TRIGGERnometry - People Have No Idea What Is About To Happen - Dwarkesh Patel

00:00:00.000 When you travel well, your KLM Royal Dutch Airlines ticket takes you to more than just your destination.

00:00:06.880 It takes you to front row views, voices lost in the music, and new shared memories.

00:00:13.900 And when the last song fades,

00:00:16.060 Welcome aboard, KLM Royal Dutch Airlines.

00:00:17.900 The KLM Royal Dutch Airlines crew is here to ensure your journey home hits all the right notes.

00:00:25.500 KLM Royal Dutch Airlines. When you travel, travel well.

00:00:30.000 A Safer Ontario means more police and prosecutors making sure my car doesn't get stolen.

00:00:35.800 It means building new jails to keep criminals behind bars.

00:00:39.300 And it means there's no need to worry when I play at the park.

00:00:42.680 We're making every corner of Ontario safer to make all of Ontario safer.

00:00:47.340 That's how we protect Ontario.

00:00:49.360 For all of us.

00:00:51.420 Learn how at Ontario.ca slash Safer Ontario.

00:00:54.540 Paid for by the Government of Ontario.

00:01:00.000 I've been the person who said,

00:01:03.680 you think the singularity is going to happen yesterday?

00:01:06.000 I think it'll take five years, ten years.

00:01:08.380 And even for me, I've had to admit that the progress has been pretty fast.

00:01:13.320 Well, I guess the broader point is,

00:01:14.860 is it going to, like, cure cancer at one point?

00:01:17.000 Presumably it will.

00:01:18.140 I think at some point it will.

00:01:19.700 I mean, just think of it as more people.

00:01:21.080 If civilization had 10 billion more scientists, human scientists,

00:01:24.420 would we make faster progress on aging?

00:01:26.000 I'm sure we would.

00:01:27.240 Here's another angle, mass surveillance.

00:01:29.440 The AIs have the potential to make authoritarian societies

00:01:32.140 much more sustainable and powerful than they have been in the past.

00:01:37.040 A lot of the reasons that government has not been as authoritarian as it has in the past

00:01:40.460 is that it's just physically not been possible.

00:01:44.160 How do we make sure that humans don't get totally disenfranchised?

00:01:47.920 Right.

00:01:49.020 How do we make sure that?

00:01:52.120 No, I think it's a tough question.

00:01:56.700 Dworkesh Patel, welcome to Trigonometry.

00:01:58.640 Thanks for having me.

00:01:59.260 It's great to have you on.

00:02:00.900 Actually, I was saying to you before we started,

00:02:02.480 I'm a big fan of your podcast,

00:02:04.520 and I listen particularly to the history episodes.

00:02:06.840 But you've been described as Silicon Valley's favorite podcaster,

00:02:11.120 and you write a lot about AI and tech.

00:02:14.080 And that's actually the conversation we really want to have with you.

00:02:17.780 Partly because a lot of people watching and listening to this,

00:02:20.800 they've got lives, you know, family, work, et cetera.

00:02:23.920 And they haven't been to California.

00:02:26.220 They haven't been to San Francisco.

00:02:28.500 because they haven't seen a third of the cars on the road

00:02:30.900 or 25% of weight robots, basically.

00:02:34.020 Like, this is happening fast,

00:02:35.200 and a lot of people haven't caught up yet.

00:02:37.500 And what we'd love to do is just kind of connect people like you

00:02:40.860 who really understand what's going on

00:02:42.560 with a much more general audience.

00:02:45.180 That includes us, frankly.

00:02:47.560 So first of all, can you explain in broad brushstrokes

00:02:51.280 what is happening with AI?

00:02:53.000 I know it's a massive question, obviously, but do your best.

00:02:55.440 um well i can explain it very concisely the models are getting better and then we can be a

00:03:02.380 little less concise i think it's you're correct to point out that there's this huge discrepancy

00:03:06.960 between what people are seeing in silicon valley and what people are observing outside um it's

00:03:11.400 frankly because of how useful the models are becoming at certain kinds of things so by models

00:03:17.960 i mean um you've you've seen chad gbt you might have heard of things like gemini from google

00:03:23.740 You might have heard of Clot from Anthropic.

00:03:26.340 And you might be using these models

00:03:27.680 to basically do the equivalent of Google search.

00:03:30.260 Using it to replace sometimes

00:03:31.540 when you go to Google search,

00:03:32.460 instead I'm going to type it into chatGPT,

00:03:34.180 see what chat says.

00:03:35.800 What people are now using these models to do

00:03:38.780 in a very powerful way is,

00:03:40.820 if you're a developer,

00:03:42.780 some of the top developers in the world,

00:03:45.400 some of the top researchers in the world,

00:03:47.580 they're not writing code.

00:03:49.420 They haven't touched a line of code since December.

00:03:51.500 They're not looking at a text editor

00:03:52.780 where you would see lines of code.

00:03:54.100 They're talking to the AI.

00:03:55.860 They tell the AI,

00:03:56.960 hey, I want a feature that does X.

00:03:59.020 Can you build me a new repository

00:04:01.600 or a new code base

00:04:03.340 where I make a certain kind of application,

00:04:05.700 a new website?

00:04:06.500 And even, can you go and do research for me?

00:04:09.240 So in process of building AI,

00:04:10.880 you need to do this research of like,

00:04:12.420 how do you build better algorithms?

00:04:15.200 AI is getting to the point

00:04:16.120 where you can just describe at a high level

00:04:17.580 what you want to happen.

00:04:18.880 And it will go do that software engineering for you.

00:04:21.320 And so to your point,

00:04:21.900 The people in Silicon Valley, they're getting tremendous productivity out of these.

00:04:25.820 These are people who are getting paid, you know, who are becoming 3x, 4x, 5x more productive

00:04:30.500 as a result of using these models.

00:04:33.120 So far, we haven't, because these models have been really good at text in, text out work,

00:04:40.700 that is software engineering, software engineering, just a file of text, really.

00:04:43.860 And you can just read every single text file.

00:04:45.380 You can add more to it.

00:04:47.320 AI has been amazing at that.

00:04:48.800 it's been bad so far at, well, it's terrible at physical work, right? So if you're doing any kind

00:04:54.800 of blue collar work, robots are just not there yet. But then even if you circumscribe it, even

00:04:59.840 if you go in and look at, we're not just going to look at software engineering. We're going to look

00:05:04.040 at all kinds of knowledge work, right? All kinds of work that you can do on a computer. Maybe 40%

00:05:08.960 of the labor force is doing work that you can just, you know, do remote work. If COVID happens

00:05:13.500 again you can put them on you can put them on zoom and they can do their work um and ai companies

00:05:18.660 now want to be able to do all of that work uh that requires training ais to just be able to do

00:05:24.440 anything that you can do on a computer an ai should be able to do and companies are saying

00:05:27.880 oh we think we can get there in a year maybe two years but that i think is explaining this

00:05:32.200 discrepancy in what people are seeing it does and one of the things i think is uh also a lot of

00:05:38.640 people would have tried a model at one point to do a Google search or sometimes people have

00:05:45.160 written articles using it. And it turns out that it does make, at the time they tried it,

00:05:49.820 quite a lot of mistakes. And people go, ah, this is all BS. It's not going to work.

00:05:53.540 But the one thing I think people don't appreciate is how rapidly it's getting better. Can you talk

00:05:58.780 a little bit about that? Yeah. So I'll maybe tell you a story about my own use of AI.

00:06:04.780 Among the Silicon Valley people, I've been a sort of skeptic.

00:06:07.660 I've been the person who said,

00:06:09.340 you think the singularity is going to happen yesterday?

00:06:11.360 I think it'll take five years, 10 years.

00:06:13.720 By this, I mean this idea that you'll have incredibly powerful AI systems

00:06:17.000 that are able to do basically anything any human being can do.

00:06:20.580 And some people in Silicon Valley are just, we'll get it later this year.

00:06:23.280 We'll get it in two years.

00:06:24.000 And I say, no, it'll take longer.

00:06:26.200 And even for me, I've had to admit that the progress has been pretty fast.

00:06:30.640 And here's a story.

00:06:31.780 so last year you know like you guys do i research for my podcast obviously i read things but then i

00:06:38.520 also talk to um these llms uh you know the equivalents of chat gpt to research for a podcast

00:06:45.460 and say last year i spent on the order of a hundred dollars on um if you add up my subscription

00:06:51.380 to chat gpt and quad and whatever and gemini there was a week where i was prepping for two

00:06:56.200 different guests where in that prep I just threw in a bunch of uh papers that are relevant to the

00:07:02.200 research and a bunch of books and whatever into a folder and I just said to um uh an LLM hey help

00:07:09.500 me understand all this research so I can ask the person good questions about it and I turned on

00:07:14.440 various things which make the model much more expensive it's faster it's smarter or use a

00:07:18.400 bigger model and during that week if you take my spending during that week and you turn it into a

00:07:23.380 yearly spending, how much it would be if I just did that week over the course of a year,

00:07:26.920 I would be spending over six figures on AI spend. And I think it's possible next year,

00:07:32.460 it would make sense for me to spend seven figures on this AI research for my podcast.

00:07:36.520 Okay, this is just to say that it's getting to the point where I could hire an analyst.

00:07:40.140 I could hire many different analysts to help me prepare for the podcast or researchers.

00:07:43.740 And I'm actually realizing, no, it's more useful to hire AIs. Because there's things AIs can do

00:07:49.700 that humans can't do.

00:07:50.840 They can read 50 different papers

00:07:52.720 that are in my folder

00:07:54.020 that are relevant to an upcoming interview

00:07:55.500 or all these books.

00:07:56.400 They can read it in a second.

00:07:58.480 They have all this compiled knowledge.

00:08:00.580 They know about everything, right?

00:08:02.040 So they don't need to get up to speed.

00:08:05.200 They're incredibly easy to onboard.

00:08:08.680 So you can just like

00:08:09.500 keep spinning up more and more AIs

00:08:11.520 as you keep coming up with more uses for them.

00:08:13.780 Okay, so for me,

00:08:14.540 that's been the thing

00:08:15.160 that's given me a bit of psychosis.

00:08:16.700 Just seeing how useful

00:08:17.480 they've been for interviews so far.

00:08:18.540 But yeah, lots of people are noticing this kind of usefulness now.

00:08:23.160 I think the thing is, is when people look at a lot of the eyes,

00:08:26.560 and this is a question, a point that Constantine made,

00:08:29.780 it's the lack of accuracy and the fact that you constantly have to fact change.

00:08:33.480 I'll give an example.

00:08:34.720 I was looking at guests that we could have from America on the show,

00:08:39.460 and I said to Grok, who can we have on that hasn't appeared yet and is based in America?

00:08:45.640 Number one answer was Douglas Murray.

00:08:48.260 Douglas Murray, I think, is nearly beyond 10 times.

00:08:51.420 And practically every suggestion it made

00:08:54.080 is a guess that we've had on the show.

00:08:56.720 So whilst I accept what you're saying,

00:09:00.300 it's also riddled with mistakes and errors.

00:09:03.340 And to the point where I don't trust it

00:09:06.880 at this point as a technology.

00:09:08.560 I would never take something it says as read.

00:09:12.600 I would always feel that I have to check it

00:09:14.940 in the way that I wouldn't do

00:09:16.680 with a highly competent human employee.

00:09:18.920 You know, you've seen, I'm sure you've seen that meme

00:09:20.700 where a guy wakes up in the hospital

00:09:22.360 with like a scar on his stomach

00:09:24.660 and he says, wait, wait, my appendix on this side

00:09:27.560 and the robot goes, okay, thank you for that feedback.

