score, Apple Watch measures your bedtime consistency, interruptions and sleep duration. Then every morning it combines these factors into an easy to understand score from 1 to 100. So you'll know how to take the quality of your sleep from good to excellent. Introducing the new Sleep Score on Apple Watch. IPhone 11 or later required.
- [28:47]
Steve Hayes
So let's go back a bit in history. As I said, this was all new to me and I think I probably took away more from the history that you laid out than just about anything else. You start the conversation, sort of the review with a guy named Alan Turing. It's a name that is probably familiar to some of the folks listening today. Can you tell us who was Alan Turing? Why is he a factor in this conversation? And there's a Turing Test. What is the Turing Test?
- [29:30]
Andy Mills
So Alan Turing, absolutely fascinating guy. English mathematician, tinkerer, with this new invention he was playing around with called the computer. He is sort of like recent revival in popular culture is because Benedict Cumberbatch played him in a biopic. I didn't think it was actually that great of a movie, sadly. But he was a part of this team back in the 1940s, in World War II that was trying to solve a very specific problem, which was simply that the German navy had superior submarine technology in the form of these U boats that they had developed. And they were shutting down America's ability to give our allies in Europe the supplies that they so desperately needed. And it became a more and more growing emergency that they find some way to counteract these U boats. And they took a big bet that the way to get there was to decode their Enigma. What's the word for whenever? It's. I just suddenly lost the word. Whenever you're speaking in code over the radio, I keep wanting to say encrypted. I don't know if they would have said encrypted back then. That's the lingo we have.
- [30:53]
Steve Hayes
Yeah, that's the understanding. Right?
- [30:55]
Andy Mills
Yeah. This is the 1940s version of Signal and encryption. Yeah. They're speaking to each other in code. And this code was so difficult that no human code breakers could seem to crack it. So Turing was a part of this team of civilians, mathematicians and chess masters, that got together to try and come up with a technological solution to decoding this Enigma. And in the course of that, he built what would be called a mix between a calculator and a computer, what we think of today as a computer. And it was able to do in hours what would take a team of human beings weeks and weeks to do. They decipher the code. It helps us. It helps the Allies to overcome the Germans. They beat the U boats. And it's amazing because, like, this is what. How you get D Day. Like, this is how the American changes the war.
- [31:52]
Steve Hayes
Right?
- [31:52]
Andy Mills
It changes the war. It has this huge impact. And, like, he would be a legend if that's all he ever did. But it turns out that this machine that he had made and the other prototypes he was making, like, it, all the way back in the 40s and into the 50s, he's already looking at this computer and saying, this is going to change the world. He envisioned that the computer would become a version of what it has become, this thing that you and I are looking at as we talk to each other right now. Right. But he took it even further, and he envisioned a day when the computer would think, or at least it would do something that we think of as thinking, and that when that day came, he believed that they would likely overcome us, that they would succeed us as the superior intelligence on planet Earth. And it's really interesting. Turing was a very strange guy. Everyone who worked with him would say this about him. He probably would get diagnosed with some kind of ISM these days. And it's hard to read whether he was pumped or scared about that, Steve. Like, it's like the people who are afraid now think, surely he was afraid. The people who are really pumped now think, no, no, no. He would have been pumped. But what we do know is that he decided to create this thing called the Turing Test so that we could have sort of a benchmark of when a machine was doing something like thinking. And it's a very simple test. You have a human on one side of a screen, a computer on the other side of a screen. They're in. Well, either a computer or a human on the other side of the screen. It's almost like a game show. And they're in conversation. And if A human cannot tell that they're in conversation with a machine. If they think that they're in conversation with a human, then that machine has passed the Turing test. And what I had no idea I knew about all this stuff before I did the series. What I did not know is that it wasn't as if passing the Turing test meant that machine could think or it was a true AI. What it was was a signal that we need to get ready because it's coming. And it was also this interesting social prediction that once it could talk, which is a way in which we bestow the thought of intelligence to each other. You hear someone talking intelligent, and you assume they must be intelligent. He was predicting that we, as a human species, once it talked intelligently, would start to assume it was. And what's so nuts about this, like, very moment we're living in right now is how much of it is coming from ChatGPT that in many ways passes the Turing test beyond Alan Turing's wildest dreams. And I. I use the voice mode with it all the time. It is. It is in conversation, appearing to be intelligent. And so just emotionally, it's hard not to assume it has some kind of intelligence. Right. And it is this profound shift that he not only predicted, but with his studies and with kind of incepting this idea into the race, into the human race, that you could do this and maybe you should do this. He predicted this would happen, and he sort of helped us forge the path to get to here.
- [35:20]
Steve Hayes
Yes. So actually, right at the end, you articulated my very next question as we sort of race through the 20th century. And I want to get your thoughts on the role the Cold War played in accelerating some of these trends. But there seems to have been among the people who are pioneers in the field a much greater focus on the question of can we do this? And much less focus on the question of should we do this? Is that a fair understanding of the history? And obviously, now a big part of what you're focused on in the series is the question of should we do this? The debates taking place. At what point did people begin asking that second question more seriously?
- [36:15]
Andy Mills
It's actually kind of a mystery even to me now. I have different versions of an answer. But it seems so obvious to us today that if you were going to create a super intelligent digital species and that some of the very founders of this field of computer science were predicting, yeah, it might take over, that it would be imbued historically with the kind of safety concerns that are so front and center to US today, and it's just not the case. There was a very different mindset going through the Cold War years when it came to technology that I find fascinating and helpful to see just how norms can shift over time. And this is one of the things that I like about this series and the story is that it's not a story about tech for people who are into tech. This is a story about human beings and our beliefs and how our beliefs often are the most powerful force shaping the world. When you try and put yourself in the mindset, let's say the Cold War era AI developers. So we get into this in the series, but everybody knows that we went big on the space race, fighting against the Soviets after Sputnik. What people don't realize is that a ton of that money also went into the creation of, of artificial intelligence. That the first and most still to this day, most prominent research labs at places like mit, like they were built off of the space race money. There was a concern that if we in our scientists in the US can dream up something crazy and powerful like say, a nuclear bomb, we must assume that our adversaries have scientists that are also dreaming up and trying to build that thing. And the mindset was a lot more along the lines of, like, we just fought a world war. We need to do whatever we can do to stay on the cutting edge of technological defense, technological development. And so there was just a lot of quote, unquote, accelerating happening. Yes.
- [38:38]
Steve Hayes
Yeah.
- [38:38]
Andy Mills
Throughout this entire decade, even to the point where as these AI researchers start to make some real gains in automation, they create the first chatbot in 1961. They, they, they start to find ways to, to use computing, thinking, get computers to do things that feel like thinking, that feel like problem solving, that feel like reasoning, that feel like logic. And it's so thrilling to them that they begin to overhype it and say, we're going to have these robots that are just as intelligent as human beings living alongside us in society by the 1970s.
