Aneesh Sathe presents an infrastructure field report on 'World Machines Theory,' a framework for reading history as the interaction of long-cycle civilizational 'machines' (Modernity, Divergence, Liveness), and demonstrates the two systems built to operationalize it: the Oracle, a multi-agent RAG pipeline over a shared knowledge lake that generates oracular 'prophecies,' and the Prime Radiant, a typed knowledge graph of ~500 historical 'machine cards' that can be simulated forward to project future institutional ecologies. He frames this as an Asimov-inspired but decentralized, commons-based reimagining of psychohistory, contrasting it with Seldon's mathematical, control-oriented original. The talk opens with context from the Contraptions Book Club, introduced by its host.
0:27 We’re here. All right. Um so just to quickly introduce the um SIGs size special session. SIGs size stands for special interest group in psycho history. And it is literally the psycho history you might recognize from the term uh as in Asimov’s u foundation novels. And uh the little bit of backstory of how we came to uh this um idea of actually trying to build um something resembling the psycho history technology in Asimo is a couple of things converging. Uh, one was, um, couple of years ago, I started a book club on my newsletter called the Contraptions Book Club. And it was just kind of like an excuse to get together a group to read um, history books that I wanted to read, especially about the birth of early modernity uh, around 1200 AD. So, it initially started out as historical curiosity. And I think it actually started because I visited Istanbul and Venice and uh got very inspired by like how much you can learn looking at the history of ancient cities. And so the first year that was 2024 u we basically decided to focus on the period u oh sorry it was 2025 I think yeah 1200 to600 which um was our idea of modernity actually began earlier than we think. So we should really rewind to 1200 not 1600 as people think. So we looked at a lot of stuff that happened 1200 to600 including you know the rise of Venice uh the last nomad in invasions the horse economy all sorts of interesting things and somewhere along the way I came up with this sort of like halfass theory of how history works called world machines theory. So the short version of world machines theory is that at any given time the world is describable by one dominant machine that gets installed. So it takes about 200 years to get it installed. Um then it sort of runs for a while and then it declines. No 400 years to get installed 400 years to run in a stable plateau mode and then another 200 years to kind of decline. So that’s a world machine.
2:50 And later I refined that idea to actually there’s three world machines running at any given time. One that’s ascendant, one that’s mature and stable, and one that’s in decline. Right? It kind of makes sense. So you have three overlapping thousand-year waves going on. And the way we periodized our understanding of history in the book club was um that there are three world machines of interest. So 1 1200 to600 the modernity machine booted up and then it became stable at 1600 and it’s going into decline now. The second world machine that we uh study is called the divergence machine. It starts in 1600 ramps till 2000 and it’s now entering its plateau phase. And then the third world machine we’re interested in is called the livveness machine which kind of takes inspiration from biology and life which is starting to ramp right now. So that’s the rough idea. um we’ve all written a bunch of it. Um we’ve built this website that collects our essays roughly around the scaffolding and I know Anisha has since then uh uh refined the machines idea to be kind of more fractile where you can have all sorts of machines at different levels and all that. So this is one thread the book club. The other thread was of course AI like along the way as we were doing our readings for the last um almost two years AI got really powerful. we could all build rag bots really cheaply. And then we started batting around this idea of could we just throw all the manuscripts of the books we are reading into a rag engine um kind of like you know Andre Karpathi’s LLM Viki model but more evolved uh and also throw in our notes our own writings plus our theoretical speculations like this world machines idea and can we evolve it to a point where it becomes usable as an oracle and then can we go beyond that and turn it into something like a prime radiant for psycho history. So to connect to Tim’s talk um just now, think of it as prediction markets but for hundreds to thousands of years as [clears throat] opposed to things you can bet on in the next month or so. So that’s kind of the segue point from the last talk. So that’s my introduction and Anish and Sean have talks and then we have discussions, right? So over to you Anish.
5:26 Yeah. Okay, perfect. So, yeah. Um, so yeah, uh, thank thanks a lot, Levanka. B basically um as you said like like we start started off as a book club and then that has since since evolved into both uh like like a larger um universe of essays as well as like some sorry software and I’ll be talking about uh both the idea and and the software behind world machines and the uh and the mechanisms behind like the prime radiant as well. So, so, so but I mean like before we jump into that, I just want to I mean I mean like um I mean like ground the whole thing with this with this idea that that as as humans we have this idea of like wildness both uh as as parts of like nature and the organisms which that exist in nature. And while one of the most key parts of of your wildness is this um affinity to exploration, right?
