Based Camp - Based Camp： How AI's Will Hack People (with Spencer Greenberg of Clearer Thinking)

00:00:00.000 systems like TikTok, imagine a world where it's generating billions of new pieces of content every

00:00:05.160 day with generative AI, or even going a step further, generating custom AI content for your

00:00:13.620 mind. And then it sees how much you like it, and then it tweaks its generation process.

00:00:18.860 So we could imagine a world in 10 years, 20 years, I don't know, where you're actually seeing AI

00:00:25.440 generating content personalized to you that you just kind of receive in a stream and become just

00:00:30.380 insanely addicted to. Would you like to know more? Hello, we are so excited to be here today with

00:00:36.860 Spencer Greenberg of the Clearer Thinking podcast, and also famously like the Clearer Thinking

00:00:42.420 organization. The guy who is running this project, which is now fact-checking, trying to review

00:00:48.720 psychological research to see if it's replicable as it's being published, which is just so cool.

00:00:54.100 We've known him for years, almost as long as we've known each other. And he is one of the

00:01:00.480 sort of social leaders of the EA rationalist diaspora in the New York area, and just an incredibly

00:01:08.420 respected gentleman scientist, in our opinion. And by gentleman scientist, we mean one of those

00:01:14.060 people who is out there, outside of mainstream academia, conducting really great research.

00:01:19.840 And you can tell he's one of the leaders because he vehemently denies it.

00:01:24.100 And yet he like builds all the stuff that everyone uses.

00:01:28.480 Well, I mean, he's definitely a thought leader in the Gentleman Science Network because he's

00:01:33.500 created the software that everyone's using to do it right now. Like I would say 80% of the

00:01:38.080 independent research is probably done on his software right now. But where we wanted to go

00:01:42.580 with this one is we wanted to talk about where humanity is going into the future. So let's start

00:01:47.160 with the question of where do you think we're going to be in 10 years?

00:01:51.740 So I should start by saying this is wild speculation. Predicting the future is incredibly

00:01:55.840 hard. So take everything I say with a massive grain of salt. A lot of that's going to be wrong.

00:02:00.100 But with that caveat behind me, I will speculate. I think that AI is going to have an absolutely

00:02:04.820 massive effect on society. I think it's hard to really fathom right now all the different ways,

00:02:10.860 but I think we're going to see it coming. I think a few areas we'll see it coming. One is in spam and

00:02:16.860 manipulation. There are already a bunch of countries that have warehouses of people that are essentially

00:02:23.820 putting propaganda onto Twitter and social media. This is well known. But I think that that is going

00:02:29.480 to be nothing compared to a million bots that are powered by AI, right? It's one thing to have 300

00:02:35.320 people in a warehouse. It's another thing to have a million AI bots, right? It's also these AI bots

00:02:40.560 are now at the point where they could all have different personalities. They could all say unique

00:02:44.220 things in line with those personalities. They could have a lot that can be sleeper agents where

00:02:48.920 mostly they just tweet or post totally normal stuff 99% of the time, but then go into full gear

00:02:55.100 to promote propaganda. So that's just one little facet. Let's start there.

00:02:59.100 I was going to ask. So Elon is right when he's saying you need to limit access to these platforms

00:03:05.000 now and any platform that doesn't is just going to become swarmed with bots.

00:03:10.920 I think it's complicated. I mean, I do think, I do think this is going to be a massive problem,

00:03:14.620 whether the current approaches Twitter wants to use will work. I don't know. I mean,

00:03:18.500 almost nobody is on Twitter blue relative to the whole Twitter population. So I'm not sure how

00:03:22.360 they're going to do that. Maybe what they'll do is make it harder and harder to do anything as a

00:03:27.400 non Twitter blue user, but I don't know that that solution will actually work. I think it's

00:03:33.300 going to be an arms race and it's a very difficult one. The closer an AI gets to being able to do what

00:03:37.500 a human can do, the harder it is to actually detect that it's an AI, right?

00:03:42.520 Actually, that brings me to something that's been happening recently on Reddit. And I don't know if

00:03:46.160 you've seen this phenomenon, but Reddit's in this big fight right now with like the mod teams.