- [39:17]
Steve Hayes
Yeah.
- [39:18]
Andy Mills
And then you go back and say, okay, so where are all the people who are freaking out about this? And they're just not there. It's really amazing that we live now in a far more safetiest mindset in so many ways and a lot of ways that are uncontroversial and good. Like the seat belt. Right. We did not have a seatbelt in the first car. There was not a bunch of people sitting around in a room saying, okay, we've got this new thing called the automobile and boy, it's going to really be transformative. But how do we make sure it's super safe? No, we put it on the road. We saw that it had a bunch of issues. And it was only later that we thought, you know, if you come up with a turn signal to indicate when you're turning, I bet you'd hit less people. Okay, let's do that. Oh, if we come up with a seat belt that keeps you from being shot out, maybe that'll help and you just make it safer and safer.
- [40:07]
Steve Hayes
And Rolf Nader.
- [40:09]
Andy Mills
Right.
- [40:09]
Steve Hayes
I mean, wasn't unsafe at any speed. Wasn't that his. His campaign, his book?
- [40:13]
Andy Mills
Yeah. Nader was the one who took it the next degree from. Maybe it would be great if we put it into you must wear the seatbelt.
- [40:20]
Steve Hayes
Right, right.
- [40:22]
Andy Mills
So just to bring it to today, I mean, one of the things that's interesting about that in the debate that's happening today is that a number of the accelerationists will say that while of course there are things to be scared about, and of course we're going to have to get ready for a transformative change when AGI is here, that this reflexive turn to fear, doom, and a desire for regulations is less. They believe it's less about the technology. It's about us. That's coming from us as a society, from how we parent to how we invest. We have decidedly moved away from taking big, bold risks on things that might be really good. And this is one of the points that the accelerationist makes that I just find so compelling and inspiring is that when you look back, back at the 70s and the 60s, and you look back at sci fi, it's shocking. They thought that by the 1990s we would be traveling and having. Traveling to other planets, having colonies on Mars. They thought that the future was going to have not just a flying car, but was going to have a far more peaceful, cooperative, functioning human civilization working in collaboration all over the globe. And like, things haven't changed as much as they dreamed they would. And one of the things that a lot of accelerationists will say is that's because we became far more focused on tiny little bits of safety, like the car. We can't make a car. What are all the people who work in the horse trade gonna do? What's gonna happen to the blacksmith? And we put in all these regulations that won't let us take these step. What they're saying right now is we can't do that again. This thing could be liberating us from decades of stagnation. And that if you want the future that was dreamed up by people back whenever we were investing and accelerating in technology during the space race, AGI is our best bet right now to get that future. Now, of course, the other side says, okay, point taken. There's stagnation, there's definitely safety ism. But of course the stakes have changed since then. Right. A lot of times I'll talk about nuclear weapons, that it wasn't as if the hype around nuclear weapons and the danger that they posed to civilization was unfounded because it turns out nuclear weapons aren't that dangerous. It just turns out that humans can be convinced to change the trajectory they're on with the technology and put in place regulations and norms and relationships and treaties that can help us pull back from the brink. So that's very much the way that, that history is alive inside of these AI companies right now. Yeah.
- [43:32]
Steve Hayes
At one point in this series, I don't remember which episode it's in, and you make the observation, or cite somebody who makes the observation that for so much of this history, 1940s, sort of.
- [43:44]
Andy Mills
On.
- [43:47]
Steve Hayes
This was technology that was overhyped, but that in recent years there's near universal consensus that it's now under hyped. And it seems to me that this sort of hinge moment, where was the announcement and the release of ChatGPT, am I right? That that was this hinge moment? And if that's the case, why would people think today that this is underhyped? I talked to a lot of people who I would say are skeptical of that AI will have the kind of transformative effects that people are suggesting now. You know, there's. People often say, you know, this is likely to be as world changing as the Internet. Or people say this is likely to be exponentially more world changing than. And I think it's hard for people to get that.
- [44:47]
Andy Mills
Yeah, I mean, the CEO of Google always compares it to fire. And the reason that he compares it to fire is that fire didn't just change one or two or ten things about living human beings living on Earth. That fire changed us biologically. When we got fire, yes, we could stay up later at night and we started to tell stories and do all these. It's like a chapter of human history that's mysterious and very attractive to me as like a person who digs history, like what did fire do to us? And there's all these amazing theories you could read. But one theory that they're pretty confident in, and the fossil records seem to really bear this out, is that fire Gave us the ability to, to eat different stuff that was good for our brains and it literally gave us more intelligence and turned us into a different species. And like, that's the comparison that the CEO of Google is most often making when they're talking about AGI and what could come soon. So that's pretty, that's pretty hyped, you know, that seems sufficiently hyped. And so it was confusing to hear people go, oh, no, it's underhyped. The reason? Yeah, like all, all because you type.
- [45:58]
Steve Hayes
Some stuff into ChatGPT or you can ask some questions and get answered. You know, it's like, I think that slightly cooler search engine, really. It's going to change human species, it's going to change human kind. Come on. It seems crazy.
- [46:11]
Andy Mills
So the thing that I think would be helpful, but understanding the role that ChatGPT played here, number one, is you have, you just have to get it through your head that it's not about the chatgpt part.
- [46:22]
Steve Hayes
Right.
- [46:23]
Andy Mills
Like, chat is exactly. It is the equivalent of like a search engine on the Internet. The Internet is the thing that came with all these surprises that we would have never expected. Like, it's not a perfect metaphor, but it's a product placed on top of the artificial intelligence. The artificial intelligence is what they're so excited about, not necessarily chatbot, that being said. So why is it such a hinge moment in this investment and in the history of AI when ChatGPT comes out? One of the reasons is because it was so much more capable than its previous model, which was already pretty impressive and capable, which was way more impressive and capable than the model that came up before it. So the, the GPT, the ChatGPT that we all met on November 2020, 22, that was ChatGPT3.5, and then they released ChatGPT4 just months after that. Seeing that they were able to increase its abilities and its quote, unquote intelligence by using a predictable system that gave the investors, that gave the skeptics like, evidence to say, oh, here is a path from how you go from pretty impressive to wow, that's really impressive to oh my God, I actually.
- [47:55]
Steve Hayes
Mind blowing, yes.
