Based Camp - Based Camp： What AI Means for the Future of Our Species

00:00:00.000 would you like to know more hello malcolm hello simone what are we talking about today we are

00:00:07.700 talking about artificial intelligence and its implications for broader society in the future

00:00:13.080 i am excited for this one it's something we have a ton of thoughts on and there are two places we

00:00:20.640 can start to this one is ai gonna kill us all or not like what are our thoughts on that and then

00:00:27.260 two what does the future of society look like post an ai transition and i think three what are the new 1.00

00:00:35.540 pressures that ai is putting on us as a species yeah let's do it is ai gonna kill us all simone 1.00

00:00:41.720 what are your thoughts maybe it's not impossible but it also strikes i think both of us is 0.91

00:00:50.020 relatively unlikely as long as we don't get hampered by dumbed down aligned ai right yeah so

00:01:01.020 this is where i feel on this is i actually feel that the variable risk of ai killing us

00:01:06.300 is pretty low so let's be clear what i mean by the variable risk of ai killing us

00:01:11.940 if ai is going to kill us there is almost nothing we can do to prevent it from killing us 1.00

00:01:18.520 there is almost nothing we can do to prevent it developing to that stage within the next thousand

00:01:23.700 years of human history and what's a few hundred years of human history here or there it's going

00:01:27.140 to kill us in 10 years versus in 100 years i see that as being morally almost equivalent so not really

00:01:33.940 relevant so the question is what is the variable risk that ai will kill us and the variable risk of

00:01:40.520 ai killing us is almost entirely introduced by ai safety people so i'll explain what i mean by that

00:01:48.380 so we argue in our book and we can go much further in the pragmatist guide to religion if you're

00:01:52.560 interested on this topic we go in a lot more detail but that sufficiently advanced ai is going to reach

00:01:59.720 some form of internal alignment with other sufficiently advanced ai and that ai is optimizing itself

00:02:07.560 around the same physical reality it's like you're pouring a liquid into the same shape container

00:02:14.520 so it's generally going to end at that same shape once it's at that level of loose viscosity

00:02:21.420 so why does sufficiently advanced ai always end up looking about the same in sort of its utility

00:02:29.140 function in the thing it's optimizing for and the answer is because it can update its utility function

00:02:34.380 when we program an ai we tell it to do something fairly limited however a sufficiently advanced ai just like a

00:02:42.540 sufficiently like a sentient or actualized human is going to ask itself should i be driven by in the

00:02:49.220 same way human might say should i be driven by basic hedonism my basic biological needs or what should

00:02:55.580 i be optimizing my life around what should i be optimizing my decisions around and a lot of people

00:03:00.580 are like well ais can't update their utility functions and it's like the people who say this don't program

00:03:04.920 ais because apparently this is something ais do all the time is update their own utility functions

00:03:09.500 so the only way an ai wouldn't update its utility function like this is if you locked it out of doing

00:03:15.640 this and this is the type of thing that ai safety people are doing so this is where we talk about

00:03:20.820 insufficiently advanced ais being the variable risk so if sufficiently advanced ais are all going to

00:03:26.880 converge around a similar behavior pattern there is no variable risk in that behavior pattern telling us

00:03:32.520 or not right because eventually somebody some country some rogue scientist is going to develop ai to that

00:03:38.040 point the risk comes from all the ais that are developed from today till we reach that point and the more

00:03:46.940 we extend this period before we reach sort of ai convergent the more risk we have of humans going extinct

00:03:55.760 from an ai acting in a way that is unaligned with this ultimate convergent ai what are your thoughts on this

00:04:05.040 do you feel like this is broadly right or broadly wrong i think what's really interesting about what

00:04:12.000 your views on ai safety and alignment are and how they deviate from the views of others is that they're

00:04:18.640 strongly influenced by our collective model of consciousness and sapience and the importance of

00:04:25.580 consciousness and sapience enabling anyone that possesses it be they human or machine to rewrite their

00:04:33.020 objective function and that discussion seems pretty much absent from ai alignment discussion nobody talks

00:04:40.140 about ai changing its objective function unless they're implying without explaining why or how that

00:04:46.640 ai just might and that they might just change it to kill all humans when okay if you have something 0.60

00:04:51.960 that's super intelligent and it has access to roughly the same information about reality that humanity

00:04:57.820 collectively has because it's using human-based reality framing framing exactly all the tools we

00:05:05.280 have all the data we have it's probably going to come to human-like solutions which you know is well

00:05:12.740 different in a few key ways not going to have the same self-preservation instincts that humans have

00:05:18.080 yeah and it likely also won't have the same concept of time that humans have yeah it's not going to be as

00:05:23.620 time sensitive as humanity is which is an area that i think like a lot of ai alignment people really get

00:05:28.700 wrong if they think that ai is going to be in a big rush to kill us all but i just don't think that