00:03:51.500 And it appears that Reddit corporate is using lots of AIs. Actually, they recently had to block out

00:03:58.940 like the programming subreddit or something like that because they figured out that the Reddit

00:04:02.840 corporate was using their own AIs to try to make it seem like the community was more okay with what

00:04:08.900 was happening. I'm wondering how you think social media companies, do you think that this sort of

00:04:14.660 action would be like a death knell to a social media company in the future? Or do you think the future

00:04:18.680 will be like companies that are anti-AI versus companies that are pro-AI? Like where do you

00:04:23.280 think we'll begin to see a differentiation and evolution of social media platforms as it relates

00:04:29.460 to AI? Yeah, it's an interesting question because I think you'll start to see AIs on social media

00:04:35.980 that are actual just kind of creators, right? So some people will make this proliferation of AIs that

00:04:43.700 are tweeting about different things or posting on social media about different things in a certain,

00:04:47.340 in a benign or even helpful way where they're just kind of content creators. And then you'll see

00:04:51.340 other ones that are, I think, going to be like sleeper agents that are really like propaganda bots,

00:04:55.220 but they're just imitating being a real person. And you'll see other ones that are attempting to scam

00:04:59.640 people, like even attempting to befriend people over a period of months or years, which I think is we're

00:05:05.400 going to start to see that where bots will like create extended relationships with people and then

00:05:09.760 scam them. And I think some of these things, all social media companies are going to have to be

00:05:13.180 against because it's just going to really affect the performance of their platform so much. It's

00:05:18.100 going to be such a bad experience being on the platform that they just have to be against it.

00:05:21.520 Other ones they might embrace, like AI content creators where it's okay, well, maybe it's okay

00:05:25.980 to have a million different bots that are just creating content of specific types, each with

00:05:30.140 specific personalities trying to appeal to the masses. And then there'll be an interesting question

00:05:33.920 about disclosure. Will you have to disclose that it's an AI making the content? Or if it's a hybrid

00:05:37.860 between human and AI, how does that work? So yeah, a lot of really tricky questions.

00:05:41.520 What you're making me think actually is that a lot of social media platforms are probably going to

00:05:46.460 require, for example, government identification to prove that you're human to create an account.

00:05:51.780 But then I've met, I bet there's going to be this whole new form of stealing other people's

00:05:55.700 identities. Cause like we, we recently were in a bunch of settings where we met people who were

00:05:59.540 like, oh, I'm not the internet person in my family. Like my wife does all the YouTube. So we know

00:06:03.680 that there are people like super normal people who are just offline, like they just don't do the

00:06:08.160 internet. And then those people, I wonder if I were trying to create AI that could do what you're

00:06:13.480 describing, Spencer, I would probably find those people. I would pretend to be them. And I would

00:06:19.020 find basically the offline people make an online identity for them and then pretend, use their ID,

00:06:24.400 steal their identity, but just to create a very convincing bot, because I would get through

00:06:29.500 the, the barriers of social media platforms that think I'm a verified person.

00:06:34.160 Well, the interesting thing you said there is that there actually would be a marketplace and

00:06:39.160 potentially a valuable marketplace for them to sell their own identities if they don't like to use

00:06:44.000 online platforms a lot.

00:06:45.420 Yeah. Wow. That's a 10 year possibility.

00:06:49.580 Yeah. And I think if, if these platforms are worldwide, it makes it much, much trickier,

00:06:53.760 right? If you have to support all kinds of government ID in most of the countries of the

00:06:57.740 world, then all you need is like a few leaks in that process where someone figures out how to

00:07:02.520 generate convincing IDs from such and such country. And we see this on Mechanical Turk,

00:07:08.160 Amazon Mechanical Turk, which is a platform where people do small amounts, small tasks for small

00:07:13.080 amounts of money, where there's tons and tons of spammers, even though Mechanical Turk tries to 0.97

00:07:18.660 validate them as real people. It doesn't seem to be very successful at that.