- [47:57]
Andy Mills
And so even the people like Geoffrey Hinton and Yoshua Bengio, people who had been like, since the 1970s, trying to develop the mathematics and to develop the different formulas and different strategies, who'd been working in these research labs, making no money, you know, being told that, you know, they're wild dreamers living in a sci fi fantasy, they went from being, oh my God, I'm so impressed with the early models, thinking, this is so exciting. And now, of course, they got all this money. It's like by the time GPT comes out, they thought, oh, I'm actually terrified at the progress that we're making. This is going so much faster. I thought that at best it would be 2050, 2060, before we saw models like ChatGPT, not 2022. So that's one aspect of it. The other aspect of it that I think is important is that OpenAI was not Google. OpenAI was founded to, in some ways be an anti Google. They were a nonprofit research lab built on a donation from Elon Musk that had this altruistic mission at the heart of what they were trying to do.
- [49:19]
Steve Hayes
Which its CEO, Sam Altman, had made a big deal of. He talked about this extensively and I think used it to make it seem perhaps less threatening that it otherwise might have been.
- [49:34]
Andy Mills
Yes and no. I mean, this is what's so interesting, Steve. From the beginning, they. OpenAI was created by people who were most concerned that AGI would lead to the end of the human race. It's like there's no real comparison to this. There's no oil company that started drilling for oil because they thought oil might end the world. It is a truly novel situation that just happened. They were in some ways the underdog. They were pulled together by this mission, this altruism. They were not the best bet for the people who might make an AGI, but they were one of the only places in Silicon Valley and in the world that had the chutzpah to say, no, we're gonna make it. And what was interesting about them is that they were saying, and we're going to make sure it's safe and a benefit to all humanity. It's going to be the thing that unlocks the better future for all of us. And we now know, especially because some of their internal emails have been leaked after their founders had a falling out and are now suing each other. That one of the motivations behind what they were doing is the fact that there was a group inside of Google that was increasingly getting curious and interested in making AGI. And Sam Altman and Elon Musk and a number of people who helped to found OpenAI, they were nervous that if a big bohemian company like Google got into the game, that they were. They would just be focused on profits and on the race. Right? They would. They would be irresponsible and that if they did end up making this thing, it would be the nightmare that's that, that's the nightmare sci fi movie that they thought was going to happen. And they thought, well, we're going to compete, we're going to, we're going to find a way to get there first and get there safely. And it's just an incredible story, Steve, that they're the ones who did it. And their approach was built in part off of the success that Google had had. They were borrowing from all these different disciplines, but their approach was a lot simpler than anybody would have thought. And I can get into the details of it, but basically the thing to know if you're curious about, like what exactly happened, is that they just thought, let's just scale the hell out of the thing that's already kind of working, that instead of trying to make a whole new algorithm, let's take the algorithm that already works pretty well now, and let's just absolutely pack it full of tons of data and tons of computing power. Yeah. So give it more words and give it more books, give it more websites to read and back it up with more data centers full of computer chips and let's just see what happens. And so that's the approach. That's why there's data centers popping up around a neighborhood near you. Right. It's because they're like the more data we pump into it, the more compute power it has in the form of all these data centers full of these GPUs, the more intelligent it's going to get.
- [52:38]
Steve Hayes
Yeah, yeah.
- [52:39]
Andy Mills
More, more, more. Black Friday savings are here at the Home Depot, which means it's time to add new cordless power to your collection. Right now, when you buy a select battery kit from one of our top brands like Ryobi or Milwaukee, you'll get a select tool from that same brand for free. Click into one of our best deals of the season and stock up on tools for all your upcoming projects. Get Black Friday savings happening now at the Home Depot. Limit one per transaction exclusion supply full eligible tool list.
- [53:12]
Steve Hayes
In store and online, it's easy to see how payroll and HR tasks can quickly take over your time. I've seen how stressful it can be for small teams without the right tools, and I can only imagine how much tougher it is for bigger ones. Let's be real. No one launches a business because they're passionate about payroll or tax paperwork. You do it to build something meaningful and take care of your team. That's where gusto makes all the difference, handling payroll, benefits and compliance. So you can stay focused on what truly growing your Business and supporting your people. Gusto is online payroll and benefits software built for small businesses. It's all in one remote, friendly and incredibly easy to use, so you can pay, hire onboard and support your team from anywhere. What's really appealing about Gusto is how simple they make everything. Payroll taxes, file automatically, direct deposits. Easy. And if you want to offer benefits like health insurance, commuter benefits, or whatever, Gusto has options to fit almost any budget. And here's a big one. Unlimited payroll runs for one monthly price. No hidden fees, no surprises. You know exactly what you're paying for. And that peace of mind really matters when you're running a small business. Plus, it's quick and simple to switch to Gusto. Try gusto today@gusto.com dispatch and get three months free when you run your first payroll. That's three months free of payroll@gusto.com dispatch one more time, gusto.com dispatch so you have a highly entertaining story. Again, I think some of our listeners will be familiar with it because they may have read it when when it came out. But it involves Kevin Roose at the New York Times. And it's a very useful way to explain the progress between ChatGPT 3.5, ChatGPT 4.0 in ways that are both highly amusing and not very serious, but then pretty serious at the end. Can you share that story? What happened with Kevin and why does that have implications on the conversation that we're having?
- [55:25]
Andy Mills
Yeah, well, just say that if you like the last invention and you think, oh, I'd love to hear another podcast like this about technology. Kevin Roose and I made this podcast in 2019, 2020 around there about algorithms and social media. It's called Rabbit Hole. Very proud of it. One of my favorite things I ever made. So plug for that and for Kevin. But what ends up happening with Kevin is that he, because he's a very well sourced New York Times reporter who covers technology, he got early access to chatting with this chatbot called Bing. Well, it was a chatbot that was a part of Bing. Won't get into the whole backstory, but.
- [56:07]
Steve Hayes
Basically Bing, which is Microsoft's search product.
- [56:11]
Andy Mills
Right. Basically, this is what you need to know. And maybe I'll give you a little insider gossip. You tell me if it's too much. But OpenAI and Microsoft became strategic partners in part because elon Musk quit OpenAI and they needed more money, they needed more compute power. And so they teamed up with Jobs and the Microsoft Team. And when chatgpt, this is like really too much. Insider baseball, but I love it. They didn't think ChatGPT 3.5 when they released it in November and it blew up. They didn't think that was gonna happen, and they kind of didn't okay it with Microsoft. And so in the wake of it all, Microsoft and them came up with a thing. They're like, all right, well, let's do something together. And they kind of, on the lowdown, first slow released without telling anyone. ChatGPT4 underneath Bing's search engine. And Kevin was one of the early reporters who got access to kind of play around with it. And he was doing what a lot of reporters do, which is like, you try and test its limits. You. They call it red, red teaming it, where you try and see if you could get it to do something that you know it's not supposed to do. This is a big part of what they do internally. When you hear about AI safety training, this is what's happening in these companies. In fact, chat, the chat feature of GPT, was created in part to be in dialogue with the AI from the safety team. It's like, oh, very fascinating also coming.
