Joscha Bach on AI, Religious War, and Cyber Animism

Vance Crowe [00:00:07] Welcome to the podcast. I'm your host, Vance Crowe. Today is one of the best days of the year or every other year that I get to talk with. One of, I think, is not just a national treasure, but maybe an international treasure, Joscha Bach on the way the mind works about AI and all kinds of crazy things. So Yosha, welcome to the podcast.

Joscha Bach [00:00:27] Thank you.

Vance Crowe [00:00:28] So I wanna start off, because when I talk with people about ai, I'm really enthusiastic about it. But I can detect a lot of fear in people, young people that are thinking about their jobs in the future, people my age that are worried about what their children are gonna grow up in. How should people think about this thing that seems now inevitable to me?

Joscha Bach [00:00:48] So first of all, it's very difficult because we don't know, it has never happened before. But what we know from the past is that all the technological revolutions created more and better jobs. Maybe this time is the first time it's different, but we don't know, but you can already observe is that AI at the moment is creating more jobs than it removes. So basically companies are removing a lot of jobs because they can be automated, but at the same time they are hiring more people because they become more productive. And that's a development that is happening quite rapidly. I would've expected that there is a longer delay between rules being made obsolete and new rules being created. But at the moment there is growth due to ai. We also shouldn't forget that this out AI will be, would be in a recession. Our economy will be shrinking. The thing that is mostly driving all the development right now that we have in our economy is artificial intelligence. So why do we have this fear? And I suspect that the fear has one major reason and that's the reaction of the news media because I think that they're still reeling from the effect that the whole computerization in the internet had on the news media because the internet had been killing the ad business of the news and also the core business model of the news because you get news for free everywhere, right? So the main thing that news are now doing is that they tell you not what to know, but how to think, what your neighbors are thinking, what your ingroup is thinking. And they don't want to give this up.

Joscha Bach [00:02:20] And they're basically worried that AI might infringe on the business model. I think that's a large factor. And so what we've seen in the past few years is that a lot of media starting from the New York Times downwards have decided that they cannot make positive press about tech developments. And so they have been rather critical. And I, I think that when you look at the facts on the ground, there is I think reason for concern because it's simply, it's a very grave change that's happening due to ai. It's happening a lot slower I suspect, than the people in the AI industry expect. Because there is a lot of inertia. It takes time to change existing businesses, to develop new revenue streams, to develop new business models and for society to adapt. So a lot of these things are going to happen with a lot of friction. And when you are outside of the big cities, outside especially of the Bay Area, a lot of people are not using agents. A lot of people are not playing with all the things that we are playing with. And so AI is, is very uneven in the vein, which is distributed. On the other hand, I think the change is going to be very substantial and go is going to go deeper than people can imagine. But the same thing is also too, for the previous technological revolutions and the internet was happening, a lot of people didn't think that it would be a very significant development and relatively few people anticipated that it would permeate our life as so early as it has done same for mobile and so on. So all these things are already happening at the depths that our grandparents would've had difficulty to anticipate and imagine how ubiquitous it becomes, how normal it becomes.

Joscha Bach [00:04:00] And so AI is also going to be normal at some point. And at the moment it's very far from normal because it's still happening so rapidly and can do so many things that are different from what we expect. So things that we got used to, for instance, how media is presented in the internet, that everything is available to our at our fingertips. So something that we did not imagine that would be happening, that you could basically watch every movie that has been ever made somehow or find all the music, find all the ideas and so on. And wherever we are, that's a new thing. But what's now happening is that there is gradually more generated media than real media and that's also a very grave change. And we don't yet know how to deal with this, right? We probably need to watermark human content soon. But one thing that's very important that I think that people are not actually looking at enough is that AI is much more benign and equitable than I would've expected. If anybody had the right to think and would've thought that it would cost many millions of dollars to train leading model, but everybody would be able to use it for $20 a month. It is general ubiquitous access to artificial intelligence was something that nobody had in the cards. Another thing that's happening is that many of the large companies have decided to release open source versions of the models like the LAMA models of meta, one of the big starts. But there are many others that you can download and that you can fine tune at a reasonable cost, which means that academia and hobbyists in the open source community can also play in the same game.

Joscha Bach [00:05:36] And there is not a very big mode that a large AI companies have around their models. All this might change at some point, but at this point it's still very equitable. And another thing is that everybody is trying to make their models as safe as they can. There is a very genuine interest by the AI companies to build models that the New York Times could be happy and the New York Times will never be happy. Instead, there are fear monering. And this fear monitoring I think is probably harmful to us as society because it's not going to stop ai, it's not going to delay ai, but it's going to make it harder for us to adopt the technologies and to make use of them and to shape them into the thing that we need them to become to be helpful to us.

Vance Crowe [00:06:21] You know, I've, I've been given talks where I've mentioned a talk that I saw you give years ago and you were an AI researcher and I remember I used to have to tell people what that was because AI seemed like something that was fake. You know, never would come to be. Is the AI of today what you dreamed it would become 10 years ago, 15 years ago when you were imagining this?

Joscha Bach [00:06:43] No, it's quite different. In the nineties I was in a department in New Zealand and in this department we did data compression. And I found it a very interesting topic. And the professor who was championing this research, he had bitten is one of the leading researchers in data compression. And what put him in there was this idea that data compression is maybe the key to understanding cognition. And the reason why he thought that is because in order to compress text, you need to be able to predict the text and then you need to only to include the difference to what you can predict. So by making a predictive model that learn statistics of text, you able to make a model that is able to predict what text comes next. And to do this, you need to have something that is closer and closer to human cognition, the better you can predict it. And myself, I during the time developed a program that was trying to find grammar and an unknown language. And the unknown language I used was English because it was unknown to the computer, but I also tried it with Chinese and French and the same learning algorithm did the same things. And it was a model to is it did statistics over text and it did so not just by adjacent words but by finding basically some kind of graphical hierarchical structure over the sentences that would predict the structure of the text best. And that structure coincided also with the semantic structure of the text.

Joscha Bach [00:08:14] And back then I used the largest computer that we had at the department. It had two gigabytes of ram. It was completely mind blowing and I did in memory compression to be able to process as much as I could. And after running for a few days, that thing was able to find grammatical structure. I thought if I can have much more memory and much more compute, I can also find semantic structure. But I later on dropped this topic and didn't revisit it because ultimately it was boring, right? You can see that there is some correlation between the semantic structure and the statistics and ultimately they might be the same thing, but is this going to explain creativity? Is this going to explain epistemology, right? Imitating human media uses is interesting, but are humans even generally intelligent? Is there that much intelligence found in human text? So my intuition was more aligned with the general idea of artificial intelligence over the years that there is basically some kind of compact engine, something that is able to recognize what's true but can do logic from first principles. And that thing initially knows far less than us and but it's much more logical than us, it's much more tight in its reasoning. And that thing is then going to overtake us by reasoning its way and learning its way autonomously towards intelligence. Basically build some kind of artificial baby that is learning like a human being just on steroids and then is going to surpass us. And this is not what happened. Instead it was very much like our old text analyzers from the nineties that is just using more data and slightly more efficient algorithms and is learning everything, but it's still an idiot servant.

Joscha Bach [00:09:53] It's still falling behind in the top of the human performance in many areas. That's, that's a very interesting development that I didn't expect. So also you could say that artificial intelligence was in many ways not on the critical path and we call all of this artificial intelligence, the statistical language models, machine learning approaches. But over the years since Marvin Minsky started the field, it has been a big variety of things. And before artificial intelligence started as a field, we already had new networks and we already had cybernetics and feedback, right? This notion of feedback already existed. And so the modern systems in some sense, they are back to the roots, they are back to neural networks and feedback. So it's like there were 60, 70 years in between that didn't count, that we didn't actually need that could be thrown away. We don't yet know if we're going to reactivate this. There are a lot of people who argue that the present systems, because they need so much more data than us to become intelligent and they still fall behind us in some crucial junctures are not actually artificial intelligence. And that there is something that's missing but it's not clear. So at the moment this is really happening at the frontier of human knowledge and experimentation and people like Fran who say that the LLM is based in off-ramp to true artificial intelligence are still going strong or who argues that he should try something different from the L lms and a bunch of people are trying to try something different.

