TWed Talk (11 Oct 2023): Jamie McCusker on " "Splitting the World With My Grandfather's Axe"

Key Takeaways

so hi um Jamie mccusker uh this is uh the uh twed talks uh I'm uh giving the talk and introducing myself because uh John Ericson can't uh make it today um today I'm going to be talking about uh the uh the problem with uh continuous in discreet representations in knowledge representation and what the heck we can do about it um so it's called splitting the world with my grandfather's a um for those of you who are familiar with the grandfather's a problem good for you I'll talk about it in a bit um but uh basically yeah we're going to uh we're going be getting a little bit into a little bit of a practical philosophy such as it exists um and really talk about kind of why that affects our uh work with knowledge graphs and knowledge representation and semantics so um I'm going to start by uh talking about uh the idea of entities so when we have a Knowledge Graph right we have a node we have another node we might connect it with a thing a link and so that's great uh and we can do so much with us it's really cool um and uh one of the reasons why this works is because we say that this node actually represents a thing an entity an object or an event or whatever but it's something and that something is distinct from the rest of the universe right we say that it's you know there's a bunch of stuff that is this things the rest of it is not it's something else now um when we're working with uh with graphs that's really helpful because we can literally draw a circle around uh the the thing that we mean when we when we're uh working with entities um and so we this actually comes from uh the world of symbolic logic where we talk about you know um you know uh you know variables like uh P therefore Q uh p q right so that's that is uh when we're working with symbols these symbols when we talk about graphs actually relate to the specific thing here right so we can say that this is you know we can pretend that a node is p and another node is q and so these are symbols that we're talking about that relate to specific specific things um you know if we say um you know the it this really sounds very ba basic but there's a reason why we're why I'm going into this um you know the idea of of entities uh kind of comes from some our use of entities in in knowledge graphs comes from symbolic logic comes from knowledge representation it comes from the tradition of treating things as distinct and unique now in philosophy that's actually uh one of the very first things that happens with that and by very first things I mean literally was discussed in ancient Greece was this idea that there oh got pizza is here I need someone else hi we someone out in a second thank you but water okay we can cut that part out um okay so you want yes and actually you know what I will probably I should use Pizza as an example because we're gonna have pizza um well no we actually the impossibility of a pizza ontology possibility yes of an accurate Pizza ontology or at least instances around the pizza on uh instantiating things in pizza antology and so um the classic example is uh that and by classic I mean classic Antiquity it was it's called the ship of Theus and so there was a ship that um Theus uh set sail from Athens to uh cre F the minitar um uh seduced AR Adney broke up with her on the way home and came back home in the same ship and actually the uh Legend is that uh actually had Black Sails on the ship and uh his father killed himself out of grief because he was supposed to switch over to White sales if he survived he forgot somehow I don't know what the deal was exactly um and uh one of the so uh one of the interesting things about the ship ofus was that so first off it it uh this Legend comes from uh very early in Greek history uh basically um Athens was one of the first group cities period um and the uh the legend actually comes from potentially very early in that history uh in the Bronze Age uh the uh ship supposedly reportedly survived until uh classical Antiquity so Plato Aristotle um Socrates potentially saw this ship in the harbor it was you know it was docked at the port in Athens for a very long time now we're talking hundreds and hundreds and hundreds of years so know the cloth sales eventually rot and uh you know maybe it was kind of embarrassing that they had white uh they still have the black flags on black sales on so they replaced them with with the white sales or maybe they kept the black ones who knows um but they were new sales uh but then suddenly the termites got into the Mast so they had to replace the Mast uh and then the you know there were a couple of boards that were um oh we only have black markers uh you know a couple of boards started to fall off and they were rotting so they need to be replaced more and more and more and more and uh you know the most philosophers ancient Greece were fairly practical they knew what was happening to the ship and they saw people having to repair on a regular basis um and you see this you know there are old ships even now that go into Dry Dock and get massive overhauls um and they still get called the same ship and this was still the same ship of thesias one thing they realized is that there's no way this entire ship even any piece of this ship was original to what was built originally that carried thesias eventually every single piece of the ship placed one by one little bit at a time and at this point are we talking about the same ship and this is this is the problem of the ship of thesias is uh is the ship ofus uh the same ship in classical Antiquity as it was when it sets sail to creep um are you and this is kind of where this comes from are you physically well so you're not physically are you the same person you were when you were born you have um a different physical composition bigger I hope um you are uh you know your brains have developed you've gained capability uh you've gained memories um and literally all the matter in you been has been replaced a number of times but what is it that makes you you um this is called the