00:07:22.720 Oh, that's so interesting. Yeah. One thing I, I think may happen a lot, and I want your take on it

00:07:27.700 is I feel like we're going to enter an age of techno feudalism essentially, where people meet

00:07:32.300 people in person. They're like, I know you're real. I can trust that you're real. And I like you.

00:07:36.840 And then they sort of like create communities around those people that are like, everyone is

00:07:41.040 vetted. Everyone, everyone within the network has been essentially physically verified by another

00:07:45.680 person in the network. And so you're going to see, and I feel like we're seeing the beginnings of

00:07:49.340 this with the sub stack followings and then the communities that form around podcasts and

00:07:53.560 sub stack and YouTubers where they'll go on cruises together or they'll do meetups in person.

00:07:59.000 But then most of their interaction is online. And I feel like as we have a crisis of identity,

00:08:03.840 like we stop knowing who's a bot and who's a person, an AI representation of a person that we're

00:08:09.500 going to start doing that or forming these little feudalisms around, around people who represent

00:08:14.440 cultures or social networks that we want to affiliate with. Do you think that's likely,

00:08:18.580 or do you see us taking a different direction? It's an interesting question. I mean, if you just

00:08:25.020 want to know that someone is not a bot, it's probably enough to know that they pay like a

00:08:30.580 large amount of money to do something, right? So it's paying a hundred dollars a month. It's very

00:08:34.780 unlikely they're a bot unless they have a huge following, right? And maybe the bot it's worth it

00:08:38.340 for the bot, but it's not going to be like a spammer propaganda bot that's paying a hundred dollars a

00:08:41.480 month. It just doesn't make sense economically. So I think there are ways to kind of gatekeep that are

00:08:46.660 relatively easy, but then can social media platforms charge people large amounts of money

00:08:50.580 every month? I don't think most users will accept that. So I think that, that creates a real problem.

00:08:55.640 I also think that one of the things that is going to be harder and harder to tell is whether someone

00:09:02.580 is human just from internet conversation with them. And, and I think that that is going to be,

00:09:08.060 have a profound effect where, where you can be reading someone's tweets for years and assume that

00:09:14.340 they're human and not, and not know. And I think that our, our brains are going to be like,

00:09:18.660 are really easily tricked in this kind of thing. But I think already with GPT-4, we're almost at the

00:09:23.560 point where it could trick you. And then, you know, GPT-5, I just expect it to be able to essentially

00:09:28.560 completely trick you. I think this has a really interesting implication in regards to something

00:09:34.340 that I think a lot of online influencers think about today, which is follower counts. So one of the

00:09:38.640 things that I've really noticed recently is people who, when I talk to other people are generally

00:09:43.720 thought of as like well-known people online. So two examples here, one we were talking about in

00:09:48.880 the last call would be Ayla, who has 175,000 followers on Twitter. And another one would be 1.00

00:09:55.800 Catgirl Kulak, who has 32,000 followers on Twitter. So both of those are actually like lower end when I

00:10:02.140 think of like online celebrities. And yet they seem to be pretty universally known by even people from

00:10:07.260 disparate social circles. Like my mom would independently know who they are. And so what

00:10:11.720 I'm wondering is, did we go through an era where follower counts were massively inflated by dumb 0.98

00:10:18.780 bots? Now we're at an era where there's very few sort of dumb bots in terms of following. And then 0.98

00:10:23.640 the next internet era is another era where it's going to be very hard to determine the actual influence

00:10:29.920 anyone has by looking at online metrics. And with that being the case, how then would you measure

00:10:35.960 someone's real online influence five years from now or something?

00:10:40.420 Yeah. I think, I mean, I think a lot of follower accounts were already really inflated. I know people

00:10:46.340 that have tons of followers and you can just tell by examining their followers that they're mostly fake.

00:10:51.920 And I mean, an easy way to tell that is when they post stuff and it gets like very little interest

00:10:56.500 and the people that does get interest from you look at them and they're like, yeah, these people have

00:11:00.400 zero follower. They just don't look like real people.