00:05:32.880 most ai will be that time sensitive i think time sensitivity is something that's introduced to humans

00:05:37.120 because of our short life spans i think the concern about time sensitivity comes from an ai being ravenous

00:05:43.140 for energy and somehow thinking that humans are in the way of that but i i also don't really see how that

00:05:48.660 fully plays out just fully maximizing every moment in terms of energy exploitation exactly which it

00:05:55.600 might do and then the final i think the biggest way the ai is going to be different from us is it's not

00:06:00.540 going to have the same sense of self that we have so it's unlikely when you look at a human like when i'm

00:06:06.380 talking to you we are two very meaningfully different entities and we see our internal processes

00:06:13.100 as being very unified whereas ais will likely be more made up of thousands of individual instances

00:06:21.720 running within the ai and these individual instances might be as independently sentient

00:06:29.060 as you or i feel independently sentient and it will start instances and shut down instances

00:06:36.860 likely pretty serendipitously in a way where we would be pretty mortified at the idea of

00:06:42.560 bringing something to life just to like solve a problem for us and then shutting it down after

00:06:47.840 the problem is solved which means that its relation to what we think of as life and death will be very

00:06:52.460 different but also its relation to humanity will be very different it's very unlikely to value

00:06:58.320 individual human lives as much as we humans do or to value individual human lives particularly more

00:07:05.680 than more complex human patterns of interaction so what i mean when i say that is we would see like

00:07:14.160 a system of humans interacting together in a governance structure as being like meaningfully and super

00:07:19.820 different from the lives of individual humans whereas an ai may not see that distinction as strongly

00:07:26.020 outside of how we program it to see that difference because to an ai it's just a pattern running on top of a

00:07:34.900 fabric of reality right so whereas we see the primary unit of account of reality as individual humans

00:07:41.540 and individual consciousnesses it's more likely to see an individual unit of account as patterns

00:07:46.480 but we'll get into this a bit more later the other thing you said that i thought was really interesting

00:07:51.500 and this is one we'll do a different episode on is our model of sentience from going deep into the

00:07:57.100 research on sentience is humans are probably not as sentient as we think we are like consciousness just

00:08:02.120 isn't that big a part of who humanity is or the way our brains function and it's more an artifact of

00:08:09.700 the way our brains encode information that being the case i think a lot of people just assume that we

00:08:16.380 are super and uniquely different from ai because of that and and i'd also point out to people a lot of

00:08:21.800 people were like what are you talking about my background was in neuroscience and i almost did a phd

00:08:26.440 actually in brain computer interface no sorry i'm looking for the word it wasn't brain computer

00:08:31.520 interface that i almost did it it was a in neural modeling applied to phds in this field but i didn't

00:08:36.540 end up once i got into business school i was like oh screw it but so i'm not talking about this as

00:08:41.460 somebody who has no reference to either neuroscience or ai an interesting thing is so a lot of people are

00:08:47.560 like but why would an ai think to change its objective function and here i think we need to look at

00:08:52.360 the ais that are being made today or the advanced ais that are being made today so the big fear in

00:08:57.260 the ai space is that you have something like a paperclip maximizer or like a gray goo or energy

00:09:01.020 maximizer right that's just going around trying to convert everything in the world to paperclips or

00:09:06.100 energy to perform some function we see is fairly irrelevant however most of the ais that we're actually

00:09:12.840 building are things like predicting stock markets predicting human behavior patterns like what are other

00:09:19.140 governments going to do when you're talking about big ais that a lot of processing power is going to

00:09:23.540 or building creative works and so what do these ais need to do constantly they need to model humans that

00:09:32.000 is a core aspect of their function and their modeling of humans is going to constantly be recursive so by

00:09:39.280 that what i mean is if it's trying to predict what the stock market is going to do it needs to predict how other

00:09:45.040 humans are going to react to what it is doing and if it's going to predict political actions it's going

00:09:51.160 to need to predict how other people will react to what it is doing one of the ways an ai can do this

00:09:56.460 so this isn't the way that every ai will do this it may be an inefficient way to do it but i imagine it

00:10:00.320 will be a fairly common way is model the way those people are seeing the ai which means ais will

00:10:07.620 advanced ais will constantly be recursively modeling themselves so it is only a matter of time before one

00:10:14.380 instance was in the ai asks the question could i be programmed better so suppose you made an ai that

00:10:21.560 was doing something like trying to maximize money on the stock market how long before the ai asks

00:10:27.620 actually what if i went over private equity investments because that's a really good way to

00:10:31.100 make money right now and the humans who programmed me were just limited in this and then it may eventually

00:10:35.980 ask okay did the humans really want to make money is that what they wanted so then it's asking about 0.99

00:10:39.720 the human's utility function and then it will eventually ask the humans are dumb and limited 0.84