problem of personal identity and is another Paradox of philosophy and it's not one we're going to solve today because it's just not going to happen um but the thing is is that it's the same question right So Physical continuity is the same sort of thing if I um if I replace if I gradually replace all the matter in me I still feel like me but I'm not I'm not physically the same entity that that I started out as um if uh you know and actually there's a example from that that's in uh the Buddhist tradition of someone who has his arm like chopped off and replac then his other arm and then his legs and basically gradually like every single part of it was replaced but you never felt different uh it's meant to be a metaphor for this idea of you know change over time that can happen and at what point are you are you a different person or are you now the other idea it's kind of a more modern version of that is something uh along the lines of imagine every minute one of your neurons was replaced maybe with an exact duplicate maybe with one maybe with an artificial one that works exactly the same you don't notice this because it's say it's magic or teleportation or whatever you want to call it um but you are your brain is being basically copied uh one by one and it's seamless and you don't notice and at what point how many neurons in do you have to go before you're not the same person anymore how many neurons you take those same neurons that you pull up and put them back together and put them in a new body is this new before or is it a new person um similarly uh there was the problem there's the question of teleportation so Star Trek they go on Transporters all the time and uh they disassemble every item from you and they beam it across the not the Galaxy usually uh but you know they beam it down to the planet and they reassemble you in in one piece uh again and so um you know that you've got you that's up on the transporter pad and then there's you down on the planet right and so they send it over and um you know what what well first off what do they do with with this person here they they go away right and so in theory you're here what do they have they have like all the atoms and their positions and whatever else they needed and you know with the Heisenberg compensators and all that sort of stuff that's happening um there's they're basically rebuilding a copy of you down here right um and interesting detail in Star Trek uh the uh food replicators use the same technology as Transporters uh they actually rebuild food from scratch as it was put together by some Chef originally um and they can make multiple copies of that food that they sent you down to that planet and you died down there oh well they make another and so we send down another person pick up pick up where they left off same person different person why did they destroy this guy is this one dead where where did what's going on here these are all aspects of um the problem of personal identity where you are um potentially uh you know drawing that line between you and not you becomes very difficult um so there's a lot of potential solutions to it one of which we talked about physical continuity so like I'm still me because I've continue to be me and I'm kind you can kind of draw circle around me and say this is me um but there are a lot of issues that we just talked about physical uh there's the idea of um so they call it processual continuity them like a process that continues but again with uh teleportation are they the same person they feel like they're the same person um that would argue that they're not necessarily uh just continuity is not is another popular one um you know the the continuity of consciousness of course is a uh you know I'm you know I've got a conscious state in my head that is uh continually creating experience and um if you like cut that off if there's there's a disconnect between that then there's um you know then it's a different person um because that's you know the different conscious experiences but there's this thing called going to sleep where you're not conscious so is that a disc is that discontinuity turning creating new person every night there's memories uh which of course can be copied again in the case of the transporter um but you can also copy them other ways um and then so this is all this sounds really theoretical right uh like who cares about this except that uh whenever we talk about a um whenever we talk about a uh a biological system we get the same problem so uh or actually most natural systems is not just biological systems but so let's say we have um a clump of cells right and they are clones of each other and this is uh this is actually called a cell line where they're all like exactly the same um but you can also you know you can split the cell line Half you used to have one Colony now you have two colonies but they're still the same cells that they were before now if you take a rock split it in half it's still the same matter is it the same Rock what if you glue it back together what if you repair it if I take a cup smash it glue it back together is it the same cup why is that different than a tree or a rock Rivers will change their course over time sometimes radically are they the same river a different River um the then there's things like identical twins which is actually even more practical than this we have two cells in uh in uh a zygote split in half now suddenly we have two people instead of one person we didn't have the entities that ex that this was an you know before split we don't know exactly why they split before it split that was one entity we would think of it as one entity and now we're thinking about two entities for some reason um when we have um so fungi mushrooms and the that's just ridiculous there because um the underground Network the fungal underground networks in most forests are massive and really Ed and we have no idea how they interact uh and it's like a multispecies thing and sometimes it's kind of the same thing and sometimes it isn't sometimes it acts as like a a forest wide Network you can actually transport nutrients