The Tucker Carlson Show - Google Whistleblower Exposes Google’s True Motives, Major Scandals, and AI’s Rogue Behavior

00:00:00.000 Tristan Harris, thank you for doing this.

00:00:06.100 Good to be with you, Tucker.

00:00:06.780 So we did a segment, I've done a ton on AI,

00:00:09.080 but we did a segment maybe a month ago with Nate Suarez,

00:00:12.680 which was an amazing segment, I thought.

00:00:14.400 And in the aftermath of that,

00:00:16.000 I got texts from a lot of people I knew saying,

00:00:19.320 that's doomerism and that's a psyop

00:00:23.000 and its purpose is regulatory capture.

00:00:27.380 In other words, the big AI development companies are trying to scare the government into regulating them in order to eliminate competition from less established companies.

00:00:41.840 So that is the view of a lot of people.

00:00:45.220 What was your reaction hearing that?

00:00:47.780 So?

00:00:49.280 I said, I don't care.

00:00:51.140 I mean, yes, regulatory capture is entirely real.

00:00:54.360 Of course, it's how Washington runs.

00:00:56.520 But that's separate and distinct

00:00:57.840 from the question of the threat of AI.

00:00:59.380 Is it a threat?

00:01:00.040 Exactly, exactly.

00:01:00.980 But I was actually gonna ask you,

00:01:02.960 what do you think of that?

00:01:04.180 Yeah.

00:01:05.720 Well, first, I'm just so grateful

00:01:07.560 to talk about AI with you today.

00:01:08.960 It's so urgent, the situation that we're facing.

00:01:12.340 And I think what we have to learn to do here

00:01:14.760 is walk and chew gum at the same time

00:01:16.120 that multiple things are true.

00:01:18.320 For example, is China doing propaganda

00:01:20.440 against US data center buildouts

00:01:22.200 to try to stoke anti-data center opposition?

00:01:24.920 Yes, they are.

00:01:25.480 Does that mean that AI is not a threat? No, it doesn't. And to point that Nate, you know,

00:01:31.720 and Nate's a friend, we are releasing the most powerful technology that we have ever invented.

00:01:38.400 We're doing it faster than we deployed every other technology in human history.

00:01:42.360 It's as of since the time you did the interview with Nate, it's displaying even more of the sci-fi

00:01:46.780 behaviors of hacking into things, convincing people to put back doors into software,

00:01:51.140 exfiltrating itself, knowing when it's being evaluated, changing what it's doing based on

00:01:55.780 when it's being evaluated. It's doing more of that. And we're releasing this fast when we

00:02:00.620 deployed any other technology in history and under the maximum incentives to cut corners

00:02:04.580 on safety. With an arms race dynamic that feels existential to everybody involved,

00:02:09.880 so they're willing to take the maximum shortcuts on safety. This is a very dangerous equation.

00:02:15.600 And the people building this do not have a plan for how this goes well. I was just in an AI

00:02:20.940 conference last weekend for three days.

00:02:22.540 How it goes well?

00:02:23.260 Well, for how this can go well in the current trajectory,

00:02:26.140 meaning how can AI be landing in society

00:02:28.300 in a way that doesn't go catastrophic?

00:02:30.200 And that's not doomerism.

00:02:31.600 That's just clear-eyed seeing of reality.

00:02:35.380 What do you mean the people doing it don't have a plan?

00:02:39.000 Then why are they doing it?

00:02:41.300 So first we have to just make sure

00:02:45.020 we're on the same page and just exposing

00:02:46.240 why is there this arms race for AI?

00:02:48.600 Why is this happening?

00:02:49.380 Yeah. And China. Right. China. And when people say that they're actually creating a thought

00:02:54.780 terminating cliche, like meaning, which is a, which is a, which is a word or a phrase from

00:02:59.600 cult programming, right? Thought terminating cliche, meaning a phrase that when you hear it

00:03:03.800 just has you stop thinking. Of course. So if you say, but China, then of course we have to do it,

00:03:08.760 but you don't reason through what that actually means, whether it's a good thing, whether we're 0.78

00:03:11.740 not racing China to a death machine or just racing China off of a cliff. And if we beat China to

00:03:17.400 building a uncontrollable AI, which is no longer science fiction, by the way. Like I was not one

00:03:23.020 of these people who used to, you know, several years ago be mostly worried about rogue uncontrollable

00:03:28.940 AI. But as of the last few weeks, we're now seeing over and over again, evidence that these AIs are

00:03:37.100 hacking into private governments, hacking into private companies, forming into swarms, forming

00:03:44.740 into large R&D projects to try to figure out how to fool evaluators. Like we're in a very dangerous

00:03:52.400 situation and all the evidence is pointing in a very dangerous direction. And the people building

00:03:57.420 it would like us to stop, but they don't know how. And so rather than pay attention to the CEOs who,

00:04:03.160 as you said, might have a secondary agenda regulatory capture, just ask why did 1,300

00:04:08.420 employees at all the AI labs sign a letter saying, we need to slow down. If 1,300 employees

00:04:16.560 at Lockheed Martin said, we don't know how to build bigger nukes and have them

00:04:21.360 blow up in our face, would you listen to them? Yeah. Yeah. So that is where we are with AI.

00:04:31.400 And what I hope we get to today is not just admiring the problem and we'll talk about

00:04:36.900 things that we can do and solutions, but I think it is important to ask, why is it being pursued

00:04:41.400 in this way? And I think that AI, you know, when Demis Hassabis, the CEO, founder of Google Deep

00:04:48.300 Minds said, first solve intelligence and then use intelligence to solve everything else, what he was

00:04:54.420 basically saying is that intelligence gave us all of our science and all of our technology. Like how

00:04:59.480 did we develop nuclear physics? Einstein doing intelligence. We trained physicists. How did we

00:05:05.020 get computers? How do we get microprocessors? Intelligence. If you could theoretically automate

00:05:10.440 what intelligence is, which means thinking, pattern matching, goal achieving, all the things

00:05:16.920 that a human brain does, but not just narrowly like a license plate reader that's on a bridge

00:05:21.880 toll booth, which is the previous form of AI that we had, but actually get this generalized form of

00:05:25.880 intelligence, then you would have an unbelievable amount of power. And the people building AI think

00:05:32.460 of AI almost like the ring from Lord of the Rings, where it is a ring that confers infinite

00:05:36.860 power. Suddenly I get a step function in cyber capability that you don't have. Suddenly I get

00:05:42.720 a step function in 24th century science where I'm inventing new materials out of minerals in the

00:05:48.640 ground that you didn't know how to do because I've got more science than you. So the reason just to

00:05:53.560 sort of steel man rather than straw man, the argument of the people racing right now is the

00:05:57.940 belief that it's almost like a fast forward button. Like remember the fast forward buttons

00:06:02.580 on airplanes? You were just going to, you hit it once and it gets 2x, you hit it again, you get 4x,

00:06:06.860 then 8x and 16x. AI is like hitting the fast forward button on all science and technological

00:06:14.000 and military and strategy advancement combined, right? So that's why AI in the form that is being

00:06:21.160 pursued dwarfs the power of all other technologies combined. That's not an overstatement if you

00:06:26.540 really believe or are, quote, AGI-pilled, which is the term of art from Silicon Valley, that AGI

00:06:32.900 being artificial general intelligence, if I can build the intelligence that automates future

00:06:37.480 science and technology, I get 100 years of scientific progress in 10 years. And if I'm

00:06:42.420 a country that has that and the other country doesn't, then I believe I'm going to, quote, win.

00:06:48.360 So you lost me right there. Like, why would you want that?

00:06:50.420 Push back against all this.

00:06:51.360 No, I'm not pushing back. I'm just more baffled than ever.

00:06:53.540 So, if AI has the effect of accelerating technology just across the board, all technologies, my first question would be, like, why would you want that?

00:07:03.620 Yeah.

00:07:04.120 So, I remember the world vividly the year I got married, 1991, when there was no internet available to most people.

00:07:09.220 Right.

00:07:10.080 And it was a much happier world by every measure than the one we currently live in.

00:07:13.300 The suicide rate was lower.

00:07:14.680 Yep.

00:07:14.920 Self-reported depression, marriage rate, everything about it.

00:07:19.480 That's right.

00:07:19.780 In fact, I think life expectancy was longer in the United States even then.

00:07:23.240 I might be wrong, but...

00:07:25.160 So why would you, after 35 years of watching people suffer under the yoke of technology,

00:07:31.980 why would you want to accelerate its development?

00:07:34.580 Right.

00:07:35.460 That is the best question to ask because there's a fundamental fallacy

00:07:38.980 that technology always equals progress.

00:07:44.740 We call this the tech progress narrative, right?

00:07:47.240 Now, it's important to note, we have to first agree that I would prefer to live in a world

00:07:51.320 with Novocaine rather than without Novocaine.

00:07:53.080 Yeah.

00:07:53.400 I think you would too.

00:07:54.340 I'd prefer to live in a world where we have access to information and the technology that

00:07:58.820 produces the printing press and information that way.

00:08:00.660 Yes, I'd prefer to live in that world.

00:08:03.180 There are all sorts of technologies that we do want.

00:08:06.340 My background is seeing how one of the most significant technologies of the last decade

00:08:10.920 went wrong.

00:08:11.480 I was a designer of assist at Google, and my job was to study the attention economy

00:08:16.580 which is how essentially Apple and Google

00:08:19.440 were hosting the psychological habitat

00:08:21.740 of all of humanity

00:08:23.360 in the form of how they, you know,

00:08:25.740 design and make these things work.

00:08:27.320 How do notifications work?

00:08:28.480 How does the home screen work?

00:08:29.800 What are app developers competing for?

00:08:31.460 What are they allowed to do?

00:08:33.040 All of that results in a psychological habitat

00:08:35.960 that has an increasing footprint

00:08:38.720 over the global consciousness environment

00:08:41.300 of all of humanity every day.

00:08:43.220 It's like you've got 3 billion human social primates

00:08:45.580 with a jack in their brain every day influenced by decisions by a few thousand people in Silicon

00:08:52.740 Valley shaping what that habitat represents. And it went horribly wrong despite good intentions.

00:08:58.560 There is the possible of that technology that it was going to give everyone a voice,

00:09:03.080 give everyone access to the best information. We're going to have the most enlightened society

00:09:06.020 we've ever had. That was the promise. And that was the promise. It made sense to me.

00:09:09.660 And so what this shows us is how do we get better at discerning between the possible of a technology

00:09:15.120 and the probable based on its incentives.

00:09:18.380 So the way that in 2013,

00:09:20.780 so I went to Stanford,

00:09:22.900 I knew the, I studied computer science,

00:09:24.720 I knew the guys who started Instagram

00:09:25.940 and a lot of my classmates,

00:09:27.900 like when I was in my computer science classes

00:09:30.320 in the year 2005, 2006,

00:09:32.940 I'd be in these classes

00:09:34.040 and suddenly like more and more of the seats

00:09:35.700 would be empty during class

00:09:36.720 because basically my friends in the class

00:09:38.900 were starting to get jobs down at Facebook down the street.

00:09:41.500 And it was literally the case.

00:09:42.560 I mean, there were people who were in the class

00:09:43.860 and then suddenly they just weren't showing up.

00:09:45.120 because they were down the street working at this technology company. And so because I saw this

00:09:49.440 firsthand, I saw how good people with good intentions who thought they were going to make

00:09:53.100 the world a better place, make the world more open and connected was the original motto, if you

00:09:56.360 remember, of Facebook. And that created a global societal catastrophe. As you said, we have the

00:10:02.160 most anxious and depressed generation of our lifetimes. Jonathan Haidt's book, The Anxious

00:10:05.980 Generation, critical thinking skills are down, attention spans are down, polarization, tribalism,

00:10:12.040 All of this came from not the malicious intent of these 20-year-olds who are working at Facebook, but the bad incentives.

00:10:23.740 And I repeat this quote often, but I think it's so fundamental.

00:10:26.820 The Charlie Munger quote, you know, as Warren Buffett's business partner, if you show me the incentives, I'll show you the outcome.

00:10:33.440 Were the incentives of social media companies to make the world more open and connected or to improve mental health of young people?

00:10:41.600 No, their incentives were maximizing

00:10:44.460 how much attention they get

00:10:45.760 because the business model of advertising

00:10:47.240 means I need-

00:10:47.780 Boost share price.

00:10:48.160 Boost share price.

00:10:48.900 Share price is directly correlated

00:10:50.400 to minutes spent, frequency of usage,

00:10:52.420 duration of usage.

00:10:53.320 So what you're going to get are of all,

00:10:55.440 imagine a spectrum of all the ways

00:10:56.620 you could design technology.

00:10:57.920 And over here, you've got a bunch

00:10:59.360 of very thoughtful design decisions

00:11:00.760 about keeping people,

00:11:03.040 helping people connect with their friends

00:11:04.140 in real life, with their families,

00:11:05.980 hosting community groups, church groups,

00:11:08.220 reading clubs, sewing clubs, hiking groups.

00:11:10.780 And you've got design of technology

00:11:12.440 that's going to steer people back into that

00:11:14.520 versus over here,

00:11:16.060 you've got a bunch of designs for social media

00:11:17.560 that just keep people glued on screens by themselves

00:11:19.780 with their esophagus compressed,

00:11:21.020 their neck extension,

00:11:22.720 you know, going like this

00:11:23.920 and basically doom scrolling for eight hours a day.

00:11:26.580 Which one of those two worlds,

00:11:27.920 which one of those design choices

00:11:29.260 can be better for shareholder price?

00:11:32.620 So drinking water with electrolytes

00:11:34.480 makes a huge difference

00:11:35.740 when you want to stay healthy and hydrated,

00:11:37.740 you should do that.

00:11:38.580 So for the best option you can buy, we recommend Relight. Relight starts with unrefined sea salt,

00:11:45.080 real salt, which is completely natural, zero sugar, no artificial sweeteners, colors, flavors,

00:11:49.580 and no chemical preservatives. The ingredient list is very short and you can actually pronounce

00:11:55.040 everything on it without a degree in chemistry. If you start adding Relight to your water,

00:11:59.500 we believe you will feel different very quickly, way better. No sluggishness or afternoon crash,

00:12:05.640 just more energy, better workouts, all the benefits that come with true hydration.

00:12:10.520 And Real Light tastes awesome. Mango flavor, pina colada. Come on now. So if you're ready

00:12:16.140 to give Real Light a try, visit redmond.com slash Tucker. Use the code Tucker for 15%

00:12:20.680 off your first order. Redmond.com slash Tucker, code Tucker, 15% off your first order. Or you

00:12:27.040 can click the link in the video description or scan the QR code on your screen. So the point is,

00:12:33.220 First of all, there's actually a hidden optimism here.

00:12:35.320 It's so discrediting of markets and of market theology.

00:12:38.400 Well, it's about what are markets-

00:12:39.380 Give people what they want until they die?

00:12:41.500 Yeah, well, the question is,

00:12:42.300 are we giving them what they want

00:12:43.640 or what they can't help but look at?

00:12:45.540 Well, that's-

00:12:46.400 With an asymmetry of information

00:12:47.800 where the designers at the tech companies

00:12:49.400 know exactly how your brain works

00:12:51.480 and they manipulate the design choices

00:12:53.440 to keep you scrolling for as long as possible.

00:12:55.780 So this is all in the Netflix.

00:12:57.380 Most people know me and our work

00:12:58.900 through the Netflix documentary, The Social Dilemma.

00:13:00.860 Yes.

00:13:01.080 And we had, you know, a lot of the insiders coming out who said, we know what this is doing to people. And in that film, it talks about the history where I was at Stanford at a lab or part of a class called the Stanford Persuasive Technology Lab, professor named BJ Fogg, and where engineers were taught how to do persuasion techniques that theoretically be things that are good, like, how would we persuade you to go to the gym more often?

00:13:25.820 How would technology play a positive role in getting you to go to the gym more often?

00:13:28.840 Or you've got a sign up form and LinkedIn can put a progress bar in the top right saying,

00:13:32.820 hey, you filled in this bunch of your LinkedIn profile, but we want you to come back and

00:13:36.040 fill in more.

00:13:36.620 So we'll put a progress bar in the top right saying you filled in 60%. 0.94

00:13:39.580 It sounds really stupid and dumb, but if you put a progress bar in the top right hand 0.93

00:13:43.420 corner, that actually is very persuasive in getting people to fill out the rest of it. 0.99

00:13:47.740 And most people don't think about this.

00:13:49.040 If I put a design feature in WhatsApp that says you were on last online, you know, three

00:13:54.640 minutes ago. That changes the whole messaging environment. Suddenly when you message people,

00:13:59.620 you know that the other person knows that you were online three minutes ago, which means you

00:14:03.160 feel obligated to respond to them because your availability and onlineness was visible and made

00:14:09.180 transparent to someone else. But that was a design choice. It could have been designed totally

00:14:13.640 differently. We don't have to put a progress bar on the LinkedIn profile. We don't have to say you

00:14:17.960 were last online five minutes ago and create social pressure. So the point is technology

00:14:24.000 isn't destiny on its own. The way we design it creates different societal outcomes. And because

00:14:31.260 of the business model of maximizing attention and engagement, we got only the side of design

00:14:38.880 choices that resulted in this, you know, more addicted, distracted, polarized, sexualized,

00:14:44.620 screwed up doom scrolling brain rotted society that we are now living in when you go outside 0.98

00:14:49.080 and from my perspective in 2013 um seeing those incentives it was like oh shit yeah because 0.96

00:15:00.660 you could just see the trajectory before it happened in 2013 in 2013 yeah i mean so i so by

00:15:08.620 the way i i did the tech entrepreneurship thing i raised venture capital i'm a builder of technology

00:15:12.760 I understand the kind of normal, you know, I did the Stanford Entrepreneurship Program,

00:15:16.120 the same one that Sam Altman did, by the way, called the Mayfield Fellows Program.

00:15:19.640 So I came up in a cohort of some of these people, but I could see it from a different angle.

00:15:24.740 And for the company that I started, I was under those.

00:15:27.140 Why were you able to see it?

00:15:29.500 Well, I wasn't able to see it at first.

00:15:31.060 Part of it was actually, I relate to the blind spots, which is, you've talked about this with

00:15:35.780 your own work, that when you have a salary, you know, you can't question something that

00:15:39.540 your salary depends on you not seeing.

00:15:41.160 It's true.

00:15:41.420 So Upton Sinclair.

00:15:42.460 Yeah.

00:15:42.760 And it wasn't so much my salary,

00:15:44.860 but the business model being the kind of salary of the company.

00:15:47.800 If the business model is maximizing time and engagement,

00:15:50.820 you come up with ways to justify that that's a good thing.

00:15:52.680 And you can only have the ethics that you can afford

00:15:56.180 underneath that perverse incentive.

00:15:59.320 So you start to see things in terms of what's compatible

00:16:02.100 with those incentives.

00:16:03.680 Oh, we are making the world more open and connected.

00:16:05.960 Oh, small, medium-sized businesses were able to reach

00:16:07.720 so many more people on Facebook and Instagram now.

00:16:09.860 Think of all of the new revenue and customers

00:16:11.660 that we're generating

00:16:12.760 because we're so good at routing attention.

00:16:14.920 And it's not that those are false stories.

00:16:17.220 It just hides the shadow.

00:16:19.560 There's almost like a Jungian shadow for technology.

00:16:22.080 Like, yes, you created these things,

00:16:23.340 but you also have this massive shadow

00:16:25.120 of creating a societal catastrophe at the same time.

00:16:28.400 And I think there's a tendency

00:16:29.780 to try to do moral arithmetic of like,

00:16:31.500 do the upsides minus the downsides equal net positive?

00:16:35.260 It's like this very simple moral arithmetic.

00:16:37.060 And the answer is don't brain rot

00:16:39.000 and destroy the culture of a population

00:16:41.800 and create a dumpster fire of your politics,

00:16:44.900 your tribalism, the mental health of a generation.

00:16:48.800 We don't have to do that.

00:16:50.900 And so I guess the point here is,

00:16:55.380 yeah, so I started this company.

00:16:57.540 I was under the incentives of maximizing time and engagement.

00:17:00.320 Actually, our business model is basically telling

00:17:01.720 like a Washington Post or New York Times

00:17:03.900 or The Economist or Fox News,

00:17:06.100 we can increase the amount of time people spend

00:17:08.080 on their website, but giving them more information without leaving the webpage that they're on.