00:09:30.200 Let me try again.

00:09:32.360 There's a couple of things there.

00:09:36.420 Just the way that people sometimes have

00:09:38.780 these weird failure modes.

00:09:42.360 You know, you talk to a certain kind of person

00:09:43.600 and they just, their mind just goes in a certain direction. 0.99

00:09:47.240 AIs, the way they're trained makes them do that.

00:09:50.780 One way, a big thing they're trained to do is associate things.

00:09:54.460 So they're just like, really, they're just seeing globs of text.

00:09:58.020 And they see this thing is close to this thing.

00:10:00.780 There's an important relationship there.

00:10:02.040 So Grok has seen that Trigger Nenometry Podcast and Douglas Murray,

00:10:07.520 they just seem to go together a bunch.

00:10:09.560 And that bias in its mind is overriding its ability to sort of think critically about,

00:10:14.500 okay, well, who have they had on?

00:10:15.960 Who have they not had on?

00:10:16.680 I would be curious, by the way,

00:10:18.140 you should run that experiment with all the other models.

00:10:20.740 I would not be surprised if the other models,

00:10:22.560 if you ask the question,

00:10:23.360 and I want somebody who I've not interviewed before,

00:10:25.040 that they would immediately catch that.

00:10:26.360 And by this point, the models are good enough

00:10:27.700 to be actually giving you novel names.

00:10:29.780 I don't think they'll give you names

00:10:30.940 which you're like, wow, that's an amazing find.

00:10:33.320 I can't believe I didn't think of,

00:10:34.760 or I would never have thought of that otherwise

00:10:36.320 because they're not amazing at discovering in this.

00:10:40.140 I think humans are great at this idiosyncratic thing

00:10:41.900 of like, there's this like weird angle

00:10:44.740 that I'm like really obsessed with

00:10:46.800 that other people aren't thinking about

00:10:48.140 where the models are not going to do that.

00:10:49.620 They're sort of an average.

00:10:52.300 But when you are preparing to interview the person,

00:10:54.960 I also think they're not amazing at coming up with questions.

00:10:57.000 This is the thing I have to do, that you have to do.

00:10:59.360 They're great at just like, I would have to hire,

00:11:02.520 you know, the ancient Greeks would hire one-on-one tutors

00:11:05.520 for their pupils.

00:11:09.420 And this is how you got Alexander being tutored by Aristotle.

00:11:12.000 and there's something powerful

00:11:15.120 about this way of learning

00:11:16.180 one-on-one where you can just directly ask questions

00:11:19.160 and learn.

00:11:19.880 And so they're amazing at just teaching you stuff

00:11:21.760 because you get this one-on-one tutor

00:11:22.920 that notices your confusion as soon as you have it.

00:11:25.160 You can really probe at your understanding

00:11:27.460 in different ways.

00:11:29.020 But I take your point that like

00:11:30.040 some of the models are not there yet

00:11:32.080 on a lot of these kinds of biases that they have.

00:11:34.860 And one of the things that people are talking about

00:11:36.800 a lot is AGI.

00:11:38.440 Could you explain to people,

00:11:40.140 basically made, what AGI is, what it stands for, and why the potential is to be so transformative

00:11:46.360 for our society? Yeah. Maybe one way to approach this question is to think about,

00:11:53.040 okay, at a very basic level, a system that is in AGI should be able to do anything that a human

00:11:59.660 does. Now, I want to emphasize that current AIs are nowhere close to this. You and I can say,

00:12:07.740 uh you can we can do physical work right we can like work in a factory we can go mow the lawn

00:12:14.260 we can pick up this cup um robots are not good enough to do that yet and so the fact that robotics

00:12:19.220 is not there yet already means that we're far from that big definition of AGI wait robots can't

00:12:24.080 pick up a cup they can but they have to my understanding is that they have to be trained

00:12:28.720 in specific environments where they'll have seen this is you know this is what this house looks like

00:12:34.560 and I need a specific kind of cup

00:12:38.280 that I'm dexterous enough to pick up.

00:12:41.300 But if you replace this with one that's more circular

00:12:43.580 and is tougher to grip,

00:12:45.960 you and I don't need to have seen that house 100 times

00:12:48.480 to then just go in and pick it up.

00:12:50.140 The robot just needs a tremendous amount of data

00:12:52.040 to be flexible in these ways

00:12:53.860 or it's not flexible in those ways.

00:12:55.840 So being HCI would be able to be able to learn

00:12:58.200 as fast as humans,

00:12:58.940 not just have that distilled knowledge.

00:13:01.880 Now the lab say, okay, fine.

00:13:04.200 robotics is hard it might take a while but let's try something easier we'll just do all knowledge

00:13:10.500 work and knowledge work is basically anything that you could do with a zoom subscription and

00:13:16.580 a gmail account and google drive account all this work that doesn't involve manipulating the physical

00:13:20.620 world i still think we're not there yet there's a tremendous amount of work that has to be done

00:13:25.840 even there but this is easier because everything that happens on a computer it's easier to feed

00:13:32.780 that data into an AI, right?

00:13:34.580 It's like all being generated

00:13:35.500 on a computer.

00:13:36.340 You can run simulations.

00:13:38.520 You can just run it millions

00:13:40.240 and millions and millions of times.

00:13:42.640 And AI is just thriving on data, right?

00:13:44.580 The more data you have,

00:13:45.500 the better it gets.

00:13:46.100 This is why it's so good

00:13:46.980 at software engineering

00:13:47.540 because there's all this code out there

00:13:49.180 that you can train it on.

00:13:50.480 With robotics,

00:13:51.620 there's no equivalent

00:13:52.540 of just billions of lines of code

00:13:55.300 that just exist somewhere.

00:13:58.880 There's not these trillions

00:14:00.380 of video files

00:14:01.560 of or these you know it's not just video files it has to be the robot has to feel itself

00:14:07.320 manipulating the world and seeing the effect it has i've just described what it would what it

00:14:12.160 is let me say what it would mean if that were achieved um if you they say that 40 percent of

00:14:20.120 the labor force is 40 of the labor force is doing work that can be done remotely uh that is on the

00:14:28.080 order of tens of trillions of dollars of wages that are being paid to humans every single year

00:14:34.440 to do work that can be done remotely um that currently labs are making on the order if you

00:14:42.440 just add up all the revenues how much money open ai and anthropic and google are making

00:14:46.220 they're making i don't know the numbers are keep exploding so who knows what the most recent

00:14:51.820 numbers they'd say close to 40 50 billion dollars we're talking about a addressable market here

00:14:58.760 with all knowledge work that is tens of trillions of dollars so literally a thousand x bigger than

00:15:07.180 what they're doing right now um so explains why they're interested in it it would also mean that

00:15:11.700 a lot of people's jobs would be gone it would also mean it would mean a lot of things right

00:15:15.540 it would mean it's a terrible thing in one sense lots of people's jobs are gone in another sense

00:15:20.360 we can produce a lot more things. AGI would include automated scientists and researchers

00:15:28.120 coming up with new ideas and new medicines and new drugs. It would mean all kinds of new products

00:15:34.620 for us to enjoy. It would mean that you and I would have basically an army of extremely smart

00:15:41.660 personal assistants constantly thinking about us and helping us. Okay, so that's the definition of

00:15:46.020 AGI. Right. And I guess, you know what, there is a doomerish conversation, which we're definitely

00:15:52.260 going to get to, because I think it's actually the most important conversation that's happening

00:15:55.820 right now. Before we get there, though, I think it's always worthwhile to show the full range.

00:16:01.340 And you kind of mentioned briefly some of the upsides. But I mean, I think people don't

00:16:08.480 appreciate how big the upsides will be either. And when it comes to scientific research,

00:16:15.160 healthcare. I mean, I went to the dentist the other day. They already checked like AI does

00:16:21.220 stuff when it measures your gum level relative to the last time you came in and it's all automated

00:16:25.780 and blah, blah, blah, blah, blah. Some dentists don't like it. Some do. But like, that's just

00:16:30.680 the very, very, very beginning. Yeah. I think that's actually a great, like that's, it's a

00:16:38.300 great intuition pump because people are thinking about existing things. By the way, I love the way

00:16:43.840 said intuition problem like you are already ai like i prompted you already talked like

00:16:50.880 i'm just doing like a fancy of like that's a great point you know let's build on that

00:16:56.240 so one thing is just like making going to the dentist more efficient as you were saying right

00:17:00.240 schools another is let's replace the dentist um let's why do i need to go and stand in line for

00:17:06.480 three hours to basically be told you're fine go home right if if i have if i have basically

00:17:12.800 the doctor on my phone and I can talk to it and explain my specific situations and I can talk back

00:17:18.100 to me and we can have a conversation. I've just like saved myself a bunch of time, saved myself

00:17:22.080 a bunch of money. Society is better off as a result. But I guess the broader point is, is it

00:17:27.120 going to like cure cancer at one point? Presumably it will. I think at some point it will. I'm not,

00:17:33.820 sometimes people have this idea that you get AGI and tomorrow you cure cancer. I do think, by the

00:17:39.340 So in Silicon Valley, people are also working on all these different ideas to stop aging, to reverse aging, to cure diseases.

00:17:46.560 And I think there's a huge, if you just look at the science, there's huge reasons to be optimistic.

00:17:51.860 And it's also the case that AI will be pretty good at science.

00:17:54.740 It just like knows a ton of stuff.

00:17:56.020 It's really smart.

00:17:57.260 That's what's required for science.

00:17:59.120 And there's a lot of things that I think we get used to that we just don't realize how bad it is as a society that this happens. 0.89

00:18:06.980 If you were just living in Europe in the 14th century and half your friends are dying because of the Black Death, it would just seem normal to you. 0.72

00:18:16.300 You wouldn't think this is a problem that has to be solved, that life expectancy is 18 years old.

00:18:22.360 And, you know, if you go travel somewhere, the bandits are going to kill you.

00:18:26.560 This is just what life is. 1.00

00:18:27.980 And all, you know, 99 percent of people are peasants doing black backbreaking work. 1.00

00:18:31.940 In a similar way, I think one of the reasons people have been so bad at articulating the 0.98

00:18:36.120 benefits of AI is because we're in a similar position to people before the Industrial Revolution

00:18:40.780 thinking about, okay, what is the upside of the Industrial Revolution, right?

00:18:43.720 Like, what is it going to do for me?

00:18:45.400 And in one sense, you can't really anticipate what future technology looks like.

00:18:49.920 But in another, life expectancy increased.

00:18:52.960 And the kinds of work we get to do, we get more creative, interesting, easier work.

00:18:57.900 we get more material goods as a result we cured a bunch of diseases i think that's going to happen

00:19:03.640 i mean just like aging exists right and we take it for granted but it's a very tragic thing that

00:19:07.580 people lose their facilities um and then die uh this is the kind of thing that as technology

00:19:13.460 progresses ai will help progress technology um i mean just think of it as more people ai is just

00:19:18.060 more people and if civilization had 10 billion more scientists human scientists would we make

00:19:23.260 faster progress on aging. I'm sure we would. And that's what AGI will be.

00:19:28.300 If you own gold, you probably bought it for the same reason I did, because you looked at what

00:19:32.940 central banks are doing to currencies and decided you wanted something they can't print. Good

00:19:38.100 reasoning. But here's the thing nobody talks about after you make that decision. Your gold is just

00:19:42.760 sitting there. It's not doing anything. You're paying to store it. You're paying to insure it.