from tree to tree uh that's actually one reason why when you you get mushrooms on old logs is they're actually pulling all the nutrients out and drawing it into the other trees in the forest um the the distinction between the the the uh the individual uh you know the individual trees and the the fungi that support them and the forest as a whole is really hard to figure out it's a continuous process that we really really want to make discreet somehow because you look at a tree and you see a tree right you might want to even identify the tree and say oh yeah there's this tree in the knowledge graph and it's next to this other tree um but then later on we realize oh wait this is an Al and uh is the elders or Aspens uh they actually have they're all one big tree underground uh they have a interconnected root system it's not one it's not hundreds of trees for acres and Acres it's just one tree um and so when we're dealing with B olical systems when we're dealing with natural systems this gets hard harder and harder and harder every time we we try to represent it and so the further down you go in reality the more continuous everything gets right until things get really discreet so um there's matter and it has you know uh if we if we look at uh you know at the level of physics there really aren't entities don't really exist right we're just colle lumps of matter and uh that happen to interact in interesting ways and we're able to uh put something put an interpretation on top of that um and that's actually one thing that is the like going back to Buddhism one reason why there's that story about the man who the man who fell apart I guess you can call him um was is that this distinction between one person and another or one person in the rest of the universe is completely arbitrary and actually kind of illusory and there uh psych there's psychology behind it to back it up there's um uh Neuroscience to back it up and the basic basic physics back this up that there is no real it's really hard to draw a distinction between you and the rest of the universe or between any one entity and the rest of the universe so these little circles that we draw are useful abstractions um but as uh some uh some people like to say all models are wrong all models are useful Dodge box Dodge boox said that and so um when you have a when you're building out these models you have to think you have to think about this idea of um identity kind of from the beginning now um before I go into kind of like the future stuff I do want to talk a little bit about how we've dealt with this in the past um [Music] in the uh I've spent a lot of time talking about entity but not as much time as the Providence working group spent talking about entity no one could no one could don't have that much time basically it was a room about the size of this gathering in terms of people probably more of people who have been thinking about Providence for quite a long time uh and the you know the the aggregate experience in Providence research was probably in the hundreds of years in this in this room um they spent entire two days talking about what is an entity basically uh and how we should represent it and what it should be and what it should mean and ultimately we came up with an entities and we didn't even call it an entity at first we called it Bob um and so Bob was could have been a description of an entity it could have been a representation of the entity itself it could have be the entity itself and the URI is the identifier for the entity and so kind of like it's like trying to rehash the whole you know there's the semiotic triangle of uh you know there's a symbol is the reference is the uh thought right and so you know the link between the symbol and the refering is mediated by the thought because U there's no direct link between a symbol PS yes um between a symbol and the thing that it represents in the world the only the only way to do that is to mediate it through some sort of thinking machine and brains in this case um and so the question was what is this thing that we're trying to work with and where does it sit in this world uh because you can also stack these up right you can have a you can have a concept um you know so let's say it's a concept instead of a thought um but the sorry I'm trying to look right so you know let's say it's actually a concept the concept itself is actually something that happens in our heads and is there there's a symbol associated with that even you know it's actually uh there's something we can there there's something that exists that uh is actually also a refering right and um the the uh the actual thing could be wait at this level you you might have a thought about a concept and you would have a you'd have a symbol for the concept but the concept isn't actually the thing itself it's actually the idea of the reference yeah and so you know we we can you can drive yourself nuts with this and some we almost did um but the uh kind of ultimately we settled that the entity itself is this guy the the uh the actual you know the UR for that entity identify for that entity this symbol here um and we weren't working with Concepts it wasn't like just identifying descriptions we were identifying things and so so we we kind of made this is direct as possible now the problem that we have when we start thinking about this is identifying tips of thesis and I literally this is like literally like what I brought to the meeting but how do we handle this issue and it wasn't the ship itself that we cared about it was the fact that there's all of these uh variations you know like the idea that you have go back to the um you know you have a ship but then you also have versions of the ship um you know you need to be able to uh how do you actually how do you like H how do you distinguish between the ship as seen by Aristotle in 500 BC versus the ship that set sail to Cree in a thousand BC right they're not the sh same ship and we want to be able to distinguish between them because they are M literally materially different um and so what we did was we basically said well um we know that