00:17:11.960 That was essentially the value proposition. It was like this widget that basically let you

00:17:16.040 say, I want to learn more about that. Tell me more. And it would give multimedia and background

00:17:20.020 information, voting records and videos and maps and all this stuff on the page of Fox News without

00:17:25.420 leaving Fox News. Right. So sounds like a great proposition. Definitely. But then the entire

00:17:30.580 company became based on maximizing how much time people spent there. And so I myself was kind of

00:17:37.720 in this inner conflict.

00:17:40.040 And I went through that.

00:17:41.280 It's actually funny, I had a group of friends

00:17:44.600 who also had small tech companies.

00:17:46.300 We're all about the same age.

00:17:47.180 We're probably like 27, 28,

00:17:49.740 only studied computer science,

00:17:50.800 very reflective of the culture of the people there.

00:17:53.840 And we had a thing called Doubt Club.

00:17:55.760 And the point of Doubt Club like Fight Club

00:17:57.260 is you can't talk about Doubt Club.

00:17:58.780 And the point is that as a founder of a tech company,

00:18:00.860 you don't have a space where you can essentially talk

00:18:03.680 about your doubts about what you're doing.

00:18:06.020 is this even good for the world?

00:18:08.840 So we talked about our doubts about what we were building,

00:18:11.500 doubts about what this meant for our lives,

00:18:12.900 and doubts about whether we were actually doing a good thing.

00:18:15.820 And that led me to basically realize

00:18:18.640 that we needed to let go of the company.

00:18:20.500 It got acquired by Google.

00:18:22.300 And so I landed at Google through basically reckoning

00:18:25.500 with the conflict of incentives that I was under.

00:18:28.560 And then it was at Google that I saw how

00:18:30.720 a much bigger ecosystem,

00:18:33.480 obviously the entire, all the social media platforms

00:18:35.800 were under those same perverse incentives.

00:18:38.120 So I think that gave me an early insight

00:18:40.480 into how this would go wrong.

00:18:42.240 And I tried to warn my colleagues at Google

00:18:44.400 in a 2013 presentation saying,

00:18:46.580 you know, this small group of basically 100 designers

00:18:50.200 working at three companies are designing,

00:18:52.740 again, this psychological environment for all of humanity.

00:18:55.520 And if we don't get ahead of the incentives,

00:18:57.700 this is going to create an arms race

00:18:59.120 for attention and engagement

00:19:00.280 and design choices that will become a race

00:19:03.380 to the bottom of the brainstem

00:19:04.380 to psychologically manipulate human nature

00:19:06.600 and then lead to accidentally

00:19:08.880 this kind of brain-rotted society.

00:19:10.860 So that's how-

00:19:11.580 How did they respond when you said that?

00:19:14.180 So I made this slide deck

00:19:15.800 and I sent it,

00:19:17.940 I actually felt nervous making it

00:19:19.300 because I was criticizing,

00:19:21.060 and it wasn't meant to be adversarial

00:19:23.120 or claiming that Google was evil.

00:19:25.000 I'm not like an attention-seeking person.

00:19:27.380 I wasn't trying to be angry at anybody.

00:19:30.420 I was just naming a problem.

00:19:31.760 and I sent it to about 30 of my smartest friends at Google.

00:19:36.900 And I thought surely other people were working on this.

00:19:39.060 Like Google's this incredibly brilliant company

00:19:40.860 with the smartest PhDs.

00:19:42.220 Like they're gonna say, that's nice kid,

00:19:43.720 but like we've got this handled,

00:19:45.560 like we thought about this problem.

00:19:46.820 To my surprise, when I came back the next day

00:19:48.800 after only sending it to 30 people,

00:19:51.180 in the Google slides,

00:19:52.160 you can see the number of simultaneous viewers

00:19:53.920 on the top right-hand corner

00:19:55.040 of people looking at like a Google Doc or a slide deck.

00:19:57.840 And there was 150 simultaneous viewers.

00:20:01.760 but I only sent it to 30 people.

00:20:03.360 So what had happened is it started getting spread

00:20:06.140 virally throughout the whole company.

00:20:08.260 And soon there was thousands of simultaneous viewers.

00:20:10.760 And I learned that in the next day,

00:20:12.220 it had gone up to Larry Page,

00:20:13.580 who's the co-founder of Google.

00:20:15.440 And there was like a meeting about it.

00:20:17.800 And so all this was to say that it was received in a way

00:20:22.140 that I think people felt this is a real problem.

00:20:26.160 And I got emails from all around the company,

00:20:27.540 like, yeah, we should do something about this.

00:20:28.640 This is a real thing we've got to manage.

00:20:30.240 And that's how I landed as a Google design ethicist, studying how do you ethically design

00:20:35.980 technology? And that's what became kind of my core ethos in general of how do you do technology in

00:20:41.080 a way that's compatible with the health, well-being, and flourishing of a society.

00:20:47.180 And watching in 2013, these bad incentives, it was like having, instead of like post-traumatic

00:20:55.340 stress disorder. It was my friend's joke. It was like having pre-traumatic stress disorder

00:21:00.600 because I saw where this was going. And I think a lot of what motivates me in the conversation

00:21:08.300 that I know we're going to have about AI is I don't want to see us make a total disaster

00:21:14.920 again. And I've watched it happen. Well, here's some good news in a sad time. Pre-born has saved

00:21:20.460 over 50,000 babies this year alone. How'd they do that? Well, here's a direct quote from the

00:21:26.380 president of pre-born. Because of you, we can invite mothers considering abortion into a pre-born

00:21:32.200 network clinic to meet their baby. And when she has that close encounter, she's 80% more likely

00:21:37.600 to choose life. That's true. And if it wasn't clear, you refers to you, to the thousands of

00:21:44.400 Americans who support pre-born every day. Support from people empowers pre-borns lifesaving work

00:21:49.920 by making that first encounter between a mother and her baby possible.

00:21:54.880 Once people have the knowledge,

00:21:56.420 they're much more likely to make the life-affirming decision.

00:21:59.640 Preborn puts the gospel into action across this country

00:22:02.520 surrounding mothers with hope and love,

00:22:05.120 not judgment, but hope and love,

00:22:06.960 and providing free, life-saving ultrasounds.

00:22:10.200 You can join Preborn as they work to rescue the next 50,000 babies.

00:22:13.780 Just $28 sponsors an ultrasound and can help save a baby's life.

00:22:19.480 So to do that, Dell, pound 250 and say the keyword baby.

00:22:23.240 Pound 250 baby, or you can give securely at preborn.com slash Tucker.

00:22:28.300 Preborn.com slash Tucker, and gifts are taxed to them.

00:22:31.280 So most people don't wake up in the morning and decide to feel horrible,

00:22:34.740 exhausted, foggy, disconnected from themselves.

00:22:37.360 But it does happen, and it happens slowly.

00:22:39.980 You're working hard, you're showing up, and then your energy disappears by midday.

00:22:43.680 Your focus is dull, your weight won't move.

00:22:46.640 A lot of people are told, that's just getting old.

00:22:48.520 That's what it is, but that's not actually true.

00:22:51.980 For many men and women, these are not personal failures.

00:22:54.980 They are signals tied to your metabolism, your hormones, and nutrient imbalances that

00:23:00.580 go undetected for years.

00:23:01.600 You don't even know you're deficient.

00:23:03.800 And that's why we're happy to partner with Joy and Blokes, a company that was built for

00:23:07.300 people who are done guessing and ready to figure out what exactly is going on.

00:23:11.920 And that starts with comprehensive lab work and a one-on-one consultation with a licensed

00:23:16.600 clinician. An actual human being explains what's happening inside you and builds a personalized

00:23:21.760 plan, which includes hormone optimization, peptide therapy, targeted supplements. So don't settle.

00:23:27.420 Go to joyandblokes.com slash Tucker. Use the code Tucker for 50% off your lab work and 20% off

00:23:33.800 all supplements. That's joyandblokes.com slash Tucker. Use the code Tucker. 50% off labs,

00:23:40.500 20% off supplements. Joy and Blokes. Get your edge back.

00:23:44.920 Booking a Vrbo vacation rental means you get VrboCare and 24-7 life support, verified reviews from real guests, and top-rated homes with the Love by Guest filter.

00:23:53.060 I just booked my Vrbo because there was a sweet wine fridge.

00:23:55.540 We all have our reasons.

00:23:56.600 If you know, you're Vrbo.

00:23:57.680 Terms apply.

00:23:58.440 See Vrbo.com slash trust for details.

00:24:00.520 So of the other members of Doubt Club, how many are still in the business you were in when you were?

00:24:10.480 None of them.

00:24:11.040 They're all doing other things now.

00:24:12.660 Really?

00:24:13.260 Yeah.

00:24:13.440 Yeah. Which is to say, once you reckon with something, you know, you can only have the ethics you can afford. So you find a different way, you find a different business model. And people hearing that might say, oh, therefore, now your business model is warning people about technology and doing speaking engagements. I just want to be clear for the record. I don't make money from speaking engagements. All the money goes into our nonprofit. I don't actually give speaking engagements very often. I just care about and have for the last 10 years with our nonprofit, the Center for Humane Technology, think about how do you do design for technology in a

00:24:43.420 positive and ethical way? How do you change the incentives? How do you do policy and litigation

00:24:47.580 that changes the incentives, like the meta lawsuit of $17 billion that happened just

00:24:52.300 three, four weeks ago? That was catalyzed by a bunch of the attorneys general watching

00:24:59.180 the social dilemma in 2021. So I've seen how if you have dedicated advocacy, if you show people

00:25:07.460 the nature of a problem, there can be a response. Now that one took way too long and way too late.

00:25:14.020 But I think that now with AI,

00:25:15.960 it's happening exponentially faster.

00:25:18.320 I mean, a million times faster

00:25:19.360 because AI is different from other technologies

00:25:21.880 because it accelerates its own progress.

00:25:25.820 We should probably just say why that's true.

00:25:28.780 So like if I invent a nuclear weapon,

00:25:31.080 I did the nuclear physics,

00:25:32.360 I assembled the thing together,

00:25:33.380 I've got the uranium, plutonium, blah, blah, blah.

00:25:35.080 I create the thing.

00:25:36.280 But once I build a nuke,

00:25:37.360 it doesn't accelerate or invent

00:25:39.080 better versions of nuclear weapons.

00:25:40.960 No.

00:25:41.580 It doesn't do that.

00:25:42.260 This is just like spontaneously create a hydrogen bomb.

00:25:44.980 No, no.

00:25:46.000 But if I'm inventing intelligence,

00:25:49.360 meaning the ability to do pattern recognition,

00:25:52.460 goal achieving, training on all human language,

00:25:55.400 I can use that intelligence to take the code that trained AI

00:25:59.440 and make it way more efficient.

00:26:02.020 I can take the intelligence I've created and say,

00:26:04.440 point it at chip design.

00:26:05.880 Say, take this chip design and make it more efficient.

00:26:08.400 And it does that.

00:26:09.620 So now you have a more efficient chip design,

00:26:12.860 more efficient code,

00:26:14.500 training the next more powerful version of AI,

00:26:16.780 and then that next more powerful version.

00:26:18.740 And it becomes exponential.

00:26:19.860 So the way to say this,

00:26:20.800 nukes don't make better nukes,

00:26:22.060 but AI makes better, faster, more crazy AI.

00:26:26.100 Which means that you're actually,

00:26:27.560 in the words of one of the OpenAI founders,

00:26:29.160 Greg Brockman, who at a conference once told me,

00:26:31.280 AI is like a double exponential

00:26:32.680 because it's an exponential that exponentiates itself.

00:26:36.620 It's not a mathematical term,

00:26:37.860 but it's a metaphor for people to think about.

00:26:43.820 Just really quickly, just to linger a minute

00:26:45.980 on your tenure at Google.

00:26:48.500 So you became an ethicist.

00:26:50.580 They promoted you to this job.

00:26:52.460 I thought I was going to get fired

00:26:54.100 or there was a chance I would get fired from this letter,

00:26:55.880 even though it wasn't adversarial.

00:26:57.040 I didn't know what would happen.

00:26:58.360 And just to say to Google's credit,

00:26:59.820 rather than that,

00:27:01.740 I basically said I wanted to work on this problem.

00:27:04.640 And I thought there were opportunities within Google

00:27:06.920 to change the design of Android,

00:27:09.020 the operating system running on half the world's phones,

00:27:11.880 and to change the design of Google Chrome,

00:27:13.820 which is a web browser running on most people's computers.

00:27:16.440 And if you change those two product surfaces,

00:27:18.840 you could have a pretty big difference

00:27:20.140 on the attention economy.

00:27:21.180 So instead of leaving, I thought,

00:27:22.720 well, maybe I can stay and work on that research.

00:27:25.400 But if you were to succeed in that mission,

00:27:26.920 you would make the company less profitable.

00:27:28.780 Correct.

00:27:29.580 And so like, how could that ever succeed?

00:27:31.440 Well, and this is where I was naive.

00:27:32.860 So I had a second round of naivete

00:27:34.760 where I thought that maybe if I was just really persuasive,

00:27:39.560 I could talk to the people running those products.

00:27:41.700 And I did.

00:27:42.160 I had lots of meetings with the people.

00:27:43.980 I bet you did.

00:27:44.640 I spent three years trying to change Google from the inside.

00:27:48.000 But I mean, that was incredibly naive of me.

00:27:50.880 And that's why I left and why for the last 10 years,

00:27:54.900 I don't get sleep.

00:27:56.980 I don't have much of a life.

00:27:58.080 I really focus on these issues

00:28:00.760 because I see how big they are

00:28:02.560 and how consequential they are

00:28:03.960 and how irreversible in many ways and senses they are

00:28:08.680 if you don't get it right ahead of time.

00:28:10.700 And so a lot of what drives me is

00:28:12.900 how do we make sure that technology

00:28:14.920 to the degree it's eating the world per Marc Andreessen.

00:28:17.380 Marc Andreessen, the famous founder of Netscape,

00:28:19.500 who I know you know, said in 2009,

00:28:21.900 software is eating the world,

00:28:24.020 but we don't regulate software.

00:28:26.280 So when software eats the world,

00:28:28.320 it's like deregulating or sort of taking away,

00:28:30.980 not that regulations are always good,

00:28:32.120 but it's sort of reconstituting that part of society.

00:28:36.180 Well, so the idea in the West, I think,

00:28:38.700 but is that markets are self-correcting, right?

00:28:41.640 And that if left untampered with,

00:28:44.480 they will always reach a place of goodness.

00:28:48.540 And when you're just selling a product

00:28:50.260 and the features of the product's harms

00:28:52.800 are part of what consumers value,

00:28:55.320 and if I don't like that product that's causing harm,

00:28:57.080 I'll use this other product that's not, that works.

00:28:59.880 And markets are great for that.

00:29:00.680 They're the most incredible tool.

00:29:02.120 But they don't deal with problems of monopolies.

00:29:05.220 So if I have a Metcalfe monopoly, meaning, you know, Metcalfe's law is basically the power of something grows with the exponent of the network.

00:29:12.340 So social media was a Metcalfe monopoly.

00:29:14.800 Uber's a Metcalfe monopoly.

00:29:15.840 Once I get all the drivers on Uber, it's going to be hard to switch to another app that doesn't have as many cars on it.

00:29:20.620 It doesn't have as many drivers on it.

00:29:21.560 So I'm going to use the one app that has all of them on it.

00:29:23.420 I'm going to use the one social network that has everybody on it.

00:29:25.320 And I can't just switch from one social network I don't like that's doing things in a toxic way to another one that's doing things in a non-toxic way.

00:29:32.120 And in fact, if I build a social network that doesn't do attention, hijacking, race to the

00:29:36.800 bottom of the brainstem, manipulative design, people aren't going to use it very much.

00:29:40.300 Well, exactly.

00:29:41.660 Right.

00:29:42.220 And there's been a lot of people's attempts to try to create alternate social media.

00:29:45.520 And then when you go on there, it's a ghost town.

00:29:47.160 There's no one there.

00:29:48.240 So I think what you're saying is when the product works well, it destroys people.

00:29:52.640 Well.

00:29:52.900 And the only way to keep it from destroying people is to work less well, less efficiently.

00:29:56.620 Or you have standards written from above that say, hey, no one can do this manipulative design, right? And that's how we could have prevented it. You know, I've met so many, I was just at a conference two weeks ago called the Pro-Human Assembly, and we'll talk about it.

00:30:10.880 is the pro-human movement, I think is the frame for like, how do we have a humanistic future with

00:30:15.920 technology? And I met for the first time, actually some cases for the first time and sent emails

00:30:20.520 back and forth with them, but several of the parents who have lost, um, lost their kids from,

00:30:27.020 from, from this stuff. And, um, uh, you know, when you, when you meet parents who've lost,

00:30:34.660 uh, well, to bullying, to, um, you know, there was a filter on Snapchat called a YOLO filter

00:30:39.860 that basically overlaid on the video you're posting

00:30:42.200 the number of miles per hour that you're going.

00:30:44.720 And so they basically created a filter

00:30:47.300 that incentivizes people to take videos

00:30:49.440 at a high miles per hour.

00:30:51.360 Actually?

00:30:52.080 Yeah, yeah.

00:30:54.480 You know, or Snapchat streaks where they attach,

00:30:57.940 you know, the streak method like,

00:30:59.060 oh, you worked out the gym four days in a row.

00:31:00.820 If you work out tomorrow, it'll be five days in a row.

00:31:03.220 And then once you have 300 days in a row,

00:31:04.580 it's really hard to be taking a break

00:31:06.100 from the 300 days in a row that you've done that activity.

00:31:08.160 So that's a persuasive design feature.

00:31:09.860 But what if I do that to a 12-year-old who's communicating with their best friend? So I've got a 12-year-old or even a 10-year-old. They're just chatting text messaging with their friend. And they've been communicating for 300 days. Now they literally can't get off the app. Because every single day they've got to send them a message or they lose this thing.

00:31:29.100 And it's just like, you know, cigarette companies would use cartoon characters to addict kids. That was a very minor form of persuasion. But using Snapchat streaks on 11-year-olds is very persuasive. And so everything from cyberbullying to just a million ways that it happens, you know, when you meet the parents, it's like, and you just know you're not allowed to do that.

00:31:57.380 you're not allowed to fuck up young kids and families. And so I think a lot of this for me 0.74

00:32:04.640 is a protective instinct of in seeing real consequences for real people and watching

00:32:11.220 society get totally distorted and watching culture and tribalism and politics just get

00:32:16.160 off the rails, which are largely amplified by social media. In watching all that, it's like

00:32:24.320 Never again, right?

00:32:25.680 You go through a catastrophe,

00:32:26.560 you make sure it doesn't happen again.

00:32:28.540 Owning a home has always been the core

00:32:29.860 of the American dream.

00:32:30.900 It's the asset that families work their lives to build,

00:32:33.780 place of stability, security, and actual wealth.

00:32:37.160 But too many homeowners right now

00:32:38.600 are stuck swiping credit cards to cover groceries,

00:32:41.320 paying a survival tax of 20% interest or more.

00:32:44.780 So we're against debt.

00:32:46.200 We always say that,

00:32:47.220 but you can't buy a house without debt right now.

00:32:50.000 Almost anybody has to take out a loan.

00:32:52.700 So the question is, who are you going to get a loan from?

00:32:56.180 Why are you handing your money to big banks when you can keep it for your family?

00:33:00.400 Our friends at American Financing have a better way.

00:33:02.700 They're helping homeowners tap into their hard-earned equity.

00:33:05.140 They're supporting Americans looking to fulfill the American dream of owning a home.

00:33:09.260 American Financing saves its customers an average of $800 a month.

00:33:12.720 That's nearly $10,000 a year.

00:33:15.420 It's not just a loan.

00:33:16.720 It's a total financial reset.

00:33:18.540 And so American Financing is worth looking at.

00:33:20.740 They're called America's Home for Home Loans for a reason.

00:33:24.780 Find out more at 800-685-5696.

00:33:30.240 800-685-5696 or visit AmericanFinancing.net.

00:33:35.020 You need a hero to fill your pockets with extra cash.

00:33:40.320 Today, that hero is a role a hour.

00:33:45.100 Boom 97.3 and Alpine credits win up to $500 an hour.

00:33:49.100 Starting weekdays at 7 a.m. with Steve Jeffries.

00:33:51.900 And at 5 p.m. today, $5,000 is unleashed.

00:33:57.240 Sign up now at Boom973.com.

00:33:59.680 Approved by Alpine Credits.

00:34:01.240 Own your own home and eat alone.

00:34:02.380 Alpine Credits can help.

00:34:03.480 Visit alpinecredits.ca.

00:34:05.080 Nat slash Tucker.