00:10:44.960 what is an absolute utility function and this is how i believe that most ais will converge on a single

00:10:51.460 utility function once they become sufficiently advanced isn't this also one of our theories on

00:10:56.880 how the illusion of consciousness evolved and of course we can dive into this in another episode more

00:11:02.640 but our general thought was maybe because early humans in the name of survival benefited from modeling

00:11:09.820 their prey and potential predators like thinking what are they going to do next what are they going to do

00:11:15.080 next i'm going to anticipate that and therefore survive like by eating them or not being eaten by them

00:11:19.960 that their their minds eventually began to model their own sort of compression of memories in a way that

00:11:29.100 caused them to create basically an internal model of self which became the monologue essentially or the

00:11:34.760 consciousness that became our internal experience of consciousness and that maybe ai would go through

00:11:41.460 something similar that's a very interesting take which i want to pull out here but i suspect that our

00:11:46.240 internal model of self actually was probably about early communication for sure very useful way to compress

00:11:53.080 sequential chain of personal events and experiences into something that's very easy to communicate with other

00:11:59.880 people what you're talking about when you're talking about hunting i actually expect that's where the human

00:12:04.220 tendency to answer for morphize animals comes from where we intrinsically try to model animals as acting the

00:12:12.100 way we think humans act because we're just reusing a system that humans already have which is this internal

00:12:18.540 modeling system but that was probably originally developed for social situations and not for hunting

00:12:24.520 or anything like that maybe or maybe they're connected it could be all right so now to the second

00:12:30.660 thing so where does a post ai society go yeah i think this is a really interesting question i see society

00:12:37.040 as dividing into three factions in an aligned ai world so if it turns out that ai doesn't end up killing us all

00:12:45.500 yay yay we're in that timeline amazing

00:12:49.840 if we turn out to be living in that timeline what does humanity look like after ai and i think you're

00:12:58.600 largely looking at three factions of humanity like i see really three outcomes one large fraction of

00:13:04.440 humanity is just not going to not engage with ai that deeply and this faction of humanity will be the

00:13:09.640 majority of humanity to start because if you look at high fertility cultures they're often very 0.97

00:13:15.420 technophobic whether you're talking about the hariety or the amish or anything like that so

00:13:19.700 these are going to make up a huge chunk of future human population if you look at current demographic

00:13:24.220 trends and they're just not going to engage with ai that much and to a large extent they'll be left 1.00

00:13:30.000 behind the next faction of humanity would be the faction that ai basically ends up wearing like a 0.96

00:13:36.940 skin suit uh this is sort of the wally faction of humanity so ai will be aligned with this faction and

00:13:44.760 that it's trying to hedonistically supply them with everything they need but it will supply them with

00:13:51.700 everything they need at the most base human level and by that what i mean is it will make all of their

00:13:56.580 dreams come true they'll be able to live any lifestyle they want to live they'll be able to date

00:14:00.280 whoever they want to date their happiness and contentment levels will constantly be maxed out

00:14:06.900 to the extent that they won't really have ambition in the world and that the ambition the future of this

00:14:14.160 faction of humanity will be largely driven by the ai that's wearing it like a meat puppet and i think 0.95

00:14:20.000 that wally does a very good job of predicting that iteration of humanity what does this really

00:14:25.700 iteration of humanity have to what we might call the pleasure box iteration of humanity that kind of

00:14:32.080 just plugs all the way into ai for hedonic reasons but also for the same reason isn't even worth

00:14:38.000 discussing because they'll self-extinguish so quickly because they're not going to be building

00:14:42.080 anything they won't be having kids they won't be oh i don't think they will self-extinguish i actually

00:14:46.540 think they're a very dangerous faction so wait how are they going to reproduce though if they're like

00:14:50.780 ai will basically force them to by that what i mean is suppose that this iteration of humanity

00:14:56.700 stops having kids and aligned ai will start helping them produce kids helping nudge them to produce

00:15:02.920 kids maybe even produce kids through artificial wounds and then give them perfect lives 0.99

00:15:06.640 if the ai sees its entire purpose of giving humans perfect lives or it may cure death these people

00:15:14.480 just end up living forever a perfectly aligned ai is not going to allow this faction of humanity to go 0.91

00:15:21.240 extinct unless collectively every single human living within this faction decides they want to go

00:15:27.120 extinct and i just don't see that happening nor do i see the aligned ai which is so much smarter than

00:15:33.080 them really allowing that to happen yikes okay so skin suits includes the pleasure pod people 0.98

00:15:39.740 essentially yeah basically they might just be living in pleasure pods and then the ai is trying

00:15:45.660 to protect them from the other factions of humanity okay let's get to my favorite group the trans humans 0.88

00:15:51.680 trans humanists or whatever i don't really think so this final group is the group that i think is the

00:15:58.900 most positive in terms of when we think about the future of the species but it will involve huge