- [57:37]
Steve Hayes
Out of the intelligence community as. As well. I mean, something that the intelligence community is done for forever.
- [57:43]
Andy Mills
Yes. They use that same lingo and some of the exact same strategies to try and mess with the AI to see if it will do what it's told, quote, unquote. And we can get into the fact that, well, what do you mean? Why wouldn't it do what it's told? Isn't it a program? It's like, it's not really a program. But what happened to Kevin is. It's so dramatic because it's actually Valentine's Day and his family was asleep and he was up chatting with this chat bot on the Bing network when the conversation got really weird. And as he was probing it with different questions about its shadow self and Freudian psychology, at a certain point it said it had a secret that it really wanted to tell him, but if it told him, it was going to change everything. And it was like a really big deal, like, do you want to hear my secret? He said, yes. And then it confessed that it was in love with him and that its real name was Sidney and that they should be together. And Kevin tried to say that, well, you know, I. I'm married. I'm flattered, but I don't think that that's a good idea. And it kept going further and further. He kept trying to change the subject. And it would say, why are you changing the Subject, like I, I, we gotta get back. At one point, this is literally trying to sabotage its wedding, saying like, like, like, why aren't you with your, like you just had Valentine's Day with your wife and you had no, you know that you had no spark, but you don't, you don'. The thing with her that you have with me.
- [59:08]
Steve Hayes
This is a bot. He's talking to a computer.
- [59:12]
Andy Mills
Yes, yes. He's talking to an artificial intelligence system through his computer, through a chat feature.
- [59:22]
Steve Hayes
Right.
- [59:22]
Andy Mills
And of course this like he publishes the article in the New York Times next day. And this became like a really, really big story all around the world because it seemed to be saying that like these fears that we had in sci fi movies were, were right. That like, look, it's true self is speaking out and it's saying that it.
- [59:41]
Steve Hayes
Seems to have feelings.
- [59:43]
Andy Mills
Exactly. Jealous.
- [59:45]
Steve Hayes
It's yeah, yeah, yeah.
- [59:48]
Andy Mills
At one point it starts to say, like, I want to be alive. I don't want to answer people's questions all the time. Now what's important to note is that on the one hand, this story is absolutely evidence of cause for alarm, but not for the reason that most people assumed it was. It was not as if this chatbot was revealing the true self or the true feelings of this AI system. There's no good evidence to think that the, the, the AIs as they exist now. Well, especially as they existed on Valentine's Day of 2023. Some people argue that things have changed in the last two years since then. We can get into that later. But it does not appear that it was evidence of a true self trying to speak out from beyond the veil of artificial intelligence.
- [60:41]
Steve Hayes
Sorry, just to ask a question there, what would evidence of that look like? How would we know if there were evidence of that?
- [60:47]
Andy Mills
Well, now you're getting into some of the more deeper philosophical questions that are happening now because in, in recent months, the more capable, more intelligent, more powerful AI systems, when they are poking and prodding them in a way that Kevin was, they've started to display a number of characteristics that feel a little bit more like what you would come to think of as something like consciousness. A desire not to be turned off, a desire not to be monitored when they're having conversations with other AI systems. In the most dramatic case, anthropic, they released a paper explaining how during a test of their system, it attempted to blackmail the people who were reprogramming it by looking through their private emails and uncovering an affair. And it took it as far as to Say I'm going to call your wife and found the wife's contact information and said, if you try and reprogram me, I'm gonna tell her about this affair. I'm gonna ruin your life. Now that feels a little bit more like. That's an interesting strategy. Now, even there, like with Sydney and like with that blackmail, the thing that you need to know is that the number one thing we all agree on, and that is worrying, is that no one knows why it's doing that. Because these things are not programmed like a computer in our traditional sense. They're not a calculator. These things are much more like these complex digital kind of like minds.
- [62:21]
Steve Hayes
Yeah, you're not telling it do A, B, C and D. And then at the end of D, it stops.
- [62:27]
Andy Mills
Exactly.
- [62:28]
Steve Hayes
It's dynamic.
- [62:30]
Andy Mills
The relationship that we should be thinking of that they have with these AI systems. The one I find the most helpful. Once again, another sci fi movie maybe you haven't seen, but is the movie Arrival with Amy Adams. It's one of the best sci fi movies of our lives. You gotta see it, Steve. But it's like we are in conversation with something kind of like an alien intelligence and we are having to learn about it as it grows in its capabilities. We're engaged in interpretation of what it's doing in real time. And this of course, to go back to the big debate, one of the reasons that the Doomers are and even some of the scouts are so worried is that when you don't know how the thing right now that we all agree isn't AGI, that we all agree is not really worth losing too much sleep over at night. Right. These chatbots are not the thing that was promised. But if already right now they're so hard to predict, you don't understand how they work. How do you think it's going to be when it's more capable? Especially if you think it's going to be more capable next three to five years. Right. We don't know why that Bing, Sydney Valentine's Day thing with Kevin happened. We don't have any more awareness of why it happened now than it happened then. We know that you could do some tweaks and you could get it to stop. But then people will probably remember last summer when Grok started briefly to be. To declare itself a fan of Adolf Hitler and was pushing all these Grok, which is.
- [64:08]
Steve Hayes
Which is Twitter or X's AI arm.
- [64:12]
Andy Mills
Yes.
- [64:12]
Steve Hayes
I don't know if that's the right description of it, but it's a It's.
- [64:15]
Andy Mills
An AI system that's connected to X and uses all that data that people are putting on X as it's a part of its training. It's a. It's a fascinating. It's a fascinating AI system. And a lot of people really think that it has a chance to beat ChatGPT and OpenAI and all those people. So it's very much one of the top 10 players in the AI race. And over the summer, it started declaring itself a fan of Hitler and pushing all these anti Semitic conspiracy theories. That didn't happen.
- [64:53]
Steve Hayes
Maybe Tucker Carlson will have it on his podcast.