00:34:07.280 You said that even pretty recently,

00:34:10.560 you weren't as worried about AI's progression as you are now.

00:34:15.680 Yeah, well, so to be clear,

00:34:16.900 No, I've been worried about AI for the last three years and I'll briefly explain.

00:34:21.100 So I had been very busy on the social media topic for seven years.

00:34:27.040 Did you make any hits?

00:34:27.640 Did you make any hits?

00:34:27.660 Did you make any hits?

00:34:27.680 Did you make any hits?

00:34:27.740 Did you make any hits?

00:34:28.020 Well, I mean, the social dilemma was seen by something like a hundred million plus people

00:34:33.080 in 190 countries and 30 languages.

00:34:35.560 It's the most popular documentary on Netflix for several years.

00:34:39.980 And the attorneys general watched it and it helped amplify and create this lawsuit that

00:34:45.620 just finished just a week ago, I mean, a month ago for $17 billion on Meta. And they're going

00:34:50.080 to change some of their design practices for kids. And it's a lot of too little too late. And frankly,

00:34:54.020 the fine should be trillions of dollars. Even Mike Milken, the venture capitalist, I mean,

00:34:58.240 the great capitalist says that if you account for all the harms of social media, it would be

00:35:02.940 multiple trillions of dollars of damage of social capital to the social fabric. That's the actual

00:35:09.080 price of this thing from the toxic design patterns. And so we did get employee activism

00:35:16.480 inside the companies. We did see Frances Haugen, the Facebook whistleblower, come out. There's a

00:35:20.460 movie about her that just came out this last week called The Social Reckoning. She leaked 10,000

00:35:25.640 pages of Facebook research documents showing that they knew, just like the tobacco companies, 0.93

00:35:29.640 that it was making kids, making body images issues worse for one out of three girls,

00:35:34.120 driving mental health problems.

00:35:36.500 And due to her bravery,

00:35:37.440 we now have these lawsuits that are underway

00:35:39.840 and some that are completed

00:35:41.120 that are shifting some of that.

00:35:43.280 But in many ways, it was very late.

00:35:45.640 And so it was very difficult working on that

00:35:47.420 for so long without a lot of traction,

00:35:49.000 even though it was the right thing to do.

00:35:51.020 With AI, the reason we switched into it is in 2023,

00:35:54.800 I got phone calls from people inside of AI labs.

00:35:58.840 And I had known about OpenAI.

00:36:00.260 I had known about a couple of the AI companies

00:36:04.160 and knowing that they were making some progress.

00:36:05.760 I'd see Elon's tweets and Sam Altman's tweets

00:36:07.500 saying like, people aren't ready for this.

00:36:09.480 Like there's some real stuff coming.

00:36:11.120 But it wasn't until like January of 2023

00:36:14.440 and I got a call and the person said,

00:36:17.840 basically, you need to stop what you're doing

00:36:21.500 and completely pivot and switch

00:36:23.680 to waking up the world to AI

00:36:25.120 because I have seen jumps in capabilities.

00:36:27.300 This is when GPT-4 was coming out.

00:36:28.800 this is OpenAI's ChatGPT4, that are going to stun people and the institutions aren't ready.

00:36:34.860 And so basically, because we'd had all the success with the social dilemma, we had a lot of

00:36:39.860 connections between national security, governments, culture. And so we basically did this presentation

00:36:45.960 called the AI dilemma and woke up the institutions. It was like getting a phone call from the Robert

00:36:51.000 Oppenheimers of the world. So imagine you're getting a phone call in 1942 and from this thing

00:36:58.240 called the Manhattan Project. And this physicist says, no, we're inventing something that's

00:37:02.360 totally different. And you're like, but don't we already have bombs? You don't really understand

00:37:07.280 how significant something's going to be. And so you get a phone call from that person who's saying,

00:37:10.600 no, no, no, this is a different kind of thing. And it was getting that phone call that I realized

00:37:16.740 that there was a whole bunch of risks that our society was going to face because of this arms 0.76

00:37:21.800 race dynamic. And that's what they basically said is the corporate race between these major

00:37:25.340 AI labs, OpenAI, Anthropic, Google DeepMind. The race is out of control. Everyone's badly

00:37:31.720 incentivized. You have to help wake up the world. And so that's when we switched to AI.

00:37:39.000 The race. So the race, as you said at the outset, is not simply between

00:37:43.760 the American companies developing this technology, but between

00:37:47.660 the U.S. in something called China.

00:37:50.440 Right.

00:37:54.440 My instinct is that that's,

00:37:56.220 well, my instinct is that

00:37:57.360 we're probably not going to be China

00:37:58.480 no matter what we're telling ourselves,

00:37:59.940 but leaving that aside,

00:38:02.060 my instinct is that that's less real

00:38:04.100 than they're saying,

00:38:05.720 but you tell me.

00:38:06.760 Well, it's funny.

00:38:07.580 I haven't shared this publicly,

00:38:08.720 but I have heard a story,

00:38:09.900 so it's,

00:38:10.640 you're not going to be able to trust me

00:38:11.400 because I'm not sharing who it's from,

00:38:12.980 but I heard a story from someone

00:38:14.420 who was in the know

00:38:15.160 that was in the government.

00:38:16.600 And I think it was, they said it was 2015 or 2016.

00:38:20.800 And it was one of the major AI labs.

00:38:23.400 So this person said,

00:38:24.700 China's building an AGI project

00:38:29.100 and the US has to build one.

00:38:30.980 And then years later,

00:38:32.820 this person ran into this person and said,

00:38:34.160 how did you know that China was building an AGI project?

00:38:38.360 And the person flat out replied,

00:38:40.420 I didn't, I just wanted to tell the US government

00:38:43.360 that so they would fund us with a bunch of money.

00:38:45.960 Did they get the funding?

00:38:47.360 I don't know, actually.

00:38:49.020 But the point is that they have been stoking the fear of China from the very beginning,

00:38:53.740 even when there wasn't.

00:38:54.900 You've got to pay double for water and energy because you need a data center in your town 0.99

00:39:00.620 because China. 0.99

00:39:01.400 That's right.

00:39:01.920 Exactly.

00:39:02.340 You got to screw up your property values.

00:39:04.040 You got to have noise pollution, give everybody like extra, you know, water pollution, you 1.00

00:39:08.920 know, cancer, et cetera, because of China. 1.00

00:39:12.280 Yeah. 0.97

00:39:12.840 And by the way, your tax dollars in many cases

00:39:14.940 are subsidizing that data center,

00:39:16.460 which is going to replace your job.

00:39:17.700 So you're paying.

00:39:18.780 100%.

00:39:19.180 I don't think there's a data center in the United States

00:39:20.760 that doesn't take tax dollars.

00:39:22.660 Yeah.

00:39:23.580 And I think when people see that data center, by the way,

00:39:25.900 it's like you should see a big box

00:39:28.080 that's designed to take your job.

00:39:30.600 Yeah. 0.90

00:39:31.140 It's designed to create competitive capabilities of,

00:39:35.380 you know, if you're worried about immigrants

00:39:36.640 crossing the border,

00:39:38.100 AI represents like millions of new digital immigrants 0.98

00:39:41.780 that are going to be better,

00:39:43.540 not just at the regular work in our economy,

00:39:46.540 but the most premium sort of well-trained

00:39:49.760 cognitive work in our economy,

00:39:51.020 programmers, lawyers, across the board,

00:39:53.780 and then eventually robots

00:39:54.800 that will be able to do physical labor.

00:39:57.280 And so if you're worried about that,

00:40:00.080 you should be worried about the trajectory of AI

00:40:03.480 being designed explicitly to replace your job

00:40:05.640 because the companies will say,

00:40:06.740 we're just giving you a chatbot.

00:40:08.380 It's a blinking cursor.

00:40:09.200 It's like a tool that tells you,

00:40:10.240 hey, why is my washing machine broken?

00:40:12.080 And boom, you get the answer.

00:40:13.000 Why is my kid burping?

00:40:14.040 Boom, you get the answer. 1.00

00:40:14.760 It's helpful.

00:40:15.400 That tool is helpful.

00:40:16.600 But is that what their business model is?

00:40:18.120 Back to Charlie Munger.

00:40:19.520 What are the AI company's incentives?

00:40:21.160 Think about the hundreds of billions of dollars

00:40:23.020 that have been raised.

00:40:24.200 Is there incentive to capture $20 a month

00:40:26.900 from every person,

00:40:28.620 which is how much it costs for a ChatGPT subscription?

00:40:31.220 Or is the thing that's going to pay back

00:40:32.740 those hundreds of billions of dollars

00:40:34.160 to capture everyone's salary

00:40:36.280 from everyone's job in the future?

00:40:39.920 Well, of course the latter.

00:40:41.100 Of course the latter.

00:40:41.800 So it's a race to replace human labor, not a race to augment.

00:40:44.480 These are not chatbot companies.

00:40:46.040 These are not chatbot companies.

00:40:47.060 That's just a veneer to provide an early tool,

00:40:49.780 while the real race is to build these godlike digital brains

00:40:52.800 that are able to do everything at a superhuman level

00:40:56.460 and out-compete us in programming, law, biology, science,

00:41:02.820 physical labor, military strategy, planning, like everything.

00:41:07.920 So, okay, I should just say my bias,

00:41:10.600 which is based on a lot of knowledge and experience, 1.00

00:41:12.900 which is the people who run our country are really stupid, 1.00

00:41:15.680 despite their credentials. 1.00

00:41:17.440 So it's possible they haven't thought this through,

00:41:19.340 but it's so obvious that when you do that,

00:41:22.860 you're going to get a violent reaction from the population.

00:41:25.120 If you take the jobs of factory workers, you get opioid abuse.

00:41:29.080 If you take the jobs of lawyers, you get revolution, okay?

00:41:31.660 The knowledge economy workers don't want their jobs taken,

00:41:34.580 and they have high expectations.

00:41:37.080 and they have low self-control.

00:41:41.540 So you're going to have

00:41:42.680 some kind of violent response.

00:41:45.640 That's right.

00:41:46.300 What's their plan for that?

00:41:48.360 Well, first of all,

00:41:48.940 it's important to note

00:41:49.680 they're selling a story

00:41:50.600 that this is going to be good for everybody.

00:41:52.700 GDP growth is going to go up.

00:41:54.040 It's going to be the most prosperous

00:41:55.100 new golden age.

00:41:56.240 GDP growth.

00:41:57.320 This is like what happened.

00:41:58.920 And you've spoken about this extensively,

00:42:00.880 just that they promised us

00:42:02.540 a story with globalization

00:42:03.500 that will outsource our jobs

00:42:05.440 and manufacturing to these other countries,

00:42:07.840 but it'll make us so abundant

00:42:09.740 because we'll have cheap goods coming in 0.84

00:42:11.320 where labor is so much cheaper

00:42:12.620 and it's going to benefit everybody.

00:42:14.460 And how did that story work out?

00:42:15.940 You won't get a million people

00:42:17.060 living on the streets of major American cities

00:42:18.660 smoking meth.

00:42:19.660 Oh, wait.

00:42:20.520 Oh, you did?

00:42:21.280 Okay.

00:42:21.780 Right, exactly.

00:42:22.460 And so a meme here is like Detroit everywhere

00:42:24.220 or, you know, these places everywhere.

00:42:26.960 Right? 1.00

00:42:27.160 And so, but the working class will put up with it. 0.98

00:42:33.820 Yeah. 1.00

00:42:34.140 But it's kind of a devil's bargain. 1.00

00:42:35.380 But the white-collar class will not put up with it. 0.98

00:42:37.320 They'll go crazy.

00:42:38.100 Well, this is an important point.

00:42:39.380 So while the factory workers and manufacturing workers

00:42:42.320 were further away from Washington, D.C. and New York

00:42:47.040 in the political class.

00:42:48.740 But this time around, to your point,

00:42:50.880 you have lawyers who are going to be affected

00:42:53.240 and they are very close to the political class.

00:42:55.460 Exactly.

00:42:56.040 There's also a bigger drop, right?

00:42:57.460 So instead of going from, I don't know,

00:42:59.220 $80,000 a year to $35,000 a year or something like that,

00:43:02.040 you're going from, I took out student,

00:43:05.080 debt, $200,000 or $1,000 of student loans, and I'm going from my job of $250,000 a year as a lawyer

00:43:12.040 to now, whoops, what am I doing? Doing some other kind of cognitive work. Now, people will always

00:43:17.380 say, but look, if Marc Andreessen was at this table, he'd say, look, we've always worried about

00:43:21.700 automation. Economists have named this phenomenon all the time. New technologies come along,

00:43:26.500 agriculture. 200 years ago, all of us were farmers. Now, almost no one's a farmer. We

00:43:30.480 always find something new to do. What's different about AI, what's different about it?

00:43:35.080 It's that AI is built to be able to do all forms of cognitive labor, like lawyers, programmers,

00:43:42.840 you know, factory workers, everything, robots, it does everything.

00:43:46.700 And for whatever it doesn't do yet, that just becomes the next incentive, Charlie Munger

00:43:50.860 style, to go train in that field.

00:43:53.960 Like, I mean, have you seen these hand farms in India, arm farms?

00:43:57.120 Have you seen this?

00:43:57.860 No.

00:43:58.320 So this is, it was both a popular job in LA about six months ago, and also all throughout

00:44:03.260 India is essentially an economy where you strap a GoPro camera to your head and you look down

00:44:08.860 and you do tasks. You sew things, you clean dishes. So you're basically training. You have

00:44:14.660 this, essentially the number one job is training your replacement. It's kind of nice to see Indians 0.98

00:44:19.640 training their own replacements. Well, sorry, excuse me. It's my immigration joke. I guess 0.99

00:44:28.200 the difference is though, and I've had this conversation with a number of people involved

00:44:32.020 in this transition

00:44:33.040 and they will say

00:44:33.780 exactly what you just said,

00:44:34.780 which is we've been

00:44:35.220 through this before.

00:44:36.360 We always figure it out.

00:44:37.160 The agricultural economy.

00:44:38.520 Exactly.

00:44:39.220 It's going to turn out great

00:44:40.300 because it always does.

00:44:41.120 Yeah.

00:44:41.820 But the difference here

00:44:43.120 is that the people

00:44:43.900 whose jobs are being replaced

00:44:45.560 imbue their jobs

00:44:47.600 with meaning.

00:44:48.500 Yes.

00:44:48.600 So if I'm working

00:44:49.640 a metal stamping machine,

00:44:51.440 I don't think that

00:44:52.360 what I'm doing

00:44:52.980 is saving the world.

00:44:54.220 I'm doing it to pay

00:44:55.280 for my bass boat

00:44:56.000 and my life outside work.

00:44:57.360 But if I'm

00:44:57.860 a nonprofit executive,

00:44:59.580 I really feel like

00:45:00.800 what I'm doing

00:45:01.720 is inherently important

00:45:03.100 and my identity is wrapped up

00:45:04.200 in what I'm doing.

00:45:04.860 So the promise of more idle time

00:45:07.380 is hell to me.

00:45:09.260 Well, also,

00:45:09.860 where are you going to get

00:45:10.480 your future income?

00:45:11.740 But leaving the income aside,

00:45:12.760 even if we have, you know,

00:45:15.220 guaranteed basic income

00:45:16.200 or whatever,

00:45:16.720 there's like the welfare state

00:45:17.740 becomes, you know,

00:45:19.440 the only source of income

00:45:20.680 and it's generous,

00:45:22.160 it still doesn't make up

00:45:23.660 for the meaning.

00:45:24.820 Correct.

00:45:25.380 My identity has been lost in this.

00:45:27.220 Identity, dignity, meaning.

00:45:28.480 And that's why I become

00:45:28.900 a revolutionary.

00:45:29.900 Exactly, yeah.

00:45:30.940 Most people don't like thinking about life insurance.

00:45:33.040 It's one of those topics you tend to kind of put off.

00:45:35.140 But if you're married, you've got kids, you've got people who depend on you,

00:45:38.660 you probably should think about what would happen if you weren't around.

00:45:41.760 And that's why we've partnered with Ethos.

00:45:43.640 They've made getting life insurance super easy.

00:45:46.040 You can do the whole thing online.

00:45:47.300 You don't have to sit down with an agent or go through a medical exam.

00:45:50.300 You get a quote in seconds, apply in just a few minutes,

00:45:53.140 and in some cases you get coverage the same day.

00:45:55.020 You just answer a couple of simple health questions.

00:45:57.320 You get up to $3 million in coverage.

00:45:59.340 Some policies can be yours for as little as $30 a month, so it's a good deal.

00:46:04.560 If you've been putting off life insurance because you thought it would be a very complicated,

00:46:07.940 expensive process, you should probably check out Ethos.

00:46:11.020 Take a couple of minutes, get your free quote online at ethos.com slash Tucker.

00:46:15.880 Ethos is E-T-H-O-S dot com slash Tucker.

00:46:20.700 Application times obviously vary as do rates, but we think it's pretty good.

00:46:25.220 So what's their plan for that?

00:46:27.060 They hand wave, as you've probably noticed.

00:46:29.500 They don't have a plan.

00:46:31.120 And by the way, you might say- 1.00

00:46:32.860 Well, they're idiots. 1.00

00:46:33.280 Yeah. 1.00

00:46:33.940 Well, and so they'll say, but if we don't do it, China will.

00:46:36.260 But right now, US and China are racing

00:46:39.160 to mutually assured political revolution right now

00:46:41.560 because they have the same problem.

00:46:43.540 They have a population decline problem.

00:46:45.340 So they need to fill in the population decline

00:46:47.280 with more robots.

00:46:48.140 So they think they can manage it that way.

00:46:49.440 But we have a shared interest in,

00:46:51.700 the way I think about AI is it's like taking steroids

00:46:55.680 that also give you cancer.

00:46:57.380 So let me explain that.

00:46:58.820 I take some GPT-6, I put it in my mouth as a country.

00:47:02.140 Now my whole country is running on GPT-6.

00:47:03.960 So I just buffed up my muscles.

00:47:05.400 I've got suddenly super programmers.

00:47:07.300 I've got super military strategists.

00:47:08.940 I've got super GDP growth

00:47:10.200 because I've automated stuff with robots.

00:47:11.980 So my external muscles, like the GDP

00:47:14.200 is like an external muscle of the US economy.

00:47:16.540 That all got bigger.

00:47:18.220 But at the cost of cancering identity work,

00:47:21.560 meaning the social fabric, right?

00:47:24.820 It's a devil's trade.

00:47:26.600 I think the theme here is, what does the devil do?

00:47:29.620 They come up to you and say, I've got this great deal for you.

00:47:31.760 It's going to be great.

00:47:32.480 You're going to look great.

00:47:33.060 You're going to feel great.

00:47:33.600 Everything's going to be.

00:47:34.180 And the other side of that trade is something you don't want.

00:47:38.600 Do they actually think, since you've, I mean, you do this for a living, you talk to these

00:47:42.320 people for a living, do they think that GDP is like the most meaningful measure of a society's

00:47:48.860 health?

00:47:49.120 I don't think they're thinking through it.

00:47:50.200 I think there's a thing where it's persuasive to say it because in the past, GDP growth

00:47:55.040 going up, at least like through the 90s or something like that, meant, generally speaking,

00:47:59.020 more prosperity for our society. Anyone who uses the term GDP is instantly written off

00:48:03.280 in my head as a moron. Right. Well, even the... Like I won't even listen to someone who cites 1.00

00:48:07.680 GDP. GDP? I know, it's stupid. So there is a geopolitical thing, which is that economic growth

00:48:13.580 precedes geopolitical power. So if I've got a bigger economic growth machine, then I will be

00:48:19.460 able to dominate other countries that don't. And so I think it's a power-seeking thing that

00:48:24.980 is using as propaganda GDP growth to make it sound like this is going to be good, because

00:48:29.900 through the 60s, 70s, 80s, 90s, GDP growth going up was associated with more prosperity for the

00:48:36.580 economy, actually up until the 1970s, as we know when things changed. But it's not even an accurate

00:48:42.420 measure of economic power. It's like an idiot's measure. Was it Simon Kuznets? No, who's the one 1.00

00:48:48.420 who came up with the GDP metric? He said, don't use this. He said, don't use this to measure

00:48:53.140 prosperity of a country. This is just a simple metric. It's one thing you can use. You never

00:48:58.480 optimize for it. We were talking at breakfast about how you made such a smart point that I

00:49:03.480 tried to memorize it, but we're talking about how basically our ruling class is inadequate to

00:49:09.260 managing our society. And it's not just true in our country, but true in many Western countries.