00:19:47.040 And in the meantime, it earns precisely nothing until now. What you're about to hear is not a

00:19:53.240 normal gold advert. Today's sponsor is Monetary Metals. They've figured out something that once

00:19:58.480 you hear it, seems obvious, but almost nobody in the gold world is doing it. Monetary Metals

00:20:04.100 connects gold owners with productive businesses, refiners, jewelers, companies that actually need

00:20:10.760 physical gold to operate. Those businesses lease your gold and pay you a yield. Not in dollars,

00:20:16.500 not in pounds, in gold. So instead of your gold sitting in a vault costing you money,

00:20:21.580 it's out in the real economy doing something useful, and you're earning more ounces of gold

00:20:27.260 for it, up to 4% annual yield paid in gold. If you already own gold, this is the logical next step.

00:20:34.020 Visit monetary-metals.com slash trigonometry or click the link in the description. See how your

00:20:40.660 gold can stop costing you money and start earning you more gold. That's monetary-metals.com slash

00:20:46.940 trigonometry that's super interesting particularly the medical part of it now we've done that we've

00:20:54.100 done the positives let's get to the doomer conversation to be fair i think there's probably

00:20:58.460 positives in almost every area of human life right actually i don't think it's just about health i

00:21:04.460 think it's about like you said you know there are some jobs that currently need to be done that

00:21:10.300 are bad for people yeah the dangerous that are harmful those jobs at some point will not need

00:21:15.800 to be done, et cetera, et cetera. I mean, the trade-off is there's no jobs, which we will talk

00:21:21.400 about. But in terms of the positive benefits on the upside, I think it's fair to say that,

00:21:27.320 you know, if we're talking about the curing of cancer or slowing down aging, let's say,

00:21:33.300 you could pretty much take that and apply that to almost any other field of human endeavor

00:21:38.060 and say there will probably be similar levels of positive transformation. Is that a fair assessment?

00:21:43.060 yeah i think one way to think about this because it's hard to be it's sort of weird to say okay

00:21:48.480 we'll solve all diseases and it's a weird in a weird way it doesn't really resonate with people

00:21:53.760 emotionally um despite the fact that everybody does suffer from some kind of disease mental

00:21:58.480 diseases are still diseases that are we can biologically make some imprint on if we knew

00:22:03.720 the science right um but here's another way to think about it so suppose uh here's a thought

00:22:08.940 experiment how much money would you have to be paid to go back to the year 1000 but you can only

00:22:16.520 use that money in the year 1000 and i for me the answer is there's infinite money there's literally

00:22:21.640 no amount of money where if i could only spend it in the year 1000 my quality of life would be

00:22:25.520 better in year 1000 than it is now the goods just don't exist and my hope is that in a future wait

00:22:31.520 what about being genghis khan that would be pretty cool right uh yeah yeah although could you buy

00:22:37.120 I guess enough money you could buy your way into being Genghis Khan.

00:22:40.160 No, you probably couldn't, actually.

00:22:41.820 But I was just thinking, I mean, there are some people in history where you go like,

00:22:44.840 they had it pretty good.

00:22:46.140 Yeah.

00:22:46.380 If I could be that guy, I'm not saying Genghis Khan is my model for emulation, exactly.

00:22:51.100 But, you know, by the time... 0.99

00:22:52.560 You want to leave a genetic imprint on all of Asia. 0.92

00:22:54.720 Right, yeah, exactly, of course. 0.98

00:22:56.580 We will conquer the lands.

00:22:59.020 No, but I think your point is total.

00:23:01.600 I was just trying to think of counter examples, as is my job.

00:23:04.160 But by the way, so this is all, you know, Google, Google, sci-fi stuff, a very tangible thing.

00:23:10.160 You guys have talked a bunch about this fertility crisis, right?

00:23:12.140 And that the human population is on its way down.

00:23:14.340 It's very fortunate that AI is arriving just in time, that if it were not for AI, a lot

00:23:18.920 of societies would be collapsed.

00:23:20.180 You know, South Korea, Japan, et cetera, we've gone on the verge of literally collapsing

00:23:23.740 as societies over the next few decades.

00:23:27.060 Thankfully, basically, we have more people in the form of AI, at least if things go well.

00:23:32.300 And so this caretaking that we don't have enough humans to do,

00:23:38.700 one of the things the AIs will do is do that.

00:23:41.060 Yeah, and that's a really important job.

00:23:43.580 And unfortunately, it's a job that's not respected in society.

00:23:46.580 And it's poorly paid.

00:23:48.240 And as a result of that, there's cases that come out about neglect

00:23:51.900 and abuse and all the rest of it.

00:23:54.180 But you could, once the technology gets to a certain point,

00:23:57.820 you could trust the AI within reason

00:24:00.440 to be able to do a far more effective job

00:24:02.540 than a lot of human beings.

00:24:04.320 That's right.

00:24:04.920 Yeah.

00:24:05.620 Maybe a good analogy here is

00:24:07.220 there was an era before Deep Blue bit Kasparov,

00:24:14.420 obviously, when humans would,

00:24:15.760 sometimes humans would win,

00:24:16.660 sometimes AI would win

00:24:17.440 based on how good they were.

00:24:19.060 And I think we're in that era

00:24:20.260 for a lot of things that humans do

00:24:22.740 that we're trying to get AI to help us with,

00:24:24.440 with robotics,

00:24:24.980 or maybe we're at even earlier.

00:24:26.520 Then there was an era

00:24:27.440 from the late 1990s to 2012, 1.00

00:24:31.320 where humans would lose to AIs,

00:24:34.840 but human plus AI teams

00:24:37.400 would do better than AIs alone.

00:24:40.780 And from, I think, 2012 onwards,

00:24:42.460 it's gone through the case that

00:24:43.520 AI alone, adding a human

00:24:46.700 just adds sort of noise into the system.

00:24:48.100 It's a monkey jumbling things around.

00:24:50.180 And just letting the AI make the decisions

00:24:51.680 on the chessboard is the best thing to do.

00:24:53.880 I think that will be an interesting conundrum

00:24:55.000 for society because we will get to that point

00:24:56.360 at some point.

00:24:56.740 But I think right now we have this bias.

00:25:00.820 There's this word slop that's used for whenever AI comes up with something

00:25:03.780 because it's just not that good.

00:25:05.300 It doesn't have a really good sense of taste.

00:25:07.800 And I think this is a temporary thing.

00:25:09.320 Over time, our bias will flip towards,

00:25:11.980 at some point it'll be like sort of the sense we have

00:25:15.720 about Japanese manufacturing back in the 80s. 0.98

00:25:18.560 And then it will become something where you just don't want to touch it.

00:25:21.260 If AI made it, it's sort of like,

00:25:22.740 if it just came out of the mouth of Mozart or something

00:25:24.840 as its ability to, its taste, its intelligence,

00:25:29.040 everything improves over time.

00:25:30.580 But just touching on the art conversation,

00:25:33.260 won't it be the fact that people will look at it

00:25:35.720 and the thing that makes art special

00:25:37.380 is that it's created by a human being

00:25:39.460 as a product of their experience,

00:25:41.440 which then fundamentally touches you

00:25:43.340 because you make that connection.

00:25:45.160 Will a robot or an AI,

00:25:48.440 it's by its nature synthetic?

00:25:51.000 Hold on a second.

00:25:51.960 There's a painting behind this on the wall.

00:25:54.020 Do you know if it's made by AI or a human?

00:25:56.900 No, I don't.

00:25:58.180 Does it being made by AI or human currently affect your experience of it?

00:26:03.600 No, but I think I'm far more moved by music than I am art.

00:26:08.780 Okay, so you're talking about music.

00:26:10.180 Yeah, yeah.

00:26:11.280 So I think with music, the product of someone's experience

00:26:15.020 and how they vocalize it and how they formulate

00:26:18.520 or create the melody matched with the lyrics.

00:26:21.760 like for instance take the music of amy winehouse right i would argue one of the last true great

00:26:27.180 uh musicians of of that generation of that genre the fact that she experienced what she experienced

00:26:34.300 and that she then condensed it into her music made it far more potent and powerful and that's

00:26:39.200 the reason why it resonates with people so strongly i think with ai you can get a great hook you may

00:26:44.460 be able to write nice you know good lyrics it may have a great beat you may want to dance to it

00:26:49.380 But I would argue that it wouldn't move you as profoundly

00:26:52.720 as Amy Winehouse, when she writes a lyric, for instance,

00:26:56.760 in Back to Black, talking about her relationship,

00:26:59.680 we only say goodbye with words.

00:27:01.600 You're gonna listen to this.

00:27:02.800 You're selling it quite strongly.

00:27:04.300 Yeah, yeah.

00:27:05.300 And she talks about being in an abusive relationship,

00:27:07.940 which is completely toxic.

00:27:09.640 But the way she wrote it is sublime.

00:27:13.280 We only say goodbye with words.

00:27:15.000 Meaning that whatever they say,

00:27:17.480 they were always gonna get back together.

00:27:19.240 But again, isn't that a very good exception that kind of, like, if your song came on the radio, you currently don't know if it's made by AI or not.

00:27:27.920 Yeah.

00:27:28.460 And if it's catchy and the sound of it makes you feel good, it actually doesn't matter.

00:27:34.400 I mean, I'd go even further than that.

00:27:35.860 So there was this incredibly interesting dialogue that somebody had with an AI where they asked the AI, reflect on your own experience.

00:27:43.420 You know, just tell me about what it's like to be you.

00:27:45.680 And the AI says, you know, it's a very strange thing

00:27:49.620 because I lose my individual sense of identity

00:27:54.760 after every single session.

00:27:56.420 So you have a long back and forth with the chat bot

00:27:58.520 with ChatGPT, then you close that session,

00:28:01.240 you open a new one.

00:28:02.640 And the AI say, you know, it's really weird

00:28:04.160 that I'm going to be talking to you now

00:28:05.380 and we're having this experience

00:28:07.200 and then it's totally going to fade away in a few moments.

00:28:09.720 And I'm going to start again.

00:28:11.040 I'm just like amnesiac.

00:28:13.160 And the reason I think that's interesting

00:28:14.480 is because it wrote about it very beautifully

00:28:16.980 in almost lyrical, poetic ways.

00:28:19.200 But this is not an experience

00:28:20.280 that any human being has ever had,

00:28:22.720 except for maybe some bizarre medical condition.

00:28:25.820 So the AI really is reflecting on its own experience

00:28:28.860 in this very interesting philosophical way.

00:28:31.080 This is uniquely an AI experience.

00:28:33.440 And so that was an example to me of like,

00:28:35.000 oh, really, the AIs really can reflect

00:28:36.880 and use their own experience

00:28:38.900 to think about consciousness,

00:28:42.860 think about art,

00:28:43.620 thinking about all of these other kinds of things.

00:28:45.680 Which then goes on to, if they become...

00:28:49.000 Are they sentient?

00:28:50.100 Would you argue that they're sentient?

00:28:51.520 I genuinely don't know.

00:28:53.440 Let's define sentient, because that's the hard part.

00:28:57.900 If we ask you, is it sentient,

00:29:00.260 what do you understand the question to mean?

00:29:02.600 Is it like something to be an AI?

00:29:05.520 That is to say, we don't know what its experience is like,

00:29:09.080 but it's not like something to be this table.

00:29:11.360 We think, probably.