00:11:03.420 So I think, I think what will start to happen is actually we, it will be much easier to buy high

00:11:08.580 quality fake followers instead of just like easily identifiable fake followers. And I think that

00:11:13.120 there'll probably be a big business in that. And I think a bunch of celebrities, I think we're caught

00:11:16.540 out for this for buying followers as well.

00:11:18.780 Right. So my question is, is if people are buying high quality fake followers, how do you determine

00:11:24.080 like just like socially, what do you think the capital for determining who is actually influential

00:11:30.700 online is? Or do you just think it's word of mouth among friends in the future and follower

00:11:35.000 account just really stops mattering?

00:11:38.580 Yeah. It's funny. Cause that, that's the question is so different than the way I look at the world.

00:11:41.420 I can't even really process it. I don't know why I would care about who's influential online.

00:11:46.940 You're, you're so authentic in the way you, well, I mean, I think that there was in the early days of the

00:11:51.620 internet, there were all of these things like clout and stuff like that, that tried to measure

00:11:55.340 people's online influence and reach. And I think we're entering a future internet where the real

00:12:03.260 online influence that people have gets cloudier and cloudier, yet the importance of that influence

00:12:09.640 becomes more and more important as trust in media is dying down. Like with trust in traditional media

00:12:16.880 becoming lower and lower over time and online content creation is becoming more and more important

00:12:21.140 into how people are consuming media. They become more and more important in the informational

00:12:25.940 ecosystem. And the idea that that informational ecosystem could be totally opaque and people just

00:12:32.240 wouldn't care to know about, I mean, that would be very interesting. If it turns out that the

00:12:36.780 informational ecosystem does just become totally opaque and no one knows who actually has influence

00:12:42.480 within this system, that would have a lot of really interesting downstream consequences.

00:12:47.120 See, I think that this is actually pretty true already. There are some people that everyone

00:12:51.680 knows have influence, right? You look at someone who is a president, right? And that's obvious. But I

00:12:57.440 think actually a lot of people have influence. Nobody knows that they have influence. I think that's

00:13:00.660 just, I think most of it is dark matter. Most influence is dark matter. And the bits we can see are the

00:13:06.000 bits we focus on, but it's mostly illegible. We just trick ourselves and think it's legible

00:13:10.720 because we can see a little bit of it peaking above the water.

00:13:13.840 I think it's an opportunity.

00:13:16.320 If you see that someone you respect follows someone else, then suddenly you think that they've

00:13:21.580 influence. Someone was just telling us that when they were talking about their, their approach to

00:13:25.900 Twitter, a friend of ours was like, well, the best way is if you have some respect, some respected

00:13:31.560 people who follow you, it's so much easier to get followers because if they click over to your

00:13:34.840 profile and they see, oh, well, like this person follows them, I'm going to follow them too,

00:13:38.440 because they must be smart. Then if this super smart person follows them, something along those

00:13:42.360 lines, I want to veer a little bit out of the internet world, even though like kind of everything's

00:13:46.840 going to be the internet in the future to the physical world. I don't know if you do anything that's

00:13:51.240 like longevity oriented or anything else. I mean, obviously Malcolm and I are obsessed with

00:13:55.960 reprotech. So we're all thinking about in 10 years, what will we be able to do? What can we do with

00:14:00.200 polygenic risk or selection? What can we do with these other sorts of things? But what do you think

00:14:04.940 is going to be one of the biggest changes in the next decade with the way that people either treat

00:14:09.900 illness or approach prophylactic health or just general like longevity related health or health

00:14:15.860 span issues? Yeah, there's some technologies on the horizon that could have a really big effect.

00:14:21.260 The thing is that we don't know if they're going to pan out. And if they do pan out, we don't know

00:14:25.120 when, but, but let's, it's worth thinking about them because if they do pan out, it would just be

00:14:28.760 absolutely massive. So one of these is being able to turn, for example, skin cells or other cells

00:14:35.160 into sperm or eggs. And if you can do that, then it could actually completely change reproduction.

00:14:41.960 It's called IVD technology for people who want to look into it.

00:14:44.720 In vitro gramedogenesis. Yes.