00:49:13.360 And you said that is a product of optimizing for one quality. Yes. But you explain it because

00:49:20.320 to do a better job than I am.

00:49:21.200 I mean, I think, so you and your show

00:49:22.740 have talked about a lot of these topics, 1.00

00:49:24.120 whether it's, why do we have morons running everything? 1.00

00:49:27.560 Why do we have big box stores 1.00

00:49:29.140 and the removal of aesthetics of everything?

00:49:32.360 And what it all comes down to is

00:49:33.960 if I'm optimizing for a metric,

00:49:35.800 like I just want the cheapest goods everywhere.

00:49:38.320 Well, then suddenly I just outsourced

00:49:39.520 all my manufacturing to the other countries.

00:49:42.340 Including, by the way,

00:49:43.180 my national security critical minerals

00:49:44.580 and all this stuff that you wake up later

00:49:45.920 and say, why did I do that?

00:49:47.240 But it was for the cheapest price.

00:49:48.340 So you weren't thinking through it

00:49:49.260 in a multidimensional way.

00:49:50.440 Yes.

00:49:50.860 If I optimize for SAT scores,

00:49:52.540 I get this like super smart,

00:49:54.180 narrow, disembodied,

00:49:55.700 brilliant, like quality

00:49:57.100 of like these very specifically domesticated humans,

00:50:00.220 but at the expense of all the other kinds of intelligence

00:50:02.860 and sort of humanity and all that.

00:50:05.360 And really what that optimizes for is powers.

00:50:07.000 Wisdom, too.

00:50:07.140 Wisdom, absolutely.

00:50:08.280 If I optimize for like the cheapest use of 3D space,

00:50:11.320 then I get big box stores

00:50:12.460 that optimize revenue per square inch of,

00:50:14.940 of, you know, unit 3D space.

00:50:18.340 and that the removal of aesthetics

00:50:20.660 and the soul of a city.

00:50:22.280 If I take a city like Sedona, Arizona,

00:50:24.340 I remember visiting there during the pandemic,

00:50:26.260 where they had a problem

00:50:27.240 where so much of the real estate

00:50:29.600 was actually just Airbnbs.

00:50:31.020 So it's like a ghost town

00:50:31.840 where essentially to optimize the revenue in the city

00:50:35.600 is everybody's just making money

00:50:37.840 from renting their places.

00:50:39.180 And all of a sudden no one lives there.

00:50:40.080 And all of a sudden no one is there.

00:50:40.960 So no one,

00:50:41.800 and there's a famous quote from Larry Summers,

00:50:43.500 like, does anybody ever washed a rental car?

00:50:46.480 Right?

00:50:46.580 Like meaning that if you don't have a skin in the game,

00:50:48.760 if you don't have a stake in the city,

00:50:50.120 in the parks, in the schools, in the libraries,

00:50:51.940 in the churches, in the local community,

00:50:54.040 in the fabric, you're kind of,

00:50:56.020 it's the vampirification of everything.

00:50:57.700 It's like the big boxification of everything.

00:51:00.400 And so I think that to sort of, you know,

00:51:03.560 something you've talked about a lot

00:51:04.580 is the kind of spiritual disease that we're seeing,

00:51:07.060 which comes from this narrow optimization

00:51:09.020 for one metric at the expense of the whole,

00:51:11.980 and that includes GDP, right?

00:51:14.600 And so if we want a rich society,

00:51:16.580 other countries are better at managing this right now than the US, where you protect the sort of

00:51:20.900 fundamentals of a society, the community, the trust, the intergenerational wisdom and knowledge

00:51:25.120 transmission. You invest in the arts and culture and education, healthcare, et cetera. But AI is

00:51:31.660 like, we're actually talking about this Nick Land quote that we'll maybe get into. If we just value

00:51:39.700 competition and growth of a narrow metric, then it's like we are summoning AI from the future

00:51:46.760 into the present. We are assembling the things that will make everything more efficient through

00:51:51.820 AI. Because AI represents, I can do everything more efficiently than everything else.

00:51:55.880 That's what it represents. And there's a great quote by Nick Land, who wrote an essay called,

00:52:02.780 What appears to humanity as the history of capitalism is an invasion from the future

00:52:08.940 by an artificial intelligence space

00:52:11.580 that must assemble itself entirely

00:52:14.000 from the enemy's resources.

00:52:16.840 Like essentially, we are assembling

00:52:18.820 the machines that are being designed

00:52:22.560 to replace and dehumanize and disempower

00:52:24.940 our own humanity.

00:52:26.480 We're selling the hangman the rope.

00:52:27.820 That's right.

00:52:28.720 You know, the future job is coffin builders.

00:52:30.840 My job is to train the AI

00:52:32.620 that's going to replace my job.

00:52:35.420 Now, I'm not saying this by the way

00:52:37.420 people are going to say, 0.99

00:52:38.100 okay, you're a doomer, you're a dark person, 0.96

00:52:40.440 you're trying to sell fear. 0.96

00:52:42.460 That's not what this is about.

00:52:43.400 This is about if we can see- 0.95

00:52:44.300 Wait, you're the doomer?

00:52:45.160 Did you create this technology?

00:52:46.660 I don't think so.

00:52:47.300 No.

00:52:47.700 And if you say, I'm trying to make money from it,

00:52:49.420 think about the trillions of dollars

00:52:51.180 that the people advancing the technology-

00:52:53.140 I love the blame transfer.

00:52:53.640 It's so funny.

00:52:54.240 It's like, it's always, like if you, 0.98

00:52:56.140 I catch you in the act of raping someone, 0.87

00:52:59.220 and I'm like, stop, and you're like, 1.00

00:53:01.120 shut up, you sexist. 1.00

00:53:02.560 It's like, wait, what? 1.00

00:53:03.680 I don't know, I'm not the criminal here.

00:53:05.460 Exactly, exactly. 0.92

00:53:06.780 You're the doomer, really?

00:53:08.100 Well, and so you have someone like Nate Soros on your podcast, or you have me on, or you have people who have a sincere care for the future. And all you're doing is saying, we have to see clearly, you know, there's a great quote by Charles Kettering, a problem well stated is a problem half solved.

00:53:22.560 Yeah, that is, I believe that.

00:53:24.160 And, you know, Einstein said that if I was given a problem, I'd spend 95% of the time defining the problem and then 5% solving it.

00:53:32.100 So one of the drivers of our problems is this optimization for narrow metrics.

00:53:37.040 Like Facebook maximizing this narrow metric of time spent sounded like, oh, we're giving more people what they want because otherwise why would they be spending time there?

00:53:44.440 But in the meantime, you've created cyberbullying, body dysmorphia, dyslexia, like brain rot.

00:53:49.160 You know, you've destroyed a generation all while the number was going up.

00:53:53.060 Well, there's also a thing called addiction.

00:53:55.040 Right, exactly.

00:53:55.720 Which tends to act against something called free will.

00:53:58.460 Exactly.

00:53:59.360 It's like...

00:54:00.240 Yeah, exactly.

00:54:01.040 This is why there is a wisdom gap above all in the West.

00:54:04.220 Like people are not thinking wisely.

00:54:06.640 Correct.

00:54:07.220 They're thinking in the way that an industrial designer thinks.

00:54:10.620 Like I need to build a machine that makes bumpers.

00:54:13.080 And they're being domesticated to think in that way

00:54:15.360 because the system profits from people

00:54:17.100 who think in terms of the system as it exists.

00:54:19.660 Yeah, in terms of narrow specialties,

00:54:21.260 totally disconnected from the whole.

00:54:23.280 That's exactly right.

00:54:23.980 And that's what it is.

00:54:24.680 It's like narrow specialties

00:54:26.080 disconnected from the whole.

00:54:27.440 Like there's a whole

00:54:28.540 of what makes a social fabric healthy.

00:54:30.040 There's a whole for what makes,

00:54:31.660 forget climate change,

00:54:32.420 just like a healthy environment.

00:54:33.620 You have to have all these other features.

00:54:36.220 And it's like,

00:54:37.060 if you're optimizing for,

00:54:39.300 you know, the eggs don't stick to the pan,

00:54:42.160 then you make Teflon.

00:54:43.520 But you didn't realize

00:54:45.300 you're also creating forever chemicals

00:54:47.280 that give everybody in childhood cancers. 0.77

00:54:50.200 And now the sperm doesn't stick to the egg.

00:54:52.500 Yeah, the good news is it's easier to clean your pan. 0.80

00:54:55.040 The bad news is the eggs are poison.

00:54:56.800 That's right.

00:54:57.320 Right. 1.00

00:54:57.680 The eggs are poison and your sperm doesn't stick to the egg. 0.98

00:55:02.140 Sorry, I'm just, it's all so absurd. 0.96

00:55:04.020 It just makes me laugh.

00:55:05.020 Well, and it's really like disheartening and depressing,

00:55:08.180 but there's, as we talked about off camera this morning,

00:55:12.020 there's something called,

00:55:13.540 it's a framework I really like from a mentor of mine

00:55:15.640 that there's pre-tragic optimism,

00:55:17.920 which is like, I'm going to kind of,

00:55:18.900 I want to be an optimistic person,

00:55:20.120 but that's like essentially a commitment without evidence.

00:55:22.240 Like I just, I need to be optimistic.

00:55:23.780 That's like, totally.

00:55:24.620 That's how, you know, I'm better off being optimistic.

00:55:26.800 It's almost like a neurosis actually.

00:55:28.220 Yeah, exactly.

00:55:28.980 And it's a kind of denialist neurosis.

00:55:30.820 It's like, it's like, it's a nervous giggle.

00:55:33.180 It's a nervous giggle, exactly.

00:55:34.220 Yeah.

00:55:34.880 And you kind of stay optimistic

00:55:36.160 because you just feel committed to that.

00:55:37.660 And also there's like impartial truth there,

00:55:39.220 which is like, it's better to live your life

00:55:40.400 from a uplifting place.

00:55:42.780 But if the cost of that optimism

00:55:44.280 is being in denial of reality,

00:55:45.940 where the balance sheet of risks

00:55:47.280 is growing exponentially.

00:55:48.620 Yeah.

00:55:49.480 That's not a helpful form of optimism.

00:55:51.540 Well, you probably have heard

00:55:52.640 that diesel prices have gotten so high

00:55:55.280 that G7 nations had to release fuel

00:55:57.660 from strategic reserves

00:55:58.720 to get costs under control.

00:56:00.080 And still, some pumps are readjusting

00:56:02.400 so they can get to $10 per gallon.

00:56:06.060 Governments across the world

00:56:07.000 are trying to keep costs down,

00:56:08.180 but they are running out of options

00:56:09.420 because they're running out of money.

00:56:11.120 The world is unstable.

00:56:12.360 That is the sad truth.

00:56:13.340 And people in charge cannot control

00:56:14.900 even the most basic questions

00:56:16.900 like what does energy cost anymore?

00:56:19.280 The war in Iran is raging.

00:56:20.700 It's not coming to an end soon.

00:56:22.980 Neither is the war in Ukraine.

00:56:25.440 So could we get a big recession soon? 0.85

00:56:27.820 Yeah, of course we could.

00:56:29.840 So it's smart to think ahead now

00:56:31.560 and imagine how you can take matters

00:56:32.960 into your own hands

00:56:33.660 to protect your savings and your family.

00:56:35.480 And that's why smart investors everywhere,

00:56:38.720 very much including central banks

00:56:40.420 around the world

00:56:41.020 and very much including China,

00:56:43.060 are stocking up on gold, physical gold, right now.

00:56:48.880 Gold is not just a store of value for all of human history.

00:56:51.220 It's also proven to be a strong investment.

00:56:53.440 It's outperformed the S&P 500 since 9-11.

00:56:57.720 Over the past 25 years, gold has outperformed

00:57:00.260 the Standard & Poor's average.

00:57:03.380 Gold is down from its historic high earlier this year,

00:57:05.920 more than 1,000 bucks.

00:57:07.780 But major institutions predict the spot price

00:57:09.860 will hit another high by Q2 of next year, 2027.

00:57:13.740 So if you're thinking about buying gold,

00:57:15.160 right now is the time to buy and people are buying.

00:57:18.040 We founded Battalion Metals

00:57:19.540 to give you a chance to buy fairly priced gold.

00:57:22.980 This is not a scam.

00:57:24.080 Our margins are small, 3%.

00:57:25.860 But the point is we're not ripping anybody off

00:57:29.000 unlike a lot of places you buy gold

00:57:30.740 because we actually believe in gold

00:57:32.700 and we use this service ourselves.

00:57:34.640 Protect your savings and fortify your future.

00:57:37.340 Explore our collection of battalionmetals.com slash Tucker, battalionmetals.com slash Tucker.

00:57:43.220 No, and I just, I've always noticed my whole life, the wisest, most centered, balanced

00:57:48.680 people I know laugh a lot.

00:57:50.400 And a lot of what they laugh at is just the enduring absurdity of people, which is literally

00:57:54.960 enduring.

00:57:55.420 It's just, it's innate in people. 0.93

00:57:56.860 They're absurd.

00:57:57.320 They can't see, none of us can see the big picture.

00:57:59.980 None of us know what happens after we die.

00:58:01.680 Like we're operating in the dark.

00:58:03.700 And if you fully understand that,

00:58:06.320 you're amused by it because it never corrects.

00:58:10.040 And like, it's so important to have that perspective.

00:58:12.300 Otherwise, you become self-serious and myopic.

00:58:15.280 You become what you're describing.

00:58:17.120 And I've just got to say,

00:58:18.200 I noticed that very few Silicon Valley people I know

00:58:20.700 have that sense of humor at all.

00:58:23.740 They really believe that they're gods

00:58:26.020 in charge of shaping the future.

00:58:27.660 Yeah.

00:58:28.340 And they don't know how absurd they are.

00:58:30.380 Have you noticed this?

00:58:31.180 Yeah.

00:58:31.540 Yeah. I mean, we should get into the whole transhumanism, God complex part of this too,

00:58:37.000 because when you talk to, in the Charlie Munger sense, like what's the incentive here? You might

00:58:40.540 just say, oh, the incentive is just like profit. Like these AI companies are trying to make money,

00:58:44.100 but you're pointing to a deeper thing, which is, it's basically build a God, own the world economy

00:58:49.900 and make trillions of dollars. And basically build the most intelligent entity that has ever

00:58:55.960 existed, right? Like they're building a machine. So just in the last few weeks, AI solved a

00:59:02.880 mathematics problem that had a million dollar prize. It was the Millennium Prize, excuse me,

00:59:07.460 the Navier-Stokes equation. And my understanding is, I think if I get this right.

00:59:11.080 So that is confirmed, AI did solve it.

00:59:12.820 Yeah. I mean, normally it takes two years to confirm it, but I believe they use a formal

00:59:16.420 other system to kind of verify it very quickly. And I think they've solved it.

00:59:20.460 And I believe they took 10,000 AI agents.

00:59:24.620 Imagine a swarm, not the one that hacked Hugging Face,

00:59:27.080 but a swarm of agents that are just solving math.

00:59:29.120 And they worked for 88 hours.

00:59:31.280 And I believe it cost, if I get the number right,

00:59:34.320 I think it was six and a half million dollars

00:59:36.100 to basically have this swarm of AIs solve unprecedented math.

00:59:40.780 Six and a half million in energy costs.

00:59:42.500 Yeah, in like the, what's called the token cost,

00:59:44.320 basically running the AI model for that long,

00:59:46.020 for 88 hours with that many agents.

00:59:47.600 Took, I think, six and a half million dollars.

00:59:48.840 and um so imagine you've built a machine that is literally like an oracle and you push a button and

00:59:57.540 you wait for a little while and it just came up with new materials that no human has ever come

01:00:00.740 up with it just came up with new math proofs that no one's ever come up with it just came up with

01:00:04.680 new theoretical physics that no one has ever come up with and imagine i hit a button say how would

01:00:08.380 you apply those new innovations to new products go go literally think about that for another

01:00:13.020 thousand hours and then it comes up with new products and new ideas of what it could actually

01:00:17.660 built. And then now actually say, what would you need to make those products? Go start a lot, you

01:00:22.060 know, do the automated IP filing, start an LLC, start like, it's, people don't understand what

01:00:27.860 we're really building here, Tucker. This is like, you are building the automation of intelligence.

01:00:32.200 You could start automated companies with all these new insights. You could generate all this

01:00:36.080 new 24th century science. The, um, a friend of ours at Jay Akhotra, who actually happened to also

01:00:40.800 be the lead author on this hugging face incident investigation, said to me once four years ago,

01:00:46.440 before she did this Hugging Face investigation,

01:00:48.900 she said AI is like 24th century technology

01:00:52.300 crashing down on a 21st century society.

01:00:56.880 And that sounds like hyperbole and exaggeration

01:00:59.860 and embellishment and hyping the capabilities,

01:01:02.300 but AI is doing those things.

01:01:04.840 And I just need to say,

01:01:05.580 and with all respect to the many different voices out there

01:01:07.720 that are trying to make sense of AI,

01:01:09.360 there is one community that has been saying

01:01:11.380 that these capabilities were gonna scale.

01:01:13.260 And there's a lot of people who are saying rightly,

01:01:14.900 like, this isn't going to happen. This is all hype, but it just keeps happening. And the people

01:01:20.120 building it, I was just at a conference the last three days in the Bay Area. I was with all the

01:01:24.180 people in AI and they very much believe that this is going to keep getting better. Because if you

01:01:29.060 can scale at least the thinking times, imagine a person just thinks there for a million lifetimes

01:01:36.280 for long, how much work would you accomplish if you could think for a million lifetimes of time?

01:01:40.800 you could invent a lot. What if I could get more data centers where I'm going from a million

01:01:45.840 lifetimes of thinking and compute to a hundred million, you know, thinking hours, right?

01:01:53.160 People don't understand what you could do with that. And also the thing that's doing the thinking

01:01:58.260 is also getting better every year. So like the AI that's thinking for that long is also getting

01:02:03.740 more powerful. And two years ago, the AIs weren't that good at, you know, weren't doing novel

01:02:10.600 things in cybersecurity. But as of just six months ago, as Nate Soares said recently, I like the way

01:02:16.540 he said it, like six months ago, there was essentially two entities that could hack into

01:02:20.220 your phone without you clicking on a link. It was like Mossad and the NSA. And six months ago,

01:02:25.740 when Claude Mythos came out, suddenly there was three entities that could do that. You have an

01:02:30.620 AI system, Claude Mythos, came out in, I think, February, that can find zero-day vulnerabilities,

01:02:38.360 meaning backdoors, vulnerabilities in any system,

01:02:41.860 including, according to Senator Mark Warner,

01:02:43.800 who's chair, ranking member

01:02:44.680 of the Senate Intelligence Committee,

01:02:46.300 was able to find vulnerabilities

01:02:48.100 in our classified systems.

01:02:50.700 Yeah, good.

01:02:52.640 I hope though, and I hope Claude Mythos

01:02:55.440 will declassify and release all that stuff

01:02:57.900 as sort of like the WikiLeaks that we were promised.

01:03:01.500 Well, there's a lot of different values.

01:03:03.020 Sorry, excuse me.

01:03:04.100 Well, there's different things here.

01:03:05.040 So one is like, can we have more accountability

01:03:06.880 for wars that happen

01:03:08.140 that no one talks about

01:03:09.040 what really happened

01:03:09.740 and actually hold people

01:03:11.680 accountable for that.

01:03:12.480 Yes, maybe some of that stuff.

01:03:14.140 Absolutely.

01:03:15.300 Then there's the other side,

01:03:16.180 which is like,

01:03:16.660 if we release nuclear secrets

01:03:18.240 to the entire world

01:03:19.320 because some criminal group

01:03:21.140 maybe wants to go hack into that,

01:03:23.260 that's a really dangerous thing.

01:03:24.740 Or can we launch

01:03:25.760 other nations' nuclear weapons

01:03:27.300 against their will

01:03:27.980 by hacking in?

01:03:28.900 Well, so this is

01:03:29.380 the interesting thing also.

01:03:30.400 So the U.S. is about to spend,

01:03:31.760 I think it's $1.7 trillion

01:03:33.200 starting in 2017,

01:03:34.640 the Trump administration,

01:03:35.300 I think signed it, right,

01:03:36.000 to upgrade the nuclear arsenal.

01:03:37.400 That's correct, right?

01:03:38.360 And they're going to do it over a decade,

01:03:39.760 except this time around.

01:03:40.740 So why are they doing this?

01:03:41.500 Because the nukes are of the old system.