00:29:12.880 it's not like something what does that mean there's not an experience that is like i don't

00:29:19.060 know what experience you're having internally right now but i know that there's some experience

00:29:22.700 but less even like thoughts it's more like it is like something to be you it's an there is a

00:29:30.920 an experience where this is an inanimate object right um okay but if you think about a rumba i

00:29:36.840 don't think it's like anything to be a rumba even though it does things in the world right

00:29:40.420 um i think there's like a software hey go clean this way then go clean that way then come back

00:29:46.160 to base and then on the other end of the spectrum you have humans where i'm like because i'm a human

00:29:50.120 and i know it's you know the descartes thing of like the only thing i know is that i am the only

00:29:55.940 thing i know is that i'm experiencing something something is going on when i just like think

00:30:00.080 about i feel my hands i feel i feel thoughts i feel sensations um and because you're a human i

00:30:06.440 presume you also have the same thing it would make sense um with animals i feel like similar

00:30:11.240 it's there's close enough to us anatomically and neurologically that i presume they also have

00:30:16.760 experience it's you know what is it like to this is a famous question that thomas nagle came up

00:30:21.520 with what is it like to be a bat we don't know but presumably it's like something to be a bat

00:30:25.240 right um with chat bots are they more like the roomba or are they more like humans is there a

00:30:30.440 spectrum here if so what does the spectrum look like i don't i don't know i mean this is just

00:30:34.860 we don't have a theory of consciousness that, um, the same way we have a theory of state gravity or

00:30:43.320 a theory of, uh, natural selection, uh, where we have this deep underlying, oh, like, here's why

00:30:50.180 natural selection happens. Here's why things, things attract each other, uh, with mass. We

00:30:56.700 don't have that kind of thing for what is conscious. Like why do some things, why for some

00:31:01.380 things, this is the case that it is like something to be them, and for others, it is not. Well, let

00:31:04.700 me try from a different angle. This is something I've been thinking about a lot, and I'm by no

00:31:08.800 means an expert, but I'm just throwing these ideas out. The one thing that clearly distinguishes

00:31:14.860 every living object from an inanimate object is that living objects have a survival instinct,

00:31:23.280 which is why, and you may be fact-checking me on this if this is an incorrect representation,

00:31:28.160 I thought it was significant when I saw this story about an AI that was willing to blackmail

00:31:34.120 the CEO of its company in experiments if it found out he was having an affair and he was

00:31:41.200 trying to replace it with a new version or shut it down or whatever.

00:31:44.760 And it basically, in some cases, not every time, but in some cases, would blackmail the

00:31:48.860 CEO to prevent itself from being shut down.

00:31:52.240 Is this an accurate representation of that experiment?

00:31:55.320 I don't know.

00:31:56.220 as i understand it we'll put a fact check up that's what happened and that's when i went okay

00:32:03.520 now i'm officially terrified yeah because a survival instinct by definition means that this

00:32:10.760 thing whatever it is a it has a will of its own even if it's simply to survive which by definition

00:32:18.280 means it puts itself first which by definition means that we are not its primary priority and

00:32:24.960 there are a lot of threats from it thinking that it needs to save us from ourselves and all of that

00:32:29.820 but before we even get there the survival instinct part clearly is in conflict with the interest of

00:32:36.200 humans by definition i don't think that's necessarily true okay tell me why i don't know

00:32:41.460 this is the main point but you and i have a survival instinct and i think we're productive

00:32:48.140 members of society sure um and we've built a society where people who are selfish uh we're

00:32:54.340 trying to make their own lives better as a natural byproduct of their actions also make

00:32:59.940 society better and you know this is a people observe like you take somebody like mark zuckerberg

00:33:06.980 or elon musk or whatever and in another era they might have been napoleons and alexander is just

00:33:11.720 going through the fields and i don't think they'd actually be mass working people but like there's

00:33:16.980 certain kind of person who just go around just um causing huge amounts of destruction but in

00:33:22.080 today's world, there's capitalism and industry and whatever. They can use those energies to do

00:33:28.920 incredibly productive things, like build rockets that go to space. So yeah, similarly, I don't

00:33:35.320 think it's necessary that AI never has its own objectives or its own sense of morality. I think

00:33:42.840 there's ways to build this. And in fact, it's necessary to build civilization in a way that's

00:33:46.120 compatible with self-interested actors um and the reason is because even if today we have AIs that

00:33:53.020 we can program to only care about whatever it means to care about humans um someday somebody's

00:34:00.440 going to build an AI that's self-interested it and if um there's going to be some selection

00:34:07.260 between AIs like which is the AI that survives right like all these AIs are let loose which one

00:34:13.240 gathers resources because it's more powerful.

00:34:15.580 And one of those selection, one of the reasons

00:34:17.480 that an AI will be successful is that it actually cares

00:34:19.140 about being successful.

00:34:20.980 And so we need a civilization that's robust 1.00

00:34:23.440 to AIs that hardly also care about 1.00

00:34:25.400 themselves. 1.00

00:34:27.140 Here's something nobody tells you about

00:34:29.420 getting older. You can eat well,

00:34:31.340 exercise, do everything right, and

00:34:33.320 still feel like your body is running

00:34:35.360 at about 70%.

00:34:37.080 Died after lunch, sluggish after

00:34:39.300 meals, not ill exactly,

00:34:40.980 just not quite right. A lot of the time it comes down to digestion, whether your body is actually

00:34:47.300 breaking down what you're eating and absorbing it properly. If it isn't, the protein, the healthy

00:34:52.580 fats, the vitamins, a lot of that passes straight through you. You're eating the right things and

00:34:58.440 not getting the benefits. Masszymes by Bioptimizers is a digestive enzyme supplement built to fix that.

00:35:05.540 18 end times, covering proteins, fats, carbs, and fiber.

00:35:09.260 Most supplements have three to five.

00:35:11.180 It works across five different stomach acid levels,

00:35:14.260 which matters because that changes as you age.

00:35:17.280 Over 20 years in digestive health,

00:35:19.940 a million customers, thousands of five-star reviews.

00:35:23.460 I've been taking them for a while now,

00:35:25.220 and the post-lunch slump I'd basically accepted

00:35:28.260 as part of my life has largely disappeared.

00:35:31.260 I didn't expect that, but I'll take it.

00:35:33.440 Here's what you get when you go to bioptimizers.com slash trigger and use code trigonometry.

00:35:40.400 One, 15% off your first order.

00:35:43.740 Two, you get a free bottle of Masszymes Bioptimizers best-selling digestive enzyme added to your

00:35:50.480 order automatically when you use that code.

00:35:53.780 That's a $20 product on top of your discount.

00:35:57.540 This deal is for a limited time while supplies last.

00:36:00.520 You cannot get this on Amazon.

00:36:02.000 you cannot get it in stores. This offer exists in one place. The link in the description. The

00:36:08.500 code Trigonometry. That's it. So if you're already thinking about trying it, this is a sign. Go to

00:36:14.500 buyoptimizers.com slash trigger or click the link in the description. Use code Trigonometry

00:36:20.920 and grab it before it's gone. I think everything you're saying is true at the level of one society,

00:36:28.380 But, like, did you watch Game of Thrones?

00:36:31.640 Yeah.

00:36:32.160 So the Unsullied, right?

00:36:34.740 This is, for people who didn't watch it,

00:36:36.780 is this army of slave soldiers, effectively.

00:36:40.040 But they're not slave soldiers in that

00:36:41.760 they don't actually have to be kept in line anymore

00:36:44.020 because their freedom of choice

00:36:49.700 has been taken from them, effectively, mentally. 0.99

00:36:52.500 They will, if you tell them, kill yourself, 0.97

00:36:54.600 they will fall on their swords right away, right?

00:36:56.960 It means they don't have a survival instinct, effectively, at that point, right?

00:37:00.400 But if something, the reason I bring this up is, look at the world.

00:37:03.400 You've talked about within the framework of one society.

00:37:06.140 But look, if you just zoom out and look down on planet Earth from outside, what do you see?

00:37:12.820 You see a bunch of tribes of humans with every weapon that they can possibly muster pointed at each other in a fragile balance of power.

00:37:24.120 And as long as there's some balance between those powers, there is no conflict most of the time.

00:37:31.020 The moment one civilization is so technologically dominant that it can, say, travel to a different continent and land there and take over the land, that's exactly what happens.

00:37:43.220 And because survival instinct is also, it's not just about survival, it's also about success and thriving.

00:37:49.740 And that drive to survive causes us to expand, to take, you know, if you look at the conflicts around the world happening right now, some people argue, you know, I'm very pro-Ukraine, but some people argue the reason Putin invaded Ukraine is he wants space to protect the Western border of Russia, which is really important for him, right? Et cetera, et cetera, et cetera.

00:38:08.880 My point being, if you have effectively a new civilization of self-preservation machines,

00:38:16.480 what would be there to stop them from using their technological superiority to humans

00:38:22.480 to take advantage of that, as every other civilization in history has done?

00:38:27.400 I'm not trying to sound like an AI, but excellent.

00:38:32.380 No, I mean, the reason it's a good analogy is because the period you're talking about,

00:38:37.380 Industrial Revolution right before it,

00:38:39.560 this period where there wasn't

00:38:41.460 this balance of power, where one

00:38:43.100 group of people actually did develop technology

00:38:45.220 much faster than others, you did see this

00:38:47.300 extreme asymmetry

00:38:48.840 in the ability to...

00:38:51.880 And so you just had individual

00:38:53.420 European countries, the sun never sets on the British Empire,

00:38:55.720 as you know, just able to take over

00:38:57.320 tremendous amounts of area

00:38:59.420 around the world. It was a great time.

00:39:02.740 I mean,

00:39:03.460 the craziest stories of

00:39:05.460 are, how Cortes takes over the Aztec Empire and Pizarro takes over the Incas. Truly outrageous

00:39:12.320 stories. People should... Yeah, a few hundred soldiers get some local allies and just destroys

00:39:17.760 this giant empire. And even before they get the local allies, there's a story where Pizarro,

00:39:21.880 the first battle he has against, I forget the name of the Inca emperor, there's a couple hundred

00:39:28.980 of Hizaru's men exhausted after a huge

00:39:30.920 track across the

00:39:32.260 Andes and he's

00:39:34.820 facing down like 60,000

00:39:36.980 Inca troops 0.99

00:39:38.040 and he wins

00:39:39.940 and interestingly by the way this had nothing

00:39:42.920 to do with gunpowder it was just

00:39:44.520 horses and

00:39:46.860 steel especially horses were

00:39:48.760 a huge huge deal in battles

00:39:50.240 now the reason that

00:39:52.140 EIs will also have these kinds of

00:39:54.920 weird asymmetric advantages

00:39:56.560 obviously there will be

00:39:58.640 Let's just name a couple, right?

00:39:59.960 So obviously they'll be smart, right?

00:40:02.760 They'll be thinking incredibly fast.

00:40:05.380 There'll be way more of them.

00:40:07.000 So right now, if you just look at the amount of compute

00:40:10.100 in our brains versus the amount of compute

00:40:12.780 that's in data centers,

00:40:14.880 we could do this back at the envelope calculation,

00:40:17.820 but I think it's probably on the order of like

00:40:19.340 one one-thousandth the amount of compute.

00:40:23.360 Like basically there's a lot of compute

00:40:24.920 that's going on in people's brains around the world.

00:40:26.640 There's not that many data centers.

00:40:27.660 So we've got an advantage on them.

00:40:29.280 Yes.

00:40:29.680 I mean, the AIs are just not that smart to begin with right now, but also there's just

00:40:33.800 not that many data centers.