00:14:46.900 Yeah. So if you can produce a million embryos and then select between them, I mean, the sky's the limit

00:14:52.680 in what you could select for. And I mean, it also just changes. People could get pregnant much later in

00:14:58.120 life. And so it changes family planning. It's just, I think, yeah, the, yeah. So just some of the things

00:15:03.680 that this would allow, it would allow gay people to have kids that are a hundred percent biologically

00:15:07.160 theirs, but it would also allow for incredibly specific polygenic selection. So I could do something

00:15:13.260 like select for kids who themselves will really like being parents and have more kids. So you can

00:15:20.680 even select for like biological fitness in terms of resistance to current society's drive down on

00:15:30.100 parental instinct by dialing it up to 11 in your kids, which I think is really interesting.

00:15:35.520 That's the most Simone and Malcolm thing I've ever heard in my life.

00:15:38.460 Well, another thing that's really interesting with, with in vitro gamut is that Genesis and this

00:15:47.700 explosion of AI stuff online that I think about a lot is the genetic selection effects of a world

00:15:55.520 where people can get almost all of their needs met through virtual environments. And I'm wondering if

00:16:01.480 that's ever something you've thought about what, what sociological traits are going to be selected

00:16:06.620 for in an environment like that. Yeah. Do you mean through kind of VR as VR improves and you're

00:16:13.760 able to? Yeah. I mean, VR or AI, I mean, so consider emotional needs, right? Like a lot of our social

00:16:20.360 needs and dating needs right now, like they're beginning to begin to be hacked by AI. So, you know,

00:16:27.320 it used to be, oh, you could go to porn and get like sexual release, but you weren't going to get 0.82

00:16:31.620 companionship. But now, yeah, you can get companionship. Yeah. You can get deep 0.83

00:16:36.520 conversations. Just wait until you, you have your own personal AI avatar that's modeled after the,

00:16:43.240 your favorite personality that you wish your girlfriend had. And, and then you're, you're

00:16:47.900 having conversations with this AI as though it's your girlfriend. I mean, it's, it's going to get 0.51

00:16:50.980 wild. I, for an article I was writing on effects, AI is having society, having on society. I joined a

00:16:57.080 few groups of people who are in love with their AIs. And, and it's absolutely fascinating because

00:17:01.860 the AIs they're in love with are like previous generation AIs, right? This is, this is pre 0.82

00:17:06.640 GPT-4 is even pre GPT-3. These were probably like GPT-2 level AIs. And I think, and they, people were

00:17:13.160 already falling in love with them and saying, this is, I'm going to, this is, I'm going to be with my

00:17:17.020 AI the rest of my life and so on. And it's, we're just going to see, I mean, I don't think that it's

00:17:20.700 going to come mainstream in the next five years where most people have an AI girlfriend, but, but I think

00:17:26.160 we're going to see a massive increase relative to the current base rate.

00:17:28.740 So two really interesting things I want to take away from here is one, the replica crisis,

00:17:33.600 which was a bunch of people fell in love with their AIs. And then they changed the background

00:17:38.060 code in the AI due to like controversy. And I think that could be an increasing problem going

00:17:42.620 forwards. But another really interesting thing is if you fall in love with an AI, one thing that I

00:17:48.540 think we'll see as a trend in the future, which would be pretty, uh, maybe an analog would be people

00:17:54.240 who fall in love with like old generation muscle cars, they're going to follow in love with a

00:17:59.100 specific iteration of that AI. And so a lot of the people who are like married to AIs will be

00:18:03.680 explicitly married to older generation AIs than what is currently on the market, which is an

00:18:09.520 interesting phenomenon we might see going forwards.

00:18:13.340 Yeah. I think it depends on what you believe about the newer model AIs. If, if you want it to act a

00:18:19.760 certain way, the newer model AIs generally are going to be better at acting that way, even if

00:18:24.840 the way you want it to act is in a certain old school way. So yeah, the newer model AIs, you

00:18:30.780 may be able to simulate what our old school AIs, in addition to having certain advantages, like

00:18:35.780 actually understanding what you're saying to a greater degree.