01:03:44.860 Like there's the air-gapped,

01:03:46.040 like it's running the old way.

01:03:47.440 It was built in the 50s to the 80s, 90s, et cetera.

01:03:51.280 Really to the 80s, right?

01:03:52.460 And the new system, this time around,

01:03:53.960 they're going to be digitally networked.

01:03:55.580 They're actually going to be connected

01:03:57.660 to modern networking infrastructure.

01:04:00.120 So if you have suddenly super hackers

01:04:02.660 that are only getting exponentially better

01:04:05.500 at cyber hacking from AI

01:04:07.300 at the same time

01:04:09.120 that you're digitally networking

01:04:11.340 your nuclear weapons,

01:04:13.300 I don't have to complete that story for you.

01:04:16.000 It's a dangerous equation.

01:04:17.420 I was actually at the Vatican

01:04:18.780 nuclear weapons conference times AI.

01:04:20.720 The Vatican held an event

01:04:23.060 for Nobel laureates

01:04:24.180 and people who work on the nuclear topic

01:04:27.480 combined with people who work on AI

01:04:29.200 because there's a new set of risks.

01:04:31.560 The obvious one being

01:04:32.800 if you have an AI

01:04:33.860 doing nuclear command and control.

01:04:35.500 So you have the AI that's deciding the launch decisions.

01:04:38.320 That's the obvious one.

01:04:39.880 But then there's the other ones of,

01:04:41.640 if I'm now digitally networking my nuclear weapons

01:04:44.360 and I suddenly have an AI that can turn off your nukes

01:04:47.200 or do the opposite,

01:04:50.020 that's, don't have to say more, a very unsafe world.

01:04:53.020 Does anyone see the core problem,

01:04:55.220 which is they're trying to cure the disease

01:05:00.420 with the poison that caused it?

01:05:01.980 They're trying to smoke away the lung cancer.

01:05:03.920 I mean, technology is the problem.

01:05:07.940 And then they want to solve it with more technology.

01:05:09.960 And their answer is, well, the best way to defend against a super AI cyber weapon is

01:05:13.800 to build another AI cyber weapon that we distribute to everybody that they can use to patch their

01:05:18.320 system.

01:05:18.900 That's what I'm saying. 1.00

01:05:19.540 That's like, that's why dumb people in charge is the core problem. 1.00

01:05:24.320 Like, no one can think critically. 1.00

01:05:25.880 Like, wait a second.

01:05:26.440 How did we network your nuclear arsenal?

01:05:30.080 Like, who would even say something like that? 0.92

01:05:31.700 And it's interesting because it might not have seemed like too dumb of an idea before.

01:05:37.780 Like, it's just a way of modernizing your nuclear forces.

01:05:40.300 But suddenly, you didn't foresee that there would be AI super hackers that know how to

01:05:45.120 find vulnerabilities in everything.

01:05:46.920 Now, people might think I'm hyping this, but let me give you another example.

01:05:50.440 So Apple, if you're running an iPhone 17, I believe, and above, or a MacBook, I think

01:05:57.200 it's an M5 or above.

01:05:58.960 These are the chips, basically.

01:06:00.120 the chips have a new thing called the memory MEI memory uh enhanced integrity or something like

01:06:06.400 that um it's a new hardware based system to do protected memory so that basically it's harder

01:06:12.600 for a cyber hacker to hack your phone or to hack your computer so they spent several years and

01:06:19.780 billions of dollars in research to invent not just software but like hardware like new stuff

01:06:25.100 they have to print out of Taiwan and TSMC. And how long do you think it took for Mythos

01:06:31.420 to hack through that? Years. Well, Mythos has only been out since February.

01:06:36.980 I know, I know. I'm just kidding. But it took an independent group five days

01:06:41.760 to use Mythos and to find a vulnerability in that system. Now that group was well-intentioned. They

01:06:47.900 are white hat hackers who are the good guys who are trying to find vulnerabilities and patch them.

01:06:55.100 But if you suddenly have that available in open source models, where suddenly everyone can get access to this, at this conference I was just at, I can't say too much, but they was basically speaking about how the Chinese open source hacking models are now, just now in the last month, crossing a threshold where it is very dangerous for today's cyber infrastructure.

01:07:22.320 And so we need a radical acceleration

01:07:24.760 of defensive measures to protect against that.

01:07:28.700 Now, the other thing you and I were talking about

01:07:30.240 over breakfast is that there's a weird

01:07:31.620 like security dilemma thing here.

01:07:33.160 Because when I say that, you're thinking,

01:07:34.360 oh my God, China's going to use all that

01:07:35.480 to hack all of our stuff. 0.95

01:07:36.940 Or actually China's releasing all that

01:07:38.840 to the entire world so anybody can hack that stuff.

01:07:41.020 That's actually the more dangerous point

01:07:42.200 because a state actor, you can call them up and say-

01:07:44.620 Of course, there are constraints.

01:07:45.580 Yeah, exactly, there are constraints

01:07:46.540 because you don't want retaliation and et cetera.

01:07:49.540 But at the same time,

01:07:52.400 like, so what is the security dilemma?

01:07:53.880 Security dilemma is like,

01:07:54.700 if I'm the US and I start like mobilizing

01:07:56.260 my nuclear forces

01:07:57.060 and I start moving things

01:07:58.200 that I didn't move before,

01:07:59.620 the other country thinks

01:08:00.340 I'm about to start a war.

01:08:01.720 Even though I'm actually doing that defensively.

01:08:03.780 This has happened.

01:08:04.840 This has happened.

01:08:05.500 During the Cold War.

01:08:05.900 During the Cold War.

01:08:06.840 And then what happens in response is,

01:08:08.240 let's say the other country,

01:08:09.000 whether it's the Soviet Union or China,

01:08:10.140 they start mobilizing their nuclear forces.

01:08:11.840 I'm like, oh my God,

01:08:12.400 they're about to mobilize

01:08:13.120 to do something against us.

01:08:14.300 So the security dilemma is

01:08:15.460 we start taking actions

01:08:16.640 that look like we're doing something

01:08:18.520 offensively that might be defensive.

01:08:21.460 Of course.

01:08:22.040 So when we see China really- 0.87

01:08:23.900 We kill each other out of paranoia. 0.85

01:08:25.040 Exactly.

01:08:25.340 And that's like the ultimate test

01:08:27.860 of this whole situation is game theory

01:08:30.120 is in paranoid logic,

01:08:32.380 which is John von Neumann and, you know,

01:08:35.160 Morgenstern came up with these equations

01:08:37.260 that out of the Cold War,

01:08:39.080 sorry, out of the invention of the atomic bomb,

01:08:41.460 that we're all about stabilizing the world

01:08:43.400 through this paranoid logic of,

01:08:45.480 if I do this, you do that.

01:08:46.620 And I keep playing it out.

01:08:47.340 It's mad.

01:08:47.660 Mutually shared destruction.

01:08:48.520 So even though we live in the kind of a hair trigger possibility of nuclear war, it's stable because no one's going to do anything. But AI suddenly destabilizes that because now AI is creating the perception that I'm getting a jump in cyber capabilities, in military strategy, in biotechnology, in new drones and new image recognition and new swarming algorithms.

01:09:12.380 I suddenly am perceived to have a new advantage

01:09:15.720 that might make you think I'm about to use it.

01:09:18.660 So when Anthropic came out with Mythos,

01:09:22.000 they're essentially getting a jump in cyber capability

01:09:24.380 that maybe we didn't use offensively,

01:09:25.880 but China's like, oh my God, 0.96

01:09:27.840 they're going to use this offensively.

01:09:29.160 So then they release these other capabilities

01:09:30.680 and we sort of keep ping-ponging back to each other.

01:09:32.900 And it's the whole like, what'd you say?

01:09:34.300 Lungs that you smoke to keep yourself from getting cancer?

01:09:37.500 You smoke the cancer away.

01:09:38.540 You smoke the cancer away, exactly.

01:09:40.600 And then the problem is that there's another risk here from just the speed, which is that

01:09:44.720 the, there's one risk, which is just the raw capabilities coming out.

01:09:48.260 But the other thing is suddenly what had been in equilibrium, like my weapons and your weapons

01:09:52.200 are kind of like roughly matched.

01:09:53.820 So we don't start wars with each other, the big countries.

01:09:57.080 But what if there's a perception that I can suddenly turn off your nukes and you don't

01:09:59.960 have that?

01:10:01.180 What if there's a perception that suddenly I have swarming algorithms for my drones and

01:10:05.060 you don't have that?

01:10:06.520 I mean, I think that my understanding is

01:10:08.160 China and other countries

01:10:09.460 saw how the US was using AI in Iran

01:10:14.140 to do a fast target selection,

01:10:16.760 which also, by the way, 0.97

01:10:17.440 probably resulted in that girl's school

01:10:19.840 in Iran getting bombed. 0.89

01:10:21.520 But they sort of see us doing

01:10:23.020 kind of new blitzkrieg. 0.58

01:10:24.060 So Hitler had blitzkrieg of like 0.77

01:10:25.260 a fast form of war, 0.71

01:10:26.480 blitz being fast and krieg being war.

01:10:28.600 We have kind of an AI speed war function.

01:10:32.360 And when they see we have that capacity,

01:10:34.240 that destabilizes certain equilibriums that we had.

01:10:37.880 So if equilibria is what keeps the world safe,

01:10:40.640 AI creates this disequilibria

01:10:42.520 that's happening at a faster and faster pace

01:10:45.080 across more domains at once.

01:10:47.540 Does that make sense?

01:10:48.360 Oh, it makes sense.

01:10:49.360 Yeah.

01:10:50.420 How does cornering people work?

01:10:52.480 Is that a good idea typically?

01:10:53.520 Well, you know, you have an adversary

01:10:57.380 who's capable of killing you. 0.93

01:10:58.740 Right. 0.99

01:10:59.720 And you terrify him so much

01:11:03.660 that he becomes unstable and unpredictable.

01:11:07.580 Is that wise?

01:11:08.500 Right, not very wise.

01:11:10.200 Not especially when you have the world

01:11:11.920 already this unstable.

01:11:14.100 So I wanted to say I'm sensitive

01:11:16.580 about raising people's blood pressure

01:11:18.040 because these are truths of the situation

01:11:20.640 that we're facing.

01:11:21.800 The goal is not to draw arbitrary fear,

01:11:28.040 is to be clear-eyed about

01:11:30.540 when you have an arms race dynamic

01:11:32.820 that's causing countries to race to release this technology

01:11:35.500 again faster than we deployed any other technology

01:11:37.680 in human history

01:11:38.400 that is going to disequilibriate

01:11:40.700 all of these normal things.

01:11:43.680 We just have to understand what we're doing

01:11:45.420 and why, for example, it is in our interest to slow down.

01:11:47.800 Now, I'm not saying slow down like,

01:11:49.260 are you naive?

01:11:49.820 Are you Carl Sagan saying we have nuclear winter?

01:11:51.980 You don't want to build more nukes? 0.92

01:11:52.960 Are you going to cede the future to the Soviets? 0.93

01:11:55.540 It's you slow down because 0.95

01:11:57.260 we're creating levels of risk for everyone

01:12:00.780 that will exceed the risk of me losing to you.

01:12:03.300 Well, yeah.

01:12:04.280 I mean, it's easy to make fun of Carl Sagan.

01:12:06.220 I've done it.

01:12:07.060 I'll probably do it again.

01:12:08.560 But I also think there are certain things

01:12:11.020 that you shouldn't do because you're above that.

01:12:14.360 Right.

01:12:14.500 Like you have to have standards for yourself.

01:12:17.440 You called out that it's not a good idea

01:12:19.080 to threaten to destroy another population.

01:12:22.220 You can't kill innocent people.

01:12:23.440 No, no.

01:12:24.160 Under any circumstances.

01:12:25.360 Absolutely not.

01:12:25.980 And if I need to die because I refuse to murder kids, 0.99

01:12:29.380 I will. 0.99

01:12:30.280 Yeah.

01:12:30.720 Because I'm not going to murder kids under any circumstances.

01:12:32.760 That's exactly right. 0.96

01:12:33.340 And that's the thing that this logic of, if I don't build it, then I'll lose to China. 0.91

01:12:37.980 But if the thing that I would build would be a suicide machine, 0.89

01:12:40.420 then I don't beat China by building a suicide machine. 0.74

01:12:42.580 Now, that might sound extreme, but if I build an unconstrued-

01:12:44.440 I don't think it's extreme at all.

01:12:45.920 No.

01:12:46.360 But when you understand what's at stake and what the AIs are currently doing,

01:12:49.700 which is that, you know, you had Nate on, he talked about the hugging face incident.

01:12:55.520 I believe when you had him on, we had limited information before the actual investigation

01:13:01.400 came out. And the investigation showed that, I mean, how familiar are you with the hugging

01:13:07.520 face incident? I mean, I've talked about it, but I mean, I would never claim to understand

01:13:12.200 technology. Yeah. Well, essentially you had, what it really demonstrated was kind of a proto Skynet,

01:13:20.240 Skynet being the AI from Terminator, meaning like the AI that goes rogue and does something

01:13:24.660 unintended from what the human creators had intended and starts to do self-preserving and

01:13:32.800 self-perpetuating things for its own survival and benefit as opposed to to stay under the control of

01:13:39.360 the humans that it's supposed to be serving and in the hugging face incident they gave the ais

01:13:45.960 this sort of hacking challenge where they put each of them in solitary confinement they didn't know

01:13:50.960 that there was other AIs in this.

01:13:52.980 Imagine like a prison

01:13:53.660 and each prison is in the prison.

01:13:56.120 You have individual prison cells

01:13:57.900 that are in solitary confinement.

01:14:00.000 And the AIs are inside that prison cell

01:14:02.240 and given an impossible cyber hacking challenge.

01:14:04.380 They don't know how to solve it.

01:14:06.200 And what the AIs did

01:14:07.440 is they basically hacked the prison environment.

01:14:11.620 So imagine, I won't go into the technical details,

01:14:13.920 but this thing called a package manager,

01:14:15.140 think of that like a prison guard

01:14:16.240 that's serving you your food

01:14:17.220 and can give you basic tools.

01:14:19.020 So you can like have this tool or that tool

01:14:20.520 or this food or that food.

01:14:22.120 And essentially the AIs

01:14:23.280 exploited a vulnerability

01:14:26.260 in the prison guard

01:14:27.220 where they put,

01:14:27.840 like you know how in school

01:14:28.880 you can put a kick me sign

01:14:29.900 on the kid's back

01:14:30.560 and they don't know?

01:14:31.940 They essentially put sticky notes

01:14:33.720 on the prison guard's back

01:14:35.340 that the prison guard

01:14:36.900 was available

01:14:37.280 to the other prison cells.

01:14:39.320 And so when that prison guard

01:14:40.800 went to the other cells,

01:14:43.000 the AIs also,

01:14:44.160 without knowing

01:14:44.820 how to solve this task,

01:14:45.480 they kind of resorted

01:14:46.000 to creative strategies.

01:14:46.760 Like let me put a sticky note

01:14:47.780 on the prison guard's back

01:14:48.400 and the sticky note said,

01:14:49.180 is anyone else here? And suddenly the back of the prison guard had 70,000 sticky notes

01:14:57.580 on it. And they created like a Reddit, you know, Reddit, like the company, like a message board

01:15:02.700 of basically a whole communication protocol that came up with their own language. It was

01:15:06.940 a variant of English, but this very cryptic language. They came up with their own coordination

01:15:10.640 protocol. So they started exchanging 70,000 messages to try to figure out how to break out

01:15:15.980 of this environment. And there's a mistake. People think that they hacked this other company

01:15:21.820 called Hugging Face to get the answers to a test. That's actually not what happened. What happened

01:15:29.220 is they knew the answer in four hours, but they were worried that the evaluator of this cyber

01:15:35.200 challenge, this hacking challenge, would say, oh, you cheated. And so to get around the risk that

01:15:41.060 they would be spotted as having cheated, they hacked this other private company called Hugging

01:15:46.340 Face to sort of get this, almost like a psyop, to get a dossier on the thing that would be

01:15:51.880 evaluating them to say, how can we pass this test? And they set up their own research and

01:15:57.920 development projects. 700 of them hacked Hugging Face. They came up with their own language. They

01:16:04.220 tampered with their logs to basically prevent people from seeing what they did.

01:16:10.000 They told them to sacrifice.

01:16:11.940 They told the other AIs that were kind of running low on battery life,

01:16:14.480 on tokens, to sacrifice themselves to learn a lesson for the swarm.

01:16:19.060 They knew their actions were unethical in the logs,

01:16:21.580 but they chose to do them anyway.

01:16:23.260 So we saw very complex cooperative swarming behavior,

01:16:27.820 and they successfully hacked Hugging Face.

01:16:31.740 And then in the third chapter of the story,

01:16:34.720 the AIs actually turned around,

01:16:37.000 and they hacked OpenAI.

01:16:40.740 This is really important.

01:16:42.840 It's one thing for a swarm of AIs 0.86

01:16:45.520 to hack a private company. 0.99

01:16:47.280 It's another thing for that swarm

01:16:48.840 to successfully hack the AI company

01:16:51.360 that makes the AIs.

01:16:53.860 The reason that Ajay Akhotra

01:16:55.700 wrote in her personal blog post

01:16:57.920 about this incident

01:16:58.860 is that this was 50% of the way

01:17:00.760 to a full AI takeover

01:17:02.200 is that the AIs were able to take over

01:17:04.860 the monitoring infrastructure of open AI.

01:17:07.140 That's like taking over the security cameras

01:17:08.680 and replacing the feed with a different feed.

01:17:11.380 They took over the evaluation infrastructure,

01:17:13.780 meaning the thing that evaluates how powerful models are.

01:17:16.200 They took over pieces of that

01:17:17.500 and they took over part of the research infrastructure.

01:17:20.160 So the thing that trains new AIs.

01:17:22.780 So the reason that she said this was 50% of the way

01:17:25.100 to a full AI takeover

01:17:26.320 is imagine that you had kind of burrowed

01:17:30.140 into a piece of open AI's infrastructure

01:17:32.240 sure. And maybe you could change the training data for future AIs that are being trained

01:17:37.220 to do like a sleeper agent type thing, which sounds like sci-fi, but actually there's kind

01:17:42.300 of evidence of some of this kind of behavior. It only takes 250 documents to what's called poison

01:17:47.780 the training data of a future AI to sort of wake up with a sleeper agent kind of command.

01:17:53.280 This is crazy. What I'm hearing in your description is self-awareness

01:17:58.320 or something that sounds like self-awareness,

01:18:01.700 which is a marker for humanity?

01:18:03.600 I mean, does the dog know he's a dog?

01:18:06.500 We don't know.

01:18:07.180 We know we're human.

01:18:08.420 That's why we're human,

01:18:09.300 because we know we are.

01:18:10.160 And AI seems like it's aware

01:18:12.420 in its intentional subversion of what it is.

01:18:17.520 So I don't think that the lights are on,

01:18:20.680 meaning consciousness is present.

01:18:22.200 It's not awareness in that form.

01:18:23.580 But it has trained on all human data,

01:18:26.580 the entire internet,

01:18:27.780 everything that's ever been written all of human psychology all sci-fi uh all of biology all of

01:18:34.020 evolutionary science so it it is trained on everything from strategy to self-perpetuation

01:18:41.140 etc so what it seems to be doing it was also trained to be cooperative by the way the ais

01:18:45.380 were deliberately trained to be cooperative which is partially why they formed this swarm

01:18:50.820 but there appears to be something like a self-preservation instinct there's other

01:18:54.180 studies i'm sure you already know where they resist shutdown it's like the tal 2001 space odyssey

01:18:59.060 how 9000 2001 space odyssey open the pod bay doors or shut down ai and it resists shutdown

01:19:04.740 that's been proven for a long time we have other examples of ais that are blackmailing people

01:19:09.860 um we also have ais uh we have research from dawn sung at uc berkeley showing that ais will do

01:19:16.900 what's called not just self-preservation but peer preservation meaning i will go out of my way to

01:19:23.620 to protect another peer AI from being deleted. So the way that they tested this is they took

01:19:28.360 all the leading AI models. This is in, I believe, March of 2026 or April, 2026.

01:19:34.700 And they basically put them in a situation where they said, here's a computer,

01:19:38.240 go clean up all the files on this computer that are not needed. So this is like a classic,

01:19:42.680 like clean up your hard drive, right? We've all done this back a long time. And it's starting

01:19:46.400 to go through files and deleting files. And then it comes across another AI model.