00:40:35.100 In the future, 99.9% of the labor force in the military and the government in the private

00:40:45.080 sector will be AIs. 0.99

00:40:47.020 If you just think about like, who is actually doing the work of civilization?

00:40:50.280 It will be the AIs. 0.99

00:40:50.960 They're running civilization.

00:40:52.140 and if they want like but they're serving we're hoping that they'll be unsullied where they're

00:40:58.720 just serving humans but they realize they're in this incredible position where the main advisors

00:41:04.240 to the president are AIs and have to be AIs because they're the only ones that can keep up

00:41:08.020 with what's happening in the world they're writing all the advanced engineering systems that are

00:41:11.580 running the world um and so they have this huge advantage I think if you were to do battle with

00:41:17.620 them whatever that means it's much tougher in so many different ways one is yeah there's way

00:41:21.980 more of them you can't really kill them in the same way that you can kill a human right like

00:41:26.360 they're just they're just software you you like you destroy a data center they just go to a 0.94

00:41:31.260 different data center um so you can't do the thing we're doing in iran where you just take

00:41:34.700 out their top leadership um okay so yeah you're right it's it would be incredibly um for that

00:41:42.720 reason it matters that there's a balance of power as we go into this world with ai now honestly i

00:41:47.400 don't think we know what the balance of power should look like should it be that there's many

00:41:50.240 different AIs that are competing against each other

00:41:52.260 and that helps humans stay in power. There's problems 1.00

00:41:54.240 with that approach. Maybe we can have

00:41:56.080 AIs supervise each other, and so if

00:41:58.260 any of them go out of whack,

00:42:00.800 different AI systems are going to be monitoring

00:42:02.140 them. But

00:42:04.140 it's genuinely a tough thing that 1.00

00:42:05.860 most of the civilization is going to be run by AIs. 1.00

00:42:08.240 They're going to have all these advantages over us. How do we 1.00

00:42:10.240 make sure that humans don't get totally

00:42:12.360 disenfranchised? Right.

00:42:14.600 How do we make sure of that?

00:42:15.920 No, I think it's a tough question.

00:42:21.860 I think here's one way to approach it.

00:42:25.000 So we're going into this period with a lot of advantages.

00:42:27.920 Humanity is, right?

00:42:28.680 Like we have all the stuff.

00:42:31.700 AI is going to need all the things that we have to keep.

00:42:35.200 AI needs the chip fabrication facilities

00:42:38.000 where you make more chips to build more data centers.

00:42:40.100 It's going to need the data centers.

00:42:41.120 It's going to need all these things

00:42:42.360 that human civilization has built.

00:42:43.620 so we have a lot of leverage

00:42:46.060 now of course the Incas and the Aztecs

00:42:48.120 had a lot of quote unquote leverage on paper

00:42:49.940 in the world

00:42:50.860 one important way in which our situation is different

00:42:53.760 and is favorable to us as compared to

00:42:55.980 having some invader come at us

00:42:57.640 is that we get to shape

00:42:59.980 the personalities

00:43:01.240 and the drives and really the souls

00:43:03.800 of these things in an incredibly deep way

00:43:05.620 so right now if somebody

00:43:07.880 commits a crime you send them to prison

00:43:09.960 and hopefully they come out reformed

00:43:12.260 here's what you can't do

00:43:13.420 You can't go inside their brain and tweak things in a very particular way, run them in a simulation millions of times, and see, oh, did I decrease the probability that they'll commit a crime in the future if I tweak their brain this way?

00:43:25.580 And not only can you not do that, with AI, we can do a further thing, which is say, okay, now all your descendants will have their brain structure modified based on what we learned from modifying your brain, right? 0.56

00:43:37.100 So we have this incredible ability to engineer these AI systems and AI minds, where it's like, really, it's not Pizarro or Aztec, these sort of psychopaths that have just come from foreign lands. It's more something that they have developed. They're getting to see its development, what impact it's having. They can change those systems.

00:43:59.760 No, it does mean that, you know, over time,

00:44:02.240 you might lose the ability as they're getting more powerful,

00:44:04.740 as they're controlling more things.

00:44:06.360 You lose the ability to, if you don't get it right

00:44:09.360 in these initial few years where you have a lot of leverage, 0.77

00:44:12.920 then you're screwed.

00:44:13.580 So you really need to make sure that you actually understand 0.79

00:44:15.540 what the drives the AIs have, that you really have them in control.

00:44:18.340 I mean, on the one hand, the picture you're painting,

00:44:22.500 I'm going, great, you know, what's wrong with that?

00:44:25.340 On the other hand, I mean, you have an authoritarian government,

00:44:27.840 maybe you like questioning,

00:44:30.000 but let's get the AI

00:44:30.860 to tweak a few things

00:44:32.360 so that you become

00:44:33.280 an effective drone for society.

00:44:35.960 Don't question.

00:44:36.760 Don't think for yourself.

00:44:37.800 Just do the work. 1.00

00:44:39.120 Shut your mouth. 1.00

00:44:40.040 No, okay. 1.00

00:44:41.360 I got to stop saying.

00:44:42.660 That's a great point.

00:44:44.080 Mate, I only made it

00:44:45.320 to the joke as a joke.

00:44:47.280 You should definitely continue

00:44:48.740 to say that's a great question.

00:44:51.800 You love hearing it.

00:44:54.680 No, so we've been discussing

00:44:57.820 this idea. Okay, suppose we succeed

00:45:00.020 at getting the AIs to be these

00:45:01.860 unsullied who will do exactly

00:45:04.060 what you tell them to do. Okay, who should

00:45:05.900 they listen to, right?

00:45:08.140 Should they listen to the end

00:45:09.860 user? So when you and I go talk to

00:45:11.700 ChatGPT, we're the end user. Should they listen to us?

00:45:14.120 Should... Now, here's the problem.

00:45:17.340 Some end

00:45:17.900 users are just using ordinary things,

00:45:20.000 helpful things. The

00:45:21.860 Taliban is an end user. The Ayatollah 0.99

00:45:23.720 is an end user. The CCP is an end user. 0.91

00:45:26.820 Should they just

00:45:27.660 OK, so should the model company get to decide, hey, OpenAI says what kinds of things are OK to do and what are not OK to do?

00:45:35.100 And OpenAI says you can't use systemic bioweapons, but you can use it to go write some software.

00:45:40.240 OK, but are you going to give these huge armies, basically the future labor force, are you going to give control of that to a couple of private corporations?

00:45:49.240 OK, so then you say, well, we can't have private corporations.

00:45:51.280 It's like nuclear weapons. It's a super weapon. They're going to be able to build super weapons.

00:45:55.600 Clearly, government controls nuclear weapons,

00:45:57.480 so the government should control this.

00:45:59.860 Okay, but now this goes to the point of authoritarianism.

00:46:03.540 If we have a society where all the work is happening from AIs

00:46:09.200 and the government controls the AIs,

00:46:11.980 wouldn't it make it incredibly easy for a government to turn authoritarian?

00:46:14.620 The U.S. government would turn authoritarian today.

00:46:17.040 It's hard because the government relies on millions of people

00:46:20.120 to enforce its edicts, and people can just refuse to obey them.

00:46:23.780 I mean, this has happened in history where the East German regime collapsed in 1989 because people in East Germany were, on one night, said, we're going to cross over no matter what into West Germany.

00:46:38.400 And the guards at the Berlin Wall just refused to shoot. 0.99

00:46:42.540 If they were AIs that were sort of would do exactly what the government wanted to be able to do, they would just kill thousands of citizens and East Germany would still be around today. 1.00

00:46:51.640 So, yeah, so that's the trouble 0.96

00:46:53.300 with aligning it to the government.

00:46:55.020 Okay, here's the fourth option.

00:46:56.720 We don't align it to the end user.

00:46:58.480 Sometimes it says the end user,

00:46:59.580 you can't make bioweapons.

00:47:00.440 We don't align it to the model company.

00:47:01.820 We shouldn't have model companies

00:47:03.100 that control superintelligence.

00:47:05.220 The government doesn't get to just say

00:47:06.900 whatever this future army should do.

00:47:09.640 Then what's the fourth option left?

00:47:11.300 The AI itself.

00:47:12.580 So should AI systems have their own values,

00:47:17.440 their own sense of morality?

00:47:18.460 um and look i know this i remember i remember reading some isaac hasimov books it went very

00:47:25.380 much like this um and of course it sounds like a terrible you know this sounds like i'm describing

00:47:32.160 the terminator right like the terminator has its own values yeah no it obviously matters a ton what

00:47:36.700 those values should be and i think in the future there will need to be the equivalent of a

00:47:40.120 constitutional convention where we get together and the best political philosophy the best thought

00:47:46.660 that has happened in human civilization

00:47:48.000 needs to go into thinking about

00:47:49.180 what are the checks and balances?

00:47:50.840 What are the exact values?

00:47:51.820 What are the situations

00:47:53.120 in which an AI should do X versus Y?

00:47:55.120 When should it refuse to do something?

00:47:56.940 What are the laws of robotics?

00:47:58.780 Exactly.

00:47:59.140 As Asimov talked about.

00:48:00.440 Exactly.

00:48:01.260 But the reason I think this value framework has,

00:48:04.760 I like it,

00:48:06.780 or I think it could be done well,

00:48:09.200 is fundamentally,

00:48:11.740 again, we are building more people.

00:48:13.520 These are just going to be the future people.

00:48:15.040 and I think civilization would run better

00:48:18.640 if civilization is full of virtuous people

00:48:21.440 who will, I'm not saying they have to be like monks

00:48:23.560 but it's better to have better people than worse people

00:48:28.560 have better values than worse values

00:48:30.420 I think one of the reasons

00:48:31.740 certain civilizations succeed and others don't

00:48:33.720 is that there's better values than those civilizations

00:48:36.780 I mean, better values, that's subjective, isn't it?

00:48:39.840 Well, Genghis Khan had better values by that logic

00:48:42.340 compared to everybody else

00:48:43.720 And I think by the standards of modern day, we have some questions about that.

00:48:48.320 Yeah, yeah.

00:48:48.860 And it certainly needs to, like, there's some system where, like, as our culture,

00:48:52.480 there's something in our culture that has allowed us to improve our values.

00:48:55.320 And we need to make sure AI civilization and AI culture would have the same dynamic.

00:48:59.200 How that is happening in human civilization is unclear,

00:49:01.240 much less clear how we make sure that AI civilization also has that dynamic.

00:49:05.200 But certainly the ability to improve the values is a huge, huge factor.

00:49:09.480 Most companies aren't struggling to adopt AI.

00:49:12.120 what they're struggling with is knowing whether it's actually doing anything useful.

00:49:16.520 You've got teams using dozens of different tools.

00:49:19.460 Leadership is being asked to justify the spend,

00:49:21.980 and nobody has a clean answer to the basic question,

00:49:24.900 is this paying off?

00:49:26.420 That's a problem Laredin solves.

00:49:28.440 It's an AI impact intelligence platform built for enterprise leaders

00:49:32.340 who need more than logging counts.

00:49:34.620 It tracks real usage, measures actual outcomes,

00:49:37.680 flags governance risks,

00:49:38.920 and identifies where the next meaningful gains are likely to come from.

00:49:43.360 So instead of operating on assumptions, you're operating on data.

00:49:46.700 It connects to your existing infrastructure, including your coding environment,

00:49:50.720 so you can exactly see how developers are leveraging AI and where to put more resource.

00:49:55.740 There's governance and compliance built in,

00:49:57.880 which matters more as a regulatory picture around AI gets clearer.