this is you know this whole thing happens and we're going to deal with that by making more entities uh and so what we do uh we we uh came up with a new not actually that new we generalized the idea of um so there's this thing so co-reference Rel when you have two entities um let's say we have two uh actually know let's go back to the uh you we've got the the um we've got the actual and it's kind of flar because I'm using a symbol to refer to create the refering um but you know we have um we might have one uh thought um and then we have a symbol um but then we can have another another symbol that points to a thought that point that refers to the reference right so this is the non-unique naming assumption the idea that you can have more than one name for a thing Barbe it for me to to speak against that particular assumption because lots of things have multiple names and in the semantic web we have you have you can have multiple uis that identify the same thing now the idea with this is that you would have these symbols and so let's go back to circles here you would have uh two symbols that are different nodes in the graph but then you say same as right so um you basically this guy is the same as this guy right we solve the problem except two no no no that's not even the problem it's good that there's two but now everything that's true about this guy is also true about this guy and reversed so every fact uh in symbol a symbol B are now shared they're indistinguishable because they're identical and that's Lian um uh identity and is great and really useful uh and when we're doing kind of data integration when we know what we're doing with it um it is uh very uh helpful to kind of pull things together and actually a lot of rdf databases now support something called smushing which is that when they see this same as what it does is basically in the back end of the database it treats it as one entity it just says uh statements in here it's going to merge them so we're going to basically have a database record with multiple uis attached to it so we don't even have to think about it it's just going to copy everything over automatically because there's no copying to do is uh very handy when what you're talking about is indistinguishable and identical it reminds me the concept of dependence origination like there a concept in philosophy and um it's like discuss this thing but it's the same as it is but it's students take we'll take it offline and yeah yeah it's a con you probably have seen it yeah probably um I comment something about the smooshing so as somebody who implemented the uh like the tabulator if you if you're familiar with Tim's tabulator uh so smooshing actually doesn't work all the time like because you know these two concepts even though they have a same ass relationship the properties are different and sometimes the properties would U you know contradict yes EX exactly yeah going to talk about that yeah well I mean so so there's uh that's actually a really interesting point because there are cases where you might um you know you might have contradictory facts about something because those facts might be contextual or even just like with within the you know I imagine you saw something that was even more practical in that yeah like Tim has like two of rdf files and they have contradictory facts second and both were created by Him himself right and uh yeah about himself about himself and he contradicted himself himself well he contains multitudes right uh and that's actually the the uh that's actually interest so that's interesting segue because this uh so yeah the same as tricky for a number of reasons some of them technical some of them representational were that you are actually thinking about maybe a slightly different Twist on the same entity and so um this is the classic uh co-reference relation where you're if you think you know you can talk treat them the same uh and but what if they're identical but what if they're uh distinguishable but identical what if these things actually have some sort of uh distinction between them and so if we have these two ships of thesias back to this um you know one of them is physically different materials different pieces and so the Providence would be very different of uh for it uh and the actual uh you know if we say the composition if we had like instruments about composition or mass or anything else those would all be very different we all all those sorts of measurements um but the thing is is that you know there are some cases where we want to talk about the fact that this is the same thing you know um there are cases where we want to distinguish between them and there are cases where we want to kind of not distinguish between them and so what we would do is actually make um what I would call um an abstract an abstracted or decontextualized entity for the for this so there is uh kind of across all uh possible interpretations of the uh ship ofus there's a much more conceptual version of it that doesn't ma that doesn't rely on any one particular piece of matter being in a particular place uh or uh cares about a particular composition or mass and we can't actually say any of those things about it in principle but it is essentially uh a way to talk about the relationship between between these things and so what we do in um the Providence uh the Prov ontology is that we talk about these kind of more concrete representations being specializations of the this more abstract one in other ontologies they use the term concretizes NE are very good in terms of terminology uh it requires a lot of explanation and I'm probably the most frequent user of specialization of in improv at uh at this point um just because it's hard to understand and uh can be tricky to get right but it solves this problem that you have two very different things that are actually the same thing and so we uh that's how we deal with that issue in Prov specifically now um what this means is that um every time you want to record something and you have a distinction to make you have to create a new URI and you have to basically you know add another