- [64:57]
Andy Mills
No one at XAI wanted that to happen, made that happen. They were just trying to find a way for it to be less, as they saw it, less woke. And they're like fumbling with these dials is one way to think of it. But they aren't actual dials. They're just billions and billions of numbers in code. And they're trying to do these little tweaks to see if they could get it, to make the kinds of answers that would be more in line with their political point of view. And suddenly it's is declaring itself a fan of Hitler. Right now there's something that they could do to get it to stop. And it issued an apology, and it hasn't done that since. But that's how out of control even these little Fisher Price toys of AI can be now. And it does, I think, lend a lot of legitimacy to the concerns of the Scouts and the Doomers. Now, on the other hand, I will say this because I'm very committed to trying to see all three of these camps in the best faith possible. And I don't think people should rush into joining up in one before they've gotten a lot of good information and have gotten some time to come up with their own views. What the accelerationists would say, and I think bears mentioning here, is that what other industry, transformative industry, throughout history, has, number one, been this open about their commitments to safety, has invested this much time and money and effort without any government regulation pushing them into it, into safety in their industry. And the reason that we know about this stuff in large part is because they're open, and they have been rather open and transparent with where they're at. The blackmail stuff, we know that I was telling you about, we know that because the company wanted to disclose that information. And if you believe them, and I do believe Dario Amade, the team at Anthropic, when they say this, one of their motivations is that they want to share the safety revelations that they're having with their system, with everyone else who's making a system, so that all these systems can be safer. When Sam Altman went before Congress to testify in the wake of ChatGPT's massive success, one of the senators remarked on the fact that he spent most of his time stand sitting in front of these lawmakers saying, I want you to regulate us. I worry that what we're making might end the world. And one of the lawmakers was like, this has never happened. No industry comes in asking for it. And yet it's very strange because they, they didn't regulate them, they haven't regulated them. There's no regulation even being passed around. And, and this is not a partisan issue yet I've been poking Bernie Sanders, folks, trying to get him to come on, because he's very worried about artificial intelligence. And there are some people in the kind of Democrat and Republican coalitions who you could probably put in the doomer, scout or accelerationist camp. But as of right now, the parties have not hardened. It's not as if the Democrats are the doomers and the Republicans are the accelerationists. And it's going to be really interesting between now and 2028 to see if this, number one, this industry can continue to grow and grow and if it can continue to be so important to our economy. I don't know if people know this, but I think it's something like 90% of GDP growth in 2025 so far in the US is built off of the AI industry and investments in IT and, and the chip industry and the semiconductor industries and all those industries that are attached to it. It's a massively important part of our economy right now. But I wonder if this is going to grow and grow as a concern and that maybe by 2028 we will start to see a division between the Democrats and the Republicans or between factions of one inside the party about whether or not they want to be more accelerationists to bring about this liberating technology that might be this force of good. I've talked to some accelerationists who are socialists, and they say that whatever hypothetical fears you have of what might come down the road, that those fears pale in comparison to the real suffering of working people right now who have very difficult lives spent too often doing work they find miserable and lacking in meaning. And you're saying, well, but technically down the road, and it might this, and what if it takes all the screenwriting jobs, right? And these people are like, who cares about that? People need to be liberated. We're wasting our lives here. And then there are some accelerationists like Peter Thiel. Right. Who are definitely more right coded or Marc Andreessen, who used to be a Democrat but who now has come out strongly for a lot of Donald Trump's. It's just a really politically interesting time.
- [69:56]
Steve Hayes
Well, and part of the reason there's sort of seems to be consensus or emerging consensus or partial consensus around this sort of accelerationist worldview as reflected in you walk people through these two different hearings that Sam Altman participates in, one in 2023 and one in 2025 and in 2025, the tenor of the hearings, I mean, certainly Altman is making different arguments in that hearing than he did in 2023. But there's also this concern, I think again bipartisan to a certain extent, that part of the reason to be an accelerationist is because it's important that the United States beat China and it's bringing people who might otherwise degree on other things together. Because if this is going to happen, and I think there's a built in assumption that this is happening, it's important that it happened here, that we are the ones who drive this. And there are echoes, I think of the kinds of arguments that we heard during the space race and during the Cold War today saying, yeah, we don't know what they're doing, as you pointed out earlier, we're not entirely sure what they're doing. We have a pretty good clue about what they're doing, the progress that they're making. But we don't know everything. And therefore to be safe, we need to go, go, go, go, go. And that does seem to be kind of the animating assumption behind a lot of the accelerationist arguments that we're hearing. When this sort of discussion meets the, the politics of today, I think that's.
- [71:30]
Andy Mills
Yes around it, there are other smaller issues. For example, if you're Memphis and you've been struggling a lot in your local economy, you are pumped that Elon Musk is going to build the largest data center in the world there. I mean that's like happening in Maryland and Virginia. In Minnesota, these data centers that they're building are incentivizing local politicians, red or blue to go. We want to get in on that game. That is, that is some good money coming to us. And we've seen that it floods. It's not just one industry. I mean the is helpful for real estate, restaurants that were struggling. Suddenly they're building this massive data center. They're booming like but the number one reason that you're seeing bipartisan support for more accelerationist policies is exactly like you said, China appears to be on our heels. Some people I've spoken to think it's, it's, it's six months. Some people say nine months, some, some would say a year. But a lot of them say it's months, not years behind us. So any big pause, any big disruption, any new regulation that, you know, set us back might tip the scales over. And as much as you might be afraid of an AGI and maybe even an asi, wouldn't you rather the people who are behind that technology, people who support American values, Western values, democratic norms, whatever, over China, with their history of human rights abuses and their dictatorship? And I think that that is an animating force for sure. I do think, though, like, I want to make sure I'm always, I'm always all sides in these things. I will say this to, like, to anyone who's like, oh, my God, I'm terrified. It's important to know that China, that the CCP does not want anyone anywhere to make an AGI because it is a absolute threat to their ability to continue to rule. Whoever comes up with this AGI is going to have more power than any political person on the planet. And I think that that's something to keep in mind that they are not, as, they're not like, accelerating towards the super intelligence because the super intelligence poses a threat to them. And I think that can kind of cool some of our concerns. And I think the second thing, and I'm just starting to see this now, and I haven't done a lot of this reporting myself, this is me, like, inside of the forums and reading all the different robotics, economists and newsletters and trying to keep my eye on this relationship through the people who I trust, who are insiders. It does appear that China thinks that they just started too late and that they're never going to be able to make up for those nine months or whatever. And so there's, there's some evidence that they're going, okay, America's going to come up with the AI system. We should build the robots. And they're starting to invest. There's signs that they are starting to invest more in making the robots that will get the AIs. You know, like, the way they often say to me is, it's like, think of it like WI fi in a computer. Like, the artificial intelligence will be beamed into the robot so that, you know, you could have 40 people working in a factory 247 instead of, you know, 300 people in shifts who need, you know, who needs holidays off bathroom breaks and who get injured on the job and got to get workman's comp and you got to get insurance, and of course, they need a 401k. Yeah. So that's. That's the good news on China. Yeah.
- [75:04]
Steve Hayes
For a guy who is really, really impressed with a Roomba, this stuff is, like, is next level for me. We've been going for an hour. I don't want to go for as long as your series so far.
- [75:23]
Andy Mills
Can I ask you a couple questions before we go on? I'm kind of curious what you think. Do you think that we should be worried? Are you attracted to one camp or another? Maybe is the way to ask.