6:30 But but the fact is that the world has been like dangerous and and we feel fear around that and and so then basically the the the idea then uh is that we need a tool which can uh lower the activation energy required for some kind of like bravery and that’s where like say things like prophecies and then uh uh I mean these kinds of future looking tools like come into play. Uh and this operates at all kinds of scales broad idea is that we use like our imagination plus some analysis as a crane which helps us to move into the future. uh it it it essentially creates like sort of like a blueprint which we can uh then for fulfill and this like obviously um then operates at the personal scale where say individuals can use say things like astrology MBTI and and then others because when we are young we don’t really know what we want to be uh and like I’m sort of now old enough to not be embarrassed to to then admit that maybe yes like being told that I was like an Aries um did did make me uh a certain kind of like a person being told Bible Bible MBTI that hey you are like an INTJ that also um put me on some parts which I may not have taken uh but but this can also take other forms say if you’re like a cat per person and you are like oh when you have a cat you may have like toxo plasmosis and that makes you much more entrepreneurial and you go oh yeah maybe I can be like entrepreneur which I have been and and so like these are the tools we which you use to to then then effectively like move into this person who you don’t really know how to be but can they figure it out and then this also works of course at the level of of of society and it is it is interesting that that We start off with with the the the modernity by machine uh because that’s when we started to sort of formalize how how we should like co co-coordinate and while we have perfected that kind of technology, we don’t necessarily always like coordinate well and that decision to actually act right that needs some kind of like like an impetus to to then move us ahead. So that’s where the the the broad category of of prophecy like comes into play. Um and like this this like encompasses a lot of things not not just like the idea that X will happen but it’s a view into the f future based on uh some analysis like which we do. So uh uh and so effectively it it it is with this lens that we look at like world machines and the prime rad which is that like they are effectively tools uh which give us access to the future uh and then help us actually zoom out and see the woods which we live in.
10:04 uh they they gave us like an access to to to to understand how the world came to be around us and to interact with each other much more effectively to to to to then effectively um see more right uh but but then before I go way too deep um I will give like a bit of like an intro uh so so I am like a biohysicist by training um I effectively was trained to look at Babio through the lens of awful engineering and the machines uh which which which has been really fun. In my day job, I I effectively apply AI to to aid in like drug drug discovery. And so like I bring that kind of like my my mindset now to to two to to both history and uh and to time where where where I ask myself can this be like modeled either as a machine or as like a living organism or both and then like what can we do with that? Um I I won’t go a lot uh into how the whole work like came about but but but but then as Levanket said we essentially uh uh I mean like we sort of coalesed uh out of like uh the the the the the book club and and that’s been extremely fun. Um so uh on to the the the idea of like world machines then um as as as as a levanker said like uh basically we look at history um in a form where it works as a machine on the inside and in the very broad formulation we can treat essentially all of like society as one big machine which which which is born it gets installed and and then like slowly like dies. which is where the modernity um like comes into play and that’s now in the the the death phases and what I actually feel is that without the the tools which modernity gave us it may not have been possible to actually think of history as a machine at all because the modernity started to create like those kinds of processes uh and then formalizing protocols which then allow us to think of of what has happened through through that machinic sense, right? And at like as as the humans go, as everyone was was centralizing, um the fact that that centralizing created tools which allowed us to effectively recreate um what is now the the divergence machine where the the idea is that information and then activity is now distributed. Everyone has a point of view. We have to like encyclopedias and the internet which which gives like everyone like a view into how things can be and how they should be done. And yet again uh probably as as like a further rebellion against like um the reverse machine. We we we we we’re now uh into the early years of the livveness machine where at least our our conception of this is that um this will behave much more uh the way that life does where where memes which have so so far been like a technology of the mind no longer actually need the mind anymore and and then we can actually use LM uh to to to then build memes upon memes.