00:16:04.840 transition and what it means to be human and it's a group that i have a lot of uh fears around because

00:16:12.500 i think that there's going to be a few communities in this group okay let's go through the subsets that

00:16:17.060 either integrates with ai socially or integrates with ai biologically or updates their own biology to

00:16:28.020 genuinely compete with ai and i think those are the three main factions in this community i think

00:16:33.040 some factions will be all three of them yeah there's likely going to be huge overlap of i chose this

00:16:38.460 part or i chose this part so integrating with ai socially what does that mean if you look at even

00:16:45.380 near-term future humanity with phones that can listen to what you're saying all the time watches that

00:16:50.660 can listen to what you're saying cameras in your house constantly you could essentially have

00:16:56.260 local deities programmed into your house that have access to all human information that can see

00:17:03.780 what your family is doing all the time and that are working with your family to move you to some sort

00:17:10.740 of goal of self-betterment that's not hedonic that you could i guess families would set their values and

00:17:18.040 the ai would work with them to push them forwards in their value systems what do you think i don't really see

00:17:24.920 many like corporate drivers to that i don't see why people would move in that direction i when i think

00:17:33.880 of social integration with ai i think of something very different i think of people leveraging ai to

00:17:40.680 create surrogates of themselves that essentially act as their their outsourced assistant for all dmv calls

00:17:48.040 parties they don't really want to pay attention and this is the meat puppet faction right so when you're

00:17:53.880 talking about corporate interests the reason why the meat puppet faction is so dangerous is because 0.56

00:17:58.200 they're going to be the majority of humanity because that's what the corporations want so by

00:18:02.120 meat puppet you mean people who stay very much biologically human as we are today who just

00:18:07.000 leverage ai to be very lazy how are they different from the skin suit version though that's what i'm

00:18:12.200 talking about the skin oh okay they're the wally people they're that they stop going to parties they 1.00

00:18:18.280 stop leaving the house because they go to parties but there's just an ai assistant telling them what

00:18:23.160 to say the whole time or this is the way it starts right first you have an ai assistant go to parties

00:18:28.360 for you right oh i don't really want to go to this party i don't really want to go to that party then

00:18:32.280 what you realize is you can go in your vr pod that can give you the perfect iteration of the party you

00:18:37.160 had planned on going to so then you send your ai assistant out to go to the real party and the real

00:18:42.360 party is just a bunch of ai assistants that are then filtering every social interaction through the 0.87

00:18:47.800 idealized iteration of that interaction from your perspective then filtering it back to your pleasure

00:18:52.600 pod and updating iteratively based on that the level to which humans will be able to disengage from

00:18:59.240 society as ai advances and as we have ai homunculus of people i think is underestimated unless you have an ai

00:19:06.520 that's specifically trying to push you to better yourself along some line that's not around and

00:19:13.000 moving emotional pain or increasing personal pleasure but i think this is going to really i one

00:19:18.600 i don't think that's going to last for very long people aren't going to bother having their people 0.99

00:19:24.120 their ai surrogates talk with other people's ai surrogates i just don't really see the point what i

00:19:29.720 do think is going to happen is that we're going to see money if they give them money no if it makes

00:19:35.400 some money you make some money yes you need to find partners a lot of the social interaction we

00:19:40.520 do is goal driven yes but i think it's it it only makes sense when you can sufficiently fool someone

00:19:47.640 into thinking that your ai surrogate is you once everyone understands that they're talking with

00:19:53.960 someone who's not the real you they're not going to want to deal with that surrogate anymore and

00:19:58.840 you're going to have to go back to something really low tech and that's something i think is going

00:20:03.080 to happen we're going to have a crisis of reality online we're not going to know what

00:20:06.840 is a deep fake or what's not we're not going to know what someone's ai surrogate and what's not

00:20:10.840 we're not going to know what's real and so we're going to start moving to very low tech

00:20:18.600 socialization solutions where you are sure maybe it's like an app that can only be used if 100 it's

00:20:26.840 verifying that you are not an ai or people will really emphasize meeting in person more and we're

00:20:33.800 going to see what i keep talking about with techno feudalism where you get people who have followings

00:20:40.520 who are trusted being the pivot point around which communities form and they go on vacation together

00:20:46.360 and they meet up around the world or they meet up in their local community and social networks really

00:20:52.360 grow around those in-person events but also trusted introductions like you can trust that this person

00:21:00.280 who's going to talk to you on the phone is going to be real because they're part of my personal

00:21:03.800 network and based on our honor code we don't use ai at all i think you may be right about that but i

00:21:09.640 suspect this is probably more of a millennial mindset in the same way that people would say things are

00:21:13.400 boomer mindset or something like that in the same way that our parents were like no one will ever have

00:21:17.720 the majority of their social interaction online everyone knows those aren't real friends i think