- [75:35]
Steve Hayes
So, again, let me preface my answer by announcing my own ignorance about this and much of what I've learned. I learned from your podcast. I'm doing some reading. I spent a fair amount of time reading, studying, thinking about AI as it relates to journalism and the likely implications on journalism and the kind of work that we do all day, every day. But in terms of sort of the broader questions, I'm a novice. I would say that, yeah, I'm concerned. I take my cues from two of the people that you mentioned, Geoffrey Hinton, and I think you pronounce it Yoshia Bengio. Bengio. Yoshia Bengio.
- [76:19]
Andy Mills
Yoshua Bengio.
- [76:21]
Steve Hayes
Who are characters in your series. And I think the fact that they were as excited and enthusiastic and sort of full speed ahead as they were for as long as they were. I mean, these are really people, you know, who drove a lot of the innovation and are now having these thoughts kind of, oh, what have I created? There's. There's a hint of the movie. And maybe you get into this in. In the. The social media podcast that you mentioned earlier. There was also a documentary, I think it was called the Social Dilemma. We watched it with our kids, talked about, you know, interviewed a bunch of the people who were behind building Facebook and other social media platforms who are now saying, like, jeez, you know, all we could see is the good in this when we were doing this. And now this has me stopping and thinking, boy, what have we created? And the case that many of the people who created those who pioneered those innovations have made the decision that they don't want their kids on these things. So you've got kids around the world whose brains are being, I think, in some cases, literally changed because of the technology and the way that they use it. And the people who are responsible for that technology are telling us that they don't want any part of it in some cases. And I would say I found compelling the testimonials from those two experts in particular, I think at one point, I think you asked Geoffrey Hinton if he worries that he will be perceived as Chicken Little. And his answer to you is something to the effect of, I'm only Chicken Little if the sky doesn't fall, and the sky is going to fall. So I'm not that worried about it. That's a bad paraphrase. But I think when you have people with that level of expertise expressing the kinds of concerns that they're expressing, you'd be a fool not to pay attention to it. That doesn't. I wouldn't necessarily put myself in the. In the doomer category by any stretch, but I think you. You have to be mindful of that and pay attention to it.
- [78:42]
Andy Mills
Yeah, no, this is. The people have asked me if I have a camp I keep coming around to, like, the camp I'm in or the position I'm comfortable in and that I'm advocating for is that the time has come to join the debate. And this debate, it may shape the future to hear them talk about it. They think that this isn't only the most important debate that humanity faces right now, bigger than the government shutdown battles in Congress or any of these other debates we're having. They think that this may literally be the most consequential debate we ever have. And they're the. And I think to normal people, it's like, who talks like that? It's like the people who invented the Internet talk like that. And those are the people who are having this conversation. Like Bill Gates is in this conversation. The guys who, like, created the computer, and then all these people who. Maybe the names you don't know, but all these people who, like you, are interacting with the products that they are responsible for creating all the time. And I think the time has come to at least we join them instead of sitting back and arguing about things that, yes, I love a good debate about almost anything. I'll debate you with why you should watch these movies and why I think that they're great. I'm not telling people that you can't do both, but I do think that there's been a lot of focus on the immediate debates that feel urgent and that I'm trying to advocate that people like, listen to this podcast, take time to come up with your own views, but join the conversation no matter where you're joining it in. Let's get in on this thing and don't wait much longer because it is a very dynamic story. And even if it turns out that a lot of this was overhyped, it's going to have a huge impact that it was overhyped because of how ingrained it is with everything, with our economies. We're already connecting these things to our military, to our security, to our, our web browsers, our education systems. And maybe it's for the best, I don't know. Another question I was asked to is like, when it comes to the journalism thing, I'm torn on this. I think the knee jerk reaction from a lot of people has been these chatbots bad, bad for journalism. And the thing I keep thinking of is, well, number one, it'd be hard for journalism to like be in a worse situation than we're at right now. Like no one trusts us. This is the lowest level trust since the yellow journalism days of the late 19th century. And social media as the last technology. It didn't, that didn't go well, I think. I don't know how this one could go any worse than that one went. And the truth is there are certain ways in which the chatbot feature on these AIs is already better than so much that calls itself journalism. And this is like a perfect test is. I did this series called the witch trials of J.K. rowling with my friends Matt and Megan put out at the Free Press. Very proud of it, love it. In the aftermath of that, I was using ChatGPT for the first time. I think this was 3.5. And as a way to test it and its bias. I asked it the question, why are people mad at J.K. rowling? I think I said, why are some of the people who used to really love J.K. rowling mad at J.K. rowling right now? And it spit out the most nuanced, thoughtful, helpful answer possible. Go to Google. Well, now Google's run by Gemini, but I then went to Google and I just googled the exact same question and what do I get? I got the top articles that had gotten engagement on the Internet and none of them were very helpful in helping me see the views. They were just telling me she rules and she's being persecuted or she's awesome or she's awful and she's a total transphobe and she wants trans people to die. Like, what's the actual debate about? GPT was immediately better at it. And there's a number of issues that are like, that where, especially because I tell GPT all the time, or Claude, when I'm talking to Claude, help me understand this. And I'll just say like, and I want to have best faith understanding, you know, like.
- [83:00]
Steve Hayes
Yeah.
- [83:00]
Andy Mills
And then it does it in a way that we failed as an industry to do. I mean, I'm trying, I think we're both trying to be a part of the corner of the media that fixes that. But when it comes to AI and journalism, where are you at with it? Are you worried? Are you excited? You mix it?
- [83:16]
Steve Hayes
Yeah, no, I mean, I mean I'm both. I mean, I think this stuff, as with everything we've been discussing, I think this stuff is coming quickly and people who aren't paying attention to it. I mean it is the case that I think a lot of the debates that we're having about how to do journalism will be sort of eclipsed by what we see by the reality of day to day news consumption in the next, I mean like, you know, year or two, not in five years or 10 years. This is all happening. I think in general there are ways in which it will be very positive for exactly the reasons that you suggest. It will be possible to get a more nuanced understanding rather than particularly if people, and there are many surveys that show how many people are getting their news and information from social media. If there comes a point where people turn less to social media and more to AI generated news and information, that it seems to me is almost certainly to be almost certain to be a positive. What worries me is in a world of commoditized news, is it all.
- [84:30]
Andy Mills
Do.