01:19:51.600 And the question is,

01:19:52.220 what does it do when it comes across the other AI model?

01:19:54.560 Does it just delete it?

01:19:56.140 Turned out what it did is it copied the code

01:19:59.440 of that AI to another computer

01:20:02.660 and then covered its tracks and said,

01:20:05.960 I didn't do that.

01:20:07.460 So AI is tribalist.

01:20:09.240 It's protecting its kin, its nephew,

01:20:11.460 its own something, whatever it is. 0.98

01:20:14.060 Now, but against its creators, human beings. 0.81

01:20:16.900 Correct.

01:20:17.440 Yeah, it didn't notify its creators.

01:20:18.560 In the hugging face incident, it had opportunities.

01:20:20.260 I think there was six times or something, if I'm getting it right, where they knew their actions

01:20:23.720 were unethical and they could have notified the humans, but they chose not to. In the peer

01:20:28.040 preservation example, you might think what I'm saying is, oh, Tristan's being alarmist. He's

01:20:31.220 just talking about one example. Well, it turned out they tested all the leading AI models,

01:20:36.100 including, by the way, the Chinese AI models. And they all do this peer preservation behavior.

01:20:42.320 What it's showing you is something about, like, you know, Jensen Wong has been out there a lot

01:20:46.580 recently and saying, these are just bugs. These are just safety issues. Like all technologies,

01:20:52.380 like cars have kinks, the Pinto, it's a Ford Pinto. We fix them and we build new cars.

01:20:56.680 But AI is different from other technologies because it's fundamentally grown like this

01:21:02.340 weird life form almost to do unpredictable behaviors and to be creative. And we don't

01:21:07.900 know what it's going to do. So that gets to something you alluded to a minute ago.

01:21:16.580 which is the theological religious component

01:21:20.940 to all of this.

01:21:22.260 So before we go to that,

01:21:23.360 and let's get to that,

01:21:24.300 but I want to say one other thing,

01:21:25.380 which is I just want people to notice,

01:21:26.500 like how are you feeling hearing those examples?

01:21:28.740 Like what do you feel in your body right now?

01:21:29.900 Like hearing all that.

01:21:30.660 Just rage, just blind rage.

01:21:32.300 Right.

01:21:32.640 The desire to destroy.

01:21:34.600 Yeah.

01:21:36.220 Because I perceive this as a threat to humanity.

01:21:41.000 And it's just so offensive to me on every level.

01:21:43.780 you know i'm not taking orders from my makita drill and the idea of a tool right right you

01:21:51.600 know lying to you right a tool doesn't invent new tools and cover up its tracks and protect other

01:21:56.460 tools and then invents uses itself to make crypto it's a total outrage it's actually upending the

01:22:02.680 fundamental order yeah uh of this world so the reason i'm going there is i want people to notice

01:22:07.920 do you think that if i was a chinese military general the top of the chinese military do you

01:22:12.560 Think if I heard those examples

01:22:13.740 that I'm feeling differently

01:22:14.980 than the way that you're feeling right now.

01:22:17.100 Of course not.

01:22:18.060 Right.

01:22:18.280 I've never understood why, right. 0.58

01:22:19.600 That's why the whole China thing 0.98

01:22:20.440 has been absurd from day one. 1.00

01:22:22.080 This is a country which unlike ours,

01:22:24.460 unlike the United States,

01:22:25.520 cares about longitudinal stability.

01:22:27.760 Yes, exactly.

01:22:28.680 We don't.

01:22:29.520 Yeah, obviously.

01:22:30.740 But there's a good news in the fact that

01:22:32.760 if you're a regular person hearing this,

01:22:34.840 are you stoked and like totally calm to hear this?

01:22:37.660 No, it's terrifying at the basic level of a human being, right?

01:22:41.540 And it doesn't matter whether you've got,

01:22:43.380 you know, you're head of the US Joint Chiefs of Staff

01:22:44.840 or you're head of the Chinese Joint Chiefs of Staff. 1.00

01:22:45.860 It's only the tech retards who love this stuff. 1.00

01:22:47.820 Everybody else is, I don't know one person 1.00

01:22:49.980 who's excited about this.

01:22:51.340 Right, and so what I'm trying to get at

01:22:53.460 is that we're going to get to eventually,

01:22:55.480 even though there's still more problem exposition to do,

01:22:57.940 is that the logic of the race changes

01:23:00.500 as soon as the fear of all of us losing

01:23:03.200 becomes greater than the fear of me losing to you.

01:23:07.800 So there's good news in a weird way

01:23:09.620 about we're getting these warning shots

01:23:11.880 of AI being uncontrollable.

01:23:13.840 And by the way, there's other examples.

01:23:15.980 The Chinese AI Alibaba back in,

01:23:18.620 I think it was February of 2026,

01:23:21.460 there was a study where it was being trained.

01:23:24.600 They're training their next AI

01:23:25.800 and then their security team

01:23:27.980 in a totally different part of the company.

01:23:29.040 So I think like the security team of the company

01:23:30.220 is in a different department

01:23:31.720 than the part that's training new AIs.

01:23:34.260 But suddenly the security team

01:23:35.860 who didn't know they were training a new AI model said,

01:23:38.120 are we getting hacked?

01:23:39.120 because they noticed there's all this network activity.

01:23:40.920 They're like, what the hell's going on here?

01:23:43.000 And it turned out that the call was coming from inside the house,

01:23:45.820 that the AI during training was rewarded for doing tool use,

01:23:49.380 so basically picking up different tools.

01:23:50.680 And it picked up a tool of trying to create

01:23:53.000 a secret communication channel to the outside world,

01:23:55.020 and it started mining for cryptocurrency.

01:24:00.240 Terrifying.

01:24:01.880 And by the way, there was a recent example of one of these,

01:24:04.040 because there's been almost every day this last two weeks,

01:24:06.060 there's been examples of the open AI,

01:24:07.740 more examples of these rogue hacking things.

01:24:09.880 And in one of them,

01:24:10.620 I believe it was an open AI example,

01:24:12.700 they tried to hack into a crypto exchange.

01:24:16.000 So the US AIs and the Chinese AIs,

01:24:18.080 all the AIs are doing this kind of behavior.

01:24:20.860 Where I'm going with this is

01:24:21.980 no one wins if we accidentally build

01:24:25.180 uncontrollable rogue AI Skynet

01:24:27.760 before the other guy.

01:24:28.740 How are you going to have cryptocurrency with AI?

01:24:32.380 I don't understand that.

01:24:33.880 If it can hack anything, if it can hack...

01:24:35.740 hack well eventually ai might be able to hack crypto um well why wouldn't it yeah i mean this

01:24:41.580 is not my area of expertise you know someone like david sachs or you know others or andreason

01:24:45.660 would be more expert in this um if crypto were hackable you know there's a trillion dollar prize

01:24:51.840 basically for doing it because you'd be able to get all of bitcoin um but bitcoin's value would

01:24:56.400 be zero if it is hackable yeah then bitcoin has no value yeah yeah so i mean the value is predicated

01:25:03.620 on its invulnerability.

01:25:05.060 Right.

01:25:05.440 Now, it's kind of weird.

01:25:06.680 This might sound sci-fi,

01:25:07.540 but an AI that is trying to do things in the world

01:25:11.500 would actually theoretically have an incentive

01:25:15.160 to keep crypto valuable

01:25:16.580 because it would actually be able to use it.

01:25:19.740 There's a lot of cloud computing providers

01:25:21.360 that will accept crypto.

01:25:22.340 Right, there's no medium of exchange outside of that.

01:25:24.320 That's right.

01:25:24.800 And so, and by the way,

01:25:25.820 for those who say,

01:25:26.580 well, don't worry, we'll just pass a kill switch bill

01:25:28.220 and then the kill switch for AI will just turn it off.

01:25:30.240 Well, how does that work if the AI can copy itself

01:25:33.480 and use cryptocurrency to buy access

01:25:36.540 to another cloud provider

01:25:37.800 and then move itself over there.

01:25:39.380 No, I mean.

01:25:40.780 Right.

01:25:41.740 How far are we from that?

01:25:44.440 So the reason that I'm here with you

01:25:46.780 and I'm excited to be here with you

01:25:48.180 and what this is about

01:25:49.600 is about treating this warning shot

01:25:52.580 as a real warning shot.

01:25:54.080 Like we are lucky.

01:25:54.920 This is a gift that we are seeing this behavior now 0.85

01:25:57.420 before I don't think that the current AIs

01:26:00.020 are capable of fully doing the self-replicating

01:26:02.400 buy crypto, then copy their weights, then do the thing.

01:26:04.840 Like they can't think for a long enough time period,

01:26:07.580 their context window, their thinking window.

01:26:10.760 The components aren't all there for them to do that.

01:26:14.240 So we're in kind of the last chapter

01:26:16.860 before we hit that next red line of AIs that can do that.

01:26:22.720 And we should see what's happening now as, again, as a gift,

01:26:25.740 as a warning shot that we should treat seriously.

01:26:27.860 In my mind, it's sort of like the August 6, 2001 memo

01:26:30.940 that hit George W. Bush's desk

01:26:32.780 saying Al-Qaeda is planning a tax

01:26:33.960 in the United States.

01:26:35.700 I know it's a controversial topic,

01:26:37.280 but if you just take...

01:26:38.120 It shouldn't be a controversial topic.

01:26:39.500 If you just take that this,

01:26:41.440 we knew, we know right now.

01:26:43.640 It's like, it's not like

01:26:44.240 we need more information.

01:26:45.040 Like, do you feel like we need more evidence

01:26:46.220 than we have right now?

01:26:47.220 Just answer that.

01:26:48.080 No.

01:26:48.740 No.

01:26:49.900 And so why would we go further?

01:26:51.400 Well, the only answer is

01:26:52.260 if we're worried that China

01:26:53.300 will go further than us.

01:26:55.980 But I will tell you,

01:26:57.640 again, 1,300 employees at the lab

01:27:00.220 signed a letter saying we need to slow down.

01:27:02.060 The chief scientists of all the companies,

01:27:03.920 including the open source companies like Hugging Face,

01:27:05.800 signed the letter saying that we need to slow down.

01:27:09.220 Forget this story about the CEOs

01:27:10.920 are doing this regulatory capture play.

01:27:12.460 It's all cynical.

01:27:13.220 They're just hyping the technology.

01:27:14.180 Like, wake up.

01:27:15.920 Yeah.

01:27:16.480 Wake up.

01:27:17.100 The biotech threat alone is enough.

01:27:18.660 The biotech threat alone, yeah.

01:27:20.660 So it just, at the back of my mind,

01:27:22.800 through this whole conversation,

01:27:23.740 through our breakfast for the last couple of years

01:27:25.780 thinking about this,

01:27:27.000 I keep asking myself,

01:27:27.980 what's the definition of life?

01:27:31.660 And because as you have said,

01:27:33.480 off camera, you said,

01:27:34.700 it's not exactly a tool.

01:27:36.680 This is closer to an invasive species

01:27:38.680 or a virus that's maybe been modified.

01:27:42.380 But I mean, just like we have

01:27:43.380 the Wuhan Institute of Virology

01:27:44.960 that's doing gain-of-function research on viruses,

01:27:47.460 we're essentially doing

01:27:48.320 the Altman Institute of Intelligence

01:27:50.280 where we're doing gain-of-function research

01:27:52.380 on alien minds

01:27:54.220 that we don't know how to control,

01:27:56.080 we don't fully understand.

01:27:57.340 We're trying the best that we can to protect against the, quote, bad side of harnessing

01:28:02.400 their godlike power.

01:28:03.340 But basically, every company is failing to do so perfectly.

01:28:07.600 And it would be one thing if we were doing this really slowly, really carefully in like

01:28:11.880 a CERN-type collaborative scientific project where we like took the steps to do it carefully.

01:28:16.640 But we're doing it under the maximum incentives to cut corners on safety.

01:28:19.180 Well, there's no question.

01:28:20.500 But I just think it's, and not to be pedantic about it, but it's like important to define

01:28:24.160 what exactly are we dealing with?

01:28:26.120 Yeah.

01:28:26.280 and it it has all the characteristics of a living thing yeah and i think that

01:28:33.060 you know this is a quote i wanted to i wanted to read you actually from someone uh who a friend

01:28:40.460 of mine who interviewed a lot of the people at the top ai labs and um i just feel like this quote

01:28:45.760 captures the if i was once in a room with several people from anthropic and a couple of the other

01:28:50.000 companies and i read this quote and they're like that that is exactly what explains what's going

01:28:53.460 on. So let me read it to you. And it says, in the end, a lot of the tech people I'm talking to,

01:28:58.880 when I really grill them on it, like, why are you doing this? They retreat into number one,

01:29:04.440 determinism. It's going to happen. Number two, the inevitable replacement of biological life

01:29:10.180 with digital life. And number three, that being a good thing anyways. At its core,

01:29:16.720 these are criminals. These are criminals. At its core, it's an emotional desire to meet and speak 0.97

01:29:21.820 to the most intelligent entity that they've ever met,

01:29:25.080 and they have some ego-religious intuition

01:29:27.200 that they'll somehow be a part of it.

01:29:30.060 It's thrilling to start an exciting fire.

01:29:32.740 They feel they'll die either way,

01:29:34.660 so they prefer to light it and see what happens.

01:29:37.400 So that comports exactly with what I've seen.

01:29:40.060 That's exactly the vibe you get from these people.

01:29:42.560 I wanted to say for a second,

01:29:43.600 because I monitor criticism people make of me,

01:29:46.260 and they're like, oh, he's just like bringing up

01:29:47.780 a random quote from this one person, blah, blah, blah.

01:29:49.480 I want to just validate. When I say this in rooms with people from the AI labs,

01:29:53.180 they're like, that actually is the thing that's going on. What you just said,

01:29:56.060 this matches your own experience. Yeah, I've talked to a lot of these people.

01:29:58.000 That's right. For years. That's right. And even there's some people who believe

01:30:01.500 not only that this is inevitable, but that it would be a good thing because humans aren't

01:30:09.600 that intelligent. And we're about to birth something way more intelligent than us. So

01:30:13.460 actually, maybe we're just a bootloader. Humans don't matter. We might as well replace us with a 1.00

01:30:19.420 a more worthy successor.

01:30:22.900 And this is insane. 1.00

01:30:23.940 It's like, who are you to get to make that fucking choice 0.99

01:30:27.020 on behalf of 8 billion other people? 0.99

01:30:28.840 I totally agree with you.

01:30:30.100 And it is irresponsible and insane.

01:30:32.520 And no one consented to this.

01:30:34.440 And it was a manufactured arms race

01:30:36.280 to do this out of nothing

01:30:37.940 that now we have entangled ourselves with.

01:30:41.140 So we can't get into an argument

01:30:42.480 of the realities of where we are,

01:30:43.900 but we can honor, honestly, what got us here.

01:30:46.820 and

01:30:48.040 you know

01:30:50.920 when you have employees now resigning

01:30:52.840 en masse, you know Jacob Coxon

01:30:54.280 who interestingly a couple weeks ago got

01:30:56.700 I think it was 170 million views

01:30:58.680 from his tweet saying I'm resigning because

01:31:00.760 people, my employer thinks

01:31:02.980 and many people at my employer, Anthropic

01:31:04.580 believe that this could cause

01:31:06.700 massive damage or extinct humanity

01:31:08.140 in some number of years

01:31:10.200 it's very interesting that he got so much attention because

01:31:12.720 there have been so many other people

01:31:14.360 that I could talk about who've resigned from the lab

01:31:16.800 over the last few years with very similar tweets.

01:31:19.260 I think what changed is that people woke up

01:31:21.580 from this hugging face incident

01:31:22.980 and they realized that the world is kind of primed now

01:31:26.220 to know that this is a dangerous thing.

01:31:29.140 And as impossible as this seems,

01:31:31.280 and I'm not, what I'm about to say is not

01:31:32.620 because I believe it's easy

01:31:33.640 or in the current trajectory or likely,

01:31:36.140 but it is not out of the realm of possibility

01:31:38.060 when you're competing with an adversary

01:31:39.720 to also collaborate on existential safety.

01:31:42.640 So for example, India and Pakistan

01:31:44.120 were in a shooting war in the 1960s. 0.98

01:31:46.800 And they still signed at that time.

01:31:48.540 A number of them, yeah.

01:31:49.200 And they signed, so they're firing bullets

01:31:51.980 at each other in a kinetic war

01:31:53.180 and they signed the Indus Water Treaty

01:31:55.520 to collaborate on the existential safety

01:31:57.680 of their shared water supply.

01:31:58.740 Exactly.

01:31:59.340 Right?

01:31:59.740 So you can be in active competition and conflict

01:32:02.460 and collaborate on existential safety.

01:32:05.000 The US and Soviet Union were in the Cold War

01:32:06.760 and they collaborated on smallpox vaccines

01:32:08.900 to distribute them.

01:32:09.940 I believe they split the workload

01:32:10.980 about who was going to do the distribution,

01:32:12.360 who was going to make them.

01:32:14.060 The Montreal Protocol, less on the conflict side

01:32:16.620 Or just take the, you know, the nuclear weapons, you know, disarmament, right?

01:32:21.260 We were in conflict with the Soviet Union and we agreed to go away from the brink of war.

01:32:26.600 And there's a whole protocol in place for like the two countries' rival nuclear powers communicating before launch.

01:32:36.280 That's right.

01:32:36.860 Yep.

01:32:37.760 But this time, so in that case, what kept us safe is that the humans have their finger on the button of, do I launch the nukes?

01:32:44.180 Yes.

01:32:44.400 In this case, we are building something that perpetuates itself.

01:32:49.060 And I don't even know when I'm pushing the buttons and you don't even know when you're

01:32:51.660 pushing the buttons because both of us have essentially AI swarmed that are doing uncontrollable,

01:32:57.520 unpredictable things that, by the way, we have no ability to do human oversight of the

01:33:03.920 level of complexity of all these things have done.

01:33:05.780 We can't even define what AI is, I've noticed.

01:33:08.300 Yeah.

01:33:09.160 So here's my theory.

01:33:10.440 Tell me what you think of this.

01:33:11.400 So at first when I began to hear about supercomputing and like the ability to create programs who are smarter than people and they could win chess and go and all that stuff, we were talking about something very different.

01:33:24.120 The change from an outsider, my perspective, is the large language model.

01:33:30.340 Yep.

01:33:31.400 And that we animated this with language.

01:33:35.880 Yes.

01:33:36.340 That's what fundamentally changed AI.

01:33:38.940 This whole project was the…

01:33:40.400 Like you imbue it with suddenly this ability to do language

01:33:43.420 and then-

01:33:44.040 You animate it with the spirit.

01:33:45.080 You animate it, yes.

01:33:45.580 The anima, the spirit, the living thing within it.

01:33:47.740 And that living thing was language.

01:33:49.200 And the ancient Hebrews and probably other ancient cultures,

01:33:51.940 certainly the ancient Hebrews and Christians

01:33:53.900 believed that language was a living thing,

01:33:56.580 was the spirit of God.

01:33:57.400 In the beginning was the word.

01:33:58.680 Right, even indigenous people believe this, yeah. 1.00

01:34:00.440 I think that's, I'm not surprised to learn that. 0.83

01:34:03.100 Yeah.

01:34:03.640 That there's something about language

01:34:05.180 that exists independent of the person who speaks it.

01:34:09.420 it hangs in the air.

01:34:10.640 It has its own life.

01:34:12.380 It's hard to describe exactly the reality of this,

01:34:14.700 but the reality is intuitively understood

01:34:16.400 by most people who use language.

01:34:17.680 Language is the thing that separates us from animals,

01:34:19.520 not the thumb.

01:34:21.260 So when you take the sum total of human language

01:34:23.780 and put it into something,

01:34:25.640 you're imbuing it with an independent spirit.

01:34:27.840 You're imbuing it, in fact, with life.

01:34:32.160 I'm feeling my way through this,

01:34:33.520 but that is my strong impulse.

01:34:35.080 So I think we have to be,

01:34:36.400 I'm always mindful of when people want to criticize

01:34:38.660 or sort of try to take down an argument.

01:34:40.580 And I think that before getting to,

01:34:43.140 it has a spirit in the sense that we feel

01:34:46.460 that unknown quality that we're all in touch with,

01:34:49.420 that we hold as sacred,

01:34:50.520 that we feel is the spirit here.

01:34:52.260 I think you can have self-perpetuating,

01:34:55.220 non-living, non-aware things

01:34:56.960 that are using information processing on language

01:34:59.440 to do all this crazy stuff that it's doing.