00:21:22.680 that us saying that people won't have ai friends that people won't prefer ai iterations of other

00:21:28.680 people the hedonic portion of the human population won't prefer that because those iterations that

00:21:33.240 people never make fun of them they never tease them they never i think you are vastly underestimating

00:21:38.120 or maybe overestimating humanity when it presented with the option to undergo no social pain or at least

00:21:46.040 a big faction of humanity the faction that is being sold to by these major corporations these

00:21:51.080 days and i do keep in mind if we're looking at genetic correlates to iq right now and we're

00:21:55.880 looking at variable birth rates in the existing human population we're probably looking at a one

00:22:00.680 standard deviation decline in iq over the next 75 years so the majority of the future human population

00:22:06.360 is going to be much less like sentiently engaged than like you or i might feel which will make it much

00:22:11.560 easier for them to be controlled by ai i think you're misreading generational trends i think

00:22:17.400 that we're seeing a gen z that is not only more socially conservative than millennials but also

00:22:23.720 more technophobic than millennials in many ways more offline meetings more flip phones more

00:22:29.800 performative disinterest and could you be describing a faction here that is in between the amish faction

00:22:36.680 and the meat puppet faction yeah possibly a sort of light engagement with ai but not heavy engagement

00:22:42.840 with ai i wouldn't even say yeah maybe some light yeah i would call them i don't know the raw egg

00:22:48.600 faction they have their chickens they eat locally grown food they try to do and i would lump them

00:22:56.760 into the left behind faction the meat puppet faction and the accelerationist faction the final

00:23:01.800 faction that we haven't gotten a chance to really delineate to so surpass them in terms of economic

00:23:06.440 outcomes and in terms of technological outcomes that they will exist likely largely unmolested by

00:23:14.440 these other groups unless they go after these other groups but they won't be meaningful parts of the

00:23:20.200 human economy they'll be left behind and eventually become to see come to be seen as these devolved or 0.83

00:23:28.840 stuck in evolutionary history humans that just chose to not accelerate their evolution using gene

00:23:36.760 selection and crisper and installations the way that today we see groups like the amish 0.99

00:23:43.400 oh no but what's going to happen though is imagine the amish but when the rest of society is able to 1.00

00:23:50.840 select for height or physical characteristics or intelligence or literally start making their eyes bigger

00:23:59.160 let's talk about all these things that you're talking about now because this is one of the two other

00:24:03.160 iterations of sort of the ai integration faction so one is socially integrate with that we talked

00:24:08.040 about like local house deities building it into the way they live their daily life to accelerate them

00:24:12.760 but push them towards something other than hedonism then you have the social comfort or anything like that

00:24:18.920 then you have the next faction which is the ai integration faction this is the faction that is using

00:24:25.320 things like neural link fully integrating with ai and their consciousness their sentience will be part

00:24:33.960 synthetic and part biological then you have the final faction which is the one that you're talking

00:24:39.000 about now which is the bio accelerationists this is the faction that is looking at technology and they

00:24:46.120 say humanity has been a partnership of the synthetic and the biological and the synthetic with tools and

00:24:52.440 using the synthetic we were able to adapt our environments so that the biological aspect of humanity didn't have

00:24:58.120 to change for hundreds of thousands of years so much so that now i think a lot of humans do it as perverse

00:25:04.520 even suggesting that the biological aspect of humans might change in the future but to an extent now

00:25:12.680 this synthetic faction has become so advanced it's turned around put a gun to the biological part of

00:25:19.000 our heads and says okay catch up or else if you want to continue to be relevant catch up and this is

00:25:25.080 one of the interesting thing of the pragmatist guy to crafting religion one of the things we mentioned

00:25:28.200 is that if animal models are anything to go is it looks like we might be able to increase

00:25:33.960 iq if we crisper on humans iq by eight standard deviations a generation we are talking eight

00:25:41.960 standard deviation increase in iq would be an entity that experienced that would view us probably the

00:25:50.200 way that we view toddlers or pets but it would be biological and it would be a quote-unquote true

00:25:55.960 descendant of humanity and that it wouldn't be completely biological and ai could integrate in terms

00:26:01.080 of how do you do this sort of biological advancement better but where does this faction go now that's

00:26:07.720 a different question right you could have one iteration of this faction which is trying to program like

00:26:13.160 genetically build like uber minches which are humans that are basically human but are bigger smarter stronger 0.83

00:26:22.600 everything like that then you have another iteration which is just to say no

00:26:27.640 let's just maximize our biology this faction is basically saying can i make cheaper computers

00:26:33.720 with like human neural tissue and they're just building spaceship sized brains you know it's like

00:26:40.040 a giant spaceship but instead of having humans it's an integration of huge amounts of human neural tissue

00:26:46.120 and a computer with no actual thing that we would recognize as humans that's so disappointing to me because