13:53 which is like s sort of what what humbold does and then they propel that into making like new things. Um and so because all of these work as my machines as you can imagine this machine can also be run into the uh into the future but that’s more into the uh into like the next part of the talk. U first though like I’ll just talk about the forces behind like the world machines Oracle itself. So as I said we essentially like started off as like a boba boba club. each of us were then like were then like uh having chats like writing essays posting things on Substack and what we found was that I mean just the way that Tim said that each of us had like a small view of this elephant which was all of history and and we found that when we mixed those two views like like together um that blending essentially gave us like a much better view of what was happening. And so we we we we then used AI and then like uh the newer tech to to then blend these ideas uh uh into one. So I mean like uh at this point if you say wanted to to totally contribute this would be as as easy as participating in uh in the chats. um I mean like having just like a well articulated note or then like an idea all of those sort of get like a combined into one right uh and so then what we build with this whole thing is sort of like a data f fabric where we ingest everything which we can we don’t really treat treat anything as noise because our idea is um that any view even if it’s like like completely wrong the fact that the view is wrong is a data point in itself which should be considered right and so we we we just treat everything as a useful item uh and all of that in text form we then run uh the the the the LLM wiki like idea on top of this where we say can we find interconnections between these two things such that we can have some kind of like asynthesis happening um and then on top of this we then build lay mechanis mechisms which can help you query this almost like a database uh but where you also go two or three connections deep and it’s not just like a not just like a like a keyword but it’s more of like a knowledge graph on the background. So all the magic actually happens at this layer uh which you can see like at the website uh there uh and on top of all of this sits like our like our part which is essentially uh like the oracle can can query that data fabric and you you you can either ask it like questions or it’ll actually give you like a prophecy or a story which is set in a particular ular time uh or place in the world. And so if we have uh information about that in our system uh it’ll use all of those uh inputs to to then say that hey this is why what what might have happened uh in the world right and so then um I mean like as I said like this is about the like the world machines now on to like the prime radian part. So effectively as the said we were inspired much more by Asimov but but there is the differences into how like how how Asimov envisioned it and then how he applied it in the books. So in the books it is it it is basically like like a mathematical model. Um I think it was based uh around fluidly dynamics and then so that works uh in a particular ways whereas what we do is we essentially take away that the idea of of the wave and make uh sort of machinely particles or then cards uh as I talk about and then that’s what actually interacts with each other. Um then in the books again uh this was used as a system of control uh and the the knowledge about the radiant itself was was wasn’t extremely like controlled.
18:37 Whereas what we want to do is we want to make this system available sort of as a commons where where people can both use it but also like contribute to it uh through through through truly different ways of doing like analysis different ways of doing AI uh which which we like talk talk about at the end and um in the book again this was used as a way to actually direct where the world should go whereas at least I view the prime radiant as a way to sort of create like inspectable future and it will help you understand where the world can go and if you want to do anything what are the things which we can do which will have an effect and is like likely to to to to be then like um actually like I mean you you you useful as an intervention if it choose to do one. So, um just like a little bit of like a background on like how the how the system works. Basically the the idea is um that within like the broader world of world machines uh they there are like thousands of like many historical machines which work inside that, right? And then the the way that this interacts is that existing my machines and then ideas create like an like an ecology around themselves such that new machines can can then make use of those kinds of like selection pressures and the affordances uh to to then like create like a new um ecology around themselves. So, so to model this uh it it it wasn’t going to be useful if you just put put the whole thing uh into into like into like an LLM training and and just hope that like some kind of embeddings will then come out. Um and this is of of course partiality towards what I’m currently obsessed with which is which is which is around causality where I said we actually need to have some kind of a of of a representation of what’s actually happening. uh but instead of just working with like co causality we we said let us use all the possible uh people who have worked in this area and just combine everything right I mean like every way that people have thought about thinking about historical trends uh and predicting the future let’s mix all of those in now obviously this creates a situation where maybe those are not compatible and things like that. But that’s where like thankfully with the help of AI and the the new newer ideas in programming which which is say around like typing right where where we we can say that we can use the ontologies and the v variable types to to to define what is and is not allowed to to then interact. we can model time where an event in the future can’t interact with the past and things like that, right?
22:03 And so like this is where the the idea of a machine card comes into play. This is essentially the the primitive behind the whole system, right? And the the idea is that there is like an existing ecology say in the case of this which is like say piprint um education and broadcast casting right that’s the the web I mean like these these are like the oldest machines which we have now and in this environment like I mean something like chat GPT gets born and every machine has this idea of ecological input puts and and then outputs. So in the case of the chat GPT we have things like culture, we have institutions, we have infrastructure which the which which the machine you uses to stay active and because now it’s active it’ll create some kind of like an affordance and then like an opportunity where we can create slides which collude um but it it also creates some kinds of pressures on the system which we’ll see. So with with this like machine idea in place what we can actually do is we can play it ahead right like that if these kinds of forces exist what does the ecology look like if you go into the future say say like 100 years right what what can we expect to to happen u in that kind of a situation which is which is where like uh the idea of pro prophecy actually like comes into play. So this is like sort of like a pseudo simulation u where essentially the the idea is that on the right side we we have things uh which are like currently live and then on the left we have like the older machines in play. U so say with print and pay pay patenting we we have like some kind of artifacts being being like generated. Um I mean like scroll to much much more like recently history if if you have that internet like archive like coming into play we changed the way that custody of of of knowledge actually works. Um then we have the Wikipedia invented. But when Wikipedia and the internet archive actually interact that creates this new machine of the Wikipedia repair loop where when Wikipedia loses per provenence. we can use the the way way back machine as a way to maintain that and that kind of like a system allows something like an open AI to start crawling that kind of a web and so we will we will talk in the chat as to how this actually is polluting the the the the internet and so which which is where the idea that the health of the commons go goes down And so this is the the broad way um that machines interact. But uh let me take you through like this is like a animation. So sorry this is like an actual uh uh example of the way that the these things work which you can uh find at this likeu URL. And the the broad idea is that we have like simulated around 500 or so machines starting actually way before the modernity machine because machines like wheat and maze actually have like very long-term effects. Um also like rice and lip potatoes and then all of that. So, so if we allow this machine to now play ahead, what happens is that like you have other machines which are dependent on this and of course because all of this is history, we can verify it. Uh and now we see all these interactions which is like the orange lines where each machine affects the other either it’ll enable it, it’ll uh or or then it’ll stop it.