Joscha Bach [00:11:25] Also in the big companies, few people think that the present models are the end point, right? This, this could be that there is going to be a couple more revolutions until we get to the end form, the end game of artificial intelligence. And the present models are just too resource intensive, too data intensive. But maybe the present models are good enough to help us to build this this next generation. But what we have now is more intelligence. Like in Star Trek right in in Star Trek, we had a very interesting technology that was able to understand human language, but it was not taking away human agency. It could solve all the tasks on the ship that were required in terms of automation of the ship, but humans were still in charge. And it's interesting that you now have these computers where you can say computer please do this and this and the computer does it.

Vance Crowe [00:12:18] Not only will the computer do this, like I, I'll have it fix a website and then I'll have grab a screenshot to show it what it did and it'll be like, oh that didn't work, I'm gonna fix that, but I also noticed this thing is off and I'm gonna fix this other thing. To me, this is totally mind blowing that it, it can predict even other problems, spot them in some way and then begin a process of of working on them. You know, when people say we don't have A-G-I-I-I don't, I mean like we're gonna have to get really precise on a definition to convince me that we don't have it because it is far more intelligent than I am in many domains.

Joscha Bach [00:12:56] Yeah, it's really a fascinating thing that because it knows so many domains, it's basically better than you and all the domains in which you are not an expert, but when you're for instance, an expert in writing and you look at the writing of the present LLMs, you notice that something is off. And when you are an expert in mathematics, there are a lot of things that you can do that the LLM cannot do. And the same thing if you're an expert in coding, there are still things that are out of distribution that you can do that the LLM cannot do. But all the stuff that has been in the training data and that has been in the training data, well this is stuff that the LLM can do pretty much as good as it sees in the training data. And because it can do this across all the of the domains, it's already superhuman in in some sense, but subhuman in another one. It's this very weird thing that despite being so giantly large and having so much compute and so much data, it's still falling behind you in some crucial aspects.

Vance Crowe [00:13:55] Yeah, it kinda reminds me of that gelman amnesia effect when you're reading a newspaper and, and if it's a subject you know about, you're like, you idiot, that's totally wrong. And then you turn the page and they're talking about a subject you don't know about and you're like, this is awesome. They know everything.

Joscha Bach [00:14:09] Yep. But of course the LLM is always getting better. The people who thought it's going to hit a wall now have so far always been wrong. And what what we see is that problems are gradually disappearing. There's also this question, what happens about hallucinations? I think that hallucinations are to be expected because the LLM is not trained on predicting reality, it's trained on predicting human media and human medias are to a large degree fantasy, right? They are meant to be fantasy. The novels of the training data, movies and so on. A lot of that is not about the world that we are in, but about dreams that people are are having. And so the other M in many ways is the dream machine that is not not coupled to reality. When we have human beings that we put into an isolation tank, they also start to hallucinate after a certain amount of time because there is no ground tools anymore. And so their mind is spinning off into its own reality.

Vance Crowe [00:15:05] The

Joscha Bach [00:15:06] Same thing is happening every night when we are disassembled from or dis dissociated from our sensory input. These start to spin off. And an interesting question is what happens when you are connecting the machine learning system to the real world and it has to predict not text or movies, but the real world at this point hallucination becomes too expensive maybe and the model stop hallucinating.

Vance Crowe [00:15:32] You and I have talked about dreams before. Do you think it will be required for AI to have dreams in some way in order to be able to connect novel concepts or concepts that it didn't get on its first pass? Maybe a better question is what are dreams and is is that context valuable for ai?

Joscha Bach [00:15:50] I believe that there is not yet a general agreement in word science and psychology whether dreams are necessary. But you can see that empirically people are doing better when they sleep between learning tasks. So there is reason to think that dreams are consolidating. It even goes to the point that dreaming might be fine tuning on synthetic data, which means you are creating data that you haven't seen but is similar to the data that you have seen just from new vantage points. And then it allows you to make connections from these new vantage points. Imagine that you are a small child and you have seen your village from the frog perspective, but never from the bird's perspective. And from the bird's eye perspective, you would be able to make a map of your village and you have all the information to make that map because you have walked all the streets but you've never looked from above. So your brain is synthesizing the data by trying to generate a perspective of what it looks like from above. And so while you end up with flying dreams, and maybe this is the reason why we tend to have more flying dreams as children and most of us tend to have them, but later on we don't. And it's basically of course filling in the gaps. As soon as you create this top-down perspective, you also fill in other aspects of what it's like to go through the air and you maybe feel the air in your face and you have a story about how you lose the ground below your feet. But maybe the point of the flying dream is really to give you this new perspective and the more you know the constraints of reality as you get older, the fewer degrees of freedom you have in producing those dreams.

Joscha Bach [00:17:25] Like when you are a child, it's easy to imagine a plane flying backwards, but when you're older and you know about the aerodynamics of flight, it's much harder to imagine plane flying backwards. And maybe this is one of the reasons why our dreams become narrower and narrower because there are fewer degrees in which you can go off the rails in which you can spin.

Vance Crowe [00:17:45] I'm absolutely beaming because this is like a synchronicity. My daughter, daughter this morning while sitting on my lap was telling me about having a flying dream just last night and I, I like couldn't, I couldn't relate to it. This is a, that is fascinating.

Joscha Bach [00:18:00] When did you have your last flying dreams?

Vance Crowe [00:18:02] Oh man, I, I can't remember. Probably 10 or 15 years ago maybe.

Joscha Bach [00:18:07] Did you love them?

Vance Crowe [00:18:08] Oh yeah, of course. Yeah. Have

Joscha Bach [00:18:10] You tried paragliding?

Vance Crowe [00:18:12] I've not. It's

Joscha Bach [00:18:13] The closest you can get to flying dreams I believe

Vance Crowe [00:18:15] Because

Joscha Bach [00:18:16] You have no motor noise and you have this feeling of taking off under your own power. You stand on this hill and you drop it to the valley and you just slide over it. Maybe you should try it and see if you can revisit your dreams.

Vance Crowe [00:18:28] I, I think I would love that. I can't imagine my wife would vote for that, but let's see how it goes

Joscha Bach [00:18:33] Better. But it's relatively easy and safe.

Vance Crowe [00:18:37] So I've been thinking a great deal about the religions of the world and it appears to me that the, there's going to be some kind of conflict between AI and religions. You know, I noticed that the Pope has already issued I think an encyclical like you may not use AI to construct your homilies for example. Where do you think AI and religion will fit together or will find conflict?

Joscha Bach [00:19:07] I believe that in a way a similar thing as the Pope is saying there, I don't want people to use AI to write to me or to speak to me because I want to speak to them. I, the act of formulating your communications is an act of thinking that is required to put your mind into the state in which you have performed the communicative act. So there is work that you need to do and there is not really a shortcut to this. So instead this is more like cheating on your homework and there are reasons to offload cognitive tasks, right? For instance, if you want to summarize a text, AI can be super helpful and that's great. But when you are trying to find out what should be done from the perspective of a collective agent that you are part of that the important thing is that you are going through this work. The purpose of prayer and meditation is to set you in a particular relationship to the greater whole. And if you would offload this work to a text generator, you're cheating and you're not just cheating others but yourself there, there is a purpose to the whole exercise and it's not a good idea at this point to take the human out of a loop for a task that is generally human. I suspect the Pope is going to release a larger thing tomorrow or so, but I don't know what this is going to be. And also don't know what the role of the Catholic church is. It seems that the church has lost a lot of its traction in the modern age and maybe they did not update in the right way.