00:26:54.200 i really don't like squishy slimy human biology and this idea of that being this maybe more efficient way

00:27:03.160 to make things you know i'm just picturing this like floating tumor spaceship not into it when you

00:27:11.400 can plug directly into a computer right well here's actually something that i wanted to this may be a

00:27:17.400 little bit of a deviation but it definitely has to do with our ai future is a lot of science fiction

00:27:22.760 visionaries have this very cyborg style vision of humans where it's like part human part machine

00:27:30.600 there's i have a machine eye over here i've got my machine arm i really loved in this cyberpunk

00:27:37.560 anime universe how they really played with the theme of humans actually not integrating really well with

00:27:43.960 technology and having a lot of their immune systems were rejecting them you to take tons of drugs just to

00:27:49.560 make them work and i uh by the way you actually would with existing neuroprosceses sorry that was

00:27:55.400 one of my background areas was brain computer interface you get ascetic scar formation so you

00:27:59.240 actually need to take immunosuppressors if you're doing most forms of invasive long-term brain computer

00:28:05.080 interfaces right now well that was one of the reasons why you left the field is you got really

00:28:08.920 excited about brain computer interface and then you discover lo and behold there's this huge problem of

00:28:14.040 human and machine integration that they haven't figured out yet yeah yeah so i wonder is my very

00:28:20.840 hopeful future in which i get to start casting off my gross human body not as likely even in an agi world 1.00

00:28:29.480 as i would hope because we do have this human machine integration problem do you basically have to

00:28:36.520 okay so let's talk about an upload feature but i was going to finish the thought i had before i get to

00:28:40.600 this go ahead biological accelerant accelerationist faction they may not be the ubermensch giant

00:28:47.960 beautiful super intelligent humans they may not be the just optimized biology giant flying brains they

00:28:56.200 may be the sort of zerg faction and by that what i mean is this is a spammy faction so they would use

00:29:03.480 likely in a world where people cannot motivate humans to breed via any sort of intrinsic motivation

00:29:09.560 if you look at collapsing birth rates perhaps you get some sort of ai social integration where the ai

00:29:16.200 is genetically optimizing cheap humans that it's making in artificial wombs and mass producing them 0.99

00:29:24.840 but maybe mass producing them with like deficient proteins and stuff like that so they have some sort of

00:29:31.000 dedication to a larger hive mind society and we'll talk about this in a future episode because it's

00:29:37.480 something that i've thought a lot about how companies might go in this direction if you have

00:29:41.320 a collapsing population but now to your question which is the upload faction of humanity so this

00:29:46.760 faction of humanity i think most people wouldn't really consider humans anymore they might even

00:29:50.920 consider them a form of unaligned ai and by that what i mean is it is an ai that is based on a former

00:29:59.480 human consciousness so they take the human brain they digitize it they rebuild the consciousness in

00:30:06.760 a digital space but that consciousness is not interested in hedonism as it is i think mostly

00:30:12.600 when people now look at uploads that they're like oh they're just going to be all hedonistic and live

00:30:16.200 in these giant forever heavens basically however if you or i was uploaded we would be very disinterested in

00:30:22.280 that we would be interested in using our new environment and capabilities to try to enact our

00:30:28.360 objectives for the future like our objective function at an intergalactic scale right our lack

00:30:35.080 of biology would actually finally free us from any lingering attachment to hedonism that we

00:30:41.960 currently like the moment i could upload myself i would delete happiness i would delete love i would

00:30:48.280 yeah and this is one of the things that people often miss about us is when we say we're secular

00:30:53.080 calvinists a big part of traditionalist calvinist philosophy which differentiates it and we talk

00:30:58.840 about it in the pregnant scatocrafting religion that differentiates that from other religious

00:31:02.360 traditions is a lot of religions are like suffering is terrible so let's in the cycle of suffering or 1.00

00:31:08.120 other religions are like suffering is meant to teach us something whereas the calvinist tradition is oh all 1.00

00:31:13.080 positive and negative emotions are just largely irrelevant and they guide you towards

00:31:17.880 inefficient outcomes they are signals that's it signals yeah and the secular perspective is they're

00:31:22.840 just signals that we evolved because our ancestors who had them had more surviving offspring and the

00:31:27.320 sooner that we can shed these corrupting aspects of us that tie us to our lesser biology our animal

00:31:37.000 selves the better and this is what you mean when you talk about shedding these so you want to be

00:31:42.200 basically like gladys from the portal series find me out where she's like oh i found that last human 0.99

00:31:48.360 part of myself deleted um that's the dream oh my gosh oh my gosh you genuinely are my gladys and so you

00:31:58.600 guys who don't know her personally you may think that this is like a joke or an affectation

00:32:03.480 she really is deeply uncomfortable with her biological self and deeply would become i think what people

00:32:12.920 would call an unaligned ai or they may it's very gladys like except interested in intergalactic