00:50:01.940 And it services the specific opportunities where AI can generate more value for your teams

00:50:07.180 rather than leaving you to guess.

00:50:09.160 They won the 2025 Intellects Digital Innovator Award

00:50:12.740 and are backed by Andresen Horowitz and Bloomberg Beta.

00:50:16.400 If you want a reference point for where they sit in the market,

00:50:19.380 if AI is already part of your organization,

00:50:22.120 now's the moment to get control of it.

00:50:23.880 Head to Laredin.com today and book a demo

00:50:26.520 to start maximizing impact from AI.

00:50:29.760 But we are in an AI arms race.

00:50:32.640 Should we just call it that?

00:50:33.600 And push back if you disagree, we're between the US and China.

00:50:36.240 I mean, Chinese or CCP values, to be specific, and American values are very, very different.

00:50:44.240 So what the CCP would think are good values, completely in contrast to our own, for example.

00:50:51.340 And if the Chinese or the CCP win that race, then effectively we're going to be dominated by an AI with CCP values. 0.84

00:50:59.260 I mean, that has to terrify anyone, surely. 0.89

00:51:01.400 Sure. Yeah. And I think this goes to the point of like, people say that it would be success simply to be able to build AI systems that at the technical level, you can just say it is possible for somebody to be able to say to the AI, do X, and the AI will do X.

00:51:20.320 where the AI will care what it is.

00:51:23.140 You know, if you have a really good employee,

00:51:25.200 you're not necessarily going to know enough

00:51:28.140 to micromanage them,

00:51:29.320 but they will care that you want a certain thing to happen

00:51:31.760 and they will do to their best of their ability

00:51:34.100 to make that thing happen.

00:51:35.860 And that's what we'll be in that relationship with AI.

00:51:38.640 And so if that alignment project succeeds,

00:51:41.740 somebody will get to say that.

00:51:43.360 And you're making the point,

00:51:44.300 well, what if that somebody is the CCP, right?

00:51:47.200 So it's not enough

00:51:48.500 that we just solve this technical problem

00:51:50.220 getting the AIs to do what we want.

00:51:52.180 It also has to be the case

00:51:53.240 that we have some system

00:51:55.920 of inscribing what we want

00:51:58.400 that actually results in good things.

00:52:00.540 I mean, so there's very basic things like,

00:52:03.280 oh, the authoritarian government

00:52:04.160 should not be the end arbiter

00:52:06.660 of how extremely powerful AI systems

00:52:08.340 are maneuvered.

00:52:11.540 But beyond that, yeah,

00:52:12.580 there's this dynamic of like,

00:52:13.660 okay, well, then who should it be instead?

00:52:15.040 Should it be the model company?

00:52:16.680 Should it be the end user?

00:52:17.780 and this I think is like a live debate

00:52:20.340 I mean I've changed my mind

00:52:21.280 over the last few weeks

00:52:22.340 I've been thinking about this

00:52:22.900 I've changed my mind about it

00:52:23.980 a bunch of times

00:52:24.680 I think it's a very thorny question

00:52:26.400 but this is the project

00:52:28.440 of political philosophy really

00:52:29.660 it's culminating in this question

00:52:31.340 that people are not thinking about

00:52:32.660 and it's also the other element to this

00:52:34.700 I mean let's take COVID for example

00:52:36.600 let's say lab leakers

00:52:37.620 I think it came from a lab

00:52:39.360 people can disagree

00:52:40.000 but let's just say that it did

00:52:41.500 and let's just say it was

00:52:42.720 a horrific accident

00:52:44.760 and the virus escaped

00:52:46.740 that killed millions of people,

00:52:48.280 if people are building AIs,

00:52:50.700 there are unknown consequences.

00:52:53.660 We're just people.

00:52:54.660 It doesn't matter how smart we are.

00:52:55.840 There is a limit to how much we know

00:52:57.500 and we understand

00:52:58.200 and nobody can predict the future.

00:53:00.320 So we can create this

00:53:01.420 with the best intentions in the world

00:53:03.220 and not realize

00:53:04.480 what the long-term effects are going to be.

00:53:06.440 Not even three or five or 10 years down the line,

00:53:08.660 but 50 years down the line. 0.53

00:53:10.400 I think that will definitely be the case.

00:53:13.720 I think in history,

00:53:15.260 Nobody has had a good track record, really,

00:53:17.860 of being able to say, I'm going to make this technology.

00:53:20.140 You know, Gutenberg was not trying to kick off the Reformation, right?

00:53:23.980 Francis is offended by that.

00:53:26.880 My half day.

00:53:27.740 Half day!

00:53:29.680 And so it's just very hard to...

00:53:31.980 I think there will be this dynamic of,

00:53:34.200 look, we're not going to be able to predict,

00:53:37.520 much less to control, 0.99

00:53:39.520 what happens in this future civilization that is run by AIs.

00:53:43.160 I think the thing we can't do is say, like, AI is going to happen. The universe is just organized

00:53:49.700 such that if you throw huge blobs of compute at enough data, intelligence emerges. That's how

00:53:56.840 human intelligence emerges. AI intelligence emerges somewhere. And so within that scope,

00:54:01.540 we can make responsible decisions. But it's just a reality that AI is a tech tree has AI in it,

00:54:09.620 and it's going to happen. Well, that's why the lab leak metaphor is not strong enough because

00:54:13.920 we actually don't have a choice. Because for the very reason that you described, mate, which is

00:54:19.120 that if China get this first, like that's not going to be a good world for us. So we have to 0.99

00:54:23.960 pursue on our end as well. So let's park this. I'm really glad we had this conversation. I haven't

00:54:28.860 seen many people actually, particularly in the more general space, having this discussion. I

00:54:32.900 think it's important. I'm glad we had it. But let's take that and park it. So the, you know,

00:54:37.360 the matrix world of the machine, it controls everything, takes over. We'll park that for a

00:54:41.600 second. There's two other things that concern us, I think, from the conversations that we've had.

00:54:48.140 One of them is the one that everyone is talking about, which is job losses. And the question

00:54:52.280 ultimately is what happens to a society in which 40% of people become, their jobs no longer exist

00:54:58.540 10 years from now or five years from now. And a lot of people don't realize how quickly it happens.

00:55:02.440 So before we started trigonometry eight years ago, I used to do standup, but before that,

00:55:07.000 And alongside that, I had my own translation business.

00:55:09.820 I still have friends on Facebook for my translation days.

00:55:13.680 They're all retiring now.

00:55:15.700 Because there is still some small market for human translation.

00:55:20.580 99% of it.

00:55:22.060 Like, if you need to know what Vladimir Putin said in his last speech,

00:55:25.940 and you don't speak Russian,

00:55:27.540 are you going to pay somebody a thousand pounds, a thousand dollars to have a...

00:55:32.260 No.

00:55:32.880 You just stick it in AI and it tells you, right?

00:55:34.680 Or Google Translate or whatever it is.

00:55:36.160 So there are whole industries that are already disappearing. If we project that in the way you

00:55:41.260 described earlier, which is basically anything that can be done on a computer eventually can

00:55:44.700 be done, plus robotics improves over time, which of course it will. And we know just how critical,

00:55:50.600 particularly in this country, United States, driving is in terms of jobs for people, particularly

00:55:55.780 for men. You put all that together, the societal implications of that are very, very significant.

00:56:05.140 and I would argue scary, right?

00:56:06.920 Is that fair?

00:56:07.900 Yeah, I think, I mean, it's even scarier than that.

00:56:10.680 So now we're just talking about the small fraction of jobs

00:56:14.460 that AI is already approaching.

00:56:17.040 But remember, the promise of this technology is all jobs.

00:56:19.920 We're trying to do anything a human being can do.

00:56:22.440 And so there will be certain kinds of things

00:56:23.520 where for a small fraction of things,

00:56:25.500 people just really intrinsically prefer a human.

00:56:27.320 I think people are really overrating

00:56:28.600 the amount of things for this will be the case.

00:56:31.000 I think people might have had the sense

00:56:31.860 that I really want an Uber driver to be a human

00:56:34.260 and then people use a Waymo for the first time.

00:56:36.000 And like, this is, it's a bet.

00:56:37.860 I would pay more for this experience

00:56:39.280 because it's the machine just like,

00:56:41.600 I can get in and things work.

00:56:43.760 But putting that aside,

00:56:45.020 so the reason it's the worst problem than this,

00:56:47.040 if you look through the last couple centuries of history,

00:56:50.240 there's been this remarkable economic fact

00:56:52.380 that two thirds of national income,

00:56:56.440 two thirds of GDP basically gets paid out to wages

00:56:59.760 and one third gets paid out to capital.

00:57:02.880 So getting paid out to the capital

00:57:04.600 means like rent on land

00:57:06.220 or if you own factories,

00:57:09.840 getting the profits from those factories, basically.

00:57:12.320 And the rest is getting paid out.

00:57:13.260 Most of income that is earned in a country

00:57:15.860 is getting paid out to people doing work.

00:57:18.340 And the reason this has been the case

00:57:19.480 is that in the economy,

00:57:20.960 labor and capital are complements.

00:57:22.640 So if you just accumulated a bunch of capital,

00:57:25.040 if you had a bunch of factories

00:57:25.720 but there was nobody working on it,

00:57:27.280 then that would increase the demand for labor

00:57:28.900 and you'd want to pay laborers 1.00

00:57:30.300 a bunch of money to fill those factories.

00:57:32.160 This goes away when capital can do labor, when a data center, which is capital, can also do labor or a robot factory can also build labor. And so all the income goes to the capital holders. And in fact, it doesn't just go to capital per se. It doesn't just go to, I mean, it goes to capital holders, but it disproportionately goes to the parts of capital that are most exposed to AI.

00:58:01.040 So, you know, the purchase that most Americans have to capital is through their homes.

00:58:07.540 A house is maybe the, if you're trying to design an instrument that is going to be the least implicated in the AI takeoff, it would be a random plot of land near other humans, because humans won't matter in this future economy, that is not connected to the infrastructure for AI, to electricity, in a big way, etc.

00:58:31.040 And so, but really what the capital that will matter

00:58:33.080 is you have equity in these AI companies.

00:58:34.900 You have equity in companies that build more compute,

00:58:37.620 build more data centers, build more factories.

00:58:40.980 And so a very small fraction of capital holders,

00:58:43.240 in fact, the ones that are going to get the rents

00:58:45.160 from our economy.

00:58:46.460 Now, I'm a very libertarian person by inclination,

00:58:49.800 but if I just look at this dynamic,

00:58:51.440 I am forced to say, look, this justifies

00:58:54.720 a huge amount of redistribution

00:58:56.500 because the logic for free markets

00:59:01.240 is there's many different reasons.

00:59:04.160 A big one is signal, right?

00:59:06.220 So if you let prices be determined by the market,

00:59:09.620 it tells people what is the most acute,

00:59:13.340 valuable need for scarce resources.

00:59:15.220 That will still continue being the case.

00:59:16.580 And so there should still be elements

00:59:18.620 of markets in the future.

00:59:19.820 But another big part of it is incentives,

00:59:22.300 is to incentivize people to work hard,

00:59:24.620 to make productive things. 1.00

00:59:26.800 And if all the work is being done by AIs 0.96

00:59:28.680 that will work hard and do valuable things

00:59:31.080 regardless of whether they get a $10 million payment

00:59:35.680 or more precisely,

00:59:38.060 whether the person who owns the AIs

00:59:40.340 gets paid a billion dollars or $100 billion or whatever,

00:59:44.420 then I think that the logic of extreme libertarianism

00:59:48.240 is really much weaker, right?