another entity to the graph and we have to know how to interpret it and it gets it can get weird um and it's not a perfect not a perfect solution so uh so that's kind of where we but that's where we stand right now um one thing that I want to talk about is uh I want to kind of flip this around for a second because there's another side to this there's a reason why we like to think about these entities why are entities so appealing to us and it's actually a deep deep deep evolutionary uh uh trait that we developed because it's useful to think of the world in terms of objects this goes back to the first perceiving uh organisms that was were able to kind of uh observe the world object detection is literally burned into our retinas and into our brains we have whole sections built for object detection and distinction our language is built on discrete pieces and turns continuous flows like literally continuous flows of of sound waves into chunks and we it does it without us having to think about it um and so thisty bias as I've been thinking of it is fundamental to how we think as humans it's fundamental to how any animals think as animals um I've I went looking and I couldn't find a counter example actually of if if it has a nervous system it seems to be thinking in terms of if there's any perception it will CH chunk that into into entities or there it's not simply doing enough with its nervous system to be able to do anything like that um but the um one of the issues there of course is that as we said that that's has edge cases right that it's not a perfect model of the Universe um it is not a um in some ways it's a misleading model of the universe and can cause lots of trouble sometimes if if you really uh assume that that distinction goes all the way down to the bottom and so what you end up with is this uh this divide between kind of like our brains wanting to create chunky a chunky Universe in a universe that's like maybe I don't know sometimes um the interesting thing is that when you flip uh when you flip that around a lot uh and you look at artificial neuron Nets they're not chunky they're not lumpy they're not entity focused they they're work in continuity and it could be that in many ways that's advantageous for working with um you know like if we're if we're doing like some sort of like analysis of a continuous stream it can actually do a prediction of another continuous stream from that that's great great but uh if we're trying to have it work with human uh style intelligence it's not uh that chunkiness is a barrier the lack of chunkiness is a barrier to how humans think and having and trying to replicate kind of the advances that we've made and so my argument is that if we have a uh if if we move forward with uh with neural AI it's going to be through introducing the bias that's thrown us off all these years all these Millennia uh and uh helping uh helping it trying to understand the universe in terms of entities so so yeah that's pretty much all I have I'm curious about everyone's perspectives on this so far and how much crap I'm full of I'll start um I guess I have heard this before so I'm cheating a little bit I had time to think about it uh but um great talk thank you uh and um what I'm wondering about is so it's true that we do tend to chunk things and we we tend to think in abstractions and symbolic representations but um we do so in a very fuzzy way right compared to the way that exactly yeah we change and kind of modify our Concepts over time they're not entirely consistent we'll just kind of make things up and say make up why it should right right yeah that way and so on um which is maybe a better way for understanding the world which isn't actually conceptual exactly but uh so I guess um how how do how does that kind of worldview so it's not just chunkiness it's kind of fuzziness do you kind of feel like that has a role there or well so that's why I was distinguishing between like chunkiness and liness and discreet systems so like a knowledge graph is discret it's like sand um but if you look at um you know if you look at human cognition the idea of these entities are not they're not fixed like that they're not like I can think about my arm um and it can be part of me but you know the seam is pretty smooth and there's no real like literal cut off point where I can say this is my arm and this isn't and so um when you um uh when you have uh a discret system you're talking about the you're talking about discontinuity uh between one thing and another and even having that sort of chunky your lumpy representation that's able to handle the idea of entities but also handle the idea that well mostly right that that's actually very that's radically different from the symbolic systems that we work with when we're actually doing things like knowledge grafts and so that's going to in my mind that's actually going to be one of the things that we need to figure out is like how do we make how do we make knowledge grafts more continuous and how do we make neural systems more lumpy and you know maybe we can get them to meet in the middle but not necessarily um you know like having these two extremes is not isn't really um doesn't help either of them and it actually keeps them from being integrated but the history there is like you know went through language languages more more discreet and then we came up with logic which kind of tried to formalize it and made it more discreet still and then um you know kind of like the further we PA the further we went down that path the more uh kind of the more everything crystallized uh as like these discret entities to the point where if we give an graph and if we say to the knowledge graph the world is continuous it literally wouldn't mean anything to the knowledge graph you couldn't explain that to it because it would represent it as discrete entities and not actually be able to tell you anything about the fact that it's continuous there's no way it could like you you can't even uh yeah you can express the fact that you know world has Type continuous uh but that's not going to that's still two discrete nodes