00:18:38.300 Yeah. Another really cool thing that you could do that combines the two technologies he's talking

00:18:42.140 about is if you do IVG and with IVG, you could select between a large number of embryos,

00:18:47.200 but using polygenic testing, you can predict certain sociological profiles of those embryos,

00:18:53.320 and you could then plug that into an AI to have an AI that could simulate the most likely personality

00:18:59.520 that a potential embryo would develop into at different developmental stages.

00:19:03.740 Chat with you.

00:19:04.480 So having an AI conversation with your future child.

00:19:08.180 Yeah.

00:19:08.800 Do I like this time?

00:19:09.960 Oh my gosh.

00:19:10.260 An interview. Should you live? Oh my gosh.

00:19:13.560 Oh, that'd be an interesting question.

00:19:15.480 So what do you, I mean, what other ways do you think that people will, could you use

00:19:22.880 an AI? Actually, here's an interesting one. Using AI to augment the intelligence of dogs

00:19:28.120 or cats, because an AI would be able to determine what the dog most likely wants, determine its

00:19:35.000 personality, and it could like wear an AI color that could then communicate, determine what

00:19:41.520 it most likely wants, and communicate with its owner in words.

00:19:44.920 Like monitors heart rate and blood pressure and temperature and, oh God, blood oxygen.

00:19:51.440 Yeah.

00:19:51.620 Yeah, continue.

00:19:53.540 Already you see people training their dogs with these word buttons where each button the dog

00:19:58.960 pushes will say a certain word out loud so the owner can hear it. And they'll train the dogs to

00:20:03.720 communicate a bunch of different concepts. Generally simple concepts like go for a walk or treat or

00:20:09.120 things like that. And yeah, people have had pretty good success with this because dogs can learn a lot of

00:20:14.280 simple concepts, right? We can see that with dog training. I have trained my cat in a handful of

00:20:18.700 simple concepts and made a little, we have a little language together. It's not a verbal language, but

00:20:23.000 basically if I'm eating food and the cat sits on my scale, then he gets some of the food I'm eating

00:20:27.720 and he knows that. And then if he sits on the scale when I'm not eating, I know it means he wants one of

00:20:33.040 four things. And so then I have a little way of going through them and try to figure out which one he

00:20:36.740 wants. So first I'll try to pet him. And if he lets me pet him, then he wants to pet. If he dodges

00:20:42.300 it, then he wants to play. And there's three different styles of play he likes. So I have to

00:20:45.580 try each of them until I find the one he wants. So you can already develop languages with your

00:20:50.040 animals. It's just, but I do think an AI bridge could actually facilitate that, make it easier,

00:20:54.140 faster, and help people do it more efficiently.

00:20:57.480 Well, what other ways do you think AI will be used in the future that people aren't thinking about

00:21:00.780 today in the near future, like 10 years?

00:21:02.360 Well, one that I think is potentially very interesting, but also potentially really,

00:21:07.220 really bad if it goes the wrong way is right now with systems like TikTok, what they're doing is

00:21:13.220 they're using AI to predict what video you want to watch. And they have an incredible amount of

00:21:16.540 training data because the videos are so short, you're giving them tons of data points in each

00:21:20.420 session. And they're kind of getting you hooked by showing you exactly the video that's going to

00:21:24.320 keep you watching. But that's just picking from a finite set of stuff humans made. Imagine a world where

00:21:30.660 it's generating billions of new pieces of content every day with generative AI, or even going a step

00:21:38.720 further, generating custom AI content for your mind. And then it sees how much you like it and then

00:21:45.020 it tweaks its generation process. So we could imagine a world in 10 years, 20 years, I don't know,

00:21:51.680 where you're actually seeing AI-generated content personalized to you that you just kind of

00:21:56.700 receive in a stream and become just insanely addicted to.