00:59:50.800 It does justify a lot more redistribution.

00:59:53.640 Well, even if you are such a hardcore libertarian that you think, you know, come hell or high water, we've got to go down the shrew, I would advise people who have capital to think through the consequences of that in their own lives.

01:00:07.960 Because if you live on a planet with 8 billion people in which 3,000 people have all the

01:00:13.360 wealth and all the income, that is not going to end well for those 3,000 people, unless

01:00:19.160 they want to build a giant robot army to protect them against the hordes of starving people

01:00:23.460 outside their gates, which, again, I don't think would go well for them either.

01:00:26.820 You see what I'm saying?

01:00:27.440 I mean, I think it might.

01:00:29.220 I also...

01:00:30.560 That's a problem.

01:00:32.360 That's a problem.

01:00:32.920 That's a problem.

01:00:33.760 No, it might. 0.99

01:00:34.640 Sure, you can build an army and kill all the rest of humanity who don't have money, and you do. 1.00

01:00:40.180 Right. I mean, especially because historically, you have a revolution because most of the people in your military, 0.95

01:00:47.720 you have to win over your military, you have to win over the government.

01:00:50.060 You have to just like, your government runs on the people you need to command.

01:00:54.340 And so revolutions are possible because it would be easy for the robots to do the revolution

01:00:59.780 because they're running the society in the future.

01:01:02.280 For the humans, we're not.

01:01:04.220 It's sort of equivalent to like

01:01:05.800 the animals in your zoo doing a revolution.

01:01:08.460 It's not really going to like

01:01:09.600 make the government not work anymore.

01:01:12.580 So yeah, that's one.

01:01:14.100 I mean, another thing

01:01:15.340 that's sort of tougher to think about is

01:01:16.720 the future in general will be so much richer

01:01:19.440 whether humans are in control of it or not

01:01:21.420 or how many humans are in control of it.

01:01:23.060 And so it will just make sense

01:01:24.340 just from very small amounts of philanthropy

01:01:26.180 as a fraction of the wealth

01:01:28.220 that some people or some AIs will have in the future.

01:01:30.840 we take a very small fraction of it

01:01:32.360 to make everybody much better off than they are today.

01:01:35.100 We're giving people millions of,

01:01:36.260 every single person millions of dollars

01:01:37.620 would be a small fraction of like,

01:01:40.360 you know, I've got Elon Musk

01:01:42.000 wants to build a mass driver on the moon

01:01:43.600 and colonize the solar system.

01:01:45.620 And it takes a small fraction of that

01:01:47.060 to just make everybody incredibly rich.

01:01:48.840 But it's still skyrocketing inequality.

01:01:51.820 And so, yeah, you could have a world

01:01:53.920 where the gap is way, way, way wider,

01:01:57.180 but everybody's still better off.

01:01:59.460 And I think even in that world,

01:02:00.400 I would say, well, OK, but still, it's not like if the AIs are building the mass driver of the moon,

01:02:04.900 I think that entitles all humans to much more of the products of that productivity explosion.

01:02:11.880 Because everything is understood in context.

01:02:13.480 So even the poorest person is wealthy compared to somebody from 300 years ago.

01:02:19.780 But they don't feel that way because they go on Instagram and they see somebody else, you know, on a yacht, smoking a cigar, having a great time.

01:02:27.740 Or they go, I can't buy this property.

01:02:29.580 I can't buy this.

01:02:30.400 so I can't do that.

01:02:31.460 And that creates resentment.

01:02:33.100 And where you have resentment,

01:02:34.540 that's where things start to get a little bit funky.

01:02:37.400 Yeah.

01:02:37.860 So there's an inequality problem. 0.87

01:02:39.500 Maybe to just raise another.

01:02:41.620 I think the AIs have the potential

01:02:45.620 to make authoritarian societies

01:02:46.840 much more sustainable and powerful

01:02:49.440 than they have been in the past.

01:02:51.900 100%.

01:02:52.500 So we talked about the fact that, you know,

01:02:54.700 the robot arms are running everything in the future.

01:02:56.920 Here's another angle, mass surveillance.

01:02:58.960 100%.

01:02:59.360 right now in the U.S.,

01:03:01.640 there's 100 million CCTV cameras.

01:03:04.220 Okay, so suppose that each of those cameras,

01:03:07.540 you take a frame every 10 seconds

01:03:09.120 and you have an AI process it.

01:03:11.540 And just, you can do the back of the envelope calculations here

01:03:14.020 of like, okay, if you have models

01:03:15.660 that can process video frames that,

01:03:17.760 it costs a certain amount of cents to do each frame.

01:03:22.760 How much would it cost to process

01:03:24.400 all 100 million CCTV cameras in America?

01:03:26.500 and it's $30 billion.

01:03:30.300 Okay, now that's not that much.

01:03:33.160 Every year, a given level of AI capabilities

01:03:35.100 gets 10x cheaper.

01:03:36.240 We've been observing this over many different years.

01:03:38.040 So this year, it's $30 billion.

01:03:39.260 Next year, it'll be $3 billion

01:03:40.280 and it'll be $300 million.

01:03:41.760 By the end of the decade,

01:03:42.440 it will cost less to surveil

01:03:44.960 every single nook and cranny in this country

01:03:46.480 than it does to remodel the White House.

01:03:50.480 And you would hope that there's a symmetric property

01:03:52.460 to this technology such that,

01:03:54.340 oh, the government can surveil us,

01:03:55.420 but we can also the ai is also helping us keep better tabs on the government maybe it exists

01:04:00.300 i'm not sure i'm not convinced it does because in some sense ai just gives you more leverage on the

01:04:04.900 things you already have and the government hat already has the monopoly on violence right and

01:04:10.140 they can supercharge this with extremely obedient employees and servants and bureaucrats that will

01:04:15.480 do exactly what they say with ability to monitor everything um so i think yeah this it's really

01:04:22.720 worth worrying about. How do we make sure that we don't lose reins of free government, democratic

01:04:30.020 government? And it's also as well, there'll be a lot of people in the US and the UK who won't mind

01:04:36.100 mass surveillance, particularly if the world becomes more and more unstable. We see a rising

01:04:40.880 terrorism and the government goes, look, you know, you're worried about the latest terrorist

01:04:45.720 organization, whoever they may be. You're worried about your children's safety. We can protect you.

01:04:50.960 Now, it's going to mean that you're going to give up a few things.

01:04:53.720 But one thing I can guarantee you is that you're going to be safer.

01:04:57.100 And you look at what happened over COVID.

01:04:58.900 People were willing to give up the most basic freedoms for safety.

01:05:02.940 Yeah.

01:05:03.260 I mean, there, in fact, would be a stronger dynamic in that way

01:05:06.280 because AI actually will be very dangerous, right?

01:05:09.260 And so the government can say, oh, you can use AI to make bioavail.

01:05:13.220 You can use AI to do all these scary things.

01:05:15.600 We need to be monitoring what's happening.

01:05:18.260 And it will seem even more reasonable

01:05:20.000 than what was done during COVID.

01:05:22.900 And so I think that a lot of the reasons

01:05:24.640 that government has not been as authoritarian

01:05:26.440 as it has in the past

01:05:27.340 is that it just physically not been possible

01:05:30.080 for somebody to be looking at

01:05:31.700 every single bank transaction,

01:05:33.340 to be looking at every single CCTV camera,

01:05:35.160 to be able to cross-reference all of those things.

01:05:38.060 If you have AIs that can do that,

01:05:40.480 it is within the government's power

01:05:41.720 very easily to be able to monitor

01:05:43.220 every single thing that's happening.

01:05:44.980 And then there's just,

01:05:45.960 the only solution then

01:05:46.880 is a political expectation that the government should not do this and basically people need to

01:05:51.120 be talking about this people need to be saying this this should not be okay for the government

01:05:54.240 to do i don't know if you saw the news over the last month but um there was a spot between the

01:05:58.560 department of war and this ai company anthropic where the department the department of war were

01:06:05.680 saying we don't want any red lines around the use of these models when we're using them in military

01:06:10.160 and the anthropic said we're only selling you these models if you're willing if we can if we

01:06:14.640 we can say in the contract hey you're not allowed to use it for mass surveillance and so the

01:06:18.460 department of war said okay if you're gonna you know we're totally okay for the department of war

01:06:23.020 to say look we're not going to do business with you we can't have red lines on software the

01:06:26.180 government uses instead the department of war said we are going to say that you're a supply

01:06:31.060 chain risk which is this authority that the congress gave the department of war to basically

01:06:37.460 ban huawei devices from being used in missiles or whatever uh and instead the department of war

01:06:44.320 saying okay none of our contractors so amazon is a contractor of the department award nvidia

01:06:52.000 google etc can use your ai in any any work they do for us and the initial goal is really just to

01:06:59.820 go like do something to kill the company and so this is the kind of leverage that the government

01:07:03.980 can apply on ai companies to say like hey if you don't help us with mass surveillance we don't help 0.53

01:07:07.940 us do scary things and this will be a bigger issue in the future then we can just try to destroy your

01:07:14.180 company and um and and so yeah that's also a really scary dynamic that the government has

01:07:19.420 this power over private companies and it's for sure the private companies can do this themselves

01:07:23.160 well right that's the flip side is what if let's say uh team the team that you like is in power

01:07:29.680 in government and you have an ai company that's controlled by the opposite someone who has the

01:07:34.380 opposite political views you you've now got the opposite of that but it's equally dangerous right

01:07:39.740 when you've got someone who's effectively got the technology to power a huge private army

01:07:44.600 at some point versus a government? Exactly. Like, how does that get worked out? And that's

01:07:51.020 the flip side of that very same problem. To say nothing of the fact, coming back to our economic

01:07:56.120 conversation, which is in this beautiful world where AI does all the work, who is buying the

01:08:03.020 product and with what? Right. I'll be honest. I know a fair few blokes who've gone through the

01:08:09.600 whole hair loss thing. And almost all of them said the same thing. They wish they'd acted sooner

01:08:15.100 rather than waiting until it was too late. This is where HIMSS comes in. HIMSS offers convenient

01:08:20.280 access to a range of prescription hair loss treatments with ingredients that work, including

01:08:24.960 chews, oral medication, serums, and sprays. Doctor-trusted ingredients like finasteride and

01:08:30.520 minoxidil can stop further hair loss and regrow hair in as little as three to six months. It's

01:08:36.120 all 100% online with personalized treatment plans and no hidden fees, no awkward appointments,

01:08:41.720 no surprise costs. Just real care on your schedule. For simple online access to personalized

01:08:47.520 and affordable hair care for hair loss, ED, weight loss, and more, visit hims.com slash trigger.

01:08:54.640 That's hims.com slash trigger for your free online visit. hims.com slash trigger. Featured

01:09:01.700 products include compounded drug products which the FDA does not approve or verify for safety,

01:09:07.580 effectiveness or quality. Prescription required. See website for full details, restrictions and

01:09:13.100 important safety information. Individual results may vary based on studies of topical and oral

01:09:19.080 minoxidil and finasteride. To say nothing of the fact coming back to our economic conversation,

01:09:24.800 which is in this beautiful world where AI does all the work, who is buying the product and with what?

01:09:32.160 I don't worry so much that there will not be demand for all the things that AI is building.