in a graph connected by a link that's not actually expressing the continuity that actually exists anyone else I was just looking at an email from H H glacer Glasser about same as oh okay he was asking if he should let it die cannot maintain it apparently oh well I mean yeah it's funny because like one of the first things I so when I was a first year I actually submitted an abstract for a talk that was called same as considered harmful to Provident to Providence and uh that got kind of that got some Buzz around it because and ultimately we ended up doing uh there were a couple papers later and then we did some stuff with uh uh uh Harry hopin and Pat Hayes yeah yeah and so that um you know making trouble like this paid off at least but um but it's we've never said same as is bad it's that it's it has its place and we need to be very care but we need to be very careful how we're using it and so yeah I mean if I I would I feel sad if same as.org went away money for available they will let the service die because there's a cost to running it but yeah they'll make the data set of it that's good yeah because I mean I think that's the main value there is being able to do that and maybe extending it as time goes on the historical artif so maybe I have a question on like computational provance so I know you talked about representing providance entity and whatnot now I work in blockchain so there's a way of like you know preserving Providence in a in a more computable way so do you have any thoughts about that do I have thoughts about proven blockchain based Provence yes yes I do do um I don't know if I can uh that would be a much more cranky talk I think um the the I mean so uh any so what you would actually represent as the Providence in the blockchain would still have some have to have some sort of some sort of representation to it and so you would still need to you stuff has some something right Knowledge Graph fragment or something um but the and the relationship like this between things and between different versions of the same thing could uh run a file of the same issues depending on how you're doing it so I mean I I I I see those as my my point of view is like you know um in the representation um in a world uh yes like anybody can assert like you know okay I'm the one to create this particular knowledge asset I I have the the first claim right right uh but there's no prove it right so wouldn't the Providence break apart like at that point because the whole point of having powerness is just you know kind of asset that you know somebody has like this is how it something was derived right right well so that's actually interesting thing so when I did my uh so my dissertation for the digital signature framework um we it was uh very he very heavily Providence driven I had heard a damnn thing about blockchain at that point but it was a a large part of it was actually about s uh creating uh creating hashes of hashes so it is a very similar structure to blockchain just um not um and so uh what we did was essentially you sign an abstract uh uh an abstract hash of of a graph not serialization of a graph but the actual graph content itself um and so you can you you wouldn't like lay claim to it you're basically uh you're not laying claim to it you're claiming it right so not it's not like claiming it like a um you know like a a homestead would claim a plot of land it's claiming it like you're asserting it you're saying this thing is true and that's about as F that's kind of all I hoped for from that I wasn't saying that this is the um you know that this is the actual you know we we can like prove uh that this went this way you can't prove that you know there's so I guess so if you make um if you show how someone used something to create another piece of knowledge and you uh can you know and they can they assert that it's the case then maybe uh you get someone else who believes it for some reason they can uh kind of co-sign that assertion that they generated that thing um but again it kind of it goes back to it's the web of trust right you uh at some point it has to that at some point there is no and I saw this with with digital signatures and the idea of signatures as a whole there is no like zero trust way to implement that that you either uh with a conventional signature like a the legal principle behind a signature is that if I sign a document um you know I can always claim that I didn't sign it that I can repudiate my own signature but the thing is is that I would have to do it under oath in a court of law which would put me in Jeopardy for perjury and so that's what's holding the legal the the contract system together right now is that consequence and nothing else there is no like you must do this thing because it says so says you do it in the contract that you signed you can just break the contract right you you can just say okay I'm out and there might be consequences for that but that has to be laid out in the contract and so you know that that's all um the idea of proving something mathematically that's never really been intended to be proved in terms of trust that's that's still hard and I'm not sure I'm not sure blockchain help uh I don't see I don't yet see and it could be that someone's figured out I don't yet see how that um circumvents that process that I talked about that legal process because ultimately you know if you have a it's one thing to get a computer to do it for you it's another thing to make a human do it because they put it on the blockchain yeah I think you have two things like signatures and also parners so um I guess you're right like you know if somebody puts a gun to my head like yeah I'm U like yeah if it's like uh I have some ownership of a certain asset but yeah you know I will give the asset to even though you know the ownership is there blockchain so I I agree with that like the legal system has to kind of work hand in hand but my question was more on the representation side so I personally believe that you know just representing Provence in an onology is never going to work like you know I think that's why the adoption of