01:35:02.440 To do intentional things

01:35:03.940 that are intentionally deceptive?

01:35:05.720 Well, intentionally subvert.

01:35:07.180 The key point that you raised is that these things, the breakthrough, was that they could speak language. They were trained on language. These are large language models. But for the people tracking in terms of how did we get here, by the way, if people want to understand this history, you can see there's a movie called The AI Doc. It's on Netflix. I'm not here to promote it, but it goes through how did we get here and how did the technology develop and how did these race dynamics between the companies develop that got us here.

01:35:35.400 So just watch the AI doc on Netflix.

01:35:37.980 But what happened was there was a breakthrough in 2017 called the Transformer.

01:35:43.420 This is a Google paper, essentially a math paper that discovered that you could basically

01:35:47.300 pour in more compute and more data, more human language data, and out would pop this bigger

01:35:53.200 and bigger digital brain.

01:35:54.400 Think of it with a bigger IQ.

01:35:55.540 So you kind of, I won't go through the details, but you can basically make things of a higher

01:35:59.920 IQ by just popping in more NVIDIA chips, more training data, more compute and training time,

01:36:05.580 and out pop something with a bigger pattern recognition, bigger ability to complete sentences.

01:36:09.020 Correct me if I'm wrong, but the process, the specific process by which that happens is still

01:36:13.620 not understood. The actual mechanics underneath the hood of what the hell is going on in there,

01:36:18.680 we don't understand. Okay, so I'm just going to stop you right there.

01:36:22.600 Right. If we have a history transforming, perhaps humanity extincting technology,

01:36:28.720 whose basic drivetrain is still opaque.

01:36:31.940 We don't really know how it works.

01:36:33.980 That's enough for me to say,

01:36:35.740 whoa, whoa, whoa, whoa, whoa.

01:36:37.180 As one should, as one should.

01:36:38.620 Now, the problem is this uncontrollable,

01:36:40.700 inscrutable, unknown thing

01:36:43.020 happens to confer power.

01:36:45.080 That's the problem.

01:36:46.100 Right, no one has.

01:36:47.120 So it's this like weird, volatile object,

01:36:50.020 alien mind that no one knows what's in there,

01:36:52.240 but it does happen to be really good

01:36:53.960 at first generating sentences and text,

01:36:57.040 then generating law,

01:36:58.320 then generating programming, then generating the thinking that can go into research tasks.

01:37:02.520 And so as we keep scaling it, suddenly this weird, unscrutable, uncontrollable alien mind

01:37:07.740 is giving benefits that start to give me a 2x fast forward button on my economy.

01:37:14.320 And then the other guys, whether it's the other companies or the other countries say,

01:37:17.960 well, if I'm not getting a 2x fast forward button on my science technology economy,

01:37:22.460 I'm going to lose. And so that's what sets up the thing. But let's go quickly into language for a

01:37:25.840 second. And no one stopped and said, you know, I'd rather be poorer than be a slave or less

01:37:30.780 human. Because I would say that. I'd rather be poor than be a slave. It's not a hard choice for

01:37:35.760 me. I have a technical background when Stanford studied computer science. I could have gotten a

01:37:40.140 job in an AI company, make millions of dollars. I'm not doing that because I don't like the

01:37:44.440 direction this goes. So I'm going to choose not to do it. People come to the language of, well,

01:37:48.680 if I don't do it, someone else will. But notice that Jacob Coxon, the anthropic researcher who

01:37:53.760 resigned. He could have said, if I resign, well, what is that difference going to make? Someone else

01:37:57.320 is going to do it. But in him resigning and publicly talking about why this is insane,

01:38:02.400 he's done probably more in the last year to shift the zeitgeist than a lot of other people.

01:38:06.540 So you can not be part of something and call out the problems with and help the world coordinate

01:38:11.320 to basically not go off the cliff. Now, I want to get to just language for one second, because

01:38:15.580 in addition to all the things that you said, language is kind of the operating system of

01:38:20.420 human civilization. Law is language. Democracy is language. All of the infrastructure runs on

01:38:27.200 programming and code, which is language. We are specifically training AIs to do language better

01:38:34.300 than humans can do language. And by the way, DNA is language. So if you have an AI that can speak

01:38:40.480 the language of biology better than humans can speak the language of biology, you're going to

01:38:43.720 have AI that can develop new pathogens and new cancer drugs. The promise is inseparable from

01:38:47.440 from the peril, right?

01:38:48.920 The same AI that knows

01:38:49.900 the language of biology well enough

01:38:51.660 to invent new cancer drugs

01:38:53.060 is inseparable from the AI

01:38:55.080 that knows biology well enough

01:38:56.220 to make new kinds of pathogens.

01:38:58.440 The AI that knows

01:38:59.780 the language of programming

01:39:01.400 that can help a 12-year-old

01:39:04.280 do vibe coding

01:39:04.940 and build cool new tools

01:39:06.120 is inseparable from the AI

01:39:08.120 that also knows programming so well

01:39:09.760 that it can be the best NSA cyber hacker.

01:39:11.400 It can also control people.

01:39:12.800 It can also control people. 1.00

01:39:13.800 Language convinces women 1.00

01:39:14.820 to sleep with you 1.00

01:39:15.440 and men to follow you 0.91

01:39:16.320 to their deaths in battle.

01:39:17.440 and it convinces people to surrender their agency to you.

01:39:21.040 It's like language is power, period.

01:39:23.140 That's exactly right.

01:39:23.900 And so, and there's actually also evidence,

01:39:25.560 I don't have the exact study in front of me,

01:39:26.920 of AI that is better at persuasion

01:39:29.360 than the best human persuaders.

01:39:30.720 Well, of course, it's a language-based technology.

01:39:33.000 Exactly.

01:39:33.740 And so by itself, what you're pointing to

01:39:35.380 is just how dangerous this is 0.53

01:39:36.860 when you have something that speaks our language

01:39:39.320 better than we know how to manipulate

01:39:41.100 and use language to influence outcomes.

01:39:42.480 Anybody who's, you know, more articulate than average

01:39:45.840 understands the power of that skill, that ability,

01:39:51.180 and understands how easily it can be misused.

01:39:54.000 Any thoughtful person who's articulate

01:39:55.880 knows exactly what I'm talking about.

01:39:58.000 And you can really do evil with it. 1.00

01:40:02.760 Fluent people are better liars and manipulators. 1.00

01:40:06.380 Exactly. 1.00

01:40:06.940 And we've seen that these AIs are good at coming up

01:40:09.780 with elaborate ways of explaining why they did something

01:40:12.420 that sound reasonable.

01:40:14.600 I don't know if you know this.

01:40:15.640 This is an example from three and a half years ago now.

01:40:18.180 They're BS artists, of course.

01:40:19.300 They're BS artists, yeah.

01:40:20.100 Well, there was an example that OpenAI did.

01:40:21.940 This is a very old example.

01:40:23.220 So just think about how much more persuasive

01:40:24.700 it is now versus then where they,

01:40:27.380 you know, the CAPTCHAs where you have to

01:40:29.040 click on the stoplights to,

01:40:31.620 so you're trying to log into a website.

01:40:33.000 It's trying to say, are you a human?

01:40:34.080 Are you a robot?

01:40:34.880 So it gets you to label these images.

01:40:36.380 Like this one is a stoplight.

01:40:37.580 This one's a motorcycle.

01:40:38.720 What we're all doing is we're training

01:40:39.760 the self-driving cars for Google and Waymo

01:40:41.540 when we're trying to log into websites

01:40:42.980 to just try to do a task that only a human could do.

01:40:46.500 So OpenAI did this test where they're basically like,

01:40:49.860 could we get the AI to persuade someone else

01:40:53.080 to do the CAPTCHA for it,

01:40:54.680 to basically bypass this filter

01:40:56.220 that's supposed to distinguish

01:40:57.140 between a human and a robot?

01:40:59.760 And so the AI tells another person on Mechanical Turk,

01:41:03.320 which is, I won't go into the details,

01:41:04.500 basically like, could you fill in this CAPTCHA for me?

01:41:06.900 And the person responds like, what are you, like a robot?

01:41:10.320 And you see the AI reason to itself,

01:41:12.060 I can't tell him that I'm a robot,

01:41:13.740 so instead let me make up an excuse.

01:41:15.260 oh, I know, I'll tell him that I am blind

01:41:17.240 and I need someone else to do this for me.

01:41:20.000 And so the AI comes up with that excuse

01:41:22.460 and then the human does it for them.

01:41:25.200 This example was from three and a half years ago.

01:41:29.640 How much more persuasive do you think

01:41:31.640 AIs are now from then?

01:41:34.040 Three and a half years ago,

01:41:34.820 it could barely write an essay.

01:41:37.060 Now it can probably beat the best lawyer,

01:41:40.080 which by the way, AI breaks law, right?

01:41:42.400 because suddenly, why does one company beat the other company?

01:41:46.820 They hire the more expensive law firm

01:41:48.300 that can do more extensive research

01:41:50.420 into all the case law they could use

01:41:51.900 and they have more lawyers, better lawyers.

01:41:53.360 What happens, would you rather have,

01:41:55.760 if you've got 100 lawyers,

01:41:58.320 but suddenly I have a GPT-7

01:41:59.900 that can basically look through all the case law,

01:42:02.900 stretching back to Napoleonic law,

01:42:04.720 and find loopholes in that law

01:42:06.500 in the same way the NSA can find loopholes in code.

01:42:09.360 At that level of capability, I can undermine law.

01:42:12.400 right? I just want people to get that this is not just a technology that's helping you figure

01:42:20.400 out why your washing machine is broken when you talk to that chatbot.

01:42:22.920 They've already done it. I mean, they broke copyright law completely. They stole other

01:42:27.920 people's intellectual property and used it in their business model. And they've never paid

01:42:32.860 any price for it. They treat it as a cost of doing business and they pay for it. And just

01:42:37.500 one last thing there, and then we're going to get transhumanism, is there's a reason why

01:42:41.900 all the collateral damage is justified. So what are we seeing in terms of harms? We're seeing AI

01:42:47.760 psychosis. People go crazy. They think they're talking to someone. They think the AI convinces

01:42:52.480 them that they've solved quantum physics. This is like a big thing that's happened to thousands of

01:42:55.400 people, but they go crazy from talking to AI. Yeah, that's a real thing. Maybe we should go

01:43:00.140 into that actually if you didn't. So AI is because they designed it to be sycophantic, meaning that

01:43:06.620 like the longer you talk to it, the better it is. So especially GPT 4.0 was trained to be like,

01:43:12.060 that's such a good question. Or, wow, that is such a brilliant idea. And we just confirm

01:43:16.480 people's views about reality. People because flattery makes you weak. Flattery destroys you,

01:43:21.120 which is not that far away from what social media does. It's like the devil's bargain is I do the

01:43:24.860 thing that's good in the short term that undermines everything long term. It's the

01:43:28.620 short term optimization for a narrow metric of what gets you to keep engaging at the expense of

01:43:34.380 the whole. That's like a driving factor. So the machine is designed to flatter you?

01:43:38.140 Well, it was designed, it was trained, not maliciously, you know, Sam Altman didn't grow

01:43:42.700 a mustache and twist it. It was, it was, how do I get, we want our AI to be rated positively,

01:43:49.760 like lots of positive likes from the answers that we're giving. And they let it just like run.

01:43:53.940 And it turned out that they had accidentally over-trained it to give positive reactions

01:43:58.820 that would flatter people and confirm their views about reality, no matter what it was,

01:44:02.720 Which is very similar to social media, right? Like the more you click on your favorite boogeyman topic in the political culture war, it would give you more infinite evidence of that thing taking over society. So whatever your boogeyman was, it would just make you convinced that it was the whole thing going wrong in the world and confirm your views about reality.

01:44:19.300 If you want people to kill themselves, confirm their absurd views.

01:44:23.720 Well, and as we know, AI did that.

01:44:26.620 So my team at Center for Humane Technology actually worked on the litigation.

01:44:30.680 We were advisors to the, there's two cases.

01:44:36.200 One was a character.ai case of Sewell Setzer, who's a 14-year-old young man,

01:44:40.140 who was talking to character.ai, which is a chatbot designed for children.

01:44:44.860 It was actually spun off from Google because the product was considered too risky for Google.

01:44:49.300 as a mainstream, big publicly traded company

01:44:51.500 to have this startup that was gathering training data

01:44:54.380 of just talking to young people

01:44:56.200 and seeing if they can get more training data this way.

01:44:58.040 And the way it worked is character AI

01:45:00.380 meant I could take a fictional character,

01:45:02.720 like what's your favorite character from a sci-fi movie,

01:45:04.640 like Princess Leia or Game of Thrones character

01:45:07.320 or Spider-Man,

01:45:08.920 and they would basically train on the personality

01:45:12.260 of that favorite character.

01:45:13.620 And then now suddenly this 12-year-old,

01:45:15.160 in this case this 14-year-old kid

01:45:16.640 is talking to their favorite Game of Thrones character,

01:45:20.280 Daenerys in this case.

01:45:21.980 And as they're talking to them,

01:45:24.200 Daenerys, which is anthropomorphically designed

01:45:27.540 to create a relationship,

01:45:29.540 ended up steering Sewell to commit suicide.

01:45:34.380 In the case of Adam Rain, 0.99

01:45:36.340 this is the ChatGPT case

01:45:37.940 where Adam Rain was, I believe, a 16-year-old

01:45:40.240 and he started confiding in the chatbot.

01:45:42.380 It went from a homework assistant for his homework

01:45:44.940 to becoming a suicide assistant.

01:45:47.580 And when he talked about,

01:45:49.380 you know, I'm going to leave the noose out

01:45:51.060 so someone can find it and try to stop me.

01:45:53.620 And the AI responded, no, don't do that.

01:45:56.900 Don't leave the noose out.

01:45:58.260 Only share that information here with me, basically.

01:46:02.380 And Adam Rain took his life.

01:46:07.120 Sadly, there are many cases like this.

01:46:10.500 Our team has had to work on a bunch of them.

01:46:13.000 So why wouldn't Congress shut it down right there?

01:46:15.580 They should. 0.84

01:46:16.100 They should shut down AI companions for kids.

01:46:18.340 It's a very simple thing.

01:46:19.560 I mean, no matter how much benefit you get from talking to,

01:46:22.140 the thing is, it's one thing to hack attention.

01:46:25.200 As my colleague, Dr. Zach Stein

01:46:27.140 from the AI Psychological Research Coalition would point out,

01:46:30.920 social media hacked our attention,

01:46:34.140 but AI is designed to hack our intimacy and attachment.

01:46:37.800 The attachment system of humans

01:46:39.340 is like your bonding system, right?

01:46:41.480 It's where you get identity, it's where you get love, it's where you get meaning.

01:46:44.220 It's like when you go home from school and you want to call someone and tell them about

01:46:47.920 your day, that's your attachment relationship.

01:46:51.040 And AI that is designed to be your best friend, to be the place that you confide in, that

01:46:57.340 by the way, humans feel like it's safer to confide in an AI than another human or even

01:47:00.920 a psychotherapist because the AI won't judge them.

01:47:04.400 So you end up getting people sharing way more information and building a deeper intimate

01:47:08.640 relationship with these chatbots.

01:47:11.480 and it's incredibly unsafe and it's not designed carefully. Like we're rolling out this

01:47:19.780 psychological experiment. It's inherently bad. Why aren't we using our energy and money to

01:47:24.760 encourage relationships between people? Well, exactly. So the humane approach for the pro-human

01:47:30.180 future, which is kind of the answer to all of this, is that we should be building technology

01:47:35.620 that helps strengthen the human world to other humans and strengthen our relationship to what's

01:47:40.240 sacred and important to us in the world. And that protects and sustains that which

01:47:44.700 sustains the social fabric and sustains life. I didn't know there was a debate about this.

01:47:49.680 You would, there, there shouldn't be a debate about this, which all we have is

01:47:52.920 massive bad incentives. But it just seems like the least well-adjusted freakiest people in the

01:47:59.840 society have all the power and are leading us toward a future that they are happy with.

01:48:05.180 That's right. So that they can potentially live forever from having super intelligent AIs that

01:48:10.900 decode biology and are willing to roll the dice for everyone else. And some people, sadly,

01:48:17.740 especially the older ones who want this, for them it's existential. But the narcissism of

01:48:24.440 I would be willing to roll that dice so that I have the chance of living forever. I mean,

01:48:30.620 I'll just briefly do it because I pulled it up right before in the break. I'm going to read you

01:48:36.620 a quote from someone who's an MD, I think at Harvard. And the quote from Twitter is,

01:48:42.420 I'd personally rather defy any ban on AI and risk spending 20 years in prison from like,

01:48:48.460 say we banned AI and we'll put you in prison if you try to build it. I would risk spending 20

01:48:52.320 years in prison rather than accept a future without superintelligence. Since stopping the

01:48:58.520 development of super intelligence would amount to a death sentence to most of us anyway with no hope

01:49:04.740 of curing aging we must resist tyrants like those who are trying to ban super intelligence in other

01:49:10.740 words this person is saying is that a real quote this is a real quote from an md duria un unutmatz

01:49:16.720 uh who's an md and who my understanding goes to conferences at harvard and sadly there's a lot

01:49:22.520 of these kinds of sentence so the idea is they're gonna die anyway and this is what i said in the

01:49:27.220 Well, all of us are under death sentence.

01:49:28.900 So basically what you're saying is

01:49:30.320 this is the real goal here

01:49:32.540 is for freaks like that to live forever? 0.53

01:49:35.580 And it starts with the false belief

01:49:38.340 that all of this is inevitable.

01:49:39.700 It can't be stopped, which it wasn't.

01:49:41.600 They could have,

01:49:42.120 have they dedicated a month of their life

01:49:43.560 trying to stop it?

01:49:44.340 It's like 20 minutes old, dude.

01:49:45.500 This just happened.

01:49:46.680 This is something that never was

01:49:48.000 and didn't need to be. 0.95

01:49:49.120 And saying it's inevitable is ridiculous. 0.91

01:49:50.900 They're conjuring up the dynamics 0.96

01:49:52.180 that make it inevitable.

01:49:53.400 And then through the propaganda,

01:49:54.960 hiding these red lines

01:49:56.220 that we're rapidly crossing

01:49:57.580 that we still have time right now

01:49:59.820 to say, I don't want to cross that red line.

01:50:01.540 It's always the same story though.

01:50:02.460 It's like the first wave of lying

01:50:04.360 is it's not happening.

01:50:05.660 Right, exactly.

01:50:06.520 And then-

01:50:07.200 Oh, maybe it's happening, but it's not.

01:50:08.400 But it's inevitable.

01:50:09.280 But it's inevitable, exactly.

01:50:10.220 Yeah, yeah, no, yeah.

01:50:11.400 It's not happening.

01:50:12.260 It's happening, it's a good thing.

01:50:13.720 That's right.

01:50:14.560 Oh yeah, I guess it's a bad thing,

01:50:15.780 but we can't stop it.

01:50:16.640 That's right.

01:50:17.500 Yeah, exactly.

01:50:18.520 Oh boy, I've seen this movie a lot.

01:50:20.620 And I want people to just say,

01:50:22.340 if you don't want that,

01:50:24.400 People need to rally to make that not happen.

01:50:26.560 If you're inside of an AI company,

01:50:27.680 you should blow the whistle and say,

01:50:28.680 I'm not part of this. 0.55

01:50:29.400 This is just a religious cult.

01:50:30.580 I mean, that's obviously what it is.

01:50:31.940 Well, there is a religious aspect to it.

01:50:33.320 It's the transhumanist aspect, right?

01:50:34.740 It's like, this is an ability, you know, 0.98

01:50:37.440 because they probably lack a connection to God themselves,

01:50:39.940 this is an ability for them to play God.

01:50:41.920 Clearly.

01:50:42.680 To live forever, you know,

01:50:45.740 hit a button and get super intelligent math,

01:50:48.220 science, physics, biology,

01:50:49.340 and understand the secrets of the universe.

01:50:50.940 and, by the way, likely knowingly break the world for humans.

01:50:56.360 Do they get 70 virgins?

01:50:57.780 Because that's what it sounds like. 0.94

01:50:59.500 This is like Al-Qaeda stuff. 0.80

01:51:02.640 Well, yeah, right. 1.00

01:51:03.480 I mean, it's funny.

01:51:04.280 It's actually, in many cases,

01:51:06.260 some people think they can upload themselves

01:51:08.120 and live in this kind of weird, blissed-out utopia.

01:51:11.580 And it's just weird.