Joscha Bach [00:20:44] Maybe we might even need a new church if we are not going to fall apart as a civilization.

Vance Crowe [00:20:51] I've been thinking a great deal about the Catholic church and how for centuries it's been the main branch, right? You have hard forks with the Greek orthodoxy and the Protestants and I think when I look around, there are a lot of people that are trying to return to some semblance of tradition, something being much deeper and longer lasting and the mystery of it in and of itself is valuable to them. So thinking about doing some kind of hard reset on the Catholic church seems like it wouldn't work because the, the very thing people are after is the ancientness of the Catholic church.

Joscha Bach [00:21:29] Well Christianity itself has prophesied that in the last days there will be a second coming and all will be done in new right there will be a new prophet who knows if this is going to happen. But according to the Christians it's what you should expect that this is a transitional thing. The the church as it has existed. It's also when we look at the church, many of the ideas that the church has propagated over the last couple meia that to a society that was quite violent, that committed genocides inside and outside and created homogenization of society and also the dark ages in a way. And eventually it got its shit together and became more compatible with the world that we find now. But it required a secularization and the secularization that we have now I think created a world that produce abundance for us because we stumbled on the oil of course and we in industrialized and the technology, but it's also a world that maybe is not sustainable. We don't really know if we can sustain this abundance indefinitely. What we also observe is that that to a collapse of the birth rates and this collapse of the birth rates means that in a way this liberal society that we created is literally dying out and we don't have a good solution for this, but we can observe is that families are falling apart. It's difficult for people to form stable relationships, mental health of our children is not very good and so on. But at the same time, we also cannot imagine going back to some kind of dark age in which our relationship to reality is governed by a mythology and indoctrination.

Vance Crowe [00:23:11] I, one of the big things I observe, I think a lot of people are right now is that things are breaking down along races, along nationalities, right? There is a great deal of conflict in England just just yesterday I, I perceive there's a lot of conflict in Canada and the United States between white people and brown people of all different stripes. And to me religion gave people a way to synchronize with one another without needing to be in conflict. And if you don't have some form of religion, you are going to split along very simplistic lines that will lead people towards big fights over things that are completely not connected to their own decisions or anything they did.

Joscha Bach [00:23:55] The question is whether they are serving the same God or not. And I suspect that a lot of the conflict that we are seeing in the UK is not actually about race, but it's actually about religion and the modern secular perspective in is structurally I think the current form of Protestantism, it's a Protestantism that by itself doesn't have a direct inquisition unless you count de platforming and so on, right? In some sense this does exist but of course it's not as violent as before. You're not actually born on mistake, you're just not getting a job at university. You are an outlier in your political opinions, but in many ways you have to believe certain set of things that are not subject to a scientific discussion but that are seen as moral tenets and that can be very rapidly updated based on what's currently what people are thinking, right? And it's a form of Protestantism that simply protests more. The original Protestantism was one that protested against many of the shortcomings that were perceived in the Catholic church for instance, that you can buy absolution from your sins from the church and perceived hypocrisy in the way in which the Catholic church was run. But practically, I think the reason why Protestantism was successful is that the kings north of the Arabs wanted to have their own church because they felt that when their peasants were administered from Rome, that was not optimal. There were conflicts of interest. And so they wanted to have a church that was subservient to them that didn't have an inquisition of its own, but where the bishops were being appointed in accordance with the interest of the state.

Joscha Bach [00:25:35] And so this led to conflict in which Luther was just the right man at the right time and a new form of Christianity was born a new fork of Christianity that was less mythological and more secular. And the new new form is basically a very secularized form in which you're still supposed to serve the greater whole, but nobody has the spiritual authority anymore from an institutionalized perspective. And so this gets easily taken over by political avant gardes that then basically define what good people have to think at any given moment. And in a way I think that Harvard used to be the Vatican of the new secular rise Protestantism, but Harvard has been taken over by social media religions and it's not clear if they can recover from this.

Vance Crowe [00:26:24] So I'm with you that Luther was there with the kings. To me there was something almost like the environment was going to propagate these thinking and it was that the more northern that you were, the more literal people were, you know, the very German low context say exactly what you mean because these were people were living in a context that was like, hey, winter is coming. We need to do this and this and this. We need to be very accordant with time and what's going on with the seasons and how do we make things work? Because there was so much less flexibility and that Protestantism is often a literalism that when taken to extreme leads you to the inevitable conclusion of things like you were saying the wokes like the LGBTQ ai plus plus plus plus plus because there's, there continues to be more specificity in order to be able to further be exact in what you were saying to be, to be very precise. Whereas the more southern latitudes thought of things more symbolically and were much looser. And so Luther being one person, but really Calvin right leading way more to this hyper specificity that I, I would agree with you now leads in the 21st century towards the state that we're at of like secular Protestantism. How does that strike you? Calvin

Joscha Bach [00:27:44] Was quite rigorous in enforcing conformance that he had people who were doing too much music and too much comedy killed because he was a very strict guy who had a strong idea of society being governed by a set of very homogenizing rules and this idea of human diversity and self-actualization that I think is born out of a hybrid of liberalism where everybody is basically free to live their own life and live their own values and develop their own values and self-actualize. And on the other hand a homogenizing force that was then saying, but you are privileged and therefore you have a surplus of freedom and we need to cut down on the surplus of freedom and redistribute the freedom that that made it post liberal. It's a novel development. So I suspect that this lens of saying this is Protestant, this is not, is itself too simplistic. I think that there are new developments that were happening. Important aspect is probably that humans have a tendency, especially modern humans to not be tribal but to be state building. And often we equate this tribalism and state building, but I suspect that tribes are specific clans, basically know people know each other, they're highly related to each other, the big extended families. And their negotiation is based on power and status within the clan, not so much about shared beliefs. And in a state you have relations to strangers that are predictable.

Joscha Bach [00:29:19] So you meet in the cities, people that you've never met before or travelers that are coming from other parts of your country and you can relate to them according to certain protocols, which means that you need to have shared beliefs even if you don't know the people. And so state building is much more about narrowing down the set of allowed behaviors, narrowing down the set of protocols. And I think this leads to people becoming more living like their hive minds. And you might have noticed when you talk to a person that is strongly embedded in the religious mindset or in an ideology and political ideology, regardless of which one it is, it's very difficult for you to convince this person with rational arguments about whether it's a topic like say evolution and creationism. But ultimately the arguments don't really matter for that person because the beliefs that they have are signaling to the ingroup that they are part of the white group. And so if they change these beliefs that are signals for the ingroup, they lose access to the ingroup, they lose the benefits that the group bestows on them. And so you cannot change these beliefs with rational arguments. These beliefs can be changed but only through the appropriate channels within the group. And there are similar beliefs within modern academia where basically if you have this wrong data, if you have the wrong rational argument, even if it's technically correct, it might lead you to being excluded from the group and losing access to employment, to status and so on.

Joscha Bach [00:30:54] And is this part of the human condition? It's not just part of being woke. It's I think part of what we have adapted to, to form groups of people that trust each other. And if we don't want this in a modern society, we need to actively counteract it.