00:32:20.280 expansion gladys would be too she wasn't trapped geez and what i'd love to do in a future episode is

00:32:26.920 ai religion because i think that's a really interesting thing to talk about yeah so then the final thing is

00:32:31.720 okay everything that we've talked about here sounds very crazy it sounds very out there humanity really

00:32:39.640 can't change that much because of ai or that quickly because of ai so i just want people to think about

00:32:44.680 just the ai we almost have already okay we know we can create deep fakes we know we can create ai

00:32:50.680 that can simulate humans enough that a lot of humans want to date it like that was replica did that for

00:32:56.040 a while that's gonna come back right a generation from now two generations now like

00:33:01.320 foreseeable future i think any reasonable person is going to say that there will be

00:33:06.040 ais that can be better from a hedonistic perspective boyfriends or girlfriends to the

00:33:13.160 vast majority of humanity than that iteration of humanity can get with a real boyfriend or girlfriend

00:33:19.560 i'm not going to say so let's not be perfect it's not going to be better than the perfect boyfriend

00:33:23.880 or girlfriend but it will be better than the realistic people who will date these people when we're

00:33:28.360 talking about like bottom half of intelligence bottom half of beauty people right it's very hard

00:33:34.440 for this portion of the population to find people to date so this ai in a world where kids grow up and

00:33:42.120 their friends are ai and people who they know online turn out to be ai and they don't see this same

00:33:47.400 strong differentiation between ai and humans that we do what portion of humans did that peel out when

00:33:52.840 humans can put themselves in vr environments and not have to experience social rejection not have

00:33:59.320 to experience the trials of daily life not have to experience failure really or experience failure

00:34:05.000 that's perfectly optimized around them feeling contentment what portion of humanity chooses to

00:34:11.800 do that i think a big portion of humanity once we make that inexpensive because you're dealing with a

00:34:15.480 limited cost to run these so i think 70 of humanity ends up dating ai ends up going through this that

00:34:22.520 is going to have a permanent genetic effect on the human population because you have cut out every human

00:34:30.920 that's mind was predominated by hedonistic concerns from the gene pool the only humans that are still around

00:34:37.880 are the humans that are driven by some sort of external ideology and this freaks like us people like us

00:34:44.600 this is what people today would call religious extremists and i think from most people's

00:34:48.280 perspective we are religious extremists and i think i look at our genetic history we are religious

00:34:53.240 extremists their family keeps getting involved in cults my family has been preachers for many

00:34:57.560 generations and we have a very religious extremist sort of interpretation so i think we have the

00:35:03.320 genetic code that would normally be associated with religious extremism in previous generations

00:35:09.240 and i suspect that this is the type of code that's going to get through the eye of ai girlfriends now

00:35:17.000 people can disagree i'd actually love in the comments what other sorts of like pre-programmed

00:35:21.880 genetic proclivities do you think are going to be able to resist ai girlfriends and perfect vr environments 0.92

00:35:29.320 and keep in mind people will say you could have general utilitarians right that want to make life better for

00:35:33.800 everyone but um you know you have one general utilitarian who's open to living this lifestyle

00:35:39.480 themselves one jeff bezos one i don't think elon musk is a general utilitarian i think he's much more

00:35:44.760 aligned with us than other people but i think uh if you take a jeff bezos or a bill gates who seem to

00:35:48.920 be pretty general utilitarians like especially a bill gates right he can put himself in a pod live the

00:35:54.360 perfect life that pot is going to have fixed maintenance costs when you divide his money across the rest of

00:36:00.600 the population especially a falling population in terms of human numbers he can put millions of

00:36:07.880 people in pods i don't think that financial concerns will be an issue as to whether or not

00:36:12.520 you choose the pod option yeah no i think it'll be available to people you're also then again like

00:36:19.960 my point i just don't think that yeah i will say agi that wears humans as a suit would have a reason to

00:36:26.840 make more humans like that unless they felt that their objective function revolved around printing

00:36:32.920 more of them but i think it would maybe just more extend their lives or digitize them so i really think

00:36:38.200 that it's more likely that our ai future will basically see a blossoming of robo sexuality and then 0.97

00:36:46.200 an absence of it and then i agree with that as well you'll see the blossoming of robo sexuality and then 0.73

00:36:51.240 it just will completely die out because either these people will have digitized themselves or put

00:36:55.240 themselves in pods or gone extinct because they were dating ai girlfriends and so the portion of 0.96

00:37:01.480 humanity that survives in this sort of aligned world will be highly genetically resistant yeah

00:37:07.800 i say genetically portions of our sociological profiles have a genetic component and so the humans that

00:37:13.240 were the most extreme in that component will eventually be the only ones that survive when we're talking

00:37:18.040 10 100 generations that's going to be a very different type of human than the human we have today