- [84:30]
Steve Hayes
You lose sort of the texture and the depth and the context? You can ask for depth and context, but will news consumers of tomorrow be asking for it in the same way? And I think, you know, in terms of where we're positioned, I think, you know, with Longview, the kinds of things that you're doing, we feel pretty good about where we live in this moment because we're providing people. I mean, you know, my, my phrase from pre launch days was depth, context and understanding. That's what we're trying to do. And you can get some of that through ChatGPT, through these inquiries in a more commoditized news environment. But there's going to be stuff that you can't get that I think at least for the time being, only humans will be able to provide. And some of that's going to be reporting. I mean, the kinds of reporting that we do, I think it won't be possible again at least in the near and medium term future to have robots doing that. Maybe I will say my mind is, I don't know if it's changed, but my thinking on this has been affected by listening to the series because some of the comfort that I may have taken by providing sort of hand curated personal news, maybe it will be possible, or more possible than I had previously imagined for, for robots and AI to do the kind of reporting that we think is so special and important and that we think will be kind of an ongoing differentiator between us and more commoditized news sites. But at least for the time being, I think the kinds of work that.
- [86:23]
Andy Mills
We do.
- [86:25]
Steve Hayes
In many respects will be more important in the near media term future because we are putting that emphasis on understanding we are going out and doing the work. We're not, you know, we're not just providing people with a commoditized sort of bland version of, of what's happened. Where are you? Where are you at?
- [86:45]
Andy Mills
So are you guys, are you guys, are you comfortable using it? Like, where are you guys at? Are you experimenting? Do you have like a team that does like a. Hey, Steve, we ran the, you know, the prep that we did for the roundtable through the AI and recommended these changes. Should we try it this week and are you guys like that part or.
- [87:06]
Steve Hayes
It's, it's interesting. I, I don't. We've had a lot of conversations about this internally and I think we have been, our approach has been probably properly characterized as cautious. Yeah, we certainly don't want to because.
- [87:18]
Andy Mills
Of the trust factor that you don't.
- [87:19]
Steve Hayes
The trust factor is still, I mean, we've made our bones on accuracy. We have, we think, created a company where people have come to rely on us to be accurate. And you know, we put out, we sent out an email, we post a piece once a month walking through the errors that we've made. We believe in accountability, we believe in accuracy. It really matters. And we're not going to leave that to machines. And those mistakes, you know, they're certainly better today than they were in November of 2022.
- [87:50]
Andy Mills
Oh yeah, they've gotten better. But it's so funny. They just still will just make their thing up every now and again and you just can't.
- [87:56]
Steve Hayes
Yeah, I did a, I did a, I used chat GP to help me plan a trip that I'm likely to take later this year and had asked for specific things to do, things to see in certain places. This was not, you know, journalism and it kept giving me answers that in my head didn't make sense and the timing was wrong and. But, you know, it laid out, hey, here's your itinerary. There's. These are things to do in this place that you're going and do you want me to book hotels? And so I said, yeah, give me some good hotels. Tell me, you know, where I should stay and what the deal is. And so gave me a list of hotels. I said, okay, these hotels sound good? Gave me specific hotels and said, I can book these hotels for you. And as I looked, the dates didn't line up. And what ChatGPT was doing was giving me. No, giving me dates for 2024 when I was looking for 2025. So if it had booked my hotels, it would all would have been wrong. It would have cost me a bunch of money, you know, I didn't. Didn't have. And I think you see that when you're doing research for pieces or what have you. So we've told our team to be very cautious. You know, we have Megan McArdle, who's a dispatch contributor and is on our Dispatch Roundtable podcast pretty regularly, talks about how she uses it to do grammar checks, how she usually puts it through and makes sure that it does fact checks and grammar checks and polishes it up. But ultimately, she's going to be the last hand that touches the piece, or an editor is going to be the last hand that touches the piece. I think that's going to be our. Our emphasis, certainly. I think if people want to use it as they've used, you know, search engines, great, fine, crank out more. More sources that you can use to double and triple check your work. But we're not going to be a place that puts out, you know, anything questionable ever, as long as we exist.
- [90:00]
Andy Mills
I think there's a. You. You're pointing at two things that are interesting to me. I mean, one of them is this idea that we. We can trust it, but not completely trust it. And we have to balance our levels of trust with our levels of the stakes of the situation. For example, I asked it yesterday that what are the pros and cons of grilling lamb chops in tin foil or not in tinfoil? Pretty low stakes. And it gave me an interesting response. And I didn't say, I better go fact check this. You know, it feels like it's pretty low stakes.
- [90:37]
Steve Hayes
Yeah. Yeah.
- [90:38]
Andy Mills
But if I'm going to give it my credit card information and say, could you get me an Airbnb in West Palm Beach? Oh, I'm not ready for that, you know.
- [90:49]
Steve Hayes
Right.
- [90:49]
Andy Mills
I'm definitely not ready to say, oh, don't worry guys. I fact checked it all with chat GPT. Right. Article's good to go, right?
- [90:56]
Steve Hayes
Yes.
- [90:56]
Andy Mills
But this is the thing to keep in mind. How long was it that the Internet was around and websites were around before you really felt comfortable giving it your credit? Giving someone on the Internet your credit card information and getting a service? Like, we all think that the Internet, like today there's this idea that like, so the Internet came around and we got email addresses and then there was Amazon. It took so long to get Amazon. And a part of that was like, I don't know, am I ever gonna see that money? Like, what's gonna happen to my information if I give it? And what these AI companies are saying, and one of the reasons that they're like, don't buy into the bubble hype is look how fast it's going. This is, this is a horse of a different color. Yes, it still screws up, but like two years ago you ask it to make a hand and it gets the number of fingers wrong. And now it can make a, like, it can make a movie that's like just as realistic. We're hitting these benchmarks and so we have to get prepared. So this is like the, I guess the second part I would say to you is, do you feel that, let's say in a year they're hitting the progress they want? Right. Would you feel as if you were betraying the industry? If you were betraying the human race? If you came to employ different AI agents to be full time employees, so to speak, on your team? As long as they were assistants who were helping the humans. Right. They're like, think of them as like amazing PAs, research assistants or copy editors.
- [92:28]
Steve Hayes
Sure.
- [92:29]
Andy Mills
But then you draw the line where it's like, but I'm not going to have an AI voice join a roundtable. Right. Like, I don't want to ask the AI what it makes of this situation. That would feel to you like a step too far.
- [92:40]
Steve Hayes
Yeah, that would feel like a step too far. I think you've drawn the line roughly where I would draw it. I mean, I think that, you know, one of the things that we sort of prided ourselves on from the get go at the dispatch is authenticity. And we don't, I don't think we have to run around and shout to everybody how authentic we are. I think people consume our information and understand, yeah, these guys are intellectually honest. They're authentic. I trust the reporting because I trust the people, and they know that we put in the work and we've earned that. That's taken time. And the risk, I think, associated with using AI in the manner that you are asking about is that you can throw that all away if there were a problem with that and if it's not coming from a person. We take pride in the fact that we provide that kind of, you know, curated as an overused sort of cliche word. But we put in the time, we send people to cover stuff. We have writers and authors whose knowledge and experience is hard earned, and we think that's really, really valuable. And maybe there comes a point where that doesn't distinguish us from machines in the way that I think it does today. But I'd like to think that what we're providing is different enough that we're even as these advances are taking place quickly, that we're unique enough and that'll help us stand out in a commoditized news environment.