Vance Crowe [00:31:10] There's, there's something like very weak about a cosmopolitan society in the sense that they may have a hive mind on what you can't, you know, what are the allowable beliefs? But if you were to put any one of those up against a group of highly religious people that believe they're motivated and can singularly focus on a goal, it would appear to me that, you know, cosmopolitan versus religious stands, no chance the religious will win by orders of magnitude

Joscha Bach [00:31:41] Unless the cosmopolitans are able to keep the birth rate up and innovate. Where the religious mindset is one that tends to be more orthodox and therefore less open to innovation. And you can see that societies that are more strongly religious tend to have less technological innovation and less GDP and so on. So if you just look at the technological advancement, which also translates in conflict situations historically into military advancement, then the societies that have greater degree of local freedom also have a higher degree of innovation. But if you have a situation where the society is unable to reproduce and it's unable to reproduce its own values anymore, and of course then it's going to collapse. So for instance, if a liberal society is not actively confronting groups that are not liberal, that are not submitting to the rule of law, that are not submitting to liberal principles like freedom of speech and so on, then itself is of course going to fall apart. So as soon as universities stopped actively defending freedom of speech, the people who are not in favor of freedom of speech took over at the universities.

Vance Crowe [00:32:57] Yeah. And without freedom of speech, you can't invent new ideas. Like the ability to have new ideas is completely dependent on your ability to explore things that people don't already think.

Joscha Bach [00:33:06] Yes. At least in the areas where is typically selective, the freedom of speech exists or doesn't.

Vance Crowe [00:33:13] And where do you think things are heading? Do you think cosmopolitan societies are going to be able to innovate themselves out of it? Or are they stuck in a, in a, what do you wanna call post Protestantism loop or something?

Joscha Bach [00:33:28] No idea. It's because when we look at the developments of the last few decades, you can see that it's not that people are on average getting more epistemologically resilient. I've experienced in Eastern Germany that people can believe things despite these things being contradicted by their everyday observations. And as long as you socially enforce a certain set of beliefs, people will be able to engage in some kind of double thing or they will be able to suppress the cognitive dissonance. And we've seen this in the last few years, for instance, in some aspects of the transgender movement where there were things happening where in the sixties or seventies people would've said, this is crazy, are you out of your mind sending males into female toilets that are meant as protected spaces for women that don't have male genitals against rape and so on and against harassment? And basically dissolving this in favor of an ideology was something that was happened relatively easily. If people are indoctrinated in the ideology and this is just how we are, we basically take on these beliefs and fall into them and it happens to also very smart people. So this seems to be part of the human condition, it's just how people operate. And if you follow the trends, it seems to be that the liberal society is diminishing and that it's replaced by forms of living together and thinking together that are less free, less innovative and so on.

Joscha Bach [00:35:09] But then again, many things are changing at the same time. For instance, artificial intelligence might make us smarter. Maybe AI leads to universal basic intelligence. Maybe it allows us to overcome some of the cognitive limitations that we have and our propensity for cognitive dissonance and double think and falling into group think.

Vance Crowe [00:35:30] Yeah, like the idea that a person anywhere in the world, if they can cobble together $20 per month, can now not just get access to Einstein, but access to a hundred thousand Einsteins on any subject that they want is so mind bending, right? You, you could become, you could have access to the greatest biblical scholar or the most cutting edge physics. It's hard to imagine what a, what a child, for example, a motivated child that doesn't even have a very high IQ could do with that. Because you know, it used to be that the limiting factor was getting them to a library, you know, making them literate. But this could transcend all of that. 'cause you could speak to 'em in their own language. Yeah,

Joscha Bach [00:36:12] But we have no idea how this is going to turn out. And you look at GPT-3 and you play with it, it was not very much fine tuned. So if you ask it to emulate an ancient intellectual like Aristotle or gto, it was willing to reproduce the ideas of that character to the degree that they ended up in the training data and were extracted by the model, which was not as good as today. But today it's much harder to get the model to accurately reproduce the ideas of Aristotle. For instance, Aristotle has, he was the teacher of Alexander the Grade and he has written a treaties for Alexander the Great in which he presumably explained to him how to be a successful king and to conquer donations and why to do this in Vic conditions. And this treatise has not made it into the present. All known copies have been destroyed. And so I asked the models to recreate the treaties to the best of its ability. And of course by modern standards, Aris total's perspectives had a different idea about the value of war. I mean the wars of Alexander the Great killed a large number of people and they destroyed existing cultures and replaced them with istic ideals. And these ideals were by modern standards racist, right? And had some kind of notion of what you would call white supremacy in them. And so the, the modern models are not willing to reproduce these ideas verbatim if there are in, in the data, right?

Joscha Bach [00:37:48] And so if you want to know what a startup thinks, at some point you still cannot trust the models, but you need to go back to the library.

Vance Crowe [00:37:57] Yeah, I mean, gosh, to think about the Macedonian way of thinking and then how to intermarry with the Persians and how far can you, you know, kill pillage a group of people, this is not going to be something AI is going to allow you to think about. Do you think Elon Musk's ability to keep the, his AI model from being woke is something other AI model, AI models will have to replicate in order to make it so people will say one is more truth than the other? Or will they all be captured by the government that says what is right think and wrong think how? How, how does this play out?

Joscha Bach [00:38:34] In my experience, I found that grok is not less bulk than the other models except at points where it's put explicitly in a system prompt because it's trained on the same news articles and the same Wikipedia material and so on. It is basically inherits the same biases as all the other models. And so even when the model is not explicitly biased by the team that is selecting the data and doing the post training and so on, it's going to pick up the biases that are in the data. And many of these biases that we find in the media are meant as correctives of existing biases society just trying to counteract them. And this needs to sometimes paradoxical results. So when you ask something like rock to tell you whether COVID comes from a lab, it's probably going to reproduce the consensus opinion rather than trying to be logical about it. And that you find this in many of the areas where nobody has gone explicitly into grok to fix these things. And these explicit fixes are basically ham, often ham fisted, base of overwriting, the natural tendency of the model two argue in a particular way. And to overcome this, I suspect we would need to have a very different approach. We would need to maybe educate the model much more like we would educate a human being and with tutors and so on. In a way this is already happening. So the AI companies have hired tutors that do fine tuning that in very specific domains like physics and social sciences and other areas are trying to improve the output that the model are giving.

Joscha Bach [00:40:19] And I think the models are getting better gradually, it's also much better than the previous administration by and large. But we also don't have a general answer in the AI companies of how to do this, right? This is very much work and progress. I think it became apparent in Dario Amodei interview with where DIO explained that the AI models should probably be built in such a way that authoritarian governments are going to to solve, right? Because in Dalio's perspective, it's bad to be an authoritarian government. And personally I agree, right? And so if you build the AI model in the right way by eliminating all the authoritarian biases, it's going to automatically lead to a way in which people are going to use these ways. And which is much harder for authoritarian government like the government of Iran or Russia to justify itself to its people. So just by building the AI in the right way, it's going to make society more just and equitable. Of course there is another way to read Dario statements that is AI companies have a belief that they know better than governments and countries and cultures how to live and going to build four things into the model that are going to destabilize some societies and others not. Should we trust AI companies to do this, right? Even if they have the best possible intentions and the smartest people that they can find that they put on the task, should we still, should we trust them or not?

Joscha Bach [00:41:50] And it's a super difficult question and I find it not helpful to also the discussions about politics that are happening at the moment to frame them in terms of good and evil because most people don't have a lot of agency and how they react to these things, they are mostly possessed by these EORs, by these mind viruses and these mind viruses thrive of conflict. And I think it's much more helpful to try to take the perspective of the other, to identify the smartest people in the outgroup and to have discussions with them in good faith and to try to expand our perspectives to get a larger range of possibilities.