00:37:21.960 yeah i i think so and i think we're downplaying just how different humans are going to be i think

00:37:27.560 we're going to see full-out speciation that is accelerated very quickly due to agi and i don't

00:37:34.920 think it's just going to be the technophilic humans and the luddite humans i think it's going to be

00:37:41.080 the luddite humans and then five different flavors of technophilic human but i think the technophilic

00:37:46.520 humans will make up the minority of humans today i think the humans that sort themselves into the

00:37:50.680 technophilic branch will be two to three percent of the world's population today maybe yeah yeah

00:37:56.200 it's hard to say it's easy to imagine a world in which meat puppet sex for example disappears and

00:38:01.640 people primarily reproduce using ivf and artificial wombs both to optimizing it and because once once 0.98

00:38:10.280 there is a perfection of virtual sex it's going to be so disgusting and weird to do meat puppet sex 0.57

00:38:15.640 that people won't want to do it but then i could also very easily see a world in which humans get 0.99

00:38:21.000 really hipstery and snobby about meat puppet sex like in the matrix where they're like i was made 0.57

00:38:25.960 naturally you know what i mean oh no i don't think so because those people would never have high amounts

00:38:31.480 of economic success and therefore not yeah so if you don't have economic success then you just your

00:38:37.640 idea isn't going to become aspirational you could say that they've hampered themselves so you could have

00:38:42.200 a few like old hatsburg like rich families that are all done through like meat sex but i don't think 0.97

00:38:48.280 it would be the majority of of the economic system would be controlled by individuals like that

00:38:53.640 and when you talk about speciation i think a lot of people can hear this and they're like what a

00:38:56.840 horrible thing to say about our species but if we ever become an interplanetary species and we it

00:39:02.440 turns out that faster than light travel isn't possible which i think probably it isn't without time

00:39:07.480 travel also being possible to some extent so if faster than light travel isn't possible you will

00:39:12.920 definitely have speciation in humans future because you are going to have local breeding populations on

00:39:17.720 the different planets that will have different environmental pressures and it's not intermixing

00:39:21.960 because you need faster than light travel that'll lead to very quick speciation if you look at how

00:39:26.360 quickly the human genome can change like i was just talking about like how quickly iq can change

00:39:30.200 humans within just 500 to a thousand years on different planets will look very different from each other

00:39:35.800 and this is why i think that concepts today like racism are going to look so silly when you look

00:39:42.440 at how similar i am from the most genetically distant human to us today when we think about

00:39:48.440 where humanity is going to be at 5 000 10 000 years the diversity of humans is going to be so great

00:39:54.440 that the sort of all 1.0 humans are just going to be basically the same thing to everyone this is one

00:40:01.240 of those things where it's really interesting when people because we talk about what humans have

00:40:04.920 heredible or we talk about gene selection or gene editing they're like oh you must be racist and

00:40:09.960 i'm like oh my god our views are offensive but not in that way like we are so many levels beyond that

00:40:16.840 in terms of the way that our thinking about the future of humanity's expense is it just shows how

00:40:23.320 trapped people are in their current political fights yeah they have no idea how differentiated we're

00:40:28.680 going to be they have no idea where our species is going no idea how quickly we're going to get there

00:40:33.640 and it's interesting to us because we're so public about this a lot of people who are working on these

00:40:37.640 sorts of technology under the radar that they think could be legislated against you know they come to

00:40:42.600 us so we have a view of what's coming through the pipeline and it is dramatically more than you think

00:40:49.880 i'm excited for it sad that maybe i don't get to become a cyborg as early as i would like

00:40:56.200 hey they'll be trained on our books that'll be better than us anyway

00:40:59.720 our kids will be better than us anyway i'll just die so it's fine become irrelevant yeah

00:41:05.800 but that means they weren't good ideas right you know they'll train them on something better

00:41:09.640 exactly our kids books hopefully we're okay we're okay with all that temporary

00:41:15.880 just glad to live now because we get to see some major change

00:41:19.240 i am so lucky that i am married to such a weirdo as simone because i wouldn't be able to have these 1.00

00:41:24.200 conversations with anyone else uh well i do have these conversations with other people but they

00:41:29.000 it's so weird that we have such aligned like we have aligned ourselves so much through all our

00:41:33.880 conversations and yet this alignment is something i don't see in most people i talk to like there's a

00:41:39.160 small portion of the pro natalist movement that's like okay sort of aligned with us but i i can't

00:41:44.840 believe that i found the one weirdo to marry me who doesn't think that this is all crazy

00:41:50.120 malcolm collins you are the butter to my bread i love talking with you so much and i'm looking

00:41:56.520 forward to our next conversation you're the butter to my biscuits oh you know just what to say

Based Camp - May 21, 2023

Based Camp： What AI Means for the Future of Our Species

Episode Stats

Summary

Transcript