26:30 And we go go go through all of this. We we go through the various like world machines and then we enter a point which is into the future right and this is where we use these past machines to actually like simulate what is possible across say fields fields um say say like culture economy ecology etc. And now we can be like hey this is a kind of machine which is expected to occur in the future. What does it actually look like? Um and and then how will it feel like to totally live through through through it right? Oh this is interesting. Uh hold on. I think the the the other tab captures something.
27:19 There we go. Okay. So yeah now um this is now like an example of a prophecy uh which which which we can make which which in the f future um there’s a person who who’s whose then ancestors um like came from like some somewhere else but now we have this whole registry which which is like a mechanism to forget history and like like why should this be the case and this we can then actually see that like Now because we have things like judge, GPT everything can be like generated and that means that we need tools which is what Apple has has also like recently announced that now machines can actually like like report what has actually happened. But even then history needs a way to deal with with with conflicting arguments. Um, and so we need to be able to forget because what happens is that the world remembers way too much and we can’t say like put put aside our differences and and by forgetting we actually enable a society to live in the present, right? And so this is like an idea of how the future could be, right?
28:37 And so here again the prophecy itself is not as important. What’s important is that the prime radiant gives you a way to get a shape of the future ecology uh of the institutions infrastructure what they might or might not care about and what the society may need. Um until now we had to like work with experts who would be like oh maybe this will happen in the f future and now instead of needing like a cybernetics group um we we can just have have the LLMs use existing ideas like existing four forces to then build new things. Um and so if if if this sounds exciting and you would like to contribute uh there are three main ways. one is to just like give us like much more raw material bring it like your perspectives right more if you’re more on the the AI side help uh I mean help us build that a AI into play and then of course like connect these two ideas to to to help us create much better things around the machine cards so so yeah that that that that’s it um and I will hand it back to like timber I guess. Um Yep.
Simone Cicero of Boundaryless presents O2A (Open Organisation Alliance), a proposed open standard for making organizations 'composable' by replacing implicit coordination mechanisms like hierarchy and meetings with an explicit, machine-readable contract graph of units, offerings, and relationships. He argues the protocol's compact grammar—three economic primitives (purchase, revenue split, investment) plus composition mechanisms—can represent bespoke institutional arrangements like Haier's Rendanheyi model or joint ventures as 'sentences' in a common organizational language, becoming essential as AI agents enter firms and transaction costs collapse.
Amber Case presents the Calm Technology Institute's 81-point certification framework, arguing that human attention operates like a law of nature—distinguishing prefrontal-cortex-demanding 'center' attention from pass-through 'peripheral' attention—and that products (from car buttons to AI interfaces) should be designed and governed accordingly. She traces the framework's lineage to Mark Weiser's 1995 calm technology principles and discusses the legitimacy problem of building certification authority without statutory power, comparing CTI to Underwriter Laboratories and the Good Housekeeping Seal. The session closes with audience discussion, including Venkatesh Rao connecting the talk to Kairos/Chronos time and AI's 'staircase' temporality.
Venkatesh Rao presents techniques for building 'artisanal bots' — RAG systems that embody a personality rather than just a knowledge base — demonstrated through three examples: vgr_zirp (his personal archival bot built from Ribbon Farm and Contraptions writing), C3PO (the Protocol Institute's institutional helper bot), and Humboldt (an artificial researcher persona). He details protocol patterns like modeling ignorance and refusals to prevent 'leak-through' (where Claude's general knowledge bleeds through in the persona's voice), weighting sources, true-naming personas, and designing guardrails against misuse of sensitive material like his 'Gervais Principle' writing.