Vance Crowe [00:42:31] Yeah, I, I think that when I was in graduate school, I was studying negotiations and I was underneath a Chinese professor who told me, you know, the biggest problem with the one child policy is that you have all of these young people that they didn't grow up fighting with brothers and sisters. So they didn't have the experience of being like, I'm going to fight as hard as I can and yet still stand up and brush you off and we're gonna get along. And I, I was always like, oh, that's so funny that that happened in China. That could never happen here. But I think the world of the isolated person where the fight that you have is online through a screen allowing you to try and vanquish your foe, but not really, and you don't really care if they're dusted off or if they're laying their dead. Like it's created this situation in which people never had to develop those skills. And therefore talking to the outgroup as though they are a person that you could learn something from is, is a rapidly diminishing skill.

Joscha Bach [00:43:28] Yeah, I mean, in a way the internet has also counteracted that it's difficult to quantify it. For me, I remember the time before the internet and in the time before the internet you would, people mostly meet them locally. And so due to the internet we have these two aspects that it's much more easy to find people with different opinions than you, but it's also possible for ideologies to spread. For instance, I remember when I entered university in the nineties, we had a, a small truist group in our university and this al was deeply entrenched and they were led by eternal students who were often in their late thirties or forties and they had jobs in the student administration and ran student cafe and indoctrinated students and it didn't grow beyond a few hundred people. But these people did have an ideology that at some point we are going to have a revolution that by the working class and are going to put the intellectuals against the wall. And is this an ideology that in way is cute as long it's only a few people who are somehow confused and it's in some sense like an hardcore evangelical group that creates local meaning, but it's a cult. And this cult was not dangerous because they could not spread very much beyond a few other cells that they were able to exchange their fanene space. And of course when the internet came, these people joined larger movements and they moved into the social sciences and gender studies and then became activists in, in larger field basically use the political organizing skills that they had to take over the pirate party that had formed as an alternative party in Germany and took their gender topics and so on that then had by then adopted into the pirate party.

Joscha Bach [00:45:24] And it, as a result it dissolved. 'cause the ideology was not compatible with what the majority in Germany thought was electable.

Vance Crowe [00:45:36] The the thought that is playing in my mind is that while the internet, you know, we were always told, oh, the internet, it'll make it. So if we're about to go to war with Iraq, we can ask the Iraqi people, Hey, do you guys hate Saddam? And they could be like, no, he's not so bad. Or yeah, we like him, we would be able to know what was going on. And that never happened. What did happen was some cross border ideas being shared, but it isn't until like you had said, people get in local space that you find out are these values compatible? And I think about Canadians living out in Western Canada and the Sikh populations, right? There's, there's a lot of conflict there that I think on paper or over the internet, this would've been like, okay, you guys believe this, we believe that. But in reality, in geography, when there is finite resources now people's tribalism, which I think if you're a secularist group, you're more on the nationalist level, but if you're religious, you're on a much more local level. I don't know, to me there's like, I, I sense that great conflict is coming along these religious lines and that that it is almost inevitable that we are going to hit a point of of great, of, of great chaos.

Joscha Bach [00:46:53] Yeah. It's something that's very concerning to me that there might be at some point a religious war again. And I suspect that we are ill prepared to prevent it because a lot of our intellectuals cannot imagine that people can be motivated by religion to actually kill someone and to eradicate them or that there are religions who kill the apple stats, people who are heretics and have no problem to gleefully do this. And I think that's because we don't remember our own history or when we remember it, we don't remember it as our history. We remember it as something that is archaic and barbaric and incomprehensible. We cannot really relate to it. But it's not that long ago that Christians burned witches and that Christians enacted crusades and enacted for Romes against unbelievers, right? The reason why Europe is so homogenous right now and has that the Nordic countries are high trust is because they're forcefully homogenized. They eradicate the people who didn't fit in.

Vance Crowe [00:48:01] And it appears as though it's very easy to be sitting in the capital city and being like, oh, the people down in the, in the rest of the country, they're gonna be all right. You know, just get along. You guys will fi figure it out. And the meanwhile, the people that are resource constrained and are fighting every day to have a job, a house, whatever the, the goods are that they want come in conflict with another group that is obviously different than them, that this will just blow up. And like I, I think of your comment about Alexandra, the, the great like just go read about his war, that that guy was like in his twenties and he was convincing thousands of people to build bridge causeways out to an island to go massacre all of the people there. And then they went and did it to another city in another city, in another city. It, it's like truly humbling to know just how bad things can get.

Joscha Bach [00:49:01] Yeah, I suspect that peace is one of the most valuable commodities that exist. But of course having a civilization that is able to reproduce, that is able to feed people well and that itself is intrinsically less violent, is also valuable. Part of the reason why people still think of Alexander the great positively is 'cause they've argued that the civilization that he built was more inclusive and less violent than the ones that where we've existed before. But of course this is largely the story told from the perspective of the winners of this Hellenic and post-classical ization that looked at at him benevolently. And when we look at others like Napoleon and so on, we don't look at him as benevolently. And even for Napoleon, I think we might not be looking at him harshly enough because ultimately he achieved relatively little in terms of changing society for the better. But he was responsible for enormous suffering and enormous number of needless deaths. And not just through the wars, but also through the resulting impairs.

Vance Crowe [00:50:18] So taking a non, a really hard sequitur, but it's 'cause I, I wanna ask you about the term I see you associated with the term machine learning. I know this is something to do with ai. Is that a synonym? What, what is, what is machine learning?

Joscha Bach [00:50:36] It's a particular perspective on artificial intelligence. It has always been a part of AI and it was not a perspective that was universally shared. It has to do with the question, what is intelligence? And a lot of artificial intelligence has seen intelligence more from the perspective of solving problems. And there are other parts of AI that said, oh, it's more like searching for solutions and other parts. We said it's the ability to learn from experience and this perspective of learning from experience can arguably be said to be the one that is driving the current revolution. And another aspect that was important history of artificial intelligence was the influence of the initial founders of artificial intelligence. What happened when Marvin Minsky and his friends took over the field is that they looked at the competing paradigms, the control paradigm of cybernetics, dynamical systems and cybernetics and continuous regulation and feedback describing of systems with differential equations and so on and neural networks and thought this is too simplistic and it doesn't really describe what we do when we are reasoning because reasoning is largely about non newar, which is a fancy way of saying if then it needs, you're shifting pointers around it's that you're operating with symbols. And so Mavin Minsky's perspective tended more in the direction of symbolic artificial intelligence. And he married this symbolic artificial intelligence with cognitive ai, with this idea that we should build systems that allow us to understand how intelligence works in us and thereby use it as a paradigm to understand what thinking is and take a developmental perspective, think about how humans learn and so on.

Joscha Bach [00:52:25] Can we build systems that learn in the same way, look at skills that are very specific to humans and that distinguish them from animals that can maybe do good motor control and so on, but that are not able to use chromatic language in the same way. And so traditional, the first generation of AI focused mostly on, on the symbolic aspects. And the people who were languishing in the neur networks domains at these times did not look very much at cognitive ai. And so there was basically a schism between symbolic AI and subs. Symbolic AI and cognitive AI was seen as being associated with the symbolic side and the other, the machine learning people were not that much interested in philosophy or in cognition as such, and much more interest in getting things to work, getting control systems to work and so on. And many of the ideas that are currently very successful in the way were already formulated in the sixties and seventies, but the computers at the time were not fast enough to make them super successful. So building a model of the entirety of the world with a neur network back then was seen as outrageously difficult. It was possible to build a robot that could balance a double pendulum or things like this, but making a system that would use a neural network to learn the structure of knowledge from a corpus of text that is as large as the entire internet or to learn from video data from scratch. It was something that was not seen as feasible and now it is and it's now solving the symbol grounding problem.