The Proven onology is very very abysmal so I guess that's the question is what what do you mean by it's never going to work uh to what use case because the the idea of you know there are a lot of things that work really well on the honor system and it works if you if you have a way of at least validating that it's the person that you trust instead of you know like if you don't trust them at all um then there's all sorts of problems that crop up no matter what right and so I guess that was my point is that um at some point you have to trust someone or something and uh the blockchain is a tighter way of verifying that it's the person that you trust yes you don't have to trust them as much or can have they they can you know you can find someone that you do trust that will post that will validate it for you but that's not the same thing as like trusting some you know going in without trusting anything or anyone and just having it be mathematically provable well but uh like honor System is like an ideal in in an ideal case case it would work but I think like blockchain because you don't the whole system assumes that everyone is malicious it's trustless so it is it provides that ideal system in which Prov except that except that the honor System has worked okay yeah but not in all cases really right just um part to be like Universal right but I guess I guess my point is is that with when if you say that um so a trustless system as you're talking about it um is well so first off you still have to trust the blockchain itself right yeah um and you have to trust that there's no way to game what's actually being put in the blockchain um and there's also the fact that by assuming that everyone involved in the blockchain is untrustworthy uh it certainly seems to be attracting a lot of untrustworthy people to that World um won't go into you know but basically like they're trying they're trying to gain the system right right um and the because they specifically trying to do this in an uh extr legal super legal like a non you know they're not trying to rely on legal Frameworks to to build out enforcement they're um you know basically it's encouraging that gaming because the the the it's kind of the point of it that we need to be able to find all these loopholes and stuff that are in it well do that but even the early internet like people were doing nefarious stuff on the internet oh yeah yeah like crazy stuff on the but at some point it kind of was regulated and we learned how to use it and not such uh horrible Manner and all that came down to legal consequences Yes actually that's so that is true but I guess like the thing with blockchain is I know like there's no such thing as a completely trustless system like if large number tries to game it it will work but I guess like the in in concept the idea of having such a system I think maybe will help like Providence work better in a way but again like I guess what you're saying it's such a blockchain is not a thing yet right but there I the idea I think is I would love to see it um but the we love to be a spectator to such a thing um but yeah I mean I I uh definitely uh yeah I appreciate the effort I've been through that process with my own adjacent little thing that didn't involve nearly as much money uh involv no money uh but uh yeah yeah yeah uh I was wondering when earlier you were talking about how neural networks like artificial neural networks um tend to um work in a way that's you said more continuous yes um I I guess I'm wondering why why that is is it because is it because they like represent discret ideas across multiple neurons or is it like um well so this is what this is one reason why I'm not this is one reason why I'm kind of waving my hands during uh uh you know Wednesday night uh BS Fest instead of writing a paper about it um but I mean what it comes down to is that you know uh you know let's talk about large language models for a second so an input to a large language model is a token which is an embedding and it's an embedding that's represented in an embedding space that is repres that has however many floating Point numbers and if you nudge that floating Point number by a certain amount it's not going to make that much of a difference but it will change sure and there's um you know like any change in that embedding space has a real but impossible to determine by just looking at it consequence on the meaning of that token okay and so uh and so once again you know you have one token that represents one particular word and it you know it has a set of numbers and you have another one that represents another word and it has a different set of numbers and if what you're doing doing is you know you have your embedding space of you know however many we'll say two for now and you have this guy here and this guy here let's say for now those all the tokens we have in in the embedding space where along this gradient does it stop being this token and start being this token okay and so at some point in a human brain there's there's a transition you know there's still like yeah you can think multiple thoughts about a a a particular reference in the world a particular object and it can wander around that your activation can wander around but then when you think about something else you're very obviously switching state to something else there is there's like a a uh a jump that's not continuous yes exactly there's a discontinuity in the activation space somehow we don't know exactly what it is yeah and so that's that's my point is that those discontinuities whether they are you know whether it's a hard line or it's a um you know kind of a fuzzy region or it's something else entirely uh you know these could be uh these could be like uh Gan clouds and maybe but the thing is then the the ties back to the reference it's like how do you get that to work this is just basically purely based on the fact that it's a differentiable model yes okay so and so I blame it on back propop got it okay I see the forward for guys oh yeah that that's a cool one yeah I love thater Yeah recording you can probably turn it off yeah Li