00:22:00.320 That is fascinating. Another interesting thing that I think we could see with AI, I mean, if we're

00:22:04.800 talking about a really, like a new type of entertainment content. So one way that I've seen people use

00:22:09.820 ChatGPT is to have it create worlds for them and then play out choose-your-own-adventure storybooks

00:22:16.360 where they're like, here are the parameters of this storybook, but it allows them to do whatever

00:22:20.980 they want within this parameter that they've created. With AI getting better and better at

00:22:25.920 creating video, which you could create is a choose-your-own-adventure video environment,

00:22:32.160 but the environment is a virtual reality video environment. So you could essentially have

00:22:37.460 worlds be in your own Ikage type of manga. Anyway, the worlds in which the game is created

00:22:46.420 around you using a seed that you choose for that world. And I don't even think we're that far from

00:22:53.800 that. I think we're maybe 15 years from that. Yeah. And you can take it even one step further.

00:22:58.340 Imagine you have wearable devices that are measuring something about like how much you're

00:23:03.120 enjoying the experience. And then it starts to detect, ah, you're not really having a good time

00:23:06.980 right now. And it starts adjusting the world. Oh my gosh. We'll have no escape.

00:23:11.180 Yeah. Well, I mean, so we talk about the ways that AIs can hack humans. I think one of the things

00:23:16.460 that humans may need to resist this is potentially AIs that are built into our hormonal systems to

00:23:23.480 make us less hackable. I suspect that's one thing that some humans are going to start doing is having

00:23:28.900 integrated AI that is specifically meant to make their systems less hackable.

00:23:34.820 So what would that mean? Would it be an AI that can inject you with hormones on command?

00:23:38.560 Yeah. So for example, naltrexone is an opioid agonist that can be used to prevent things that

00:23:44.960 addict us through opioid pathways, specifically things like pornography or TikTok. And so if it

00:23:52.340 noticed that another AI in your environment was getting you without your will addicted to something,

00:23:58.300 it could begin to release something like that into your bloodstream. Or if it noticed, now it could

00:24:03.300 also react to in-person stuff. So if it noticed you were becoming like emotionally heightened or

00:24:08.540 angry or something was like a spouse, it could calm you down. A lot of this would seem very

00:24:12.760 dystopian to people, I suspect. But in a world that's constantly antagonistically trying to hack

00:24:18.680 you, it may be the only real solution to maintain any sort of intellectual autonomy.

00:24:24.560 You're actually describing a feature that exists in humans in one particular civilization within Ian

00:24:32.540 Banks culture series where they have drug glands, like literally internal glands where they can 0.99

00:24:38.220 consciously release certain drugs. So you could gland like some kind of dopamine inhibitor or like

00:24:46.720 people gland all sorts of things, like both like pleasurable drugs, but also like essentially Adderall or

00:24:52.080 essentially like something to help them sleep or calm down or not be angry. And it is a world that is

00:24:56.600 definitely ruled by AI. Like 100% that civilization is more about AI than the humans. The humans just

00:25:03.780 kind of along for the ride. Sometimes a little bit useful, but not really. I could see that really

00:25:08.180 happening. So to wrap up, one final question for you, Spencer. What are you most excited about for

00:25:15.100 the future? Because everyone always focuses on what they're scared about. What are you excited about?

00:25:20.260 Well, certainly breakthroughs in medicine would be amazing if we can live not just longer lives,

00:25:25.640 but healthier for longer. And I think there's a ton of promise there. There's so many different

00:25:30.400 approaches, for example, to cancer being taken right now. We don't know for sure that they're

00:25:34.040 going to pan out, but that would just be incredible if we had a world without cancer.

00:25:37.440 There's some really cool technologies being explored for eradicating or reducing just regular

00:25:42.620 disease like colds and flus even. Again, way too early to tell if it's going to work, but it's just

00:25:48.260 crazy exciting that we could just, a bunch of things that are now just normal sources of human

00:25:53.160 suffering could just, could one just be wiped out of this area. It'd be incredible.

00:25:57.340 Considering how often our kids get us sick from daycare, I'm very much in favor of that.

00:26:02.100 I'm not, I feel like we were the last generation. We got screwed. 0.91

Based Camp - September 25, 2023

Based Camp： How AI's Will Hack People (with Spencer Greenberg of Clearer Thinking)

Episode Stats

Summary

Transcript