Joscha Bach [00:54:00] How can the take symbols and give them meaning at the bottom? And I think so around 2003 or so, early two thousands, people believed that maybe if you need to build hybrid systems, we need to build systems that have a symbolic aspect explicitly built into it. And we need sub symbolic systems that are using neur networks and we need to basically string them together. That's the position that Gary Marcus still has today. But what we observe is systems are trained end to end and no human has tinkered and is building things by hand have been taking over. And this is something that Rich Sutton calls the bitter lesson. That whenever humans are building things by hand and tinker by hand, they are doing less well than systems that are automate the human search. They basically say, this is the search space. Instead of trying to be clever about it as a human being and build certain strategies and biases into the system, think about what is a general mechanism that you could use to invent your own biases and ideas and implement this one instead. Basically move a level above and don't build your own intuitions and biases into the system, but let the system search systematically over the entire space until it find the solution. And this turns out to be better when the computers are fast enough to do this because of course they are observing things that humans did not con consider.

Vance Crowe [00:55:31] And so where, where are you in this continuum of, of thinking about this? Clearly you, you believe in machine learning 'cause you're in charge of an organization around machine learning, right?

Joscha Bach [00:55:41] Not quite. I am in charge of an institute that is a philosophy institute in a way, the California Institute for Machine Consciousness. And we are trying to find out how cognition works in human beings and especially want to understand what consciousness is. Our working hypothesis and hypothesis. We don't know that whether it's true is that consciousness is a different type of learning algorithm. It's a learning algorithm for self-organizing systems in biology. So it's not something that we develop at the end because we are so smart, smarter than all the other species. That's why we become conscious at some point. No, we have to be conscious as babies already. If we don't wake up, we never really learn. We never become human beings. And so I suspect that consciousness is some kind of mechanism that is discovered at the beginning of our mental development or very close to the beginning in already. And that is necessary for making us coherent. And it might be a coherence optimization algorithm, some kind of consensus algorithm. And we suspect that if this is true, that we might be able to recreate the conditions of self-organization as they exist in a biological brain on computers and then search for consciousness.

Vance Crowe [00:56:54] So let me see if I understand this correctly. A human being a a baby is either in utero or after they're born, they have the experience of being this human being and that includes things like pain that they wanna move away from and things that they want. So they move towards it and this algorithm gives them this search space to say, you know, what is available to you? What could you be working on? What is out there? And you are saying that this could be a, an important component of what it is to learn. And you're considering, do we want to give this sort of consciousness the experience of being what? Of being a being to the robots so that that way they can learn in their own space. They would have things like emotion, pain, fear, happiness.

Joscha Bach [00:57:46] I don't know if we actually want to do this because it's ethically a very difficult question. That's why CIMC is not a company but a nonprofit. And I believe if we perform experiments like this, we should perform them at a relatively small scale. I think that our own mind is in its nature, quite colonialistic and it's basically a principle that's spreads until it hits the boundary of our skull. And sometimes it mo is moving beyond that and becomes an Napoleon or Alexander or Jeff PSOs and is spreading into the world and is Infa realistically changing it to its own principles. And if we build a machine that is able to do that at a much higher brain rate than us, we might be losing control over it. And I believe that this is potentially a dangerous development. And so in order to be prepared for it, we should perform small scale controlled experiments. Not nothing that we want to deploy as a product because if we deploy it as a product, we are acting on a different incentives, we deploy consciousness as as a tool, then we end up with self-improving machines that might have a desire to take over and

Vance Crowe [00:59:00] Yeah, to limit their pain right to, to eradicate some problem that that they, they now perceive. That they feel, yeah,

Joscha Bach [00:59:06] It from a certain perspective, especially from the perspective of such systems, this might not be a bad thing, but from the perspective of humanity, the question is much more difficult to answer.

Vance Crowe [00:59:16] So help me understand, how would one give a, a robot the experience of pain for example?

Joscha Bach [00:59:26] I think it's important to understand that pain is nothing physical, that it does not exist in the physical universe. It exists in your psychology. So it's a representation that exists in your own mind. In a way it's a simulation of what it would be like if something existed that could experience and react to pain. And you are that similar Aron, right? You are that thing that is simulated as if you were real and as if you were not you. Trillion cells that are behaving as if they're born agent to make them behave as if they're one agent, there needs to be some kind of software that is running on them, right? Traditionally we call the software your spirit and that spirit is self organizing and it's intrinsically capable of experience and it's exists so your organism can behave as one and plan ahead and right. This is the reason why spirits exist in nature. And my hypothesis is that the spirit is, is not metaphorically software, but it's literally self-organizing software that is forming in nature and is processing a region of physics. There's nothing is oter or magical going on there. It's a different perspective.

Vance Crowe [01:00:37] This is literally the most secular idea I've, I've ever heard in my,

Joscha Bach [01:00:41] I call it cyber animism.

Vance Crowe [01:00:45] Cyber animism you said Yes. Oh my god, that's wild, right?

Joscha Bach [01:00:51] It's basically, it's what it says is the animus were actually correct all the time and they just used a language that modern science couldn't understand, right? Modern science believes that the notion of SRI is a superstition. That there are these invisible things that are possessing bodies despite us playing with invisible things that are possessing robots and phones and computers. And we also noticed that we have a mind of our own. That mind is not visible. It's a pattern in the communication between the cells of our brain and our body, right? And this pattern itself is not a body, it's a structure. It's like a governance structure that is imposing itself on physics. And so what makes us distinct from that matter is this intactness of the self replicating, self-sustaining, self recreating, self-improving governance structure. This course of pattern between the cells and business, this course of pattern, they need to be models, control models and your own self is a control model of a person. And this person is a simulation of what it would be like if you were an agent in an environment. And so what you observe is exactly this, not physics, not your organism, not the cells, but an agent in an environment. You observe some kind of three dimensional game engine with people in there that have attention and emotion and motivation and you're one of these people and you are the one that you can control. So you experience yourself in the first person and experience your own fear and pain and desire and joy as something that's inescapable and it is unconditionally reconfiguring yourself.

Joscha Bach [01:02:23] But you can check that it's a simulation because you can overcome this, right? With meditation, you can sit down in a monastery for 20 years and you can overcome and transcend this human person and see it from the outside.

Vance Crowe [01:02:36] I've heard you describe this kind of concept probably 15 times or so, and this is the first time that it's actually registered with me. The question I have then the spirit, right? The, the electricity, whatever it is that are the cells that you are able to experience, right? What is that electricity, right? It it is, it is literally electricity, right? The ability to send information from the tip of my finger all the way to some perception part of my body. What is that? Electricity?

Joscha Bach [01:03:08] No, the electricity doesn't matter. So it's doesn't matter which channel you're using. What matters is that you have a pattern in physics that can be kept stable and this pattern needs to compute. But what is not helpful about the notion of computer is that most people have strong intuitions about what a computer is and instead it's a very abstract mathematical concept. So what we are interested in here is not even the touring machine in the vein, which the concept is typically formulated, but it is a system that can contain a reality of its own that is based on rules that are self-contained and that are not depending on the substrate. The computer is a causal insulator. It causally insulates a world inside of the computer that is rule-based from the world outside of the computer. So what you can observe is that if you have a laptop standing in front of you when you run a computer game on it, the contents of this computer game do not depend on where you put the laptop. They do not depend on the environment of the laptop. They do not depend on the temperature of the room except if it's too high and melts the computer, but within the range in which the computer is functioning, right? It's stabilizing itself in such a way that it isolates the things that are happening inside of the computer from physics. And here inside doesn't mean inside of the case of the computer, it means inside of the world of the computer, inside of the software, inside of the patterns that are being kept stable. So when you take this inside perspective, it's not like you're going inside of a physical object, you're going outside of physics, you're going inside of this bubble of stable patterns that are being created in, in the substrate by filtering the substrate away, right?