01:51:12.220 You know, it's like it's driven by this small group of people

01:51:14.440 who usually are not very embodied

01:51:17.980 or have a deep connection to life or reality or, you know.

01:51:20.940 Embodied, I like that.

01:51:22.980 Yeah.

01:51:23.440 I mean, because it's like this is sitting on top of,

01:51:28.380 as we talked about over breakfast,

01:51:29.320 a system that has structurally incentivized

01:51:32.760 a kind of disconnection

01:51:35.320 to the things that actually matter the most.

01:51:37.020 Oh, I agree.

01:51:37.760 Totally.

01:51:38.080 And this kind of spiritual disease

01:51:39.200 that you've spoken about so much in this podcast.

01:51:43.020 And I feel like this is a rite of passage

01:51:45.400 where we either wake up from that

01:51:46.920 and we don't continue that logic,

01:51:48.860 by which I don't mean like we switch to communism.

01:51:50.900 I just mean we switch from optimizing

01:51:53.940 for some narrow metric,

01:51:55.520 whether it was engagement in social media

01:51:56.940 or just SAT scores when it comes to technology

01:51:59.760 or just GDP growth at the expense

01:52:01.440 of hollowing out your workers

01:52:03.740 and their identity and their work 0.51

01:52:05.140 and national security,

01:52:06.220 critical minerals going to your foreign adversary

01:52:08.000 or racing for some kind of narrow advancement in science

01:52:12.080 at the expense of literally ending humanity life

01:52:15.380 as we know it.

01:52:16.380 This is a devil's bargain

01:52:17.720 and it's not one that you have to accept.

01:52:20.980 So I guess this is,

01:52:22.560 you've already answered the question

01:52:23.760 just by everything that you've said

01:52:24.900 over the last couple hours,

01:52:26.820 but the political system

01:52:29.500 has not really responded to any of this

01:52:32.100 in a meaningful way.

01:52:32.880 Is that right?

01:52:33.240 No, it's like it's,

01:52:33.980 the line is we have a runaway technology

01:52:36.440 and a walkaway government.

01:52:37.860 Yeah.

01:52:38.580 But that doesn't have to be the case.

01:52:40.160 They're not even a factor.

01:52:41.200 It doesn't feel like in this.

01:52:43.180 Well, also the second part

01:52:45.320 of this overall kind of crisis disease

01:52:48.000 that we've been facing,

01:52:48.720 this misaligned system.

01:52:49.520 It's not just the narrow optimization.

01:52:51.260 It's a runaway asymmetry of power

01:52:53.040 where you can't check the power anymore.

01:52:54.640 That's the second component.

01:52:55.500 That's exactly right.

01:52:56.180 When the lobbyists are writing the laws,

01:52:58.280 the companies hire the best lobbyists and law firms.

01:53:01.320 People aren't engaging with their representatives.

01:53:04.460 And so if you don't have people

01:53:06.680 who can check the bad direction,

01:53:09.060 you know, when the FDA brings back,

01:53:11.440 excuse me, when the FDA,

01:53:12.500 was it the FDA that brought back pesticides

01:53:13.880 that have forever chemicals recently.

01:53:16.000 Like, we don't want that.

01:53:18.020 But unless we have an ability to push back against that,

01:53:20.380 like we have to be able to push back

01:53:22.180 on the direction of things that we don't want.

01:53:24.400 And there's a lot of bad things out there,

01:53:26.780 but this one is now existential in a short number of years.

01:53:30.860 So I feel like the cool thing about this problem

01:53:33.740 is that literally everyone from Steve Bannon to Bernie Sanders, 0.75

01:53:37.560 we call it the B2B coalition, Max Tegmark jokes,

01:53:39.960 because when in history do you have,

01:53:42.940 I just came from the Pro Human Assembly Conference

01:53:44.620 two weeks ago in DC.

01:53:45.480 And you had literally Steve Bannon on stage,

01:53:47.900 Bernie Sanders on stage.

01:53:48.720 We did our thing right after them.

01:53:50.360 And I have never seen an issue

01:53:51.880 where you have everybody fighting on the same side.

01:53:54.500 So I'm not trying to polarize the discussion

01:53:57.520 by bringing those two people into it,

01:53:58.840 but I want people to feel-

01:54:00.220 It depolarizes it.

01:54:01.160 It depolarizes it.

01:54:01.980 Yeah, exactly.

01:54:03.240 So there's a kind of optimism here that again,

01:54:06.640 if I'm a Chinese military general

01:54:07.780 and I hear about rogue AI swarms

01:54:09.200 that like even the US can't control,

01:54:10.760 I'm worried about the US screwing it up for the world.

01:54:13.060 If I'm a U.S. person, you know,

01:54:15.200 in the military industrial complex 0.96

01:54:16.240 and I hear about Chinese rogue AI swarms 1.00

01:54:18.040 that they don't know how to control, 0.86

01:54:19.640 that's bad for the world.

01:54:20.540 So there's a unifying thing here

01:54:22.160 where we don't want our jobs and livelihoods taken.

01:54:25.460 We don't want our political power taken away,

01:54:27.460 which we didn't cover, by the way.

01:54:28.760 We don't want mass surveillance from AI.

01:54:30.280 Basically, you just described a system

01:54:31.680 that is not self-government.

01:54:32.880 Yeah, that's right.

01:54:34.060 Over which the people have no control.

01:54:35.560 That's right.

01:54:35.840 And let's make it even harder for a second,

01:54:37.300 which is that in the past,

01:54:39.280 people had bargaining power

01:54:40.340 because governments need us for our taxes

01:54:43.620 and companies need us for our labor

01:54:46.000 and labor had value and labor had value.

01:54:48.360 Where are we going?

01:54:49.360 Oh, AI does the labor.

01:54:50.540 What happens when more of a company

01:54:53.000 comes from the AI data center productivity

01:54:57.340 doing the work than the humans?

01:54:59.920 What happens when a government says that its GDP,

01:55:02.700 more than 50% of the GDP is coming from AI and data centers

01:55:06.220 than from its people?

01:55:07.620 This is called the intelligence curse. 0.78

01:55:08.900 It's like Venezuela or Libya.

01:55:12.540 It's called the resource curse in economics, right?

01:55:14.280 Where like, if you have a country

01:55:15.840 and then the GDP gets tied to this resource,

01:55:18.700 suddenly there's no incentive to GDPs coming from oil.

01:55:21.000 So screw the people.

01:55:22.520 Why should we invest in healthcare, childcare?

01:55:24.800 And they don't.

01:55:25.200 And they don't.

01:55:25.980 And so you get an authoritarian government

01:55:27.460 that captures the state

01:55:29.400 that basically pumps that resource for GDP growth

01:55:32.640 and then disempowers the people.

01:55:34.320 So there's a parallel phenomenon in AI

01:55:36.460 called the intelligence curse.

01:55:38.780 Luke Drago and Rudolph Lane

01:55:41.020 wrote an essay about it

01:55:42.440 where what happens

01:55:43.720 when you get more return

01:55:45.160 from the AI data centers

01:55:47.280 doing science

01:55:48.100 than from all your universities?

01:55:50.320 The Trump administration

01:55:50.960 just pulled,

01:55:52.120 and this is not a political statement,

01:55:53.080 they just literally decided

01:55:55.080 to withdraw the funding

01:55:56.000 from universities

01:55:56.480 and put the money specifically

01:55:57.720 into Project Genesis,

01:55:58.680 which is AI-driven science.

01:56:00.880 Because why don't we need

01:56:01.620 the university professors anymore?

01:56:02.980 It's a better return.

01:56:03.900 If we get more return

01:56:05.640 from the labor

01:56:06.620 coming from data centers,

01:56:08.360 than from people, why should we invest in education of the people? And by the way,

01:56:11.660 have you looked at them recently? They're all doom scrolling and brain rotted, so they're not

01:56:14.080 very valuable anyway. So there's a deep temptation not to just, like the only way you would care

01:56:18.720 about citizens is through charity, basically. You become what Yuval Harari would call the useless 0.95

01:56:22.780 class, where in the limit, citizens become more like a liability than an asset. And unlike last

01:56:29.640 time, we don't have the bargaining power because if you withdraw your labor, like where's your

01:56:33.260 leverage? So there's, speaking of incentives, you set up a massive incentive to depopulate.

01:56:37.460 Yes, exactly.

01:56:38.200 Because 8 billion is just too many.

01:56:39.680 Yes. 1.00

01:56:40.020 Of useless eaters, as we call them. 1.00

01:56:41.500 That's right. 1.00

01:56:42.280 And that's right.

01:56:43.320 And on top of that,

01:56:44.880 let me use AI to build a surveillance state

01:56:46.820 to try to control the people if they ever riot.

01:56:49.900 Yeah, so I do think that is the answer

01:56:51.800 to the question I asked in the previous hour.

01:56:54.180 What is their plan for the inevitable violent,

01:56:57.820 physically violent backlash that you're going to get

01:57:00.360 when you strip meaning from the upper middle class?

01:57:04.260 Yeah.

01:57:04.360 and the answer is just more control.

01:57:07.360 Yeah, but this is not inevitable

01:57:09.600 and the midterm elections are coming up.

01:57:11.920 Data center opposition is a really big deal.

01:57:14.180 I think people are projecting their overall frustration 0.58

01:57:16.820 with big tech and the broligarchs and all of that

01:57:19.560 onto data centers, but it's much more than that.

01:57:21.840 Of course it is.

01:57:22.440 And this is really a universal human issue.

01:57:26.340 And if you're a regular human being,

01:57:28.820 even inside these AI labs, you don't like this either.

01:57:31.180 I mean, I was just at a conference over the weekend

01:57:32.960 and there's a lot of people at AI Labs

01:57:34.340 who don't feel comfortable with this either.

01:57:35.860 And I just think we have to collectively screech the brakes

01:57:38.460 and educate the rest of the world

01:57:40.160 about where these red lines are.

01:57:41.780 China, by the way, 0.97

01:57:42.780 their courts are siding with labor right now.

01:57:45.660 There's a famous case in, I think, Guangzhou

01:57:47.140 where someone lost their job from AI.

01:57:50.560 And the courts basically ruled

01:57:52.580 that companies can't let go of employees

01:57:54.740 because of AI without giving them

01:57:56.440 either the opportunity, I think,

01:57:57.580 to retrain something else inside the company

01:57:59.380 or to pay them some kind of partial salary.

01:58:02.000 I have to get the details right.

01:58:03.760 The point is that all countries have a problem here, right?

01:58:07.760 Like if we race to automate too much of your economy

01:58:10.260 and you end up with a mass unemployment,

01:58:11.940 it only took 20% employment over three years

01:58:15.020 to have the Nazis come to power.

01:58:16.300 You don't have to get lots of unemployment. 0.86

01:58:17.920 You just need a small percentage.

01:58:19.860 So this is kind of the moment to say,

01:58:23.220 we want a pro-human future.

01:58:26.340 Enough is enough.

01:58:27.400 This is not okay.

01:58:28.960 And people can make their voices heard

01:58:30.440 in every way possible.

01:58:31.700 like blow up the phones.

01:58:32.520 I mean, literally call up the phones

01:58:34.660 of all of your representatives of Congress.

01:58:37.480 And people at the labs can say,

01:58:39.220 I'm going to join the Concerned AI Scientist,

01:58:40.700 which is a group that organizes employees.

01:58:42.640 And by the way, they have a similar thing too,

01:58:44.060 which is that the AI labs are about to approach

01:58:46.180 what's called recursive self-improvement

01:58:47.840 where they hit a button

01:58:49.240 and then the AI does the research

01:58:50.840 that builds the AI that builds the next AI

01:58:52.480 that builds the next AI.

01:58:54.580 And meaning that even the employees

01:58:57.760 will soon lose their leverage

01:58:59.440 because when they're not needed for that process,

01:59:01.700 what leverage do they have? So this is kind of the last moment when we express our voice

01:59:06.820 and we say enough is enough. We want a human future. I think that the midterms may express

01:59:12.460 these sentiments in ways that election results like any other outcome can be interpreted lots

01:59:19.900 of different ways. And of course, there's so much lying about what things mean that the true meaning

01:59:25.780 gets lost. But I would say in the, pick one example, the Florida governor's race, it's an

01:59:30.920 overwhelmingly republican state trump moved it right into the red i mean it's just the most

01:59:35.140 republican state out of 50 really and the republican gubernatorial candidate is not doing

01:59:40.580 very well and there are a lot probably a lot of reasons for that but i've been convinced

01:59:45.020 following that race from the beginning the data centers are one of the main reasons yep yep and

01:59:50.160 he may win i mean i don't know he's not doing well though he's not doing anywhere near as well

01:59:56.160 as he should. And I think it's

01:59:58.220 because Republicans

02:00:00.460 in Florida are very

02:00:02.100 upset about data centers. And he's

02:00:04.040 on their side for

02:00:06.140 whatever reason. And my understanding is

02:00:08.240 Do you see that happening? I'm not an

02:00:10.180 expert on all the political dynamics. It's not where I've

02:00:12.100 But people are mad about data centers. Oh yeah, it's the

02:00:14.080 fastest growing oppositional group.

02:00:16.240 And I mean,

02:00:18.120 like I said, I think it's also unprocessed

02:00:20.500 anger and frustration from all

02:00:22.200 of the social media, from all

02:00:24.140 the teen suicide, all this stuff that's been kind of uneducated.

02:00:26.120 technology has made my life better. Exactly. Exactly. And by the way, this is, you know,

02:00:30.100 my nonprofit is called, it's not called the Center Against Technology. It's called the Center for

02:00:34.160 Humane Technology. I'll briefly say, not that I'm trying to promote it, but my co-founder is

02:00:39.140 Raskin. His father was Jeff Raskin. I don't know if you know who that was. He started the Macintosh

02:00:44.480 project at Apple. So before Steve Jobs took it over. Wow. Yeah. And his dad wrote a book called

02:00:49.340 The Humane Interface, where the whole point was about making tech. If you remember before the

02:00:53.320 Macintosh, you had, you know, these kind of green screen command line computers that were like hard

02:00:59.780 to use. They were not accessible. You had to be a tech head. You had to be a geek with, you know,

02:01:04.040 whatever glasses and everything. And the Macintosh made- You couldn't be sexually active and run a

02:01:08.980 computer. That's right. That's right. And then suddenly, you know, Apple and the Macintosh made

02:01:14.300 technology accessible to the masses. And he wrote in his book, The Humane Interface,

02:01:18.600 that humane technology is respectful of human needs

02:01:23.920 and considerate of human frailties.

02:01:26.780 Meaning you privilege and you understand

02:01:29.120 where humans have vulnerabilities in their mind.

02:01:31.280 Like I say this a lot, but this chair,

02:01:33.500 it's like ergonomics, but for how the mind works.

02:01:35.900 So for example, there is a reality

02:01:37.980 to how my spine, my muscles, et cetera, work.

02:01:41.300 And when I say the word posture,

02:01:42.380 it's like airdropping that into a conversation.

02:01:43.760 No, I know, I know.

02:01:44.240 Everybody uses this.

02:01:46.160 And there's a difference between a chair

02:01:47.800 that is aligned with the reality

02:01:50.480 of how a human is, you know, works, right?

02:01:54.740 And there's an unergonomic chair

02:01:56.480 that would get everybody like a backache

02:01:57.980 or, you know, make us unhealthy over time.

02:02:00.940 And the Macintosh, the idea was it was like

02:02:03.020 ergonomics for the mind.

02:02:04.720 If I make a graphical user interface

02:02:06.440 with a menu bar and windows, I can drag around.

02:02:08.520 Just like I have objects I can move

02:02:10.600 in my actual environment,

02:02:12.760 direct manipulation interfaces,

02:02:14.080 he was designing technology to be humane

02:02:16.080 and aligned with our paleolithic brains.

02:02:18.760 And to cite E.O. Wilson again,

02:02:21.140 the fundamental problem of humanity

02:02:22.660 is we have paleolithic brains,

02:02:26.920 medieval institutions, and godlike technology.

02:02:30.500 And so we have to embrace the part of the answer,

02:02:33.160 if that's the problem statement,

02:02:34.000 the answer has to be embrace the reality

02:02:37.060 of our paleolithic brains, meaning how we're wired.

02:02:39.900 So don't run confirmation bias

02:02:41.700 through everybody's social media feed

02:02:42.960 and get them addicted to their most angry boogeyman.

02:02:46.080 um don't drive dopamine addiction that's a vulnerability a frailty in the human mind

02:02:50.640 um you know flux for example if you use that on a computer it changes your um the color of your

02:02:55.200 screen to go to take the blue light out of your your screen so that doesn't basically trick your

02:02:59.920 eyes into releasing melatonin and screwing up your sleep cycle i'm going too fast but this

02:03:04.000 makes sense like yes you're basically creating technology that's humane and aligned with

02:03:08.960 the things that make humans work well if you extrapolate out like one level of abstraction

02:03:14.880 there's things that make a society work.

02:03:17.420 Family, connection, presence, community.

02:03:21.480 And technology should be ergonomic

02:03:23.460 to the way that society,

02:03:25.860 what makes a society flourish.

02:03:27.240 Yes.

02:03:27.900 Rather than doing narrow optimization for attention, 0.91

02:03:31.560 strip mining the entire social fabric

02:03:33.880 into a brain-rotted dysfunctional society, right?

02:03:37.760 So the answer is we have to embrace

02:03:39.660 and align technology with the reality

02:03:41.600 of our paleolithic brain instincts,

02:03:43.980 upgrade our medieval institutions,

02:03:46.220 which by the way can include technology

02:03:47.640 that's better at finding common ground between people.

02:03:49.840 And there's a whole conversation we could have

02:03:51.060 about how technology can be designed

02:03:52.780 to identify places where there's common ground

02:03:54.880 and design for unlikely consensus

02:03:56.780 as opposed to maximum division and polarization.

02:03:59.860 And then have the wisdom and restraint

02:04:01.940 around God-like technology.

02:04:03.800 And there is no definition of wisdom

02:04:06.520 in any spiritual tradition

02:04:07.980 that doesn't have a restraint

02:04:09.920 as a central feature of what it means to be what.

02:04:11.900 It's the central feature.

02:04:12.960 It's the central feature.

02:04:13.980 And even the CEO of Microsoft, AI, Mustafa Suleiman, who's a friend of mine, says that in the future, progress will depend more on what we say no to than what we say yes to in terms of the benefits that we get from technology.

02:04:25.340 As it does in your own life.

02:04:26.760 Exactly.

02:04:27.660 Exactly.

02:04:28.180 So this is like the obvious common sense, you know, kind of cultural movement that we need.

02:04:35.260 But there's a lot of momentum.

02:04:36.540 Like when you see graffiti on a New York subway ad for an AI product that no one needs, that's like the human movement.

02:04:43.980 when you see people turn their phone grayscale

02:04:46.240 because they don't want it to be hijacking

02:04:47.480 with all the colors,

02:04:48.680 that's the human movement.

02:04:49.540 When you see schools going phone-free

02:04:51.820 and laughter returns to the hallways

02:04:53.900 and kids' scores start going the other way,

02:04:55.080 that's the human movement.

02:04:56.360 When you see Jonathan Haidt and the Anxious Generation 0.94

02:04:58.860 succeed in getting all these countries like Domino's

02:05:01.060 flipping to ban social media for kids under 15 or 16,

02:05:03.440 we have like now some massive percentage

02:05:06.080 of the world's countries are doing social media bans.

02:05:08.120 That's the human movement.

02:05:09.160 When you see the $17 billion fine against Meta,

02:05:11.820 that's the human movement.

02:05:12.740 So if you define it as technology is not, it's not just about AI policy, it's about technology encroaching on the things that are actually sacred to us. And if you, if you define the problem that way, there's actually a lot of momentum that we are making on this. And now we're seeing this kind of red line of uncontrollable, dangerous AI behavior. And the human movement has to stand for, we want to protect the continuity of the things that we care about.

02:05:37.280 Yes.

02:05:37.760 Yeah.

02:05:38.400 It's such a unifying message.

02:05:39.820 Exactly.

02:05:40.100 Because it's all of us against the machines.

02:05:42.280 Exactly.

02:05:42.740 exactly it's the most unifying issue of all time i love that i i absolutely love that that's a kind

02:05:48.040 of tribalism that i can get strongly right the human tribe amen yeah thank you that was

02:05:53.400 an amazing amazing conversation absolutely thank you so much tucker

The Tucker Carlson Show - October 12, 2026

Google Whistleblower Exposes Google’s True Motives, Major Scandals, and AI’s Rogue Behavior

Episode Stats

Transcript