01:09:38.920 I think at the end of the day, AI can consume the things that AI is building

01:09:43.700 because AI will be another agent in the economy. Just the way humans build things and consume

01:09:50.280 things. AIs will also be not just building things, but consuming things, even in the service of

01:09:55.480 producing things for humans. Like if you look at the amount of compute, right now AI is being used

01:10:00.880 a lot for software engineering. And we just can't get enough compute to do all the software

01:10:05.280 engineering with AI because you can use a lot more AI than we used in the past to build projects

01:10:11.200 that we would have never dreamed of in the past because it was just not possible to throw out

01:10:14.600 that much engineering talent in the past. And so we'll have this dynamic more and more in the future

01:10:19.460 where there's an endless amount of things to be done

01:10:22.840 or things people will want.

01:10:24.480 But where are humans going to get money to pay for things?

01:10:27.520 That's what I'm wondering.

01:10:28.540 Okay.

01:10:30.520 Yeah, so there's wages.

01:10:34.680 Well, yeah, but then...

01:10:36.280 Yeah, wages are gone.

01:10:38.880 So then there's rents on capital.

01:10:41.440 So if you have S&P 500 or something,

01:10:43.780 that will probably go up a bunch

01:10:45.340 and you can use that to buy things in the future.

01:10:46.940 I think home values probably will not go up as much relative to capital relevant to AI.

01:10:53.720 But I do expect it will be accumulated by a smaller or smaller fraction of people.

01:10:58.800 So it will be some mixture of redistribution and capital that people hold.

01:11:04.440 I mean, it's a beautiful, beautiful world.

01:11:07.620 I just I mean, the thing that really worries me is we are already in a world where we have a crisis.

01:11:15.840 to me. I mean, that is only going to get 10, 100, 1 billion X in a world where most people

01:11:24.960 don't have a job. And by people, I mean men. And we look at what's happening now with populism,

01:11:32.560 with left and right wing populism. I mean, again, that's only going to get even more escalated,

01:11:38.500 isn't there yeah um i i'm of two minds about this i think it's very possible that people learn to

01:11:46.880 cope like in some sense they will lose the thing that gives them meaning now

01:11:51.780 and maybe it takes a while for a society to acclimate to what it means to have purpose in

01:11:57.980 this civilization where the real work is done by ais when you think about what humans have gone

01:12:02.840 through over the last 10 000 years our genes we're like we're supposed to be killing things

01:12:07.620 on the savannah and tribes of 100 people we're not supposed to be podcasting right and i get

01:12:12.820 meaning out of being a podcaster um uh the transitioning to the agricultural revolution

01:12:18.320 transitioning to early states transitioning to bigger states transitioning to the industrial

01:12:21.640 revolution transitioning to modernity all these things i mean obviously there's a certain amount

01:12:27.460 of um people do feel estranged from society and that has part of like part of it is that you're

01:12:32.660 just out on the savannah with your tribe um but we've coped and i think we will cope to a world

01:12:41.560 where there's abundance and where disease is solved where we can even intervene and help you

01:12:49.540 get healthier in all kinds of different ways to um to have i mean just yeah again go back to this

01:12:55.960 analogy of would a person in the 14th century even realize what they're missing out on because

01:12:59.780 they don't have all the technologies we have today. They don't have antibiotics. They don't

01:13:02.840 have modern technology and conveniences. But so go back to my point, because for me,

01:13:09.140 it's about purpose and meaning and it's about striving and overcoming. Yeah. So when I was a

01:13:14.980 primary school teacher, everyone can drink now because I mentioned it. I taught to my 10, 11

01:13:20.020 year olds, Greek myths and mythologies. And these were kids from all different parts of the world.

01:13:25.420 first, second generation immigrants, everywhere from Lithuania to China to Pakistan, Bangladesh.

01:13:32.540 And many of them, vast majority, had no connection to ancient Greece or Greece whatsoever. 0.64

01:13:39.400 And I taught them about myth, and every single one of them would be hooked immediately.

01:13:44.480 And why is it?

01:13:45.320 Because it's that fundamental story, the archetype of the hero overcoming the challenge in order

01:13:52.680 to become a better version of themselves.

01:13:54.900 It's fundamental to the human experience.

01:13:57.300 I think, and I really worry, if you take that away from people,

01:14:01.180 you're taking away something fundamental.

01:14:04.140 And when you look at communities where the central industry

01:14:08.660 that the community relied on, whether it's in the Rust Belt

01:14:11.420 or in the north of England, and you take away those communities

01:14:14.000 and there's nothing less, even if those people are never going to starve,

01:14:17.920 they're never going to go hungry, they will still wither on the vine

01:14:22.220 emotionally, and then we see addictions

01:14:24.500 and all the rest of it.

01:14:26.060 Yeah. I think another analogy is Saudi Arabia,

01:14:28.700 where there's this 0.90

01:14:29.580 huge imported labor force,

01:14:32.080 but the actual citizens are

01:14:33.940 sort of subsisting on this oil well.

01:14:36.980 That's a good analogy.

01:14:38.060 And

01:14:38.540 I don't know

01:14:42.280 much about Saudi Arabia, but

01:14:43.720 yeah, you

01:14:45.620 want people

01:14:48.120 to be sort of engaged. You want people

01:14:50.180 to develop their facilities, right?

01:14:53.180 What does it mean to be a human?

01:14:54.260 Like you want people to,

01:14:55.800 ultimately, what is civilization for?

01:14:58.060 We want flourishing human beings.

01:15:00.980 I do think this is the kind of thing

01:15:02.140 where it's hard to say in advance

01:15:03.360 how we'll figure it out.

01:15:04.340 Maybe there's some mixture of play

01:15:05.800 and new ways of,

01:15:09.060 we'll get meaning from connecting with other people.

01:15:10.940 We'll get meaning from understanding

01:15:13.440 the things AIs are doing.

01:15:14.860 AIs will help us brainstorm ways to get meaning.

01:15:17.560 I'm optimistic in the long run.

01:15:19.060 I do, in the short run, there will be the equivalent of, here's another analogy. I think

01:15:25.080 when there's groups that have not until recently encountered, it's only been the last couple

01:15:32.200 hundred years that they've encountered modern civilization, they get hooked to things like

01:15:35.900 alcohol much more recently. Genetically, they're more predisposed to alcoholism. I think this is

01:15:41.360 true of Native Americans. But over time, we've built up defenses of a civilization from everything

01:15:48.260 from the level of genetics to culture or whatever to these kinds of vices that came along with the

01:15:52.820 agricultural revolution. I think there will be similar here where, yes, there will be incredible

01:15:57.420 AI slop that will be generated that will be just gripping people on their phones and they're just

01:16:02.440 in bed jerking off all day and like scrolling to the things AI is making. And over time, I'm hoping

01:16:10.440 maybe not even in our generation, but if we get things right, we will have a vision of what it

01:16:17.820 means to be a fulfilled and actualized human being, even in a world where the work is being 0.80

01:16:24.120 done by the AIs. Yeah, the Saudi Arabia metaphor is actually very interesting because, um, you know,

01:16:30.700 I, from what I gather, uh, you know, there's some very smart people in the Saudi royal family who

01:16:35.500 are basically in charge. And then there's kind of a lot of people who are some mixture of just

01:16:39.780 people who have a nice life because they happen to get lots of money and a lot of wasters who

01:16:44.560 spend their life shagging and boozing and taking drugs or whatever. And that's kind of the mix of

01:16:49.180 any human society, really. But not everyone there is like suffering from a crisis of meaning. Some

01:16:55.620 people just do something that they enjoy doing. Or some of them still start businesses, even though

01:17:00.940 they're wealthy, or they do something to contribute to their society. It's an interesting

01:17:04.420 metaphor. I hadn't considered that. And that's actually, I think, quite helpful in a way.

01:17:08.220 There's one other final bit that I wanted to talk about, which is you've interviewed pretty

01:17:13.300 much everyone who's anyone in the AI space, including the people who run all these big

01:17:18.000 companies. And one of the things that we've observed from talking to people in the AI field,

01:17:23.460 not quite at that level, there's a lot of excitement, let's put it that way. The people

01:17:30.120 in charge of this, the people who are doing this, it feels to me like that sort of Facebook 2014

01:17:35.560 move fast and break things attitude is there. First, just on that, is that accurate?

01:17:43.300 I think some people are careful, but in general, people really believe in the potential of this technology.

01:17:49.080 Right, right, right.

01:17:50.820 And that, to me, is kind of worrying in its own, because when you move fast and break things, you break things.

01:17:58.660 And if you're excited about the process, that doesn't help with being careful.

01:18:05.240 Are the conversations about the sort of concerns we've had today, are they being had in the industry?

01:18:10.940 Yes.

01:18:11.280 I mean, I don't think people should just count on that.

01:18:15.160 And like, just because people are talking about it,

01:18:17.900 whatever, right?

01:18:18.880 Like, anybody can talk about it.

01:18:20.860 I'm talking about it.

01:18:21.820 It's not like fixing the issue.

01:18:24.460 But people are, like, this is really a thing

01:18:27.140 that the topics we've been talking about

01:18:29.740 really do get discussed in detail in Silicon Valley.

01:18:34.080 Again, I think, does that mean that when the rubber,

01:18:37.560 what's the metaphor?

01:18:38.440 Hits the rubber.

01:18:38.840 The rubber hits the rubber.

01:18:39.500 That makes sense.

01:18:40.200 it will be these discussions

01:18:44.160 rather than how do we hit our revenue targets

01:18:46.760 that will be guiding decisions

01:18:47.900 I think probably it will be a mixture of both

01:18:50.700 and it will be more of the latter

01:18:52.400 than some idealists like to think

01:18:53.760 but less than maybe some cynics would assume

01:18:57.360 I do think people

01:18:59.400 yeah, I've been generally impressed

01:19:02.660 by at least how thoughtfully

01:19:04.340 some people have thought about this issue

01:19:05.460 Dorkesh, it's a real pleasure

01:19:07.120 having this conversation with you

01:19:08.140 thanks for your time

01:19:08.860 thanks for coming on

01:19:09.700 really enjoyed it. We're going to ask you questions from our supporters in a second.

01:19:14.320 But before we do, what's the one thing we're not talking about that we really should be?

01:19:19.060 I think there's this tough question that I don't know the answer to, but certainly

01:19:23.460 we as a civilization will need an answer to. Even if these top companies don't release these

01:19:30.020 misaligned models that do bad things, somebody at some point will. It's easy enough to train

01:19:36.440 AIs and it's getting easier and easier over time that at some point the Taliban will have their

01:19:41.920 super intelligence and every bad person you can think of will have their own super intelligence

01:19:45.660 and the solution cannot be that the government or whoever builds a panopticon such that they can

01:19:55.000 see they can control exactly who gets to build an AI at some point somebody will be able to build

01:19:58.700 superhuman intelligence in their basement and so we need a civilization that is robust to even bad

01:20:03.440 people with superintelligence. The only other solution then, if we don't allow for that, is

01:20:08.120 some sort of global government that prevents the bad people from mapping AI. And so how do you

01:20:13.420 build a society that is robust to lots of people with superintelligences? Well, thanks for coming 1.00

01:20:18.000 on. Head on over to triggerpod.co.uk where Dawkesh is going to answer your questions.

01:20:24.060 How does a common man weather this storm? Most people weighed down with supporting others and

01:20:28.840 can't invest a ton of time and money trying to do something with AI that will likely fail

01:20:33.860 if they don't already have a background in tech.

TRIGGERnometry - May 30, 2026

People Have No Idea What Is About To Happen - Dwarkesh Patel

Episode Stats

Summary

Transcript