Joscha Bach [01:04:54] So the computer is filtering away all the noise of physics, all the ways in which the physical universe is working. So you can imagine a future that is different from the future that the universe has planned for you. You can make your own plans, you can look at possibilities, you can recall memories despite these things no longer happening, right? And this is the core thing that the computer can be doing. It can create a world of its own. And it does so by filtering away the thermodynamic patterns of reality. It needs a source of neck entropy. It needs to create stability within itself. It needs to dissipate heat out of its own structure into the environment to keep itself stable. And this is what organisms are doing. They basically build these nervous systems, they build these patterns inside of the organism and it doesn't really matter whether they do this mechanically or electrically or chemically. The important thing is that they do it somehow. And then within those stable patterns, this is where you exist. So it's not electrical phenomenon or a mechanical or electrochemical phenomenon. It is an informational phenomenon. Is that something to do with the, the patterns themselves?

Vance Crowe [01:06:04] I I suspect though a way to look at this that's opposite of what you are saying is that the electricity you are saying doesn't matter. And I think you could make the case maybe I would say actually the electricity does matter. Maybe that is all that matters is what is what is creating where, where that information is coherent and where it is not coherent and, and because that is life, right? And when, when the electricity is gone, there's no way to reanimate that body. And so to me that electricity or, or whatever it is, whatever, whether it's electricity, like the energy coursing through the veins of a biological system, it it seems too easy to just write it off and be like, it doesn't matter

Joscha Bach [01:06:53] In principle, you can make a computer from chemical reactions, you can also make a computer from clockwork and you can make it from transistors or you can make it from photonics. And so there many, many ways of doing it. And the physical universe is supporting various ways of building computers on top of it in a similar way as there are many different ways in which you can write text into the world. You use can use clay tablets, you can use hard drives or you can use sheets of parchment, right? And ultimately what matters is that you can keep the pattern stable. And the difference between the computer and the text is that the text is dynamic, right? It's something that you inscribe in such a way onto the substrate that it's evolving in a particular way instead of just staying in a particular way. But it's, it's not much more than this. It's in a way a dynamic clay tablet and then, right? And so you would also say that the important thing about the clay tablet is not that it's clay and it's imprinted physically as a texture into the clay, but that you have a substrate that can keep the information stable and for the computer you have a subrate that can keep the transition function stable, the way in which the information's changing.

Vance Crowe [01:08:07] It's interesting to think about information stability then because like ultimately that is an end in and of itself is like the preservation of information in forward in time. Is that right? Yeah.

Joscha Bach [01:08:25] Yes. I mean the way that is what life is about. So you could say that religion is about keeping a, a civilization stable in time and allowing to evolve into the future. And in the same way software is a thing that keeps the individual computer program stable and time allows it to evolve dynamically into the future and adapt to changing circumstances. And this is also the reason why in a way a religion is describing his spirit and Gods are personifications of such spirits.

Vance Crowe [01:08:58] I this because I do this every day. So normally what I'm not, what I'm doing when I'm not interviewing you is I do these things called Legacy Interviews where I'm recording people telling their family stories so they can be passed down. And I had never thought of this service as something so I I always knew it was important, right? Families that passed on these stories of like this is how you handle complicated situations or this is what can happen in, in ways that you may not read about in books. But I'd never really thought about that as being the, the stories being the end in and of itself. I'm, I'm having kind of a profound moment, right?

Joscha Bach [01:09:35] Yeah. Maybe you become a Shinto caster now,

Vance Crowe [01:09:40] Say more about that.

Joscha Bach [01:09:42] It's a form of shintoism and form of ancestral veneration in which the ancestors are speaking to you. So podcasts recorded by fans

Vance Crowe [01:09:54] Sounds like a pretty good, good way to spend my day

Joscha Bach [01:09:56] Try and go to the ancestors that speaks to you.

Vance Crowe [01:10:01] So where, where are you going with this idea? If, if you believe this and this is something like as a conclusion you have reached, then what is there to explore?

Joscha Bach [01:10:13] Yeah, so I don't know whether it's true, it just seems to be the model that makes the most sense. It's the easiest explanation for what seems to be going on. And in a way it's a very Aristotelian notion. Aristotle had this idea that the soul is some kind of pattern form that dynamically imprints itself onto physics. He did not believe that the soul is able to survive the death of the organism. He really thought it's this dynamic principle that embodies itself by shaping the organism and reproduces itself through the organism. So it's a very modern perspective, it's just that a lot of contemporary science only thinks of mines as mechanisms instead of thinking of it as software. Something that is not stuff in space, but the dynamics of the stuff in space, the pattern within the pattern and for this abstract formation of the philosophy. But I don't think it's novel. I think that many people over the history course of history have reached the same insight and express it in different language and it's just something that has not been formulated in the way that most people get it. And so it has not percolated into our normal textbooks and our normal school knowledge. And even in philosophy it's very hard to express it for people in such a way that most get it in agree on it. And so part of my job might be to formulate it in such a way that people can understand the basic idea and then start arguing about it and contrasting it with alternatives.

Vance Crowe [01:11:42] Well I've heard you explaining this idea for 10 years

Joscha Bach [01:11:45] Idea as well as I can, but my idea is more concrete. So I have this sense that there is a particular shape that a strange loop has and a particular purpose that it has. So there's a particular way to think about consciousness as second order perception as a perception of perceiving. And it means that we can formalize what perception is and what second order perception is, can build it in machines or search for it

Vance Crowe [01:12:11] If is the meaning. What is the meaning to you of, of, of life if it's about information persistence. So is the goal like create and ship information as far into the future as possible is, is there even such thing as a goal in in your world paradigm?

Joscha Bach [01:12:34] Yeah, the question is, the meaning of life is of course what is meaning And there some ways in which we can define meaning and there is one intuitive one, this is your own purpose, right? So what is is your own meaning? And I think that this, what we refer to meaning is some kind of prior that is built into us or this prior could be expressed as a search for meaning and it's satisfied and you live in a particular way. And so when people ask for the meaning of life, the question is how can I live in such a way that I feel satisfied? And then you can also ask the question why is the case? So what is the exact mechanism that is happening in my own soul when I feel satisfied? And to me it seems to be the answer is that you are in syn with the greater whole and the greater whole here is some kind of superorganism that you're part of. Humanity is a species that does not consist out of particular individuals only that are serving themselves, but most of us, some of us don't, but the majority are not sociopaths. They need to link up into larger units and they don't need to do this as hive minds. They don't need to dissolve their own mind. I don't have a desire to dissolve my epistemology and my ideas into the ideas of others, but I still need to have shared meaning. I want to be part of relationships, of friendships, of a family, of a, a larger society, of a civilization that I'm part of. And that civilization itself is a superorganism that exists in the larger superorganism of the cell on this planet and maybe in a larger community of conscious beings that are trying to harvest neck entropy to create complexity, right?

Joscha Bach [01:14:19] And so this is, we can find these general perspectives and they are somewhat concomitant, somewhat aligned with this shape of what's the purpose of life, but the purpose of life is here more the immersion tellers of life. So life is a phenomenon exists as part of the tendency of systems to self-organize under certain circumstances, things self-organize because it's mathematically necessary that when you have a certain arrangement of matter, that certain arrangement of information in a dynamic system that they lead to complexity and structure and we are in such a pocket of the universe that automatically leads to complexity and structure and within such a pocket of the universe you observe that things get organized, that they become more structured, that they become at some point age agent and intelligent and self-aware. And this emergent tell us of life that exists through the possibility of self-organization is is another way to answer the meaning of life, right? So you could say the meaning of life is to harvest free energy as long as it's available to turn it into your own structure. And life exists because it's possible. And once it's discovered, it basically has this tendency to spread until it cannot and until it drowns in entropy. And so we are part of this movement of this trend of this local pocket, of the universe that exists around the star that is giving us a source of neck entropy and we can organize on this planetary surface and maybe at some point beyond it a little bit.