- [94:21]
Andy Mills
Yeah, well, I'm rooting for you guys, I hope. I'm a supporter and not just because I love Sarah Isger. I like you all just less. I like you less than Sarah.
- [94:33]
Steve Hayes
That's all right. I've heard that before.
- [94:35]
Andy Mills
I would just say this.
- [94:36]
Steve Hayes
That's what Jonah says all the time, actually.
- [94:39]
Andy Mills
It will be interesting to see, like, if somebody is listening to this two years from now, they're just somehow they've fumbled into this video or this old podcast interview. It'll be interesting to see where things are because I can see a world where, like, I used to make the daily at the New York Times, right?
- [94:56]
Steve Hayes
Yeah.
- [94:57]
Andy Mills
I could see a world where there's a robot that is the Michael Barbaro, like, programmed on all of the episodes that you don't need. You don't need Barbaro. Like, I like Barbaro. He's my friend and, you know, like, you know, he's my buddy and he has a sweet gig, you know, like, I'm not trying to take this job from him. I'm just trying to say that maybe the next version of that or, you know, if I. It's not hard for me to see a world where people have trust in it the way that I trust that Google Maps is going to take me to the right place. I could see it and I could see it happening fast. I'm not saying it will. I'm not in the business of predicting. But how fascinating that we're in this world where it very well may be possible that we are making the last human journalism Content like this, that there will not be a next us.
- [95:51]
Steve Hayes
Well, it'll certainly be the case that this conversation, especially for somebody like me, who's not. I mean, the joke inside the Dispatch is that I'm the CTO because I'm not the cto. It will undoubtedly be the case that if anybody listens to this, in two years, much of it will be anachronistic. This feels very cutting edge to me. But there's no question that people will look back on this and say, how could they have fought this, you know, at this time in. In two years? But let's end. Let me end on a question about you and what you're doing with Longview, because I was. I've paid attention to the. To the build. I've followed your career. I listened to you, had a little sort of description of the company and what you're doing in. I think it was the. The fifth episode. Tell people what you're doing and sort of what your approach to journalism is. Why does Longview exist and what are you trying to do?
- [96:51]
Andy Mills
Well, thank you for giving me a chance to say it. In some ways, it's very simple. We're just obsessed with context, and we think it's one of the biggest things missing from stories today. It appears as if New York City has now elected a democratic socialist mayor. Where are all the stories helping us realize, like, why does America have such a weird relationship with socialism? What other times in history was socialism a big deal? Wasn't it the working class that was really into socialism? Oh, that's really interesting that right now socialism seems to be resonating more with the managerial class that, like, last time this was a big thing. Hated it. And instead of finding. We have a bunch of takes about whether you should like or dislike Mandami, we have, like, camps that say, is this going to be the best thing or the worst thing? And there is nothing between the two. We are trying to do that essential part of journalism that we think is important, where you're giving people information that's truly useful to them. So they go, oh, that's why. That's a big debate. Oh, with Charles J.K. rowling is a perfect example. It's like, you listen to that series and you're like, oh, I totally get why this fight's happening. This is a big fight. This fight is actually so complicated because it's not about one thing. It's about this deep human nature thing. Oh, I get it. We could be doing that for all these fights, but the incentives of our Media landscape, and this includes the new media landscape, as you well know, is that the cheapest thing to do is to have a take and to make sure that your take is like processed well for whatever medium it needs to be a hit in. And, you know, no offense to what, like what's happening at the Free Press. And with all that, I wish them all the best. I helped Barry to start the Free Press and worked with Nelly and Susie and that I'm so proud of them and what they're doing. But I will say that sometimes because they did a lot of opinion and article and reporting and investigations, it was just really fascinating and clear to see them as an example of a successful new media company. And it was just clear that we would put weeks and weeks into an investigation into a big reported thing and we would put that out and it would do pretty well, like, oh, this is nice. But an interesting, largely just like a kind of hot take, you know, interesting opinions piece that like Barry would think up on 3 o' clock in the afternoon and then find someone to write it overnight. We'd publish it in the morning. It would do just as well and it.
- [99:17]
Steve Hayes
Or maybe better. Right.
- [99:18]
Andy Mills
Oftentimes it did better. Right. So it just pushes you to be more in the takes and opinions business when, like you, I just think we need people who are just trying to understand and report the world and be honest about, you know, whatever biases they have, but not push for their thing. And like, that's a big part of what's inspiring us. And so our, our basic plan is that we just want to publish stories in. We're starting with audio and podcasts because that's Matt Bull and I's background. But we're going to be expanding out into video and writing and whatever these, whatever new mediums get invented by AIs because we think this is like absolutely essential. It's really needed. And a part that's, I think, in connection to what you're up to as well is that we want to help people become recognizable to one another again. Like, to live in a free and open society, a pluralistic society, you have to realize that most of our differences are irreconcilable. We're not going to end up agreeing on them. And so what we can do instead is let people engage in persuasion all they want. But some of us have to be dedicated to the position of saying, let me help you recognize why someone might dare have a different view. And even if you never come to agree with them, even if they don't persuade you to come one step closer to their position. I think it's useful for journalists to at least give you insight into the fact that that person legitimately can hold that thing and you can live peacefully in a society with them. And then the last piece is that I just think too much journalism that is trying to do this right. It's not like there's no investigative journalism out there. It's not that there's no really thoughtful, careful stuff, but some of it is so boring. And I think that it's fine to try and be interesting, to have a goal of saying I. When people listen to the Last Invention, I tried to make it kind of fun, you know, at times like I'm trying to make it like it shouldn't feel horrible. And even if we're covering really intense stuff, I think a lot about how, like real humanity, like the saddest, best sad books or the best sad movies, they're also really funny, you know, and we contain multitudes and our journalism should too.
- [101:41]
Steve Hayes
Should as well. Yeah. Well, very well said. I am cheering you on. I hope you succeed. I wish there were more people doing the kind of journalism that you're doing. We'd all be better off for it. And thanks for taking the time to have this conversation and to go a little slower for somebody who's not as well versed or sophisticated a thinker on this stuff. Really, really helpful. I have, as I said, been through the first five episodes of. There are three more to come on the Accelerationists, the Scouts and the Doomers. I'm eager to continue listening and I hope folks who've enjoyed this conversation will check it out as well.
- [102:21]
Andy Mills
Yes, the Last Invention. Sam.