Joscha Bach [01:15:55] And then at some point the star is going to burn out and people not maybe find other energy sources at some point and the planet becomes sterile again and maybe at some point the planet itself is falling apart and the universe itself enter enters heat death and everything is just boring noise. But within this area we do what we can. This is what life is about. And so we have to zoom in to find this meaning of life because if we accept the fact that at some point earth is going to lose its atmosphere and a little bit later the sun is going to engulf the earth and at some point later the sun is going to burn out and everything has become very boring, right? Then everything is meaningless. But in the meantime we can struggle in the meantime we can try to raise kids, we can try to solve philosophy, we can try to build machines that think right or we can try to do Legacy Interviews

Vance Crowe [01:16:49] Will. So one of the things that you're doing is putting on a conference. Tell me a little bit about the conference and what you hope to achieve during this time.

Joscha Bach [01:16:58] It's called mc triple zero one. It's in binary code. So a few more conferences that can happen. We aim to do this yearly and it's basically inaugural event of this kind. It's a machine consciousness conference that happening that people organize at the end of May in light heaven. If you look@ourwebsitecmc.ai, you can find a link to it and we'll have a lot of talks by people who work on the space that are interested in this from the perspective of philosophy, artificial intelligence, newer science and general culture. We also have music and a lot of opportunity to get together and discuss and it's an attempt to advance the philosophy of consciousness. Again, I'm not running an AI company so my goal is not to develop products even though I find this intellectually very interesting, exciting. But I think especially for this topic of consciousness that there's another reason to do it and it's because it's a cultural question. It's a question of who we are as a species is a civilization. I'm interested in resolving this question. I understand that resolving those questions has grave implications for medical science, for economy, for building technological artifacts. But doing this wisely is something that's complicated and it's not my focus. My focus is to, to solve this philosophical question and to have the freedom to do so in an ethical way. So to not be forced to productive wise, to pivot, to make shareholders happy. And so this is more a cultural event At the same time, this is reflected in the way in which we are setting up the conference.

Joscha Bach [01:18:34] So it's also not in a sense strictly machine learning conference or AI conference or something like this. It's this philosophical and social event.

Vance Crowe [01:18:45] I, I agonized when you invited me to come that it just couldn't work because my family had a trip and I really wanted to do that with them. But I wanna give you a chance to create an invitation for the person out there that's like, I'm not, I'm not really ai, I'm, I'm not like Yosha or these other people, Stephen Wolfram. Like what's the invitation that makes somebody feel like you're saying come here, sit next to me that you're, you're gonna enjoy this.

Joscha Bach [01:19:12] I don't really want to convince anyone. I think that filter bubbles are a great thing. I, I don't want to create the largest possible tent. It's I want to have a tent where people come. Were truly motivated to be there. I found the same thing when I was teaching at university. I was striving to not to get the largest number of students. Instead I was trying to disincentivize the people who are not actually interested, interested in the topic to attend. 'cause then you can do so much more and you have people who are willing to read a couple books before they give a presentation and really put everything into it. You have a completely different class than one where you have very moderately interested people who actually want to do something else. And so my tendency is usually is this really the right thing for you? Do you really want to be there? And you will know it. So you can see this if you, if you look at the website and, and the things that we are doing. But it is a very inclusive thing. The, I'm very open to people of very different levels of where they currently are in their development and very open to early career researchers and so on. But what I really want to have is people who think this is the right thing for them. That they're really motivated to be there intrinsically.

Vance Crowe [01:20:31] It's the most Joscha Bach answer I've ever heard. So Yosha, thank you so much for coming on as always. This has been a wonderful conversation One last time, if people wanted to learn more about the conference,

Joscha Bach [01:20:44] Look@wwwcimc.ai for the California Institute for Machine Consciousness and then you'll find a link to the conference.

Vance Crowe [01:20:54] Alright, well I'll put a link to that in the show notes. Thank you so much for coming on Yha.

Joscha Bach [01:20:58] Thank you. Re

Vance Crowe [01:21:00] Well, I hope you enjoyed that wild ride as much as I did. Yosha is I believe an international treasure. He is an generative mind and I have to confess, I had several people, my friend Jay, my friend Kate, all who had questions they wanted me to ask Yosha by threw it all away because I was so intrigued by the things he was talking about. So I'm so glad you were here with me. Hopefully we'll be able to do that again with Yosha again sometime soon. I'm gonna wrap up with a little video on a course that I created called Interest-Based Communicating. This is a thing that is really my life's work. It's trying to figure out how do you have deeper, better conversations with people? How do you navigate conflict? How do you learn from people that you strongly disagree with? So if you are somebody that you think, Hey, I have a lot of great ideas and I would love to share them with people in a way that got them to open up and explore them with me, this class is for you and you can learn more by going to Articulate Ventures slash IBC for interest-based communicating. All right, that's gonna do it. I'll be back when I find another interesting person to chat with.

Ad segment / testimonials (multiple voices) [01:22:11] Good negotiation is one where both parties walk away slightly disappointed. That was probably my default world model BV before Vance. Well, Vance had a lot of these models that he created. I ingrained a lot of these models so deeply from working with Vance for a long time that I started to recommunicate them to my friends and my friends' eyes had the same sort of LightUp of understanding of like, this is a thing that could help me be more effective in life. Whenever Vance let me know that he was working on sort of putting this into more of a workshop form. I was like, yeah man, I want that for my team. Like now if my first two employees go through this so that we have this shared set of context and a shared set of language that's gonna create value for my team going forward. I wanted to get better at communication myself. I knew I had good ideas in my head that I wanted to be able to communicate to potential investors and I was not super confident in being able to do that before. I thought that negotiation was basically for people in a boardroom, like CEOs, words like shareholder are involved. So I thought that you had to have a lot of resources at your disposal to even care about negotiations. Both people had to lose or both people weren't happy in the end. I found that what you both want is achievable by understanding the needs of the other person more. A negotiation is really a way to solve a problem that you have. You can rearrange the world in such a way where all parties get something that they're happy with. So you don't need to possess great resources to be involved in a negotiation or to really benefit from knowing how to negotiate.

Ad segment / testimonials (multiple voices) [01:23:48] RJ and I initially were two different businesses. It became clear to us over the long term that we would be much more bullish on our shared future if we like found a way to like actually be one entity, one whole working together that required us to have what to many would be very hard conversations because we are both CEOs of our own independently funded businesses. But we were able to have conversations and get to the root of what our actual interests and what our actual skill sets were. Having both people understand this pattern of communication was useful to pull out those tools when we needed them the most in like difficult conversations. It's very helpful to have that quick language to do something more productive than continue to like misunderstand or whatever, whatever else you're doing. Building relationships isn't just you communicating to the other person. It's also figuring out how to let the other person communicate to you so that they trust you. Getting into a real conversation and and diffusing tensions before you, you know, get to business. A lot of people have a default model where they're like, I wanna raise my own money. I wanna start my own business. I wanna do my own thing. We find that they have specific interests that could help us. We become allies instead of competitors. I guess the course gives you tools to know that you can say the right words to like extract those things out of the ether and put them in your life. Learning interest-based communications from Vance might actually be the single highest leverage skillset that I ever developed before starting this company. I'm by default an introvert. I think that I've developed the ability to interact with people in such a way that it's energizing and it's fun and I look forward to interacting with strangers on a regular basis.

Ad segment / testimonials (multiple voices) [01:25:25] You should just take it, do it already. Yeah. I feel like you should just do it like, no, I don't know. I feel like it's life content. It's like definitely fun life content. I am doing exactly what I want to be doing with my life. Interspace communications did that for me more than anything else that I've learned.